Background
Estrogens, whether from endogenous or exogenous sources, are potent mitogens in the endometrium and thus constitute a major carcinogen in this tissue [
1]. Endometrial cancer is therefore a good model for investigating clinical effects of estrogen signaling. Estrogen receptor alpha is the principal estrogen receptor expressed in the endometrium [
2,
3] and it is considered to be crucial in the development of endometroid endometrial carcinoma, the most common histological subtype among endometrial neoplasms [
1]. Since altered ER function could have impact on the risk of endometrial cancer, the estrogen receptor alpha gene (
ESR1) is a plausible endometrial cancer candidate gene which has been investigated in a few studies [
4‐
6]. We have investigated whether polymorphic variation in the
ESR1 is associated with invasive endometrial cancer risk, overall and in subgroups defined by other hormonally related factors.
Results
We successfully genotyped 695 cases and 1564 controls for the rs2234670, 698 cases and 1561 controls for the rs2234693, 698 cases and 1562 controls for the rs9340799, 701 cases and 1563 controls for the rs4986934, and 702 cases and 1563 controls for the rs1801132 marker. In controls, SNP genotype frequencies were in HWE but the rs2234670 genotype frequencies were not, whether in dichotomised form (p = 0.003) or when all alleles were considered (p < 0.0001).
The two intron 1 markers were in strong LD while there was weaker pairwise LD between the remaining markers (Table
1).
Table 1
Pairwise linkage disequilibrium between polymorphisms in the estrogen receptor alpha gene
| | | | |D'| | |
rs2234670a
| | - |
0.790
|
0.734
|
0.046
|
0.188
|
rs2234693 | |
0.571
| - |
0.998
|
0.825
|
0.338
|
rs9340799 | r2
|
0.349
|
0.590
| - |
0.742
|
0.245
|
rs4986934 | |
0.001
|
0.023
|
0.011
| - |
0.822
|
rs1801132 | |
0.008
|
0.029
|
0.009
|
0.090
| - |
Risk factor exposure differences between cases and controls reflected established associations (Table
2). Ninety-three percent of the tumours in the following analyses were of endometroid histological type. Myometrial invasion was available for 79 percent of the cases, out of which 31 percent had invasion through more than half of the myometrial thickness. Thirty-eight, 45 and 17 percent were of histopathological grade 1, grade 2, and grade 3, respectively. We found no convincing associations between genotypes and any of the covariates (data not shown). There was no difference in genotype frequencies between cases who donated blood samples and those who participated via tissue samples (P = 0.96, 0.66, 0.48, 0.56, and 0.51 for rs2234670, rs2234693, rs9340799, rs4986934, and rs1801132, respectively) or between primary selected controls and controls included after they had been genotyped for the parallel breast cancer study (P = 0.52, 0.98, 0.77, 0.31, 0.29 for rs2234670, rs2234693, rs9340799, rs4986934, and rs1801132, respectively). Further, genotype frequencies did not differ between those included in multivariate models and those excluded due to missing information about one or more of the included covariates (data not shown). Cancers in women participating via tissue samples were more often of more advanced stage as measured by degree of myometrial invasion (no invasion 2.8 vs 10.5 percent, less than half of the myometrial thickness 65.3 vs 61.7 percent, more than half of the myometrial thickness 26.4 vs 25.9 percent, through the serosa 5.6 vs 1.9 percent, p = 0.05).
