Allele-Specific Deletions in Mouse Tumors Identify Fbxw7 as Germline Modifier of Tumor Susceptibility

Genome-wide association studies (GWAS) have been successful in finding associations between specific genetic variants and cancer susceptibility in human populations. These studies have identified a range of highly statistically significant associations between single nucleotide polymorphisms (SNPs) and susceptibility to development of a range of human tumors. However, the effect of each SNP in isolation is very small, and all of the SNPs combined only account for a relatively minor proportion of the total genetic risk (5–10%). There is therefore a major requirement for alternative routes to the discovery of genetic risk factors for cancer. We have previously shown using mouse models that chromosomal regions harboring susceptibility genes identified by linkage analysis frequently exhibit allele-specific genetic alterations in tumors. We demonstrate here that the Fbxw7 gene, a commonly mutated gene in a wide range of mouse and human cancers, shows allele-specific deletions in mouse lymphomas and skin tumors. Lymphomas from three different F1 hybrids show 100% allele-specificity in the patterns of allelic loss. Parental alleles from 129/Sv or Spretus/Gla mice are lost in tumors from F1 hybrids with C57BL/6 animals, due to the presence of a specific non-synonymous coding sequence polymorphism at the N-terminal portion of the gene. A specific genetic test of association between this SNP and lymphoma susceptibility in interspecific backcross mice showed a significant linkage (p = 0.001), but only in animals with a functional p53 gene. These data therefore identify Fbxw7 as a p53-dependent tumor susceptibility gene. Increased p53-dependent tumor susceptibility and allele-specific losses were also seen in a mouse skin model of skin tumor development. We propose that analysis of preferential allelic imbalances in tumors may provide an efficient means of uncovering genetic variants that affect mouse and human tumor susceptibility.


Introduction
Susceptibility to cancer development is consequence of a combination of genetic and environmental factors. Whereas a wide range of environmental agents may contribute to cancer development, the genetic component of risk is difficult to dissect [1]. Genetic risk factors for cancer are either in the high penetrance category, such as mutations in BRCA1/2 that cause familial breast and ovarian cancers, or are classified as low penetrance variants that work additively or in complex combinations to increase risk [1,2]. Relatively little is known about the genes in this latter category, which are nevertheless likely to be very important determinants of the developmental risk of many cancer types [2,3]. Present approaches to the detection of low penetrance tumor susceptibility alleles in humans involve association studies using DNA samples from hundreds or thousands of cancer patients, and an equal number of well-matched controls. Such studies are plagued by confounding factors such as population heterogeneity, weak effects, and genetic interactions, and require very large numbers of cases and controls to reach statistical significance [4][5][6]. For many complex trait diseases, including cancer, the total number of significant SNP associations detected can only account for a very small proportion of the total genetic risk [7], leading to considerable discussion of the best ways to discover the majority of disease-causing alleles in the human population.
Mouse models offer an important alternative approach to the study of cancer susceptibility. Studies on mice have revealed that tumor predisposition in different strains is controlled by multiple loci that exhibit complex genetic interactions [8,9]. Importantly, in these mouse crosses only two alleles at each locus are segregating in the population, and the large differences in susceptibility are therefore due to combinations of common alleles. However, the resolution of mapping studies, based on linkage analysis involving inbred strains alone, is poor [1,3]. Analysis of somatic events in tumors could provide more high resolution information from relatively smaller numbers of samples [10,11], due to the fact that genetic instability in tumors results in specific genetic aberrations leading to copy number gains or losses of important cancer genes. Indeed, analysis of somatic deletions was instrumental in identification of the causal genes for hereditary cancer syndromes such as Li-Fraumeni syndrome and retinoblastoma. Predisposition to tumor development caused by a germline mutation was accompanied by tumor-specific loss of the wild type allele (TRP53 and RB respectively), thus establishing a precedent for use of somatic genetic alterations for identification of germline tumor susceptibility genes [12,13]. The information obtained from somatic events in tumors combined with linkage analysis could provide the opportunity to obtain more high resolution information from relatively smaller numbers of samples. Thus, fine mapping of somatic recombination events, deletions and amplifications, together with genotyping to detect preferential allelic imbalance, could offer important clues to the localization of polymorphisms of low penetrance genes that influence cancer susceptibility [10,11,14].
