Evidence for a Hereditary Neuroblastoma Predisposition Locus at Chromosome 16p12–13¹

Retinoblastoma is the prototypic example of a cancer susceptibility syndrome due to germ-line mutations in a tumor suppressor gene. Knudson provided the seminal hypothesis that both familial and sporadic retinoblastomas are explained by two separate mutational events (1) and also noted that other embryonal cancers of childhood, such as neuroblastoma, fit a “two-hit” model (2). The subsequent cloning and characterization of the retinoblastoma susceptibility gene (RB1) validated this hypothesis and defined the paradigm of biallelic inactivation of tumor suppressor gene alleles as necessary and sufficient for cancer initiation (3, 4, 5, 6, 7).

Neuroblastoma is a relatively common cancer of childhood that usually occurs sporadically (8). The cell of origin is postulated to be an undifferentiated neural crest-derived cell committed to the developing sympathetic nervous system. Neuroblastoma is characterized by striking clinical heterogeneity, including subsets that show spontaneous tumor regression (9). On the other hand, the disease is lethal in over 40% of cases (8, 10). Several acquired genetic alterations such as amplification of the MYCN oncogene, deletions of chromosome bands 1p36 and 11q23, unbalanced gain of 17q material, and alterations in neurotrophin receptor expression have been well characterized and shown to correlate with tumor behavior including response to treatment (reviewed in Ref. 11). However, the molecular genetic events that initiate neuroblastoma tumorigenesis remain obscure.

Hereditary neuroblastoma is rare, and a family history of the disease is present in only 1–2% of cases (2, 12, 13, 14). Similar to retinoblastoma, patients with familial neuroblastoma are characterized by an earlier median age at diagnosis and a higher frequency of multifocal primary tumors (2, 12, 15). However, patients with hereditary neuroblastoma show the same clinical heterogeneity seen in sporadic cases, including cases as disparate as spontaneous regression and relentless progression within individual families (12, 15). Thus, the rarity of familial neuroblastoma may be explained, at least in part, both by the incomplete penetrance and by the lethality of the phenotype.

We hypothesize that hereditary predisposition to neuroblastoma is due to a germ-line mutation in one allele of a tumor suppressor gene. We also postulate that the de novo germ-line mutations in this gene may explain congenital and/or sporadic multifocal cases. Lastly, we suspect that somatic biallelic inactivation of this gene may initiate neuroblastoma tumorigenesis in the absence of a germ-line mutation. We have previously reported exclusion of chromosome band 1p36, a frequent site of allelic deletion and therefore a likely location for a neuroblastoma suppressor gene, as a genomic location for HNB1³ by region-specific linkage analysis (16). We now report the results of the first genome-wide screen for a predisposition locus in neuroblastoma kindreds.

Seven families with two or more individuals of first-degree relation affected with a neuroblastic tumor were eligible for inclusion in this study (Fig. 1). In addition, three families in which a proband affected with neuroblastoma and a second- or third-degree relation also had the disease (two cousin pairs and one proband-uncle pair) were studied but not included in the primary analyses. Pedigrees were ascertained through referral, and an oncologist examined all neuroblastoma patients except individual 11-007 (Fig. 1). Affected patients were defined as those with biopsy-proven neuroblastoma, ganglioneuroblastoma, or ganglioneuroma according the histopathological classification system of Shimada (17). Disease was staged according to the International Neuroblastoma Staging System (18) after review of the medical records and surgical reports. MYCN gene copy number was determined by Southern hybridization (19) or by fluorescence in situ hybridization (20). Each research subject and/or their guardian signed informed consent. The Children’s Hospital of Philadelphia Institutional Review Board approved this research.

