An imprinted locus epistatically influences Nstr1 and Nstr2 to control resistance to nerve sheath tumors in an NF1 mouse model

Abstract

Cancer is a complex disease in which cells acquire many genetic and epigenetic alterations. We have examined how three types of alterations, mutations in tumor suppressor genes, changes in an imprinted locus, and polymorphic loci, interact to affect tumor susceptibility in a mouse model of neurofibromatosis type 1 (NF1). Mutations in tumor suppressor genes such as TP53 and in oncogenes such as KRAS have major effects on tumorigenesis, due to the central roles of these genes in cell proliferation and cell survival. Imprinted genes expressed from only one parental chromosome affect tumorigenesis if their monoallelic expression is lost or duplicated. Because imprinted loci are within regions deleted or amplified in cancer, the parental origin of genomic rearrangements could affect tumorigenesis. Gene polymorphisms can vary tumor incidence by affecting rate-limiting steps in tumorigenesis within tumor cells or surrounding stroma. In our mouse model of NF1, the incidence of tumors mutant for the tumor suppressor genes Nf1 and Trp53 is strongly modified by a linked imprinted locus acting epistatically on two unlinked polymorphic loci, Nstr1 and Nstr2. This interaction of an imprinted locus and polymorphic susceptibility loci has profound implications for human mapping studies, where the parental contribution of alleles is often unknown.

NF1 is an autosomal dominant disease affecting 1 in 3500 people and is caused by mutations in the NF1 gene (1) (Nf1 in the mouse). A study comparing monozygotic twins and more distant relatives has shown that polymorphic loci unlinked to NF1 alter the severity of the disease (2), although these loci have not yet been identified. NF1 is characterized by benign tumors of the central and peripheral nervous systems, such as optic pathway gliomas and neurofibromas, respectively. In addition to benign tumors, NF1 patients have an increased risk for developing particular malignancies, including malignant peripheral nerve sheath tumors (MPNSTs) and malignant astrocytomas, including glioblastoma (1). MPNSTs arise in the nerves of NF1 patients and have been shown to originate from Schwann cells (3, 4) often through malignant transformation of benign plexiform neurofibromas. As part of this malignant transformation from plexiform neurofibroma to MPNST, the p53 tumor suppressor (encoded by the Tp53 gene in humans and the Trp53 gene in mice) has been shown to be mutated or lost (5). Since these tumors are highly aggressive and currently incurable, a greater understanding of how they arise may lead to novel treatments. Because MPNSTs are currently incurable and evidence from humans suggests that modifier genes affect the severity of NF1 (2, 6), we have used a mouse model of the malignancies associated with NF1 (7–9) to study the role of different genetic factors that influence MPNST tumorigenesis.

We and others have developed a mouse model to study NF1 by combining mutations in the murine Nf1 gene and Trp53 gene together on the same chromosome (NPcis mice) (7–10). The product of the Nf1 gene, neurofibromin, is a rasGAP protein involved in downregulating ras signaling. Because the Nf1 and Trp53 genes are closely linked on mouse chromosome 11, loss of the wt copies of the genes in this model occurs as a single genomic event, subsequently leading to concurrent upregulation of ras signaling and loss of p53. These mice develop many of the malignancies associated with NF1 in patients, including aggressive peripheral nerve sheath tumors (genetically engineered murine or GEM PNSTs) (7, 9, 11) as well as malignant astrocytoma and glioblastoma (8, 10). Because these mice are on genetically well-defined strain backgrounds and the incidence of tumors is completely penetrant by one year of age, we are using these mice to investigate the genetic and epigenetic effects on tumor susceptibility. Candidate modifiers of tumor susceptibility can then be tested for their role in the variable expressivity of NF1.

There has been a growing appreciation over the past 20 years of the role of epigenetic changes in tumorigenesis (12). Much of the data have pointed to the role of methylation in the silencing of tumor suppressor genes or in the deregulation of imprinted gene expression as a step in tumorigenesis. In addition to epigenetic alterations, expression of imprinted genes may be altered by genomic changes such as the loss or reduplication of chromosomes. We have shown recently that an imprinted locus linked to Nf1 and Trp53 on mouse chromosome 11 affects susceptibility to astrocytoma depending on whether tumors lose the maternal or paternal copy of mouse chromosome 11 (10). This work has suggested that linkage of an imprinted locus to a tumor suppressor gene can influence the rate of tumorigenesis (for further discussion see ref (13)).

