Genetic Variation in the Chromosome 17q23 Amplicon and Breast Cancer Risk

Molecular cytogenetic studies have revealed common regions of DNA amplification in breast tumors (1). These regions contain oncogenes such as Erb2 and c-Myc that are frequently overexpressed as a result of the amplification (2) and contribute to tumor progression. Recently, common genetic variation in oncogenes such as FGFR2 (3) and AURKA (4), which are amplified and overexpressed in primary breast tumors (5-7), has been associated with an increased risk of breast cancer. This suggests that genes contributing to breast tumor progression as a result of amplification and overexpression may also influence breast cancer susceptibility through inherited genetic variation.

The chromosome 17q23 region is gained or amplified in >40% of breast tumors (8-10). We (11) and others (12) characterized the 4-Mb amplicon on chromosome 17q23 and identified several independent peaks of amplification-containing genes such as T-box transcription factor (TBX2; ref. 13), Peanut-like 2 (SEPT4; ref. 14), and protein phosphatase 1E (PPM1E; ref. 15).

In view of the possible link between genetic variation and amplification of oncogenes in breast cancer (3), we hypothesized that common genetic variation in selected amplified genes from the 17q23 amplicon (11), TUBD1, SEPT4, PRKCA, TBX2, TBX4, TEX14, TLK2, and YPEL2, which are overexpressed in breast tumors, is associated with breast cancer risk. Interestingly, these proteins are thought to influence the process of cell division and might be associated with cancer through induction of aneuploidy and/or polyploidy (14, 16-21) much like AURKA (22). In addition, because of a recent report suggesting an association between expression of mitotic protein kinases and luminal breast cancer (23), we examined associations between genetic variation in the kinase-encoding genes, PRKCA and TLK2, and estrogen receptor–positive (ER⁺) and progesterone receptor–positive (PR⁺) breast cancer.

The Mayo Clinic Breast Cancer study is an ongoing clinic-based case-control study initiated in February 2001 at Mayo Clinic (Rochester, MN). The study design has been presented previously (3). Consecutive cases were women ages ≥18 y, without a prior history of cancer, diagnosed with histologically confirmed primary invasive breast cancer within 6 mo of recruitment. Women with no history of cancer were frequency matched on age (5-y age category) and region of residence to cases. Controls were recruited from those seeking general medical examinations at the outpatient practices at Mayo Clinic. The 798 cases and 843 controls in this study were Caucasian women (99% of study participants) who completed a self-administered risk factor questionnaire and provided blood samples for genomic DNA extraction.

Tag single nucleotide polymorphisms (SNP) and all putative functional SNPs for TUBD1, SEPT4, TBX2, TBX4, TEX14, TLK2, and YPEL2 were identified according to previously published criteria (24). Two functional SNPs were included from PRKCA in place of the large number of tagSNPs in this gene. SNPs were genotyped with the GoldenGate Assay (Illumina, Inc.; refs. 25-27). Only samples and SNPs with call rates >95% were included in analyses. Concordance between 100 duplicate samples was >99.99%. A single SNP (rs16943326) from the PPM1E gene in the 17q23 amplicon that displayed a significant inverse association with postmenopausal breast cancer in stage I of the Cancer Genetic Markers of Susceptibility (CGEMS) genome-wide association study⁶

The Studies of Epidemiology and Risk Factors in Cancer Heredity (SEARCH), an ongoing population-based study of invasive breast cancer cases of European ancestry ascertained through the Eastern Cancer Registration and Information Centre in England (29), was used to further evaluate the association between TLK2 rs2245092, the SNP most significantly associated with risk in the Mayo Clinic sample, and breast cancer. Successful genotyping was achieved for 97.6% of 4,470 cases and 4,560 controls with the Taqman assay.

