Background

Transforming growth factor beta 1 (TGFB1) can act as a tumor suppressor by mediating growth arrest via the CDK inhibitors p15INK4B [1] and/or p21CIP1 [2, 34]. This study was approved by the IRB of the Brigham and Women's Hospital.

Results

We did not detect any deviation from Hardy-Weinberg equilibrium at either SNP (p = 0.59 for -509 and 0.49 for L10P in controls). Risk assessments using conditional logistic regression models were similar to unconditional analyses, therefore we will report only the unconditional analyses to increase power. Though no overall association was found between L10P and breast cancer risk, a marginally significant an inverse association between the -509 SNP and breast cancer risk was detected (Table 1). This association was limited to women diagnosed with estrogen receptor (ER) positive tumors (p-heterogeneity in risk between ER+ and ER- breast cancer = 0.002) (Table 2). Compared to controls and using the C/C genotype as a reference, women heterozygous at -509 had an 18% decrease in risk of ER+ breast cancer (OR 0.82, 95% CI 0.67 – 1.00), women homozygous for the T allele had a 38% decrease in risk (OR 0.62, 95% CI 0.42 – 0.90), and there was a highly significant trend in decreased risk across these two genotypes (p = 0.04 for L10P and p = 0.005 for -509). The association was similar among progesterone receptor (PR) positive tumors. No difference in risk was observed upon stratification by menopausal status at diagnosis, body mass index (BMI <30/30+), or postmenopausal hormone (PMH) use (ever/never) for either SNP.

Table 1 Association between TGFB1 SNPs and breast cancer risk in the Nurses' Health Study
Full size table
Table 2 Association between TGFB1 SNPs and breast cancer risk in ER+ breast cancer analyses in the Nurses' Health Study
Full size table

The genotype frequencies of the TGFBR1 alanine microsatellite (TGFBR1*6A) were similar in cases and controls (Table 3); since there were so few rare variants (<1%), these were removed from the analyses. Again, no deviation from Hardy-Weinberg equilibrium was detected among controls (p = 0.65). Following the classification of high, intermediate, and low signalers proposed by Kaklamani et al., the L10P polymorphism in TGFB1 and alleles of the TGFBR1*6A microsatellite were combined. No difference in risk was observed among the high, intermediate, and low signaling types (Table 4), and no difference in risk was observed upon stratifying the cases by estrogen receptor status of their tumors (data not shown). In order to clarify previous reports of association between this polymorphism and breast cancer risk, we have also performed a meta-analysis including our results with previously reported genoty** results of this polymorphism in breast cancer cases and controls. We have separated out all the participating populations (i.e. the two populations represented in ** et al. [35], genotypes attributed to Reiss in [31], in addition the genotypes attributed to Offitt in [31] were removed, as they were included by Kaklamani et al. [32]) in order to more clearly evaluate and display the data. The summary odds ratio was 1.10 (95% CI 0.89 – 1.38) from the random effects model (Fig. 1), however, there was evidence of significant heterogeneity in risk estimates (p-heterogeneity < 0.01).

Table 3 Association between the TGFBR1 alanine microsatellite and breast cancer risk in the Nurses' Health Study
Full size table
Table 4 Association between TGFB1 L10P/TGFBR1*6A hypothesized signaling levels and breast cancer risk in the Nurses' Health Study
Full size table
Figure 1
figure 1

Random-effects meta-analysis of TGFBR1*6A under a dominant model.

Full size image

Discussion

Although the TGFBR1*6A polymorphism has been associated with breast cancer risk in a meta-analysis (with a total of 1,420 cases and 3,451 controls), we did not see any evidence of a relationship between this microsatellite and breast cancer in our large nested case-control study. Our study has 80% power to detect a log-additive per allele odds ratio of 1.27 at the alpha = 0.05 level. We have added our results, and to prior genoty** reports [31, 32, 35–39]. This new meta-analysis consists of 3,459 breast cancer cases and 4,557 controls. While the summary odds ratio does not show a statistically significant change in risk associated with this polymorphism, there is statistically significant heterogeneity in the risk estimates. One possible explanation of this heterogeneity is variation in the specificity of genoty** methods used. However, the most likely explanation for this heterogeneity is random sampling variation, despite the fact that the studies presented are all composed of mostly Caucasian populations.

Prior studies have shown increases, decreases, or no change in breast cancer risk associated with the L10P polymorphism of TGFB1. Recently, the Breast Cancer Association Consortium pooled data from case-control studies examining this polymorphism, including 5,587 breast cancer cases and 6,863 controls, and found a very moderate per-allele increase in breast cancer risk associated with this polymorphism (OR 1.08, 95% CI 1.02 – 1.14) [40]. Our study is underpowered to detect such an association, although the association we observed between this SNP and breast cancer risk was in the opposite direction. If we combine our risk estimate for the L10P polymorphism with those of the Breast Cancer Association Consortium (BCAC), significant heterogeneity in risk estimates (p-heterogeneity = 0.03, random effects model) is observed, and the summary odds ratio would be 1.02, 95% CI 0.88 – 1.17). It is unlikely that population differences would explain the heterogeneity between our results and the BCAC, as in a recent genome wide association scan performed on a subset of the NHS breast cancer cases and controls >99% of the subjects did not have genetic contributions from populations other than Caucasian [41]. One possible explanation for this heterogeneity is that the BCAC is largely composed of prevalent cases, as compared to only incident cases in the NHS, and therefore case-specific variables which may effect the association between this polymorphism and breast cancer risk overall could have different distributions in the BCAC as compared to the NHS. More than likely however, this heterogeneity is due to sampling variation.

More interesting is our observation that TGFB1 polymorphisms are inversely associated with ER+ breast cancers. TGFB1 colocalizes with ERα in mouse mammary epithelial cells, and there is a higher proportion of ERα-positive proliferating mammary epithelial cells in mice with only one copy of the tgfB gene, which therefore have significantly lower TGFB1 protein levels [42, 43]. This is evidence that TGFB1 may prevent proliferation in ER-positive breast epithelial cells. The T allele of the -509 SNP in TGFB1 has been associated with higher levels of secreted TGFB1 [21]. Our association between this allele and decreased risk of estrogen receptor positive tumors is compatible with the hypothesis that increased TGFB1 levels decrease the potential for proliferation in ER+ breast cells. As these results are not our original hypotheses, our observation that TGFB1 polymorphisms are inversely associated with ER+ breast cancers should be considered hypothesis generating, and needs further replication.

Conclusion

In conclusion, polymorphisms in the promoter region of TGFB1 are not likely to be associated with large increases in breast cancer risk overall among Caucasian women. However, alleles in this region associated with increased TGFB1 levels may reduce the risk of estrogen receptorpositive breast tumors.