In addition to interactions between imprinted loci and major tumor suppressor genes, such as Trp53, we wanted to test whether there were interactions between imprinted loci and low-penetrance polymorphic modifier genes. Modifier genes can alter the phenotype of mutations in tumor suppressor genes (or oncogenes) depending on the genetic background of the individual (14). Unlike the highly penetrant mutations in tumor suppressor genes, polymorphic modifier genes exert a more subtle effect through their normal functions, by changing the efficiency of different steps in tumorigenesis. These steps could occur within the tumor cell, such as the efficiency of cell cycle checkpoints or DNA repair, or systemically in the organism, such as the efficiency of anti-tumor immune responses. We have further explored the role of the imprinted locus in the context of polymorphic modifier genes in the NPcis mouse model by focusing on GEM PNSTs. In this mouse model, GEM PNSTs arise at younger ages than astrocytomas and are easier to identify and to score as a binary trait, allowing us to map modifiers more quickly. Similar to human MPNSTs, the GEM PNSTs in NPcis mice show mitotic figures, spindle cells, nerves within the tumor, and are variably positive for the tumor markers S100, p75 (low affinity NGFR), desmin, and collagenIV (Figure 1) (15). The majority of these tumors are negative for the rhabdomyosarcoma markers Myf4 and MyoD1 (data not shown) (16), although we occasionally see tumors with regions of Myf4 and MyoD1 immunoreactivity (4 out of 20 examined). This suggests that these tumors may also include malignant Triton tumors, a variant of MPNSTs seen in NF1 patients.

Figure 1.

NPcis mice develop GEM PNSTs with characteristics of human MPNSTs. (a) Low magnification view of a GEM PNST forming along a nerve root ganglion of the lower spinal cord (sc). Arrows indicate the tumor in the ganglion and arrowheads indicate tumor invasion into surrounding tissue along the spine. (b) High magnification view of the tumor from (a) with a nerve visible within the tumor (n) and a mitotic figure (arrow). (c) Positive immunoreactivity in a GEM PNST for S100 including a nerve (n) as a positive control for staining. (d) Positive immunoreactivity for p75. (e) Rare positive immunoreactivity for desmin. (f) Positive immunoreactivity for the basement membrane marker collagen IV.

Materials and Methods

Generation of NPcis mice on different genetics backgrounds

We bred NPcis mice at MIT and NCI-Frederick as described previously (10) according to the guidelines and regulations of the Animal Care and Use Committee at each institution. We compared cohorts of C57BL/6J-NPcis and C57BL/6JXA/J-NPcis bred at MIT and NCI to determine if the phenotype was reproducible and found no significant differences between the different cohorts. The results shown in Table 1 and Figure 2 are pooled from the two institutions. We genotyped the NPcis mice by PCR as described previously (10) and aged the mice until tumors developed as described previously (8).

Table 1.

Frequency of GEM PNST in NPcis Mice

Progeny	Total	With GEM PNST	% Affected
NPcis, B6	138	77	56%
NPcismat, B6*	47	18	38%
NPcispat, B6†	91	59	65%
NPcis, [B6XA]	100	29	29%
NPcismat, B6XA*	80	20	25%
NPcispat, AXB6†	20	9	45%
NPcis, [B6XAXB6]	254	131	52%
NPcismat, (B6XA)XB6*	110	42	38%
NPcispat, B6X(AXB6)†	144	89	62%

^*NPcis^mat mice inherit the NPcis mutant chromosome from their mother.

^†NPcis^pat mice inherit the NPcis mutant chromosome from their father.

Figure 2.