Individual SNP associations for breast cancer risk were assessed using unconditional logistic regression to estimate odds ratios (OR) and 95% confidence intervals (95% CI) using ordinal (log additive) and codominant genetic models. Analyses were adjusted for age and region of residence in the Mayo Clinic sample. Observed P values in the Mayo Clinic sample were compared with empirical P value distributions using Monte Carlo simulation methods (30) that permuted case-control status using 2,000 iterations. The “corrected” P value for the SNP with the smallest observed P value was then calculated by comparing its value with the corresponding empirical distribution of permuted P values. We compared ORs and 95% CIs observed in the Mayo Clinic sample with age-adjusted estimates in the SEARCH sample for TLK2 rs2245092 and with estimates adjusted for age, race, and postmenopausal hormone use among postmenopausal women in the CGEMS sample for PPM1E rs16943326.

In secondary analyses, interactions between pairwise combinations of the SNPs with statistically significant individual associations were evaluated by including into statistical models a product term of the coefficients for the log-additive genetic models of both SNPs and assessing significance with a likelihood ratio test. We also subset cases defined by ER⁺ and PR⁺ tumor status and compared SNP associations in the kinase-encoding genes, PRKCA and TLK2 (23), between cases and all controls using unconditional logistic regression.

A total of 60 SNPs (Supplementary Table S1) were examined in Mayo Clinic cases and controls. Covariates that differed in distribution by case-control status (Table 1) did not affect risk estimates when evaluated in statistical models.

Eight SNPs in four genes (PRKCA, TBX4, TLK2, and YPEL2) showed significant dose-responsive (P_trend < 0.05) associations with breast cancer risk in the Mayo Clinic sample (Table 2). Decreased risks were associated with each copy of the minor allele for PRKCA rs7342847 (P_trend = 0.02), TBX4 rs11867179 (P_trend = 0.04), and TLK2 rs733025, rs12451705, rs6504112, and rs2245092 (P_trend ≤ 0.02), whereas increased risk was associated with each copy of the minor allele for TBX4 rs3744448 (P_trend = 0.03) and YPEL2 rs16943468 (P_trend = 0.03). Two other SNPs in SEPT4 and YPEL2 displayed associations with breast cancer risk in codominant genetic models (Supplementary Table S2). Overall, the number of observed significant associations (n = 10) was greater than the number expected by random variability (n = 3). When compared with the permuted P value distribution, none of the associations with breast cancer was statistically significant at P < 0.05. SNPs with observed nonsignificant associations with breast cancer are listed in Supplementary Table S3.

Given the similarity in ORs and the correlation between the four TLK2 SNPs (pairwise r² = 0.57-0.81; Table 2), we used backward elimination regression to evaluate the independent effects of TLK2 SNPs with breast cancer. Only TLK2 rs2245092 was retained by the model at P = 0.01, suggesting that rs2245092 or a variant in strong linkage disequilibrium accounted for these associations with breast cancer. We subsequently evaluated TLK2 rs2245092 for an association with breast cancer in the SEARCH study. The association was not confirmed: each allele was associated with an OR of 1.0 (95% CI, 0.9-1.1; P_trend = 0.98).

In exploratory analyses driven by the hypothesis that coamplified genes may cooperate to promote tumor development, significant interactions were observed between SEPT4 rs758377 and TEX14 rs302864 (P_interaction = 0.0003) and between TLK2 rs733025 and YPEL2 rs16943468 (P_interaction = 0.05) under a log-additive genetic model.

We also evaluated associations between SNPs and hormone receptor–positive breast cancer in the Mayo Clinic sample. ER⁺/PR⁺ tumors were inversely related to PRKCA rs7342847 (OR, 0.7; 95% CI, 0.6-0.9; P_trend = 0.002) and to TLK2 rs22450925 and rs2245092 (both: OR, 0.8; 95% CI, 0.7-1.0; P_trend = 0.03) under log-additive genetic models (Supplementary Table S4). None of the SNPs in either gene was associated with hormone receptor–negative breast cancer (data not shown). Data from 2,018 SEARCH cases with ER⁺ tumor status (PR⁺ status was unavailable) showed no association with TLK2 rs2245092 (OR, 1.0; 95% CI, 0.9-1.1; P_trend = 0.98).

Finally, the CGEMS study reported associations with postmenopausal breast cancer risk among women with one (OR, 0.96; 376 cases) or two (OR, 0.37; 2-degree-of-freedom P = 0.00004; 29 cases) copies of the minor allele of PPM1E rs16943326 (minor allele frequency = 0.23 among CGEMS controls). A nonsignificant trend toward increased risk of postmenopausal breast cancer was observed for this SNP among homozygote minor allele carriers (22 cases) in the Mayo Clinic sample (Supplementary Table S3). Significant associations of other SNPs in the Mayo Clinic sample by menopausal status are shown in Supplementary Table S5.

