Nature Genetics
34, 403 - 412 (2003)
Published online: 27 July 2003; | doi:10.1038/ng1220
Identification of Stk6/STK15 as a candidate low-penetrance tumor-susceptibility gene in mouse and humanAmanda Ewart-Toland1, 9, Paraskevi Briassouli2, 9, John P de Koning1, 8, 9, Jian-Hua Mao1, Jinwei Yuan1, Florence Chan3, Lucy MacCarthy-Morrogh2, Bruce A J Ponder4, Hiroki Nagase5, John Burn6, Sarah Ball6, 8, Maria Almeida6, Spiros Linardopoulos2, 3
& Allan Balmain1, 71 UCSF Comprehensive Cancer Center, University of California, San Francisco, California 94115, USA. 2 The Breakthrough Breast Cancer Research Centre, Institute of Cancer Research, Fulham Road, London SW3 6JB, UK. 3 Cancer Research UK, Centre for Cancer Therapeutics, The Institute of Cancer Research Haddow's Laboratories, 15 Cotswold Road, Sutton SM2 5NG, UK. 4 Cancer Research UK, Human Cancer Genetics Group, University Department of Oncology, Strangeways Research Laboratories, Worts Causeway, Cambridge CB1 8RN, UK. 5 Roswell Park Cancer Center, Buffalo, New York, USA. 6 Institute of Human Genetics, Newcastle Upon Tyne, NE1 3B2, UK 7 Department of Biochemistry and Biophysics, University of California, San Francisco, California 94143, USA. 8 Present addresses: Department of Physiological Chemistry, University Medical Center Utrecht, P.O. Box 85060, 3508 AB Utrecht, The Netherlands (J.P.d.K.); University Hospitals of Leicester NHS Trust, Leicester Royal Infirmatory, Leicester LE1 5WW, UK (S.B.). 9 These authors contributed equally to this work.
Correspondence should be addressed to Spiros Linardopoulos spiros@icr.ac.uk or Allan Balmain abalmain@cc.ucsf.eduLinkage analysis and haplotype mapping in interspecific mouse crosses (Mus musculus  Mus spretus) identified the gene encoding Aurora2 (Stk6 in mouse and STK15 in human) as a candidate skin tumor susceptibility gene. The Stk6 allele inherited from the susceptible M. musculus parent was overexpressed in normal cells and preferentially amplified in tumor cells from F1 hybrid mice. We identified a common genetic variant in STK15 (resulting in the amino acid substitution F31I) that is preferentially amplified and associated with the degree of aneuploidy in human colon tumors. The Ile31 variant transforms rat1 cells more potently than the more common Phe31 variant. The E2 ubiquitin-conjugating enzyme UBE2N was a preferential binding partner of the 'weak' STK15 Phe31 variant form in yeast two-hybrid screens and in human cells. This interaction results in colocalization of UBE2N with STK15 at the centrosomes during mitosis. These results are consistent with an important role for the Ile31 variant of STK15 in human cancer susceptibility.Studies of cancer predisposition have largely concentrated on the role of high penetrance susceptibility genes. Less than 10% of the total human tumor burden, however, is accounted for by mutations in these genes. More genetic variation in cancer risk is probably due to more common but less penetrant alleles1,
2. In man, such genes will be difficult to find as they do not segregate as single mendelian traits. The mouse offers a powerful system for studying polygenic traits, such as cancer, and has been widely used for this purpose3,
4,
5,
6. But problems at several levels must be overcome in any strategy to find tumor modifier genes: mapping the locus at high resolution, identifying the critical gene and finding the functional polymorphism. Standard linkage analysis methods for localizing quantitative trait loci (QTLs) detect loci within intervals of at least 10−30 cM, and the usual method of refining the region involves generation of congenic mice. This is an expensive and time-consuming process that can be problematic because multiple QTLs may colocalize within the same interval or because interacting genes must be present for the disease phenotype to be observed. Finally, identification of the relevant polymorphism is difficult because all of the genes in the immediate vicinity may have sequence variants that correlate with the phenotype7. We used a new multi-step approach that helps to resolve many of these problems by exploiting the genetic diversity between tumor-susceptible M. musculus and both outbred and inbred strains of tumor-resistant M. spretus with the high recombination and low linkage disequilibrium (LD) found in humans. Using this approach, we identified a common polymorphic variant in STK15 that modifies cell transformation in vitro and tumor aneuploidy in vivo and may be a low-penetrance tumor-susceptibility gene acting at the level of genetic instability.
Results
High resolution mapping of Skts13
We have previously identified by QTL analysis 12 loci that control skin tumor susceptibility in crosses between outbred M. spretus and inbred strains of M. musculus6,
8 (NSP cross). By further analysis of the same cross (in over 300 mice) using more densely spaced markers, we identified a new skin tumor susceptibility locus on distal chromosome 2, Skts13, that confers resistance to papilloma development (P = 3.4 10-3). We also carried out two independent backcrosses, each involving over 100 mice, between inbred strains of M. spretus (SPRET/Ei from the Jackson Laboratory and SEG/Pas from the Institute Pasteur) and the same strain of M. musculus (NIH/Ola). The cross with the SPRET/Ei mice (NSJ cross) showed an even stronger linkage with papilloma resistance at this locus (P = 3.7 10-5), whereas the SEG/Pas cross (NSE) showed no significant linkage (Fig. 1a). The fact that the results from the original outbred cross were intermediate between those from the two inbred crosses suggested that outbred mice segregate different alleles, whereas SPRET/Ei and SEG/Pas mice are homozygous with respect to resistance or susceptibility alleles, respectively.