Frequency of GEM PNSTs on different F1 strain backgrounds compared to C57BL/6J-NPcis (B6). The incidence of GEM PNSTs in F1 progeny of C57BL/6J-NPcis fathers crossed to different wt mothers is shown in black bars and the incidence in F1 progeny of C57BL/6J-NPcis mothers crossed to different wt fathers is shown in white bars. Error bars indicate the 95% confidence interval. C57BL/6J-NPcis^pat (NPcis-pat B6), N=91; C57BL/6J-NPcis^mat (NPcis-mat B6), N=47; A/JXC57BL/6J-NPcis^pat (NPcis-pat AXB6), N=20; C57BL/6JXA/J-NPcis^mat (NPcis-mat B6XA), N=80; CBA/JXC57BL/6J –NPcis^pat (NPcis-pat CBAXB6), N=46; C57BL/6JXCBA/J –NPcis^mat (NPcis-mat B6XCBA), N=22; 129S4/SvJaeXC57BL/6J –NPcis^pat (NPcis-pat 129XB6), N=30; C57BL/6JX129S4/SvJae –NPcis^mat (NPcis-mat B6X129), N=16.

Pathology and Immunohistochemistry of GEM PNSTs

Hematoxylin and eosin stained sections were scored for the presence of spindle cell sarcomas by K.M.R. and R.T.B. A subset of the sarcomas (N = 20) was characterized with antibodies. We stained sections of tumors fixed in 10% formalin and paraffin-embedded with 1:1000 anti-S100 (#Z0311, Dako), 1:100 anti-p75 (#MAB365R, Chemicon), 1:100 anti-desmin (#10570, ICN-Cappel), 1:50 anti-collagen IV (#10760, ICN-Cappel), 1:30 anti-Myf4 (clone L026, Novocastra), and 1:50 anti-MyoD1 (clone 5.8A, Dako). Slides to be stained with anti-collagen IV were predigested with pepsin. All antibodies were detected using biotinylated secondary antibodies (Jackson ImmunResearch), the Vector Elite ABC kit, and DAB substrate for peroxidase (Vector Laboratories). For phenotyping of backcross progeny for genome-wide mapping of modifier loci, mice were considered to have a GEM PNST based on the characteristic spindle cell morphology of these tumors in hematoxylin and eosin stained sections. All backcross progeny were scored by K.M.R. for consistency.

Genome-wide SSLP genotyping

To map modifiers in (C57BL/6JXA/J)XC57BL/6J and C57BL/6JX(A/JXC57BL/6J) backcross progeny, we genotyped tail DNA at simple sequence length polymorphism (SSLP) markers (ResGen) listed in Supplementary Table 1, run on 3% Metaphor agarose gels in 1XTBE buffer.

Linkage analysis and statistical methods

We performed multipoint, binary trait linkage analysis (17, 18) separately in individuals who received the mutation from their father and from their mother, respectively, with the genetic mapping software, R/qtl (19), an add-on package to the general statistical software, R (20). We used a permutation test (21) to establish statistical significance. P-values, adjusted for the genome-wide scans, were estimated as the proportion of permutation replicates in which the maximum LOD score, genome-wide, was greater than or equal to the observed LOD score.

Assessing allelic imbalance by SNPs and SSLPs

We designed Taqman PCR assays for single nucleotide polymorphisms (SNP) using Assay-by-Design (www.appliedbiosystems.com) or Primer-Express software (Applied Biosystem, Inc, CA) (Supplementary Table 2). The ratio of C57BL/6J to A/J DNA in the tumor was determined by normalizing the samples to a standard curve of multiple ratios of C57BL/6J genomic DNA to A/J genomic DNA (The Jackson Laboratory, Bar Harbor, ME) using SDS2.1 (Applied Biosystem, Inc, CA). We ran SSLP markers on chromosome 15 and 19 on tumor DNA as described above. Densitometry on the ethidium stained bands was performed using the AlphaEase FC software (version 4.0.1) on the AlphaInnotech ChemiImager 5500. Between two and five replicates were analyzed for each tumor sample and between five and fourteen C57BL/6JXA/J F1 controls were analyzed to determine the reproducibility of the assays. Samples were determined to have allelic imbalance if they showed a greater than 30% change in C57BL/6J from the average C57BL/6J value in F1 controls and if the C57BL/6J value of the tumor DNA fell outside the 95% prediction interval (22) of the F1 mean.