Genomic DNA was extracted from whole blood, lymphoblastoid cell lines, or bone marrow mononuclear cell pellets and from snap-frozen tumor specimens by anion exchange chromatography (Qiagen, Valencia, CA). A modified high salt precipitation methodology was used for extraction of genomic DNA from paraffin-embedded tissues when no other DNA source was available. Genome-wide genotyping was performed with the Weber v9 panel of 387 short-tandem repeat polymorphic markers (Research Genetics, Huntsville, AL). The sense primer was fluorescently labeled, and PCR products were pooled in panels of 8–10 markers for each individual before electrophoresis. Data were analyzed with the GeneScan v2.1 and Genotyper v1.1 software (Applied Biosystems, Foster City, CA). Each allele assignment was confirmed by visual inspection. Additional markers for region-specific analyses were obtained from the Genome Database and ordered according to the Marshfield comprehensive genetic map. Fluorescence or ³³P-labeled genotyping with these markers was performed as described previously (13, 16). Physical distances between markers were determined by analysis of sequence data according to the National Center for Biotechnology Information Build 30 (June 2002). Finally, cytogenetic localization of marker data was referenced to the human genome sequence using eGenome.⁴

Expected LOD score calculations were performed using the SLINK software program as implemented in FASTLINK version 2.3P (21) with 100 simulation replicates and assuming no heterogeneity. These data suggested sufficient power to detect linkage with expected LOD scores >5 for the seven families selected for these analyses. Genome-wide genotyping was therefore performed on two of the most informative three-generation families (FNB01 and FNB02, Fig. 1). Genotyping data were analyzed with the Genehunter software package (version 1.2) using both parametric (LOD) and nonparametric (NPL) multipoint algorithms for each autosome individually (22). All analyses assumed no interference or sex differences in the recombination fractions and that hereditary neuroblastoma was attributable to a single biallelic autosomal dominant gene with a population disease allele frequency of 0.000022 (2). We also assumed no linkage heterogeneity and an incomplete penetrance function of a 63% lifetime probability of manifesting disease in carriers. This was based on our own observations (15),⁵ which agreed completely with published data (2). Lastly, both the disease gene allele frequency and the penetrance function assumptions were varied in final analyses between 0.0002, 0.00002, and 0.000002 and 0.5, 0.6, 0.63, 0.7, 0.8, and 0.9, respectively.

Results of the genome-wide search were used to identify regions with LOD scores of >1 for further analyses. At each of these regions, genotyping was performed in the five additional families with each marker showing a positive LOD score. Also, additional markers were added if necessary to improve informativeness at each region.

Paired tumor and constitutional DNA samples were available for 11 familial and 336 sporadic neuroblastomas. The latter cases were patients enrolled on C.O.G. clinical trials. We screened for LOH using simple tandem repeat polymorphic markers D16S404, D16S3075, D16S500, D16S3103, D16S3041, D16S3046, D16S403, D16S3068, and D16S3136 spanning 44 cM at 16p12–13 in all specimens. Sense primers were labeled with FAM, HEX, or NED, and antisense primers were modified with a reverse tail to avoid single nucleotide overhangs (23). Thereafter, individual specimens were analyzed with selected additional markers by conventional radiolabeled uniplex PCR to confirm 16p12–13 allelic status and to map the common region of LOH, as described previously (24, 25). LOH was assigned when comparison of the allelic intensity of fluorescence electropherogram or densitometric peaks from autoradiograms gave a score of ≤0.5 or ≥2.0 (50% reduction in intensity of one tumor allele compared with constitutional allele). Breakpoint defining data were repeated at least twice to unequivocally map regions of LOH. Analyses for LOH at chromosome bands 1p36 and 11q23 were performed as described previously (24, 26). Comparison of clinical variables was performed using χ² testing, with the exception of age, for which a two-tailed Student’s t test was used.