Table 2
Distribution of endometrial cancer risk factors among participating endometrial cancer cases and population controls
Age | 702/1563 | 64.1 (6.80) | 63.4 (7.03) | 0.0001 |
Age at menarche | 653/1427 | 13.5 (1.37) | 13.5 (1.42) | 0.14 |
Age at last birth | 604/1405 | 29.4 (5.08) | 30.4 (5.34) | 0.0002 |
Age at menopause | 614/1503 | 51.0 (4.06) | 50.1 (3.91) | <0.0001 |
Recent BMI | 701/1546 | 27.4 (5.31) | 25.5 (4.27) | <0.0001 |
| | Percent | |
Nulliparous | 158/97 | 14 | 10 | |
1 childbirth | 289/146 | 21 | 18 | 0.008 |
2+ childbirths | 1116/459 | 65 | 71 | |
Family historya
| 667/1397 | 10 | 5 | <0.0001 |
Diabetes Mellitus, self-reportedb
| 702/1441 | 10 | 8 | 0.17 |
Ever use of oral contraceptives | 699/1550 | 24 | 36 | <0.0001 |
Ever use of menopausal hormonesb
| 690/1541 | 28 | 28 | 0.97 |
Smokingc
| 702/1563 | 35 | 43 | 0.0004 |
Non-participants, i.e. those that were eligible for this study but that either were not selected or did not consent to participate, were on average older than participants. Non-participating cases more often had cancer with more advanced myometrial invasion; 50 percent had cancer invasion to more than half of the myometrial thickness as compared to 31 percent among participants (p < 0.0001). Participants were also more likely to have been oral contraceptives users. Due to the over-sampling scheme, participants were also more likely to have used menopausal hormones. Despite our over-sampling of women with diabetes mellitus, participants were less likely to have this disease than non-participants. Participants and non-participants were similar with regard to smoking habits, family history of endometrial cancer, BMI, age at last birth and age at menopause (data not shown).
Association with endometrial cancer
In univariate analyses conditioned on the selection variables the rs2234670, the rs2234693 and the rs9340799 were associated with endometrial cancer risk so that subjects homozygous for the rarer alleles (denoted 11 in Table
3) were at decreased risk as compared to homozygotes for the most common allele (denoted 00). None of the covariates appeared to be confounders. However, including information about parity, BMI, smoking status and use of oral contraceptives significantly improved the model fit. Age at last birth and age at menopause further bettered model fit but since adjustment left point estimates unchanged these variables were excluded due to missing information in many individuals. Multivariate analyses yielded slightly stronger results compared to those from univariate analyses (Table
3). The difference could not be attributed to any one variable. Heterozygotes were at intermediate risk. Excluding cases with DNA derived from tissue left estimates largely unaltered and associations using cases with tissue derived DNA did not speak against pooling all successfully genotyped cases in the further analysis (Additional file
1, Supplementary table 3).
Table 3
Odds ratios (OR) for endometrial cancer and 95% confidence intervals (CI)
Genotype | 00a
| Case/controls | | 232/441 | 211/382 | 324/573 | 617/1276 | 406/834 |
| | OR (CI) | | 1 (ref) | 1 (ref) | 1 (ref) | 1 (ref) | 1 (ref) |
| 01a
| Case/controls | | 307/618 | 336/670 | 280/632 | 44/89 | 226/468 |
| | OR (CI) | Univariateb
| 0.95 (0.75–1.19) | 0.88 (0.70–1.11) | 0.79 (0.64–0.98) | 1.02 (0.68–1.54) | 1.02 (0.82–1.26) |
| | | Multivariatec
| 0.92 (0.72–1.16) | 0.86 (0.68–1.10) | 0.75 (0.60–0.93) | 1.03 (0.67–1.58) | 1.00 (0.80–1.26) |
| 11a
| Case/controls | | 123/309 | 115/316 | 58/163 | 1/3 | 30/66 |
| | OR (CI) | Univariateb
| 0.72 (0.54–0.97) | 0.65 (0.48–0.87) | 0.59 (0.41–0.85) | 0.94 (0.09–9.40) | 0.94 (0.58–1.51) |
| | | Multivariatec
| 0.73 (0.54–0.99) | 0.65 (0.48–0.89) | 0.53 (0.37–0.77) | 1.02 (0.09–11.61) | 0.95 (0.58–1.57) |
We examined whether including more than one marker would improve model fit, indicating influence on risk by combinations of alleles at different loci (i.e. haplotype effects). The association with rs9340799 was the strongest. The addition of either rs2234670 or rs2234693 to a model with rs9340799 did not significantly improve explanation of disease status (LR test p = 0.70 and p = 0.50, respectively). Conversely, starting with rs2234670 or rs2234693 and adding rs9340799 significantly improved model fit (LR test p = 0.02 and p = 0.03, respectively). Adding any of the last two markers, rs4986934 or rs1801132, neither improved fit nor changed estimates nor confidence intervals for the first three markers considered one at the time. When adding to the model with rs9340799 also rs2234693 and the interactions between these loci, the fit was not improved (LR test, p = 0.48). Neither did adding the rs1801132 (LR test p = 0.31) or the rs2234670 (LR test p = 0.83) and the respective interactions improve fit. Only three out of the four possible haplotypes of rs9340799 and rs2234693 existed in our study population (Table
4), so that we were unable to compare the model containing rs9340799 and rs2234693 main effects and interactions with the haplotype model. Other haplotype models (under the assumption of multiplicative penetrance), using rs9340799 and one of rs2234670 or rs1801132, did not provide a significant improvement in fit over the single locus (rs9340799) model.