 | | |
Information on shared haplotypes in outbred populations of humans or experimental animals can be used to refine the locations of potential disease susceptibility genes. We therefore constructed haplotypes for distal chromosome 2 using the variation in microsatellite lengths between the M. spretus alleles in the outbred colony. Four different M. spretus haplotypes (H1 to H4) were present in these outbred mice (Table 1), and we investigated a possible association between haplotype and tumor number. We compared the mean papilloma numbers of the different haplotypes with that of mice from the same backcross that were homozygous NIH/Ola at this locus. Tumor number in mice with haplotypes H1, H2 and H3 but not with H4, each haplotype considered separately, differed significantly from that in mice that were homozygous NIH/Ola at this locus (all mice from the same cross; H1 P = 0.06, H2 P = 0.04, H3 P = 0.002, H4 P = 0.3). Therefore, we grouped haplotypes H1, H2 and H3 together. The mean papilloma number in mice with haplotypes H1, H2 and H3 combined (mean = 2.3, n = 107) also differed significantly from that in mice that were homozygous NIH/Ola at this locus (mean = 3.6, n = 113; P = 1.8 10-3).
| | |
Notably, only two small intervals seemed to be shared by haplotypes H1, H2 and H3 but not by H4: D2Mit50 at 95.5 cM and a region from D2Mit172 to D2Jau1 at 98.5−100 cM. Including the haplotypes of the inbred M. spretus strains in this analysis and looking for an interval shared by haplotypes H1, H2 and H3 and SPRET/Ei (strong linkage in NSJ cross) but not by either H4 or SEG/Pas (no linkage in NSE cross), only part of the second region qualified, refining the locus to approximately 1 cM (with the causal polymorphism predicted to map between D2Mit229 and D2Jau1).
To determine the precise location of the lod score peak, we combined data from the NSJ backcross with data from NSP mice carrying haplotype H1, H2 or H3 and calculated the composite lod scores. The 1-cM region between D2Mit229 and D2Jau1 identified by haplotyping seemed to match the part of distal chromosome 2 with the highest lod score (6.1 at 99−100 cM; Fig. 1b). This coincidence of haplotyping and linkage analysis data strongly indicated the presence of a susceptibility/resistance allele within this interval. The most telomeric end of chromosome 2 was excluded in an independent study with mice that were generated by backcrossing for four generations to NIH/Ola. Papilloma numbers in mice homozygous with respect to NIH/Ola alleles on proximal chromosome 2 but heterozygous with respect to the most distal part (from D2Mit74 at 107 cM to the telomere) were similar to those in mice that were homozygous NIH/Ola for the entire length of chromosome 2 (J.P.d.K. and J.-H.M., unpublished results).
A physical map of Skts13
Skts13 lies in a region of distal mouse chromosome 2 that is orthologous to a locus on human chromosome 20q13.2 showing frequent amplification in a variety of tumor types including breast, colon and ovarian cancer9,
10,
11,
12. Increased 20q13.2 copy number is observed in approximately 18% of primary breast tumors and 40% of breast cancer cell lines and is associated with aggressive tumor behavior, poor prognosis, cellular immortalization and genomic instability. The amplicon on 20q13.2 consists of at least three smaller subamplicons, each containing a putative driver gene for amplification, ZNF217 (encoding a putative Kruppel-like transcription factor), CYP24A1 (encoding vitamin D 24 hydroxylase) and STK15 (encoding a centrosome-associated kinase; refs. 13,
14,
15,
16,
17).
To identify the genes located in the region of the mouse genome defined by haplotyping, we constructed a physical map of the distal part of chromosome 2 surrounding Stk6 using the Celera database (Fig. 2). The minimum amplicon on human 20q13.2 (refs. 13,
14,
15,
16,
17) corresponds exactly to the 1-cM interval at 99−100 cM that we identified by haplotyping. These results indicate that the strategy we used identified a small interval whose human counterpart is amplified in tumors. Based on the location of the lod score peak (Fig. 1b) and the high frequency of 20q13.2 amplification in tumors, we decided to focus on the region at 99−100 cM, orthologous to the 20q13.2 amplicon. The possibility that the interval around D2Mit50 contains an additional susceptibility gene(s) has not yet been investigated.
Stk6 is a candidate mouse tumor susceptibility gene
We focused on the region containing Znf217, Cyp24a1 and Stk6 to look for polymorphisms that might influence protein function or expression. Preliminary gene expression analysis using a large panel of mouse skin tumor cell lines showed that although Znf217 and Stk6 were expressed at varying levels. Cyp24a1 was not detectably expressed at any stage of carcinogenesis (A.E.-T. and J.-H.M., unpublished results), and therefore this gene was not investigated further. To screen for important single-nucleotide polymorphisms (SNPs) in Znf217 and Stk6, we determined the coding sequences of these genes in NIH/Ola and SPRET/Ei mice. Polymorphisms in SPRET/Ei were considered potentially important if they were also present in haplotypes H1, H2 and H3 but absent in haplotype H4 and SEG/Pas mice. In Znf217, we identified 39 SNPs of which none fit the genetic screening criteria (A.E.-T. and J.P.d.K., unpublished results). In Stk6, we identified eight SNPs that all seemed to be silent (A.E.-T. and J.P.d.K., unpublished results). No obvious coding sequence changes in either of these genes could therefore be responsible for the linkage to tumor susceptibility in this region.