Seven neuroblastoma families with at least two individuals of first-degree relation affected with a neuroblastic tumor were ascertained for analysis (Fig. 1). The median age at diagnosis for the 24 affected patients in this cohort was 14 months, and 6 patients were documented to have multifocal primary tumors (Table 1). Patients coded as unaffected were all >10 years of age, and many were also screened for neuroblastoma by radiographic examinations and/or quantitative analysis of urinary catecholamines. Patients FNB01-001 and FNB01-010 were also affected with Hirschsprung disease (27). In addition, patient FNB01-001 had multiple café-au-lait spots due to neurofibromatosis type 1 acquired as a new germinal mutation (delT 3775; data not shown). This patient’s parents were both shown to have wild-type NF1 alleles. No congenital or development disorders were detected in the other families.

Fig. 2 and Table 2 show the results of the genome-wide scan for linkage in families 1 and 2. Using polymorphic markers at 10 cM density, we were able to exclude 49.9% of the autosomal genome as a potential neuroblastoma predisposition locus with LOD scores < −2.0. An additional 35.0% of markers studied showed negative LOD scores between −2.0 and 0. Only 15.4% of markers showed positive LOD scores, and 12.1% of these were between 0 and +1.0. Eight genomic regions showed LOD scores > 1.0 and were considered further. Only a single 39.1-cM genomic interval at distal 16p, defined by markers tel-ATA41E04-D16S769, showed LOD scores > +1.5, with a LOD_MAX of +1.81 at D16S764.

Each of the eight regions with LOD > 1.0 identified in the genome-wide scan was studied further by genotyping additional families (Table 2). Additional markers were also added to improve the information content of each analysis. Using this approach, LOD scores increased substantively at the 16p12-p13 locus, with the LOD_MAX improved to +3.30 at D16S764 (NPL_MAX = 3.17; P = 0.0012; Fig. 2). The LOD_MAX was also the maximum heterogeneity LOD score. LOD scores decreased for five genomic intervals but did not significantly change for genomic intervals on the long arms of chromosome 15 and 22 (Table 2). The regions on chromosomes 15 and 22 that continued to show LOD scores between +1 and +2 were defined by markers D15S655-D15S816 (17.8 cM) and pter-D22S345 (19.3 cM), respectively.

Thirteen genomic regions (including seven known genes; Table 3) have been proposed as candidates for harboring HNB1 (13, 15, 16). Each of these candidate regions was analyzed as described above by genotyping all seven families with markers evenly spaced throughout a genomic interval or with intragenic and/or flanking markers in the case of candidate genes. Each of these regions showed negative LOD scores, and nine could be excluded with LOD scores at multiple markers and recombination fractions < −2.0 (data not shown but available upon request).

Analysis of the 16p genotyping data showed a unique haplotype segregating among the affected patients within each family studied (i.e., no linkage heterogeneity) but no evidence for a common haplotype among families (Fig. 1). The fifth-degree relation (maternal relation) in pedigree 3 did not share a 16p haplotype with the affected sibling pair (who shared a paternally derived haplotype, Fig. 1). Identification of informative recombination events in families 11 and 14 defined a 25.9-cM region (13.1 Mb) of concordance between D16S748 and D16S3068 located at chromosome 16p12.1-p13.3. Altering the gene frequency and penetrance functions in these analyses had minimal effect on the results. When the postulated mutant gene frequency estimate was changed to 0.0002 and 0.000002 at a penetrance estimate of 0.63, the LOD_MAX at D16S764 was 3.39 and 2.83, respectively. When the penetrance estimate was assigned as 0.5, 0.6, 0.7, 0.8, and 0.9 with the frequency estimate held constant at 0.00002, the LOD_MAX at D16S764 was 2.96, 3.23, 3.42, 3.48, and 3.31, respectively. Nonparametric analyses that do not include assumptions in the model such as gene frequency and penetrance further supported the conclusion of linkage to 16p (Fig. 2, C and D). Lastly, the three neuroblastoma families with affected individuals of ≥second-degree relation each showed no evidence for linkage to this genomic interval (data not shown).