Table 4
Distribution of ESR1 four-locus haplotype frequencies as estimated through EM algorithms among endometrial cancer cases and population controls
T | A | C | C | 0.371 | 0.404 | 0.406 |
C | G | C | C | 0.288 | 0.255 | 0.252 |
C | A | C | C | 0.114 | 0.118 | 0.118 |
T | A | C | G | 0.123 | 0.137 | 0.135 |
C | G | C | G | 0.055 | 0.045 | 0.046 |
T | A | T | G | 0.028 | 0.025 | 0.027 |
Proportion accounted for by the most common haplotypes: | 0.979 | 0.984 | 0.984 |
We investigated the association between rs9340799 and endometrial cancer stratified by the covariates (Table
5). We were primarily interested in potential interactions between the marker and use of menopausal hormones and BMI, since these factors are the most important determinants of serum estrogen levels in postmenopausal women, though we also examined other hormonally related variables. We could not estimate main effects of variables that we used in our sampling scheme (see Methods section) and thus the results presented (first in Table
5) from the analyses stratified by these factors convey only genetic effects. In subdividing the cases and controls by use of menopausal hormones, the protective effect of rs9340799 emerged among never users while the pattern was unclear among users, possibly due too small numbers of observations (Table
5). P-values for interaction were 0.007, 0.35, and 0.08 for compounds containing estrogen only, estrogen plus progestins cyclically and estrogen plus progestins continuously, respectively. Despite the small p-values, there were no allele-dose response relations. Association analyses stratified by various other risk factors for endometrial cancer are given in Table
6. BMI did modify the association between rs9340799 and endometrial cancer (p = 0.91). This result was not sensitive to BMI category cutpoints (p = 0.75 and p = 0.45 using BMI>30 or BMI>31 as the highest category, respectively). The protective effect of rs9340799 was clearer among non-smokers and among non-users of combined oral contraceptives although tests for interaction were non-significant (p = 0.17 and p = 0.19, respectively). There appeared to be no modification of the association between rs9340799 and endometrial cancer risk by use of oral low potency estrogen (p = 0.50), diabetes mellitus (p = 0.91), parity (p = 0.88), age at menopause (in tertiles, p = 0.72), age at last birth (in tertiles, p = 0.46), or family history of endometrial cancer (p = 0.51).
Table 5
Odds ratios (OR) and 95% confidence intervals (CI) estimating the association between ESR1 rs9340799 and endometrial cancer by subgroups of hormone use and self-reported diabetes mellitus
Use of menopausal hormones with medium potency estrogens | | | | |
Never users | 485/1006 | 1 (ref) | 0.78 (0.62–0.97) | 0.51 (0.34–0.77) |
Ever use of estrogen only | 85/104 | 1 (ref) | 0.47 (0.23–0.97) | 1.96 (1.62–6.14) |
Ever use of estrogen plus progestins cyclically | 100/196 | 1 (ref) | 0.85 (0.49–1.49) | 0.78 (0.33–1.84) |
Ever use of estrogen plus progestins continuously | 38/145 | 1 (ref) | 0.57 (0.24–1.35) p = 0.08b
| 0.94 (0.29–3.08) |
Diabetes Mellitus | 598/1258 | 1 (ref) | 0.76 | 0.56 |
No | | | (0.61–0.94) | (0.39–0.81) |
Yes | 64/110 | 1 (ref) | 0.52 (0.25–1.09) p = 0.91b
| 0.57 (0.15–2.15) |
Table 6
Odds ratios (OR) and 95% confidence intervals (CI) estimating the association between ESR1 rs9340799 and endometrial cancer by subgroups of other endometrial cancer risk factors
Use of oral low potency estrogen (estriol) |