Next, we investigated possible differences in expression of Znf217 and Stk6 between NIH/Ola and SPRET/Ei mice. We determined allele-specific expression patterns of these genes in skin from normal (NIH/Ola SPRET/Ei) F1 mice and from (NIH/Ola SPRET/Ei) F1 mice treated with 12-O-tetradecanoyl-phorbol-13-acetate (TPA) by RT−PCR and subsequent restriction fragment length polymorphism (RFLP) analysis. Although no differences in expression of Znf217 could be detected between M. musculus and M. spretus alleles, Stk6 was expressed at significantly higher levels from the M. musculus allele relative to the M. spretus allele (J.-H.M. and A.E.-T., unpublished results).
To confirm and quantify these results, we designed TaqMan probes specific for intra-exon polymorphisms that distinguish between M. musculus and M. spretus alleles of Stk6. Repeated TaqMan analyses of cDNA derived from RNA from normal and TPA-treated mouse skin showed a small but significant and reproducible difference in expression of approximately 0.8 CT in favor of the M. musculus Stk6 allele, independent of TPA treatment (Fig. 3a). As a control, similar experiments using genomic DNA from the same F1 hybrid mice showed a 1:1 ratio by TaqMan analysis (Fig. 3a). We then examined the expression patterns in several M. musculus/M. spretus hybrid cell lines derived from skin tumors. Expression of the M. musculus allele of Stk6 was 5−6 times higher than that of the M. spretus allele in F1 hybrid cell lines B9 and A5, and this overexpression was associated with a similar level of amplification of the gene (Fig. 3a). Of 19 cell lines that were available from F1 mice, 8 significantly overexpressed the M. musculus Stk6 allele ( CT ranging from 1.9 to 15; Fig. 3b), and in 5 cases, overexpression was associated with specific amplification of the M. musculus allele (Fig. 3a,c). No cases were detected of amplification or overexpression of the M. spretus allele. This result is statistically highly significant (P = 0.0039) and lends strong support to the conclusions from the linkage data that a germline difference must exist between the M. spretus and M. musculus alleles of Stk6.
| | |
In an attempt to find the causal polymorphism, we sequenced approximately 5,000 bp of the Stk6 upstream region containing the promoter from NIH/Ola, SPRET/Ei, one of the original outbred M. spretus mice and mice representing each of the haplotypes H1, H2 and H4. We did not identify any polymorphism that could explain the linkage and haplotype data. Identification of this polymorphism might require more extensive sequencing as control elements can be located at a considerable distance from the transcriptional start site or within introns or 3' regulatory regions; further studies are in progress to address this question. We can, however, conclude that there is an allele-specific difference in the regulation of Stk6 that is in agreement with a potential role as a tumor modifier gene. Further studies will be necessary to identify the causal polymorphism and to clarify the exact mechanism of transcriptional regulation.
Human candidate susceptibility genes on 20q13.2
Evidence for the presence of genetic variants that influence human tumor susceptibility is normally obtained from linkage analysis of large family pedigrees or alternatively by association studies using DNA samples from individuals with cancer and population-matched controls18. A susceptibility gene in the 20q13 region has been implicated in prostate cancer19, but no strong candidate gene has yet been identified. We initially carried out association studies using DNA samples from individuals of Northern European ancestry with cancer and matched controls using a series of polymorphisms that were identified in STK15, ZNF217 and CYP24A1. These studies suggested a role for only one SNP, a T A change at position 91 in the STK15 coding sequence that alters the corresponding amino acid at position 31 from phenylalanine to isoleucine (Phe31Ile; A.E.-T., B.A.J.P., J.Chan, J. Ma and A.B., unpublished results). But the results in this population did not reach the significance levels required for acceptance as a true susceptibility variant20, and further analyses of additional samples are in progress and will be published elsewhere. Nevertheless, the mouse studies described above, together with suggestive data from association analysis and the functional evidence for a biochemical difference between the proteins encoded by the two alternative STK15 alleles (see below), prompted us to probe more deeply into the role of the variant STK15 forms in tumor development.
Gene amplification, tumor aneuploidy and STK15 Ile31
STK15 was originally identified as a frequently amplified gene in human colon cancers, but it is also amplified or overexpressed in other tumor types16,
17. Based on our observations that Stk6 is a strong candidate gene located in a mouse tumor susceptibility QTL and shows evidence of allele-specific changes in mouse tumors, we investigated the possibility that allele-specific amplification of the 91A allele of STK15 (encoding the Ile31 variant) might also occur in human tumors. We genotyped DNA samples from 162 individuals for whom both normal colon mucosa and colon tumor DNA was available. Of the 162 normal samples typed, 48 were heterozygous with respect to the 91TA polymorphism, and we analyzed these 48 samples for allele-specific amplification of the 91TA polymorphism of STK15 by TaqMan analysis (Fig. 4 and A.E.-T., unpublished results). Twenty-one samples showed no allelic imbalance between the two alleles, 4 samples showed gain of the 91T allele (encoding Phe31) and 19 samples showed gain of the 91A allele (encoding Ile31). Four samples showed differences of CTs of less than 0.6 and were scored as uncertain. These results show statistically significant allele-specific amplification of the 91A allele (P = 0.018,  2 test), providing additional evidence for the role of this allele in human cancer.
| | |
The possibility that the results are explained by another gene in LD with STK15 is highly unlikely, as LD in humans usually extends no more than 100 kb (ref. 21) and other candidate genes, such as ZNF217 and CYP24A1, are located much further away than 100 kb (Fig. 2). From LD analysis at the STK15 locus, we specifically showed that the SNPs identified in both ZNF217 and CYP24A1 are not in LD with the STK15 91TA SNP (D' < 0.1). Even CSTF1, which is located very close to STK15 and shares potential regulatory sequences at the 5' end of both genes, did not have any common SNPs that are in LD with the STK15 91TA SNP (A.E.-T., unpublished results). Statistically significant allele-specific amplification of the 91A allele is therefore strong evidence that the Ile31 variant is important in human cancer. The data highlight the potential usefulness of allele-specific copy number or expression changes in tumors as an alternative to linkage studies in both mouse and human systems for the detection of tumor susceptibility genes.