Diagnostic primary tumor DNA samples were available for eight of the affected patients from pedigrees provisionally linked to 16p. Three specimens showed 16p LOH, including two specimens (FNB01-003 and FNB02-002) with loss of the non-disease-associated allele at each marker analyzed (Figs. 1 and 3). Tumor from patient FNB03-012 also showed LOH, but this individual did not share 16p haplotype with the affected sibling pair of distant relation. In addition, two of the three primary tumor samples available from the affected cousin pairs not included in the primary analysis showed 16p12–13 LOH (Fig. 3).

To determine whether somatic hemizygous deletions occur in nonfamilial neuroblastomas, 336 primary neuroblastomas obtained at the initial diagnostic procedure were screened for LOH at 16p12-p13 (Table 4). These cases were representative of the known natural distribution of clinical and biological variables typically present at diagnosis (8). Sixty-eight of these samples (20.2%) showed LOH at multiple markers within the putative HNB1 locus. Although most regions of LOH spanned the entire critical region defined in the linkage studies, one stage 4S tumor sample had a 14.5-cM interstitial deletion defined by tel-D16S500-D16S412-cen (Fig. 4) at 16p12.2–16p13.12. This interstitial deletion was confirmed by triplicate analysis of each marker within and flanking the region of LOH and is approximately 9.1 Mb in physical distance according to the National Center for Biotechnology Information Build 30.

Table 4 shows the distribution of clinical and biological variables for sporadic neuroblastoma cases in relation to 16p allelic status. There was no significant correlation of 16p12-p13 LOH with any commonly measured clinical variables, except a trend toward this abnormality being more frequent in tumors from patients diagnosed at a younger age. There was a strong correlation between 16p12-p13 LOH and allelic deletion at chromosome 11q loci (P < 0.0001).

Familial neuroblastoma was described as a distinct clinical entity in 1945, and a genetic hypothesis was provided in 1972 (28). However, due to the unique features of this tumor, including spontaneous regression in some patients and lethality prior to reproductive age in others, it has been difficult to identify the families and biological reagents necessary for genetic linkage analyses. The families described in this report conform to the predictions of the two-mutation model. The median age at diagnosis of 14 months of our patients is lower than that reported in surveys of large populations of sporadic neuroblastoma cases, where the median age has typically been reported to be 23 months (8, 15). In addition, one-quarter of patients were documented to have more than one primary tumor, whereas multifocality is only rarely observed in sporadic neuroblastoma cases (8). Lastly, the seven pedigrees studied here suggest incomplete disease penetrance, with generation skipping in at least one family. Therefore, the penetrance value of 63% calculated in a previous review of hereditary neuroblastoma (2) is a reasonable estimate of the chance of manifesting the disease in those with heritable predisposition.

A combined genome-wide and candidate locus approach for linkage analysis identified a single genomic interval at 16p12-p13 con-sistent with a hereditary neuroblastoma predisposition locus. The initial genome-wide screen with two families identified eight chromosomal regions with multipoint LOD scores > +1.0. Additional genotyping at these regions showed consistent evidence for linkage only at 16p12-p13. The maximum multipoint LOD score of 3.30 is highly suggestive of linkage but should be viewed cautiously due to the genome-wide approach taken to identify the HNB1 locus (29).

Because of the relatively small size of the families available for study, we also focused attention on 13 candidate loci for HNB1 to assure that we were not missing linkage due to constitutional homozygosity at key polymorphic loci in families 1 and 2. These included five regions of consistent somatic hemizygous deletion in sporadic neuroblastomas (1p36.3, 3p21-pter, 4p15-pter, 11q23, and 14q32) and thus putative tumor suppressor gene loci. Because activating mutations can predispose to human cancer syndromes, the MYCN oncogene and distal 17q, a very common site of somatically acquired unbalanced allelic gain, were also considered candidates for HNB1. Lastly, because of the coincidence of neuroblastoma with other neural crest-derived disorders in our patients and in the literature, we also considered regions containing candidate “neurocristopathy” genes (GDNF, GFRA2, RET, EDNRB, NF1, and EDN3) as candidates for HNB1. As shown in Table 3, each of these regions showed negative LOD scores, and nine could be formally excluded as neuroblastoma predisposition loci in these seven families with LOD scores <–2.0 (30).