Since Fbxw7 is frequently mutated or deleted in a large proportion of lymphomas from p53 heterozygous mice [15], we first analyzed the possibility that these deletions may preferentially involve one parental allele. Here we demonstrate that a straindependent polymorphism in the N-terminus of Fbxw7 causes 100% selectivity in patterns of alleleic loss in mouse lymphomas. Genetic linkage studies confirmed that this polymorphism confers increased risk of tumor development in a p53-dependent manner. Extension of this approach to human samples may identify a proportion of the genetic variants that affect cancer risk, but are not detected by standard linkage of genetic association studies.

Allele-specific loss of Fbxw7 in radiation-induced mouse Lymphomas
We have previously proposed that it may be possible to identify germline polymorphisms that influence susceptibility by analysis of allele-specific deletions or amplifications in mouse or human tumors [10,11]. Thus, we carried out a detailed analysis of the directionality of allele loss of Fbxw7 in lymphomas from a series of different p53+/2 F1 hybrid mice. In contrast to tumors derived from 129/Sv6Mus spretus (Spretus/Gla) hybrids, which showed 50% loss of markers on each parental chromosome 3 (Fig. 1A), tumors derived from F1 hybrids between C57BL/6 and Mus spretus, or between C57BL/6 and 129/Sv, showed consistent loss of the Mus spretus or 129/Sv alleles respectively (Fig. 1A). Although these examples of chromosomal imbalance involve large regions of mouse chromosome 3 that include the region containing Fbxw7, our previous studies demonstrated that Fbxw7 was indeed the driver gene for these losses, since tumors exhibited intragenic Fbxw7 deletions or mutations, and lymphomas from p53+/ 2Fbxw7+/2 mice did not show any evidence of significant deletions on this chromosome [15]. These data suggested that 129/Sv and Spretus/Gla may carry similar haplotypes in the Fbxw7 region, while C57BL/6J should carry a different sequence. Sequence analysis demonstrated that Spretus/Gla and 129/sv strains share a non-synonymous coding sequence polymorphism at position 61 (GAC) that encodes Aspartic acid whereas in C57BL/6 mice there is an Asparagine residue (AAC) leading to the disappearance of a negative charge (Fig. 1B). Additional sequencing of other strains showed that this polymorphism segregates in many common laboratory strains of mice (Fig. 1C, Table 1).
Asp61Asn polymorphism in Fbxw7 is associated with susceptibility to development of radiation-induced lymphoma Having found a common allele-specific deletion involving Fbxw7 on chromosome 3, we then asked whether this polymorphism is associated with altered susceptibility to development of radiationinduced lymphomas. We previously carried out a cross between outbred Spretus/Gla and p532/2 mice of a mixed C57BL/6 and 129/Sv genetic background (1296B6 mice). These interspecific F1 hybrid p53+/2 animals were backcrossed to the 129/B6 p532/2 mice to generate a backcross population (n = 168) of which approximately half of the animals were p53+/2, while the remainder were homozygous for the p53 null allele (p532/2) [15]. Since all 3 Fbxw7 alleles were segregating within this population, we tested the possibility of association between lymphoma development and the presence of the candidate susceptibility SNP in Fbxw7. While there was no detectable influence of the Asp61Asn polymorphism in Fbxw7 on lymphoma development in p532/2 mice (Fig. 2B), a significant effect was detected at the same SNP in radiation-induced lymphomas from p53+/2 mice ( Fig. 2A). The result is in agreement with the observation that the tumor suppressor activity of Fbxw7 is also p53dependent [15,16]. To our knowledge, this is the first example of a natural genetic variant that influences tumor development in a p53-dependent manner.