Never | 533/1213 | 1 (ref) | 0.79 (0.64–0.99) | 0.55 (0.37–0.81) | |
Ever | 127/153 | 1.77 (1.20–2.61) | 1.28 (0.86–1.92) | 1.52 (0.74–3.15) | 0.50 |
BMI (kg/m2) | | | | | |
< 25 | 265/698 | 1 (ref) | 0.83 (0.61–1.13) | 0.62 (0.37–1.03) | |
25–28 | 136/387 | 1.01 (0.71–1.45) | 0.70 (0.48–1.02) | 0.59 (0.27–1.27) | |
> 28 | 261/283 | 2.90 (2.05–4.11) | 2.05 (1.46–2.87) | 1.39 (0.80–2.43) | p = 0.93 |
Smokingb
| | | | | |
Never | 434/783 | 1 (ref) | 0.84 (0.05–1.08) | 0.51 (0.33–0.80) | |
Ever | 228/585 | 0.75 (0.56–1.00) | 0.50 (0.37–0.68) | 0.59 (0.35–0.98) | p = 0.17 |
Use of combined oral contraceptives |
Never | 500/863 | 1 (ref) | 0.86 (0.68–1.09) | 0.58 (0.38–0.87) | |
Ever | 162/505 | 0.63 (0.46–0.87) | 0.38 (0.27–0.54) | 0.43 (0.24–0.78) | p = 0.19 |
Parity | | | | | |
0 childbirths | 94/137 | 1 (ref) | 0.78 (0.44–1.38) | 0.44 (0.16–1.24) | |
1 childbirth | 135/251 | 0.69 (0.41–1.18) | 0.62 (0.37–1.04) | 0.51 (0.23–1.11) | |
≥ 2 childbirths | 433/980 | 0.62 (0.39–0.96) | 0.45 (0.29–0.70) | 0.37 (0.21–0.65) | p = 0.88 |
At at menopause (years) |
< 50 | 157/447 | 1 (ref) | 0.81 (0.53–1.20) | 0.51 (0.25–1.05) | |
50–51.5 | 125/393 | 0.98 (0.64–1.48) | 0.68 (0.45–1.04) | 0.79 (0.40–1.55) | |
> 51.5 | 294/470 | 1.75 (1.23–2.51) | 1.48 (1.03–2.12) | 0.95 (0.54–1.66) | p = 0.72 |
Age at last birth (years, among parous only) |
≤ 27 | 209/364 | 1 (ref) | 0.66 (0.46–0.97) | 0.75 (0.40–1.41) | |
27.5–32.5 | 194/452 | 0.75 (0.52–1.09) | 0.53 (0.36–0.79) | 0.33 (0.17–0.64) | |
≥ 33 | 165/415 | 0.55 (0.37–0.81) | 0.52 (0.35–0.77) | 0.39 (0.20–0.75) | p = 0.46 |
Family history of endometrial cancerc
|
No | 564/1262 | 1 (ref) | 0.78 (0.63–0.97) | 0.63 (0.44–0.90) | |
Yes | 65/67 | 2.56 (1.47–4.46) | 1.28 (0.76–2.18) | 1.48 (0.39–5.66) | p = 0.51 |
Only 45 cases were of non-endometroid histotype. Excluding these from the analyses did not change the results of any of the above analyses (data not shown).
Analyses where cases were stratified by degree of myometrial invasion (Additional file
1, Supplementary table 3) indicated a more pronounced protective effect of the rs9340799 variant allele among those with invasion through more than 50 percent of the myometrium (adjusted OR 0.67 CI 0.45–0.98 for heterozygotes and OR 0.35 CI 0.16–0.75 for variant allele homozygotes). This latter analysis was based on only 139 cases. Stratifying on histopathological grade did not reveal any clear pattern to indicate a differential effect of genotype on risk for endometrial cancers of various degree of differentiation (Additional file
1, Supplementary table 4).
Discussion
We show that non-coding variation in ESR1 is associated with endometrial cancer risk. With regard to the rs9340799 locus, homozygotes for the rare G allele had an almost halved risk for endometrial cancer compared to common allele homozygotes. This protection seemed more pronounced when only cases with invasion through more than 50 percent of the myometrium were considered. The latter analysis was based on few cases and needs further confirmation. We found no convincing modification of this association by other endometrial cancer risk factors or histopathological grade.
We found no evidence of additional effects of other markers than rs9340799. However, we cannot determine whether this SNP is functional or in LD with some other functional locus. Furthermore, we cannot be completely certain that rs9340799 is more influential than rs2234693 because of the strong LD between these markers, separated by only 46 bp.