The main role of STK15 in tumor development is in controlling chromosome segregation during mitosis22,
23. Misregulation of this process results in aneuploidy, a distinctive feature of most cancers, which can be induced by overexpression of STK15 in cell lines16. We hypothesized that tumors from individuals carrying the 91A allele might show more evidence of aneuploidy than those from individuals who are homozygous with respect to the common 91T allele. Comparative genomic hybridization on microarrays (CGH arrays) is a powerful tool for the analysis of gene copy number changes in tumors14. We therefore carried out genome-wide CGH array analysis to investigate the degree of aneuploidy in tumors as a function of genotype.
We obtained high quality CGH profiles for 53 tumors (Fig. 5). Frequent gene copy number gains and losses occurred on multiple human chromosomes, in agreement with previous studies of genetic instability and aneuploidy in human colon cancers11,
24. Stratification of these results according to genotype at the STK15 91TA SNP clearly showed that tumors from individuals carrying a single 91A allele (STK1591A/91T heterozygotes, Fig. 5c) were significantly more aneuploid than tumors from individuals with the genotype STK1591T/91T (Fig. 5b). Copy number gains on chromosomes 7p and 20p in tumors from STK1591A/91T individuals were highly significantly different from those in tumors from STK1591T/91T individuals (P < 0.01), and gains of other chromosome arms (1q, 2p, 2q, 3q, 8q and 15q) were significantly different (P < 0.05). Results on the group of tumors from individuals homozygous with respect to the high-risk 91A allele (STK1591A/91A) showed a similar trend but did not reach statistical significance because of the small number of individuals (n = 7) in this category (J.Y. and A.E.-T., unpublished results). Notably, in this particular data set, no significant correlation was observed between host STK15 genotype and gene copy number losses (J.Y. and A.E.-T., unpublished results), suggesting that STK15 genotype affects whole chromosome arm gains rather than loss of potential tumor suppressor genes by deletion.
| | |
As an additional control, we grouped the individual CGH profiles according to genotype at the neighboring gene CSTF1. We examined a GA polymorphism in the 3' untranslated region of CSTF1 because it is very close to the 91TA SNP in STK15 and has a similar frequency (16% and 21%, respectively) in the population but is not in LD with this SNP. The highly significant changes on 7p and 20p that we observed between tumors from STK1591A/91T and STK1591T/91T individuals were not seen after stratification using the CSTF1 SNP (A.E.-T., unpublished results). With the exception of 8q, which showed a slight trend towards higher copy number in tumors from individuals homozygous with respect to the common allele of CSTF1, none of the chromosome arms showed any significant difference associated with CSTF1 genotype. We conclude that individuals with even one copy of the STK15 91A allele develop tumors that have on average a higher degree of aneuploidy than those from STK1591T/91T individuals.
STK15 Ile31 induces cell growth and tumorigenicity
To test the functional significance of the 91TA polymorphism in STK15, we carried out a series of cell growth assays using 293 cells and rat1 cells expressing either Phe31 or Ile31 isoforms of STK15. We selected stable transformants that showed equal levels of expression of the exogenous proteins (Fig. 6b) and tested them for growth in cell culture (Fig. 6a,c) or as xenografts in nude mice (Fig. 6d). In both 293 and rat1 cells, the high-risk STK15 Ile31 isoform induced more rapid cell growth than the low-risk STK15 Phe31 form. In addition, rat1 cells expressing the Ile31 isoform gave rise to actively growing tumors in nude mice by 28 d after injection, whereas controls expressing the Phe31 form produced only very small lesions. Thus, the STK15 Ile31 variant has more strongly transforming properties than the more common Phe31 variant.
Interaction of STK15 Phe31 with UBE2N in vitro and in vivo
STK15 has two functional domains, an N-terminal region containing the Aurora box, thought to mediate protein-protein interactions25, and the C-terminal catalytic domain. To identify any biochemical differences between the two isoforms of STK15, we carried out a yeast two-hybrid screen using as bait the sequences of amino acids 1−117 from each of the STK15 isoforms. One of these inserts contained UBE2N, the human homolog of yeast Ubc13, a ubiquitin-conjugating enzyme. We produced recombinant STK15 Phe31 and Ile31 isoforms tagged with histidine (His) and UBE2N for use in pull-down assays. We immobilized His-STK15 (1−117) on agarose beads and incubated it with recombinant in vitro translated UBE2N. UBE2N bound to the Phe31 isoform of STK15 (Fig. 7a). To confirm the UBE2N-STK15 interaction, we prepared cell lysates from 293 cells expressing the full-length STK15 Phe31 or Ile31 tagged with hemagglutinin (HA). These cells lines, constructed with the ecdysone inducible system, were induced with Ponasterone A resulting in lysates containing exogenous HA-tagged STK15. We immobilized His-UBE2N on His-agarose beads and incubated it with 1 mg of cell lysates. STK15 Phe31 bound to His-UBE2N to a greater extent than STK15 Ile31, although the lysates contained equal levels of STK15 protein (Fig. 7b).