Fbxw7 Participates in Skin Tumor Progression
Loss or mutation of FBXW7 has been identified in a wide range of human tumors of both mesenchymal and epithelial origin [29].
To clarify the role of Fbxw7 in epithelial cancer development, we investigated the consequences of Fbxw7 deficiency in the DMBA/ TPA mouse model of skin cancer development. In this model, skin tumors are initiated by treatment with a single dose of the carcinogen DMBA, which induces H-ras gene mutations in target cells. Subsequent promotion of these cells by treatment with TPA results in formation of benign papillomas, a subset of which undergoes malignant progression to carcinomas. Since complete loss of Fbxw7 is lethal in mice, we examined the effects of partial Fbxw7 deficiency on skin tumor development, both in wild type mice and in animals lacking one functional copy of the p53 gene. Groups of 19 Fbxw7+/+, 28 Fbxw7+/2, 32 p53+/2 and 34 p53+/ 2Fbxw7+/2 mice were initiated by a single dose of DMBA and promoted twice weekly with TPA. Mice that were wild type at the p53 locus showed no effect of reduced Fbxw7 gene dosage on papilloma number (Fig. 3A). However there was a strong and statistically significant increase in the number of papillomas in p53+/2Fbxw7+/2 mice compared with p53+/2 mice (Fig. 3B). We conclude that Fbxw7 is an important regulator of skin tumor progression, and that a partial germline deficiency causes increased susceptibility to papilloma development, particularly in the context of a reduced gene dosage for p53. These data extend and support the conclusion above that Fbxw7 is an important tumor suppressor gene that interacts strongly with the p53 pathway.
We next investigated the possibility that allele-specific deletions in Fbxw7 locus may occur in skin tumors. Due to normal tissue contamination, it is hard to detect allele-specific change in skin tumors by regular PCR described in lymphoma studies. Thus we designed TaqMan probes specific for Fbxw7 polymorphisms discovered above. Gene copy number changes at the Fbxw7 locus were examined by quantitative PCR (TaqMan) analysis of skin tumors from interspecific backcross mice derived by crossing male Spretus/Gla with female mice of the NIH/O strain that carry the G and A alleles respectively (see Table 1). Female F1 hybrid mice generated from this cross were backcrossed to the parental NIH/O strain to generate a backcross population (NIHBX) as previously described [34,35]. Fbxw7 deletions were detected in some papillomas and in the majority of the carcinomas studied (Fig. 4A). Analysis of the pattern of allele-specific losses in these tumors indicated that the preference was overwhelmingly for loss of the Fbxw7 allele inherited from the Mus spretus parent. These data indicate that loss of Fbxw7 is an important event in skin cancer progression, and demonstrate that the strain-specific polymorphism affects the directionality of the deletion in the same manner as was seen for lymphomas. In the lymphoma model described above, the presence of the G/A allele was linked to the allele-specific deletions and also, by SNP association analysis, to susceptibility to lymphoma development. We investigated the possibility that allele-specific losses of Fbxw7 in the NIHBX population may also predict linkage to a skin tumor susceptibility locus in the region containing the Fbxw7 gene on chromosome 3. Interestingly, although allele-specific deletions were clearly seen in tumors from this backcross, no obvious linkage was detected to markers near the Fbxw7 locus (Fig. 4B). We conclude that in this skin model, analysis of preferential allelic imbalance in tumors leads to detection of a germline susceptibility locus that is not detected by standard linkage analysis.