Generation of the haplotype map

SNPview (http://snp.gnf.org/) was used to search for haplotype blocks, ordered according to Public Assembly v.3a. C57BL/6J, C3H/HeJ, DBA/2J, A/J, 129S1/SvImJ, and CAST/EiJ SNP alleles and C57BL/6J and A/J SSLP alleles were used to build the haplotype blocks. The region between 31.52 Mb and 52.07 Mb on chromosome 15 and the region between 0 Mb to 6.25 Mb on chromosome 19 were examined.

Determining syntenic regions in human MPNSTs

By comparing Mouse Genome Build 34 and Human Genome Build 35.1, the candidate regions were matched to human chromosomal regions using Comparative Maps (http://www.ncbi.nlm.nih.gov/Homology/).

Results

We have examined how polymorphisms in the strain background affect GEM PNSTs by crossing NPcis mice on the C57BL/6J background (C57BL/6J-NPcis) to several different strain backgrounds. We have shown previously that strain background strongly affects the tumor spectrum in these mice, with C57BL/6J-NPcis mice being more susceptible to astrocytomas than NPcis mice on a 129S4/SvJae background (129S4/SvJae-NPcis) (10). C57BL/6J-NPcis are also highly susceptible to GEM PNSTs (Table 1, Figure 2). Whereas the 129S4/SvJae strain and the CBA/J strain are resistant to astrocytoma (10), we found no effect of these backgrounds on GEM PNST incidence (Figure 2). Conversely, the A/J strain caused NPcis mice to be more resistant to GEM PNSTs in C57BL/6JXA/J F1 crosses (P=0.0004, χ² test) (Table 1). We also found marked reduction in GEM PNSTs in NPcis mice on DBA/2JXC57BL/6J F1 and CAST/EiJXC57BL/6J F1 backgrounds (data not shown). We did not note any significant change in the age of onset of the GEM PNSTs on different strain backgrounds or in reciprocal crosses. The effect of strain background on astrocytoma and GEM PNST incidence suggests that polymorphic modifier genes act in tissue-specific ways. Of the seven strains compared to C57BL/6J, none showed an overall reduction in tumorigenesis that might suggest a global modification of Nf1 or Trp53. This is consistent with the idea that polymorphic modifier genes act on the context in which the cancer develops and on cell type specific functions of Nf1 and/or Trp53.

In our previous study on astrocytoma susceptibility in NPcis mice, we showed that inheritance of NPcis mutant chromosome 11 from the mother increased susceptibility to astrocytoma (10). We therefore tested whether the inheritance of the NPcis mutant chromosome from the mother versus the father made a difference in GEM PNST incidence. We crossed C57BL/6J-NPcis fathers to wt C57BL/6J mothers and selected C57BL/6J-NPcis progeny for analysis. We similarly crossed C57BL/6J-NPcis mothers to wt C57BL/6J father and selected C57BL/6J-NPcis progeny for analysis. We aged the progeny of these two groups until tumors formed and compared the incidence of GEM PNSTs in the two populations (Table 1). Whereas the progeny of NPcis mothers (NPcis^mat) are more susceptible to astrocytoma, we found that the progeny of NPcis fathers (NPcis^pat) are more susceptible to GEM PNSTs (Table 1). This effect on susceptibility is independent of strain background and can be seen within the C57BL/6J-NPcis strain (P=0.003, χ² test).