| | |
To address whether the interaction between UBE2N and STK15 Phe31 takes place under physiological conditions, we carried out a series of coimmunoprecipitation experiments. We chose the HCT116 colon cancer cell line (heterozygous with respect to the 91TA polymorphism). We transiently transfected the HCT116 cells with both STK15 isoforms tagged with a Myc epitope and immunoprecipitated the endogenous UBE2N. STK15 Phe31 was also found in the precipitates (Fig. 7c). Apart from the expected STK15 band of 46 kDa, we also detected a high-molecular-weight smear that may represent ubiquitinated forms of STK15 (Fig. 7c). In agreement with this interpretation, we have shown in an in vitro assay that STK15 can be ubiquitinated in the presence of UBE2N, E1 ligase and ubiquitin (P.B. and S.L., unpublished results). Further studies are in progress to assess the effects of this modification on stability and activation state of STK15.
We carried out additional STK15 immunoprecipitations using two lymphoblastoid cell lines, 10859A (homozygous with respect to 91T) and 07038D (homozygous with respect to 91A), which were characterized genetically (A.E.-T., unpublished results), to investigate the binding levels of endogenous UBE2N. These cell lines provide an ideal environment to study the endogenous binding partners of the two STK15 isoforms, as each line is homozygous with respect to one of the STK15 alleles. As expected, UBE2N was detected at higher levels in immunoprecipitates from STK1591T/91T homozygous cells (Fig. 7d). Consistent with the data collected from the yeast two-hybrid and in vitro experiments, the Phe31 variant was required for the interaction; although STK15 Ile31 was expressed at a level similar to that of STK15 Phe31, it did not form a strong complex with UBE2N. We conclude that the interaction between UBE2N and STK15 Phe31 occurs with endogenous as well as with ectopically expressed proteins.
Ubc13, the yeast homolog of human UBE2N, has been reported to be mainly cytoplasmic in a complex with Mms2 but can translocate to the nucleus in response to DNA damage26. On the other hand, STK15 is located at the centrosomes and shows distinct cell cycle−regulated expression. To analyze possible colocalization of the two interacting proteins, we transfected the 293 inducible cell lines expressing the HA-tagged STK15 isoforms Phe31 or Ile31 with a Myc-tagged UBE2N construct. After double staining of the cells with the appropriate antibodies, we observed centrosome localization of STK15 Phe31 and UBE2N (Fig. 7e). STK15 Ile31 showed considerably less colocalization with UBE2N, and most cells expressing this isoform showed only the cytoplasmic localization (Fig. 7f), in agreement with earlier observations26. These data suggest that preferential binding of the Phe31 isoform to UBE2N results in selective concentration of the Phe31 protein in centrosomes, whereas UBE2N remains predominantly cytoplasmic in the presence of the Ile31 variant.
Discussion
Identifying the multiple combinations of common genetic variants that modify individual predisposition to sporadic cancers is widely considered the next challenge in determining the genetic basis of cancer susceptibility1,
2. The approach we describe uses the power of mouse genetics to map low-penetrance loci by linkage analysis and haplotyping. The use of outbred M. spretus in crosses with inbred M. musculus allowed us to carry out both analyses on the same experimental data set, obviating the necessity to refine the region containing the modifier locus by traditional methods, such as generation of congenic mice. Using this combined approach, we were able to refine the localization of a tumor susceptibility QTL to a region of 1−2 Mb on distal chromosome 2. Improving resolution further to the level of individual genes, however, required a parallel study of the orthologous locus in humans, to benefit from the high recombination history and low LD between individual human samples. Another important component of this analysis is the use of mouse and human tumor DNA samples to identify allele-specific genomic alterations, markers of the presence of germline variants that influence the course of tumor development. In several mouse tumor cell lines, increased expression of the M. musculus allele of Stk6 was associated with amplification of the allele, but this amplification event frequently involved fairly large regions of the chromosome, hampering identification of the gene(s) responsible for driving the process. In human tumors, the amplification event also frequently covered large regions, but the preferential amplification of the high risk STK1591A allele in tumors from unrelated heterozygotes provided strong evidence that this gene, rather than one of its near neighbors, is the main driving force for the amplification event. Even more convincingly, stratification of tumors according to genotype at the 91TA SNP in STK15 indicated that individuals with even one 91A allele develop more highly aneuploid tumors, in agreement with the known role of STK15 in tumor progression.
This combined strategy allowed us to identify a common, low-penetrance tumor-susceptibility gene in both systems, although the mechanisms by which the genetic variants modify risk differ between mouse and human. In humans, a coding sequence polymorphism in STK15 does not alter the intrinsic kinase activity of the protein (P.B. and S.L., unpublished results) but modifies interactions with binding partners in the cell, such as UBE2N, that change its function. In mice, the Stk6 amino acid at the equivalent of position 31 in humans is the same as that encoded by the high-risk human allele, a notable observation in view of the known plasticity of the mouse genome and the ease with which mouse cells become aneuploid. The polymorphism responsible for the differential activity of Stk6 between the parental mouse strains used in these studies is unknown, but it seems to act at the level of transcription or mRNA stability, as differences were found in the expression patterns of the parental M. musculus and M. spretus alleles in normal cells and in tumors. The constitutive expression of what is effectively the high-risk human allele in normal mice suggests that mouse models will be important for testing drugs aimed at inhibition of STK15 activity for human cancer therapy. The recent development of BAC transgenic mice27 offers the possibility of testing the effect of expression of the variant human STK15 alleles on tumor development in the mouse.