Discussion
The identification of low penetrance cancer susceptibility genes is a major challenge, and fraught with difficulties in standard human population studies. In this work we have established Fbxw7 Is a p53-Dependent Tumor Susceptibility Gene PLoS ONE | www.plosone.org evidence that supports the approach of studying somatic mutation events within the tumors, in combination with linkage analysis, to identify polymorphic allelic variants that have cell-autonomous effects on tumor development. Analysis of allele-specific deletions of Fbxw7 locus in mouse lymphomas showed that the pattern in three different sets of F1 hybrid was 100% specific: the allele contributed by the C57BL/6 parent was selectively retained, while alleles from either Mus spretus or 129/Sv were preferentially lost, in lymphomas from the respective F1 hybrid animals. An association study in a backcross in which all 3 alleles were segregating confirmed that the AAC (Asn) allele inherited from the C57BL/6 parent conferred greater risk of lymphoma than the GAC (Asp) allele from either 129/Sv or Mus spretus. An important conclusion is that while the genetic effect of the polymorphism on overall cancer risk is small and this variant is clearly in the ''low penetrance' category, the effect shows 100% penetrance in the pattern of allele-specific loss of heterozygosity in lymphomas. Application of this approach to human cancers may therefore identify genetic variants that only have very small and possibly undetectable effects on cancer risk in GWAS, but have strong effects on somatic genetic changes.
Our results on the skin tumor model lend strong support to the conclusion that Fbxw7 variants contribute to cancer risk. In this model, loss of heterozygosity at the Fbxw7 gene is specific for the Spretus allele, and no examples were seen of loss of the parental NIH/O (AAC) allele. In addition, germline deletion of one copy of Fbxw7 clearly increased skin tumor susceptibility, particularly in the context of p53 deletion. Nevertheless, in the same backcross population in which the allele-specific losses were seen, no evidence was detected of a significant germline skin tumor susceptibility gene at this locus using a range of markers on mouse chromosome 3. The reasons for this observation are unclear, but it is possible that additional germline susceptibility or resistance alleles for skin cancer map within the same linkage region as Fbxw7. If these act in the opposite direction, the effect due to susceptibility through the Fbxw7 AAC allele may be masked by the opposing genetic effects of these linked variants, as discussed previously [1]. The results of linkage analysis may therefore depend on the tissue and tumor type under investigation, where in some cases closely linked genetic variants that act in opposite directions may confound detection of susceptibility alleles. Similar Fbxw7 Is a p53-Dependent Tumor Susceptibility Gene PLoS ONE | www.plosone.org tissue-specific effects may underlie the differential effect of p53 status on germline deletion of Fbxw7.
It has been reported that monoallelic deletion of Fbxw7 in different tumor systems, in human and mouse, is a frequent event during cancer progression [15,29], and that Fbxw7 acts as a haploinsufficient tumor suppressor gene [15,29]. In the light of these results on mouse models of cancer, it is possible that specific genetic variants in human FBXW7, or indeed of other common human tumor suppressors, may also affect the patterns of allelic losses in tumors. Such changes may be more difficult to detect in human tumors than are allele-specific gene amplifications [10] which occur over a wider range of copy number changes and are less subject to confounding by the presence of contaminating normals cell that increase the signal-noise ratio. Nevertheless, tumor microdissection in combination with genome-wide analysis of allelic imbalances may reveal new opportunities to detect germline variants that influence cancer risk.
In preliminary studies we have been unable to detect any effect of the specific C57BL/6 or Mus spretus alleles on intracellular localization of epitope tagged Fbxw7 proteins or any reproducible differences in protein stability (data not shown). While a human non-synonymous coding sequence polymorphism has been found in the same region of the human gene, this is only seen at low frequency in populations of African descent (rs6842544: AGARGGA) (http://www.ncbi.nlm.nih.gov/snp). Further studies will be required to investigate possible allele-specific loss or retention of this allele in tumors from heterozygous patients. Elucidation of the mechanisms by which this and other Fbxw7 genetic variants act may allow us in future studies to recognize patients who are at high risk of cancer development, or who may be more likely to respond to therapies targeted at this pathway. The recent identification of mTOR as a downstream target of Fbxw7 [27,28] suggests that germline or somatic changes may affect tumor responses to known inhibitors of the mTOR pathway such as rapamycin, and such investigations are presently in progress in this laboratory.