We also observe increased susceptibility to GEM PNSTs in NPcis^pat progeny when the strain background is varied. When C57BL/6J-NPcis fathers are crossed to wt A/J mothers the NPcis^pat progeny are more susceptible to GEM PNST than the NPcis^mat progeny of C57BL/6J-NPcis mothers crossed to wt A/J fathers (Table 1, P=0.08, χ² test). As a final test, we crossed C57BL/6JXA/J-NPcis fathers to wt C57BL/6J mothers and compared the NPcis^pat progeny to the NPcis^mat progeny of C57BL/6JXA/J-NPcis mothers crossed to wt A/J fathers. In this experiment, the A/J strain background is inherited from the opposite parent as the NPcis mutant mouse chromosome 11, controlling for imprinted effects in the strain background as opposed to imprinted effects linked to mouse chromosome 11. In the reciprocal backcrosses, increased susceptibility to GEM PNSTs is again associated with inheriting NPcis mutant chromosome 11 from the father, not the A/J strain background (Table 1, P=0.0002, χ² test). This evidence shows that the impact of the reciprocal crosses on tumor susceptibility is not due to the mitochondrial genome or to the maternal influences on the embryo because these effects would be dependent on strain differences. We conclude that an imprinted locus on mouse chromosome 11 affects susceptibility to GEM PNST.

Because the F1 crosses demonstrated that polymorphic modifier loci can affect the susceptibility to GEM PNSTs (Figure 2) and the reciprocal crosses demonstrated that an imprinted locus on mouse chromosome 11 can also affect susceptibility to GEM PNSTs (Table 1), we wanted to determine whether there was any genetic interaction between the imprinted locus and the polymorphic loci responsible for the strain-specific effects. We identified the polymorphic modifier loci by binary trait linkage analysis of backcross progeny (14, 19). We chose the A/J strain as the resistant test strain because of the availability of the genome sequence for comparison to C57BL/6J (23), as well as the availability of chromosome substitution strains (24) to aid in future refinement of the modifier loci. We backcrossed C57BL/6JXA/J-NPcis to wt C57BL/6J mice and phenotyped the backcross progeny for the presence or absence of GEM PNSTs when the mice were sacrificed due to tumor burden. A total of 254 backcross progeny were phenotyped and genotyped at 105 SSLP markers with an average sweep radius of 8 cM. Binary trait linkage analysis of the pooled backcross progeny found no significant linkage of SSLP markers to resistance to GEM PNSTs. However, when the backcross progeny were split into NPcis^mat and NPcis^pat progeny, significant linkage was found on two different chromosomes (Figure 3) using a permutation test to determine the genome-wide adjusted significance threshold (P-value) of 5% or less.

Figure 3.

Linkage analysis of resistance to GEM PNST in (C57BL/6JXA/J)XC57BL/6J-NPcis and C57BL/6JX(A/JXC57BL/6J)-NPcis backcross progeny. (a) LOD plot across all autosomes for backcross progeny. Progeny of (C57BL/6JXA/J)-NPcis fathers and wt C57BL/6J mothers are shown in blue. Progeny of (C57BL/6JXA/J)-NPcis mothers and wt C57BL/6J fathers are shown in red. (b) Linkage was found to chromosome 15 (LOD = 2.6, P = 0.05) for NPcis^mat progeny. The red bars underneath the peak indicate the syntenic regions of human chromosomes 15p13-15 and 8q22-24. (c) Linkage was found to chromosome 19 (LOD = 3.0, P = 0.02) for NPcis^pat progeny. The green bars under the peak indicate the syntenic regions of human chromosomes 11q12-13 and 9q21.

In NPcis^pat backcross progeny, resistance to GEM PNSTs maps to the SSLP marker D19Mit59 at the proximal tip of mouse chromosome19 (P = 0.02 by permutation test for genome-wide significance). We have named this locus nerve sheath tumor resistance QTL 1, or Nstr QTL 1 (Nstr1). Progeny that are homozygous for the C57BL/6J allele at Nstr1 are more likely to develop GEM PNSTs, consistent with the presence of a resistance allele in the A/J strain (Table 2). Interestingly, this locus has no effect on the resistance of NPcis^mat backcross progeny (Figure 3, Table 2).

Table 2.

Affected NPcis Backcross Mice by Genotype

	D15Mit111			D19Mit59
Progeny	N	with GEM PNST*	%	N	with GEM PNST*	%
NPcispat, B6/A†	76	44	58%	58	25	43%
NPcispat, B6/B6‡	66	44	67%	72	54	87%
NPcismat, B6/A†	52	12	23%	52	21	40%
NPcismat, B6/B6‡	57	31	54%	48	18	38%

^*with GEM PNST for a given genotype.