STK15 is a member of the Aurora/Ipl1p family of mitotically regulated serine/threonine kinases that are key regulators of chromosome segregation and cytokinesis22. A wealth of functional data exists showing that overexpression of STK15 leads to centrosome amplification, chromosomal instability and transformation15,
16,
17,
28. Immunohistochemical analyses showed overexpression of STK15 in 94% of invasive ductal adenocarcinomas of the breast, which is notable because genetic instability is an early event in the development of ductal breast carcinoma29. All these findings support our hypothesis that elevated expression of STK15 modifies cancer risk, perhaps by leading to aneuploidy. During mouse skin carcinogenesis, initiation by mutation of the gene Hras1 is followed by trisomy of chromosomes 7 and 6 (refs. 30,31), an event that could be facilitated by increased expression of Stk6.
The consequences of overexpressing the two allelic variants in cell lines support the identification of the STK15 91A allele as a high-risk allele: both cell growth in vitro and tumorigenicity in nude mice are enhanced by this variant but less so by the alternative 91T variant. We also show that the weak Phe31 variant preferentially binds to UBE2N in human cells and is colocalized with UBE2N in the centrosomes. UBE2N belongs to a family of ubiquitin-conjugating E2 enzymes, with over 17 members characterized to date. E2 ligases have not been reported, apart from a few exceptions including Ubc2 (Rad6) and Ubc9, to bind to the ubiquitin substrate directly32. The direct binding between UBE2N and STK15 could suggest that the kinase is a direct substrate for ubiquitination by UBE2N, and further studies are in progress to address this question, as well as to identify other components of the UBE2N- STK15 complex. The outcome of this interaction, which may lead either to activation of STK15 (and subsequent cell death) or to inactivation and degradation, is also unclear. Nevertheless, this interaction correlates with attenuation of the ability of STK15 to induce cell growth and transformation. The mechanisms of cell transformation by this mitotic kinase are clearly complex and may involve kinase substrates that lie outside the centrosome, as shown by the observation that kinase-defective mutants of STK15 can induce centrosome amplification but not transformation33. Further elucidation of the UBE2N-STK15 complex might identify a pathway that regulates the essential functions of STK15 activity in cell cycle progression, providing additional targets for the development of new cancer therapies.
Our preliminary analyses of the importance of the STK15 91TA polymorphism in human cancer susceptibility by association studies in cancer cases and controls provided some suggestive evidence of greater risk in homozygotes but not in heterozygous individuals (A.E.-T., B.A.J.P., J. Chan, J. Ma and A.B., unpublished results). To reach formal statistical significance for an allele with a 6% homozygote frequency and low relative risk, several thousand cases and controls would be required. The cross-species strategy described here based on the mapping of loci in mouse models and allele-specific alterations in tumors, may lead more rapidly to the identification of candidate human low-penetrance susceptibility genes. The gene PTPRJ was also recently identified as a potential mouse and human tumor-susceptibility gene, based on combined data from mouse models and human tumor analysis34. The use of appropriate mouse models to guide the prioritization of candidate low-penetrance susceptibility loci for testing in human association studies could lead to significant advances in understanding the polygenic basis of cancer.
Methods
Mice and tumor induction.
We purchased inbred NIH/Ola mice from Harlan Olac and obtained outbred M. spretus and inbred SEG/Pas (derived from M. spretus) mice from S. Brown (Medical Research Council, Harwell, UK) and J.-L. Guenet (Institute Pasteur, Paris, France), respectively. We purchased inbred SPRET/Ei mice from The Jackson Laboratory. We carried out three backcrosses with different M. spretus strains: NSP, (NIH/Ola outbred M. spretus) NIH/Ola; NSE, (NIH/Ola inbred SEG/Pas) NIH/Ola; and NSJ, (NIH/Ola inbred SPRET/Ei) NIH/Ola. We used the same breeding and skin tumor induction protocols in all crosses and estimated papilloma susceptibility by counting the number of papillomas 20 weeks after initiation in 320 NSP, 163 NSJ and 106 NSE mice as reported previously8. All animal experiments were approved by the UCSF Committee on Animal Research.
Mouse genotyping, linkage analysis and haplotyping.
We prepared DNA samples from tails and carried out genotyping using microsatellite markers by standard methods. Marker positions given are based on the Mouse Genome Database. We used negative binomial regression analysis to identify QTLs that control skin tumor susceptibility, as reported previously8. For fine mapping of QTLs, we constructed haplotypes in outbred M. spretus mice for association studies by using the variation in length of microsatellites between the different NSP alleles.
SNP identification.
We used published and database sequences of CYP24A1, ZNF217 and STK15 to design primers for PCR amplification of these genes and employed an ABI 3700 for sequencing of PCR products. We analyzed the resulting sequences using Sequencher (Genes Codes Corporation) and confirmed any sequence discrepancies by reverse reads and RFLP analysis where possible.
Human genotyping.
At current estimates, there are 15 genes mapping to the region of the human genome that corresponds to the minimal locus defined by linkage analysis and haplotyping in the mouse35, 7 of which were mapped to the locus when this work was initiated (C20orf17, ZNF217, UBL2, BCAS1, CYP24A1, STK15 and MC3R).