Common genetic variation in amplified oncogenes such as FGFR2 has been associated recently with risk of breast cancer, suggesting that certain genes can both predispose to (3) and contribute to progression of (6) breast cancer through independent mechanisms. We observed that breast cancer risk in the Mayo Clinic sample was associated with eight common polymorphisms in the selected candidate oncogenes PRKCA, TBX4, TLK2, and YPEL2 on chromosome 17q23, a region we previously identified as amplified frequently in breast tumors (11). Moreover, the SNP by SNP interactions observed between SEPT4 and TEX14, in particular, are intriguing given that both genes participate in the separation of the dividing cell at the end of mitosis (19, 20). This supports the notion that interdependent mechanisms involving genes from the 17q23 amplicon may contribute to breast cancer.

We also observed significant associations between PRKCA kinase rs7342847 and TLK2 kinase rs733025 and rs2245092 and hormone receptor–positive tumors in the Mayo Clinic study. This observation raises the possibility that these variants predispose to specific subtypes of breast cancer, as has been observed for rs2981582 in FGFR2, rs13281615 in 8q24, and rs3803662 in TNRC9, which have recently been verified as genetic risk factors for breast cancer (31). Our findings are also intriguing given the recent suggestion that certain mitotic kinases contribute to luminal breast cancer (23). A larger sample of ER⁻/PR⁻ tumors, however, is needed to confirm the absence of association observed in our study with these SNPs.

We were unable to confirm the association between TLK2 rs2245092 and breast cancer using the SEARCH sample. Possibly, one of the other TLK2 SNPs with which TLK2 rs2245092 was in strong linkage disequilibrium in the Mayo Clinic sample may have shown an association with breast cancer had it been genotyped in SEARCH. We were also unable to confirm the association reported in CGEMS between PPM1E rs16943326 and postmenopausal breast cancer using the Mayo Clinic data. In post hoc evaluation of the CGEMS data, SNPs in TUBD1, SEPT4, TBX2, and YPEL2 were represented in both CGEMS and the Mayo Clinic samples; however, only YPEL2 rs16943468 was significantly associated with postmenopausal breast cancer in both studies, and similar to PPM1E rs16943326, the ORs were in opposite directions. At a minor allele frequency of 0.06 (YPEL2 rs16943468) or 0.22 (PPM1E rs16943326), if the true OR among homozygotes is in the range of 1.2 to 1.5, then a much larger study than Mayo Clinic or CGEMS would be needed to have sufficient power to confirm the association at either the conventional P < 0.05 or genome-wide (P ≤ 10⁻⁷) level of significance (32). It also remains possible that the distribution of other exposures may explain the differences in genetic associations with CGEMS (e.g., 60% of Mayo Clinic cases did not use postmenopausal hormones versus <30% in the Nurses' Health Study; ref. 33), emphasizing the need for careful interpretation of results between study populations.

The observed significant P values for SNPs shown to be associated with breast cancer in the Mayo Clinic sample were not supported by distributions obtained using mathematical permutation methods. However, the biological plausibility of our hypothesis and the support from published literature of examples of oncogenes within amplicons that contribute to cancer at the level of commonly inherited variation (e.g., FGFR2; ref. 3) suggest that the findings are at least as plausible as the explanation of random variability (34). Replication in similar populations of adequate sample size, therefore, should be the standard that determines the significance of other potentially promising associations cited in this report.

It is plausible that a proportion of women at risk of breast cancer due to chromosome 17q23 polymorphisms may also have gene amplification in this chromosomal region or that other biological or clinical features may modify the genetic associations. Although tissues and clinical data were unavailable for evaluation in the present report, we are pursuing these hypotheses for future investigation.

In conclusion, this study may suggest a role for genetic variation within candidate oncogenes from the 17q23 amplicon in predisposition to breast cancer. Further studies will be critical to clarify the importance of these findings.

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.