^†B6/A indicates mice heterozygous for C57BL/6J and A/J at the specified marker.

^‡B6/B6 indicates mice homozygous for C57BL/6J at the specified marker.

In NPcis^mat backcross progeny, resistance to GEM PNSTs maps to SSLP marker D15Mit111 on mouse chromosome 15 (Figure 3; P = 0.05 by permutation test for genome-wide significance). NPcis^mat progeny that carry the A/J allele at this locus are less likely to develop GEM PNSTs (Table 2), again consistent with the presence of a resistance allele in the A/J strain (Table 2). We have named this locus nerve sheath tumor resistance QTL 2, or Nstr QTL 2 (Nstr2). Whereas Nstr1 on mouse chromosome 19 has little effect on NPcis^mat progeny, Nstr2 has little effect on NPcis^pat progeny (Figure 3, Table 2).

If a polymorphic modifier gene acts cell-autonomously and dominantly to block tumor formation, the allele responsible for the resistance is more likely to be lost when tumors form than the allele that does not confer resistance (25). To determine whether Nstr1 and Nstr2 are lost in tumors and are likely to act within the tumor cell during initiation, we characterized 20 of the rare GEM PNSTs from the resistant C57BL/6JXA/J-NPcis background for allelic imbalance on mouse chromosome 15 and 19 (Figure 4). MPNSTs are known to involve hyperproliferation of many stromal cell types, in addition to the tumor cells (1). We used quantitative methods to assess allelic imbalance, because we expect the tumor samples to be contaminated with significant numbers of non-tumor cells. Quantitative real-time PCR was used to analyze allelic imbalance at 110 single nucleotide polymorphisms (SNP) throughout the genome (S. Tsang, D. Sun, J. Diehl, K. Reilly, and D. Munroe, manuscript in preparation), and densitometry of ethidium stained gels was used to further analyze 15 SSLP markers on mouse chromosome 15 and 19 to confirm the results (Supplementary Figure 1). We found no allelic imbalance on mouse chromosome 15 in the region of Nstr2. In the genome-wide SNP scan for allelic imbalance, mouse chromosome 15 was the least affected of any chromosome with six out of thirteen tumors analyzed showing small regions of allelic imbalance (Figure 4). None of the tumors showed loss at SSLP marker D15Mit111. Allelic imbalance was found on mouse chromosome 19, with nine out of 20 tumors examined showing imbalance. Of the nine cases of allelic imbalance at the SSLP marker D19Mit59, four showed reduction of the A/J allele, consistent with loss of a resistance modifier allele from the A/J strain. On mouse chromosome 19 we observed allelic imbalance in tumors from both NPcis^pat mice (3 out of 7 tumors examined) and NPcis^mat mice (6 out of 13 informative tumors examined). Although one might expect Nstr1 to be altered only in NPcis^pat tumors, since NPcis^mat backcross progeny do not show linkage to Nstr1, the allelic imbalance measures an end state of the tumor, whereas the linkage to Nstr1 in the backcross progeny measures a rate-limiting step in tumorigenesis. Loss of Nstr1 occurs in GEM PNSTs from both NPcis^pat and NPcis^mat, but may only be rate limiting to tumorigenesis in NPcis^pat progeny. These data on allelic imbalance suggest that Nstr1 shows allelic imbalance in roughly half of tumors examined with equal preference for loss of the A/J allele and the C57BL/6J allele.

Figure 4.

Allelic imbalance in C57BL/6JXA/J-NPcis F1 GEM PNSTs. Each row represents the genotype of a single tumor along the chromosome. Reduction in the A/J allele by greater than 30% is indicated in black. Reduction in the C57BL/6J allele by greater than 30% is indicated in white. Samples scored as heterozygous are hatched and samples that gave no data are indicated as NA (not available). The markers nearest to the mapped loci are highlighted in boldface, with the marker nearest the Nf1 indicated on chromosome 11.