We genotyped all SNPs using the ABI PRISM 7700 sequence detection system. PCR reactions contained genomic DNA (4−10 ng), 1 TaqMan universal PCR master mix, forward and reverse primers (900 nM), 200 nM VIC-labeled probe and 100−200 nM FAM-labeled probe. Amplification conditions on an MJ Tetrad thermal cycler (GRI) were as follows: 1 cycle of 50 °C for 2 min, followed by 1 cycle of 95 °C for 10 min and 40 cycles of 95 °C for 15 s and 60−64 °C for 1 min. Completed PCRs were read on an ABI PRISM 7700 Sequence Detector and analyzed using the Allelic Discrimination Sequence Detection Software (Applied Biosystems). We constructed double-stranded artificial template controls using a long forward primer specific for each SNP with a long common reverse primer that overlapped the forward primer. Primers were filled in using standard methods. We designed TaqMan primers and probes using the Primer Express Oligo Design Software v1.0 (Applied Biosystems). Probes were MGB probes designed specifically for TaqMan Allelic Discrimination (Applied Biosystems). Primers and probe sequences are available on request. All human genotyping studies were approved by the UCSF Committee on Human Research.
Allele-specific expression and allele-specific amplification assays.
We measured allele-specific expression and amplification using the ABI PRISM 7700 sequence detection system (Applied Biosystems). Amplification conditions and reactions were as described above, except that PCR reactions for allele-specific expression (50  l) contained 50 ng reverse-transcribed RNA or 50 ng genomic DNA. We carried out whole-genome amplification of normal colon mucosal DNA and colon tumor DNA before the TaqMan analysis. PCR was done in triplicate for each sample and experiments were repeated at least three times. CT values were normalized to the average normal genomic CT difference in each experiment. The CT value differences between the two probes for the triplicates were then averaged.
CGH array analysis.
We collected colon tumors for CGH array analysis from 53 individuals undergoing surgery at Newcastle University Hospitals. Tumor genomic DNA was extracted from frozen specimens containing at least 75% cancer cells. We labeled 400 ng of each tumor DNA sample with Cy3 by random priming (Bioprime random priming kit) to incorporate Cy3-dCTP and used normal male genomic DNA (400 ng) labeled with Cy5-dCTP as a reference probe for every experiment.
We carried out CGH array analysis using HumArray2.0 version of the previously published BAC array36. The array contains 2,464 BAC and P1 clones printed in triplicate and provides genome-wide coverage with clones at  1.4 Mb spacing. CGH hybridization was done as previously described37.
We collected 16-bit, 1024 1024 pixel DAPI, Cy3 and Cy5 images using a custom-built charge-coupled device camera system38 and used UCSF SPOT software to segment spots, do local background subtraction and calculate log2 ratios of the total integrated Cy3 and Cy5 intensities for each spot. A second program, SPORC39, was used to associate clone identities with a mapping file to plot the data relative to the position of the BACs on the August 2001 freeze of the draft human genome sequence.
Cell culture, transient transfections and tumorigenicity assay.
We grew 293 cells in Dulbecco's medium supplemented with 10% fetal bovine serum and glutamine in an atmosphere of 5% CO2. We constructed 293 inducible cell lines using the ecdysone system according to the manufacturer's instructions (Invitrogen), containing either Phe31 or Ile31 STK15 tagged with HA. We maintained these cells lines under zeocin (0.2 mg ml-1), selected them with geneticin (0.4 mg ml-1) and induced STK15 expression with Ponasterone A (1 g ml-1). We obtained the EBV-transformed human lymphoblastoid cell lines homozygous with respect to 91T (10859A) or 91A (07038D) from the NIGMS Human Genetic Cell Repository, Coriell Institute for Medical Research, and cultured them in RPMI 1640 medium with 15% fetal calf serum. For transient transfections, we used Lipofectamine Plus reagent (Invitrogen) according to the manufacturer's instructions and collected cells 36 h after transfection.
Western blotting and immunoprecipitation.
We lysed cells in lysis buffer (50 mM Hepes pH 7.5, 250 mM NaCl, 0.5% Nonidet P40 with protease inhibitors), incubated the lysates on ice for 30 min and centrifuged them at 13,000g for 15 min. We collected the supernatant and estimated protein concentration with the Pierce BCA assay reagent. For western blotting, we separated samples on Novex Nupage Tris-Glycine gels (Invitrogen) followed by electrophoretic transfer on PVDF membrane (Millipore) and blocking in 5% nonfat milk. We carried out immunodetection using chemiluminescence detection (Amersham). For immunoprecipitation studies, we incubated lysates with the appropriate antibody in lysis buffer. After 2 h rotation at 4 °C, we added Protein A or Protein G (Santa Cruz) and incubated them overnight. We washed samples extensively in lysis buffer before resuspending them in sample buffer.
Immunofluorescence confocal microscopy.
We grew cells on 13-mm coverslips in 6-well dishes and, after appropriate induction or transfection, fixed them in 4% paraformaldehyde in phosphate-buffered saline for 30 min, permeabilized them with 0.1% Triton-X in phosphate-buffered saline for 10 min, rinsed them and incubated them with primary antibodies for 1 h, all at room temperature. We added secondary antibody after extensive washing (anti-mouse Alexa 568 or anti-rabbit Alexa 588) for 40 min and then washed the cells in phosphate-buffered saline and stained nuclei before mounting onto glass slides using Vectashield (Vector laboratories). Optical sections were taken using a Leica TCS-SP2 confocal imaging system.