LOH data can add information to linkage analysis because the two-hit model holds that the non-disease-associated allele should be deleted in the tumor specimen. Our observation that two of seven available tumor specimens from affected and 16p-linked individuals in these families showed 16p12-p13 LOH, and that in both cases the allele retained was from the parent predisposed to the disease, further supports our conclusion that neuroblastoma predisposition occurs due to heritable mutation in a tumor suppressor gene located on the short arm of chromosome 16.

Three additional families were available for analysis but were distinguished from the 16p12-p13-linked families in that affected individuals were of ≥second-degree relation. These three families consisted of two first-cousin pairs and one uncle-proband pair, and in none of these families was neuroblastoma predisposition linked to 16p12-p13. This suggests the possibility that hereditary neuroblastoma is genetically heterogeneous, like other pediatric cancer predisposition syndromes such as familial Wilms’ tumor and the Li-Fraumeni syndrome (31, 32, 33, 34). Alternatively, it is possible that some or all of these neuroblastoma pedigrees are actually examples of chance occurrence of a relatively rare disease within the same family.

The short arm of chromosome 16 had not previously been identified as a site of recurrent genomic alterations in nonfamilial neuroblastomas. However, if HNB1 functions as a tumor suppressor and is analogous to RB1, one would expect somatic biallelic inactivation in at least a subset of sporadic neuroblastomas. We showed LOH at multiple 16p12-p13 polymorphic loci in 20% of a representative panel of sporadic neuroblastomas. This suggests that somatic deletions in this region might be targeting the second allele of a gene harboring a hemizygous mutation, but proof of this awaits identification of HNB1. Consistent with the hypothesis that inactivation of HNB1 would occur as an early event in tumorigenesis, there was an equivalent distribution of the frequency of 16p LOH among various clinical parameters. Of as yet unknown significance, there was a strong correlation between LOH at 11q loci and LOH at 16p12–13. This is of interest not only because deletion of the long arm of chromosome 11 is a relatively common somatically acquired abnormality in this disease, but also because constitutional deletions of 11q14-qter have been shown to be associated with two separate cases of neuroblastoma (35).⁷ Thus, deletion of a gene or genes at 11q may cooperate with the oncogenic effects of loss of a 16p gene.

The vast majority of hemizygous deletions identified overlap the critical region for HNB1 defined by the linkage data. However, a single small interstitial deletion mapped completely within the critical region defined by linkage analysis. Based on these data, we expect that the location of HNB1 is within a 14.5-cM region (9.1 Mb) at 16p12-p13.

In summary, our data support a model in which hereditary neuroblastoma occurs due to germ-line mutation in a gene located at 16p12-p13. Furthermore, our LOH data suggest that a second, somatically acquired event may be necessary to initiate tumorigenesis in these families. This model is consistent with the observation of significant intrafamilial clinical heterogeneity in neuroblastoma pedigrees, suggesting that additional somatically acquired events determine the ultimate tumor phenotype. Lastly, our LOH data from sporadic cases of neuroblastoma are consistent with the model that somatic biallelic inactivation of a tumor suppressor gene at 16p12-p13 may initiate tumorigenesis in at least a subset of neuroblastoma cases. Cloning and characterization of HNB1 will be necessary to determine whether hereditary predisposition to neuroblastoma is genetically homogeneous or whether additional predisposition loci exist.

We thank Drs. Gerald Feldman, William Morris, Paul Woodard, and Stuart Siegel for pedigree referrals, sample ascertainment, and patient outcome measures and the C.O.G. for providing research specimens. We also thank Drs. Kevin Shannon and Lucy Side for performing NF1 mutation analysis and Drs. Richard Spielman and Dawn Teare for advice.