We have used a haplotype prediction approach (26) to confirm that Nstr1 and Nstr2 carry polymorphic regions consistent with our observations of different strain phenotypes and to focus on regions of likely modifiers within the mapped loci. We have compared these regions to syntenic regions of the human genome to see if they are implicated in human MPNST tumorigenesis. Using available haplotype maps (http://snp.gnf.org) we have looked for regions where the resistant A/J strain has a different haplotype from the susceptible C57BL/6J and 129 strains (Supplementary Fig 2). On mouse chromosome 15 between D15Mit201 and D15Mit151, three regions fit this pattern at 37.67 Mb, at 41.92 Mb, and at 44.57 Mb. At 41.92 Mb, A/J shares a haplotype with DBA/2J, another resistant strain, and at 44.57 Mb A/J shares a haplotype with CAST/EiJ, which is similarly resistant. In addition, these regions are syntenic with human chromosome 8q22-23. Human chromosome 8q has been found to be amplified in MPNSTs (27, 28), a translocation between chromosome 5q11 and chromosome 8q22 (29) and a translocation between chromosome 5q13 and chromosome 8q23 (30) have been reported in MPNST, and amplification of chromosome 8q22 has been observed in an MPNST (31). While these alterations are not common events in human MPNSTs, as might be expected for alterations in a highly penetrant tumor suppressor gene or oncogene, they are consistent with the presence of a low-penetrance polymorphic modifier gene. These data suggest that the syntenic region of mouse chromosome 15 and human chromosome 8q22-23 contain a polymorphic modifier gene that affects GEM PNST and MPNST tumorigenesis.

A similar analysis of haplotype structure on mouse chromosome 19 between the centromere and D19Mit109 found one candidate region at 5.22 Mb, very close to the peak at D19Mit59 (Supplementary Fig 2). At this location, the A/J strain shares a haplotype with DBA/2J. This region also overlaps with Mtes1, a modifier of breast cancer metastasis identified in mice (32). Interestingly, this region is syntenic with human chromosome 11q13, a region found to be specifically deleted in two MPNSTs (30, 33) and specifically amplified in one MPNST (33). In addition, translocations between human chromosome 11q13 and chromosome 15p11 (30), chromosome 19p13 (34), and chromosome 19q13 (30) have been reported. Taken together, these data are consistent with the presence of a polymorphic modifier gene affecting both GEM PNSTs and MPNSTs on proximal mouse chromosome 19 and human chromosome 11q13.

Discussion

The data we present here, together with our previous data (10), demonstrate that NPcis mice are more susceptible to GEM PNSTs or more susceptible to astrocytoma depending on whether the NPcis mutation is inherited from the father or the mother, respectively. We confirm here that these effects are independent of strain background and are therefore linked to inheritance of mouse chromosome 11, rather than strain-specific factors such as maternal cytoplasm or mitochondria. The effect of mouse chromosome 11 on susceptibility is likely due to changes in dosage of mono-allelically expressed genes during tumor initiation. An imprinted gene expressed from the wt mouse chromosome 11 may be lost when the wt copies of Trp53 and Nf1 are lost during tumor initiation. An imprinted gene expressed from the NPcis mutant chromosome 11 may be reduplicated through subsequent genome instability after loss of Trp53. Alternatively, the chromatin marks that maintain the silenced chromosome at the imprinted locus may change the genomic stability on mouse chromosome 11, such that the imprinted chromosome is more or less likely to be lost or mutated. Recent data in NPcis mice have demonstrated increased loss of wt mouse chromosome 11 when the genome is hypomethylated through mutations in Dnmt1 (35). This raises the possibility that the differential methylation on the two parental chromosomes at imprinted loci may affect their rates of loss. Whether the imprinted locus drives the loss of wt Trp53 and Nf1 or the mutation at Trp53 and Nf1 drives loss or duplication of the imprinted locus remains to be determined. However, this distinction has important implications for targeting therapies against imprinted genes, depending on whether the effect of the imprinted locus is through effects on chromatin structure or effects on gene dosage.