Radiation-induced Lymphoma. This cohort of mice was described before for radiation-induced induced lymphoma susceptibility [15,36]. Briefly, F1 hybrid mice were generated by crossing females p532/2 129/B6 with males p53 wild type Spretus/Gla, and F 1 backcross mice were obtained by mating the female F 1 hybrids with male p532/2 129/B6 mice. The 5-weekold 129/Sv, F 1 and F 1 backcross mice were exposed to a single dose of 4 Gy whole-body irradiation and observed daily until they were sick, then sacrificed and autopsied.
Mice were bred and treated under University of California at San Francisco (UCSF) Laboratory Animal Resource Center (LARC) regulations.
Allele specific LOH determination PCR Analysis. For DNA preparation, tumors were ground into a fine powder in liquid nitrogen. DNA was isolated using standard phenol/chloroform extraction following over-night 55uC incubation with Proteinase-K. All tumor DNA samples were examined together with normal DNA from the same mice, to take account of variation in Fbxw7 alleles. Tumor and corresponding normal DNA was used for LOH studies with microsatellite markers at the Fbxw7 locus [15]. PCR amplification was performed in 20-mL volume containing 16 PCR buffer (Bioline), 200 mM of each dNTP (Pharmacia), 6.6 mM of each primer (Qiagen Operon), 1.5 mM MgCl2, 2 units of Taq polymerase (Bioline), and 80 ng of DNA. Amplifications were initially denatured for 3 min at 94uC, followed by 35 cycles of 30 sec at 94uC, 30 sec at 55uC or 52uC, and 30 sec at 72uC. PCR products were electrophoresed in 4% (3% NuSieve/1% normal) agarose gel and visualized by ethidium bromide staining.
Quantitative Real Time PCR (QPCR) Analysis. Primers and TaqMan probes were designed using the ABI Primer Express Software (version 1.5). The absence of non-specific amplification was confirmed by analyzing the PCR amplification by 3% agarose gel electrophoresis and ethidium bromide staining. The sequences were as follows: Fbxw7 forward primer: GCGCGGAA-TGGTGAACTT; Fbxw7 reverse primer: CGTTCTGGTCTC-CAGGCCT; probe for Fbxw7-GAC allele: 59-FAM-TGGGTGCAGACCCTA-MGB-39; probe for Fbxw7-AAC allele: 59-VIC-TGGGTGCAAACCCTA -MGB-39. The probe for the detection of amplified products was labeled with 6carboxifluorescein (FAM) in the 59end, all the reporters were quenched by 6-carboxy-tetramethyl-rhodamine (TAMRA) conjugated at the 39end. All probes were purchased in Applied Biosystems. PCRs for each sample were carried out in triplicate and repeated at least twice in 50 uL volume consisting in: 25 uL TaqMan 26 PCR Mater Mix (Applied Biosystems), 100 nM TaqMan probe, forward and reverse primers (100 uM). A master mix of the components including the equivalent of 125 ng of DNA per well (5 uL at 25 ng/uL) was made and aliquoted into a 96-well optical plate (Applied Biosystems). PCRs were performed according to the thermal profile: 1 cycle 50uC 2 min, 1 cycle 95uC 10 min, 40 cycles 95uC 15 sec and 60uC 1 min. For quantification of gene relative copy number, we used comparative Ct method [37,38]. The Ct values for each set of triplicates were averaged. For Fbxw7 copy number |DDCt|.0.5 were considered to be losses.

Statistical and linkage analysis
The Kaplan-Meier method was used to compare the tumor development after irradiation of mice carrying different alleles of Fbxw7, and the Mann-Whitney test for comparing number of papillomas between different genotypes of mice. Statistical analysis was performed using SPSS version 12.0 (SPSS, Chicago, IL). Linkage analysis was carried out using R/QTL.

Ethics statement
The animal protocols were approved by University of California at San Francisco (UCSF) Laboratory Animal Resource Center (LARC) (AN084982).