Yeast two-hybrid.
Because the full length STK15 protein is toxic in yeast, we subcloned a fragment spanning amino acids 1−117 and encoding either Phe31 or Ile31 isoforms into the NcoI/EcoRI sites of the GBKT7 yeast cloning vectors. We transformed these constructs into the AH107 yeast strain and isolated the resulting clones on Trp-selection plates. We assessed protein expression of the truncated STK15 isoforms by western blotting and used a clone for each construct (Phe31 or Ile31) to carry out the screening using the mating method. Briefly, we purchased a pre-transformed library from Clontech containing fetal brain cDNA library in the Y187 yeast strain. After an overnight incubation of AH1071-117 (Phe31) and AH1071-117 (Ile31) in selection medium, we mated the cultures for 24 h with an aliquot of the pre-transformed library and plated them on 50 150 mm dishes selective for Trp-/Leu-/His-. After 5−7 d of incubation, we streaked positive clones on high stringency selection plates (Trp-/Leu-/His-/Ade-/X-gal). From each separate screening, the positive colony yield was 22 independent clones for Phe31 and 19 for Ile31. The mating efficiencies of each screen were comparable. Positive clones were reconfirmed by retransformation of the yeast, and unique clones for each screen were cross-transformed among the STK15 isoforms. The functionality of the system was proven by the p53/T7 interaction controls provided by Clontech.
Plasmid constructs and protein expression.
We subcloned full-length STK15 (91T or 91A) into NcoI sites of the pGEX-KG cloning vector (Pharmacia) and transformed the clones into the BL21 Escherichia coli bacterial strain. We grew the bacteria to an optical density of 0.7−1 at 37 °C, induced them with 1 mM of isopropylthiogalactoside and incubated them for 5 h more at 25 °C. We washed pellets in phosphate-buffered saline, stored them at -80 °C and purified the protein on GSH-Sepharose beads. We amplified UBE2N cDNA by PCR from an IMAGE clone (number 277925) and subcloned it into BamHI/EcoRI sites of the pRSETB vector (Invitrogen). We then transformed the plasmid into BL21 bacteria and purified the protein using nickel-agarose beads (Qiagen). We subcloned STK15 (91T or 91A) and UBE2N cDNA into BamHI/EcoRI sites of the pcDNA3.1/Myc-His vector (Invitrogen).
Antibodies.
We obtained rabbit polyclonal antibodies to HA, His, GST and Myc from Santa Cruz, mouse monoclonal antibody to IAK1 from Transduction Laboratories and rabbit polyclonal antibody to UBE2N from Imgenex.
URLs.
The following web addresses were used during the course of these studies: Celera database (http://www.celera.com), Ensembl human genome project (http://www.ensembl.org) and Mouse Genome Database (http://www.informatics.jax.org).
Received 19 February 2003; Accepted 25 June 2003; Published online: 27 July 2003.
REFERENCES
- Balmain, A., Ponder, B.A.J. & Gray, J. The genetics and genomics of cancer. Nat. Genet. 33, 238–244 (2003). | Article | PubMed | ISI | ChemPort |
- Ponder, B.A. Cancer genetics. Nature 411, 336–341 (2001). | Article | PubMed | ISI | ChemPort |
- Fijneman, R.J., de Vries, S.S., Jansen, R.C. & Demant, P. Complex interactions of new quantitative trait loci, Sluc1, Sluc2, Sluc3, and Sluc4, that influence the susceptibility to lung cancer in the mouse. Nat. Genet. 14, 465–467 (1996). | Article | PubMed | ISI | ChemPort |
- Balmain, A. & Nagase, H. Cancer resistance genes in mice: models for the study of tumour modifiers. Trends Genet. 14, 139–144 (1998). | Article | PubMed | ISI | ChemPort |
- Dietrich, W.F. et al. Genetic identification of Mom-1, a major modifier locus affecting Min-induced intestinal neoplasia in the mouse. Cell 75, 631–639 (1993). | Article | PubMed | ISI | ChemPort |
- Nagase, H. et al. Distinct genetic loci control development of benign and malignant skin tumours in mice. Nat. Genet. 10, 424–429 (1995). | Article | PubMed | ISI | ChemPort |
- Cormier, R.T. et al. The Mom1AKR intestinal tumor resistance region consists of Pla2g2a and a locus distal to D4Mit64. Oncogene 19, 3182–3192 (2000). | Article | PubMed | ChemPort |
- Nagase, H., Mao, J.H. & Balmain, A. A subset of skin tumor modifier loci determines survival time of tumor-bearing mice. Proc. Natl. Acad. Sci. USA 96, 15032–15037 (1999). | Article | PubMed | ChemPort |
- Iwabuchi, H. et al. Genetic analysis of benign, low-grade, and high-grade ovarian tumors. Cancer Res. 55, 6172–6180 (1995). | PubMed | ISI | ChemPort |
- Kallioniemi, A. et al. Detection and mapping of amplified DNA sequences in breast cancer by comparative genomic hybridization. Proc. Natl. Acad. Sci. USA 91, 2156–2160 (1994). | PubMed | ChemPort |
- Korn, W.M. et al. Chromosome arm 20q gains and other genomic alterations in colorectal cancer metastatic to liver, as analyzed by comparative genomic hybridization and fluorescence in situ hybridization. Genes Chromosomes Cancer 25, 82–90 (1999). |
|