The reciprocal effect of mouse chromosome 11 inheritance on susceptibility to astrocytoma and GEM PNST is interesting and raises questions about the nature of the imprinted gene causing the effect. A recent study has identified 37 candidate imprinted genes on mouse chromosome 11 using a bioinformatics approach (36), in addition to the two confirmed imprinted genes on mouse chromosome 11, Grb10 (37, 38) and U2af1-rs1 (39). It is possible there are two unrelated imprinted genes on mouse chromosome 11 affecting astrocytoma and GEM PNSTs independently, with opposite imprinting. Alternatively, a single imprinted gene on mouse chromosome 11 may act as an oncogene in one tissue type and a tumor suppressor gene in the other, such that, for example, loss of the paternal chromosome leads to increased tumorigenesis in one tissue and reduplication of the paternal chromosome leads to increased tumorigenesis in the other. Another possibility is that a single gene is expressed from different parental chromosomes in the two tissue types, but has a similar function (as an oncogene or tumor suppressor gene) in the two tissues. In the example of an imprinted tumor suppressor gene, loss of the wt paternal mouse chromosome 11 would lead to increased astrocytoma, whereas loss of the wt maternal mouse chromosome 11 would lead to increased GEM PNST. It is interesting to note that Grb10 has been shown to be expressed from the paternal chromosome in the brain and from the maternal chromosome in other organs in a highly isoform-specific manner (37, 38). Finally, differences in gene silencing on mouse chromosome 11 between the two tissues types may cause sequestering of the maternal or paternal chromosome. This may lead to differences in the mutation rate or kinetics of chromosome loss for a particular parental chromosome, leading to a secondary effect on the rate of mutation of the wt copies of Nf1 and Trp53.

The importance of epigenetic changes in human cancer is now well established (see (12) for review), with changes in dosage levels of imprinted genes such as IGF2 (40, 41) affecting the rate of tumorigenesis. The monoallelic expression of imprinted genes acts similarly to the first “hit” in Knudson’s two-hit model of cancer (42), with the second hit occurring either through epigenetic mechanisms (loss of imprinting) or genetic mechanisms. An example of genetic changes acting to alter imprinted gene expression was first shown in human rhabdomyosarcomas (43) where maternal human chromosome 11p15 is preferentially lost and paternal human chromosome 11p15 is duplicated. More recently, susceptibility to succinate dehydrogenase D (SDHD)-linked paraganglioma was found to be inherited by paternal transmission of the mutant SDHD gene. Because SDHD is biallelically expressed it has been proposed that a linked imprinted gene on human maternal chromosome 11 is lost when the wild-type allele of SDHD is lost. In this way, inheritance of SDHD mutation from the father has a long-distance effect on a linked imprinted gene. Our data in the NPcis mouse model shows a very similar effect, with concomitant loss of the wt copy of Trp53 and Nf1 with another imprinted locus on mouse chromosome 11.

We have analyzed here for the first time the genetic interaction between an imprinted locus and polymorphic loci and their effects on cancer susceptibility. The imprinted locus on mouse chromosome 11 interacts epistatically with a network of other loci to affect resistance to GEM PNSTs. In the case of NPcis^pat mice with increased susceptibility to GEM PNSTs, a locus on chromosome 19 on the A/J strain background acts to increase resistance. In the case of NPcis^mat mice with reduced susceptibility, a locus on chromosome 15 on the A/J strain background further increases the resistance to GEM PNST. The relevance of these results to human NF1 patients is supported by the overlap of these loci with regions altered in MPNSTs. Consistent with these loci being low-penetrance modifiers for MPNST susceptibility, we see evidence for changes in these genomic regions in a relatively low numbers of mouse and human tumors. However, the observation that these regions are translocated in multiple human tumors suggests that these rearrangements are not random. Identification of modifiers of NF1 in mouse models will allow these modifiers to be tested directly in human association studies, specifically in cases where tumors can be tested for the change in expression of candidate imprinted genes. These data demonstrate that polymorphic modifier genes affect tumorigenesis under very specific conditions. The understanding of these conditions will allow for more accurate risk assessment in the future and genetic counseling for individuals at high-risk for cancer.