The association between rs9340799 and endometrial cancer was not convincingly modified by exposures that alter the availability of estradiol or estrone such as use of menopausal hormone preparations or high BMI. In our parallel breast cancer study [
25] the
ESR1-disease risk association was stronger in women with high BMI but this appeared not to be the case in endometrial cancer. The effect of rs9340799 was largely the same in any stratum with sufficient sample size.
Our study is population-based and the largest investigating the relation between ESR1 and endometrial cancer to date. Furthermore, the study population is genetically homogenous, which limits concern about population stratification and which also preserves power as no additional stratifications based on ethnicity need to be made. The vast majority of cases in this study were of the strongly estrogen related endometroid type, and thus constitute a sample in which functional ESR1 alterations are expected to become evident.
The markers that we have studied here were selected because they were known to be polymorphic according to the literature available when the study was planned (around 1996). They were not selected, as is usually the case with current studies, in an attempt to represent the entire gene. This is a weakness but it should not discredit the reported association.
It appears as if the
ESR1 gene is composed of three different LD blocks [
26]. All markers that are typed in the present study are located within the first region, towards the 5' end of the gene. We did not find evidence of any interactions between markers, which would have indicated that functional combinations of alleles exist in the gene. We initially intended to investigate if we could establish whether any SNP was functional or instead in LD with a functional locus, by comparing a model with genotype main effects and interactions, to a haplotype model, using SNPs rs9340799 and rs2234693. If the haplotype model had exhibited a significantly better fit then there would have been indication that other, linked, un-typed, SNPs were responsible for the association [
27]. However, only three out of the four possible haplotypes of rs9340799 and rs2234693 exist in our data, so that we could not make this comparison. In contrast to our findings in the parallel breast cancer study [
25], a haplotype model with rs9340799 and rs1801132 did not provide any additional explanatory information.
Our main results replicate our own findings in a separate but equivalent Swedish study population where we observed associations of similar strength and magnitude [
6]. They also corroborate results of a small Japanese study where rs9340799 GG and rs2234693 CC genotypes were associated to a lower risk for endometrial cancer compared to those homozygous for the common allele [
4]; OR 0.26 (CI 0.09–0.79) and OR 0.23 (CI 0.07–0.82), respectively. Another Japanese study found no such association [
5] but a decreased risk with allele-dose effect with the rare allele of a
ESR1 codon 10 SNP.
Weel and colleagues found the rs2234693 CC genotype to be associated with an earlier onset of natural menopause [
28], which would entail a lower risk of endometrial cancer, but we did not find any similar association, nor any interaction between
ESR1 genotype and age at menopause.
There is no evidence in the literature that rs9340799 affects the amount or function of the estrogen receptor protein. Herrington and colleagues found that the adjacent rs2234693 C allele produced an additional binding site for the myb family of transcription factors [
29]. Binding of B-myb to this site appeared to have the capability of enhancing
ESR1 transcription. Since B-myb is in itself estrogen responsive there is a possibility of a positive feedback loop that could amplify estrogenic response. It is possible that in fact rs2234693 is the functional variant and that our finding was due to the strong LD between the markers in addition to random variation in the data. Ongoing studies that include dense mapping of
ESR1 variants will give guidance as to what variants need to be further functionally validated.
Although our results, which are in line with a priori theories, might indicate a true influence of
ESR1 variation on estrogen dependent phenotypes they could conceivably be due to selection bias, a potential problem in case-control studies. Our participation rates, 88 and 76 percent among cases and controls, respectively (67 and 64 percent using those eligible for the parent study in the denominator) could lead to bias if participation were related to genotype. Non-participating cases had a more advanced disease as measured by myometrial invasion. If, for example, a genetic variant is associated to severe cancer but not to less severe cancer and the more severe cases are less likely to participate because they have died, any association with cancer overall would be biased towards the null. Furthermore, genotype frequencies were not different between participants who donated blood and those who took part via a tissue sample, the latter of which were more often deceased. Our results cannot be generalized to include all endometrial cancer because there is data that indicates differing etiologies between endometroid and non-endometroid endometrial cancer [
1].
Another possible mechanism for selection bias is if an
ESR1 variant influences a phenotype that differentially affects ascertainment or recruitment probability among cases and controls. In this case one would expect to find case-control associations with multiple otherwise unrelated phenotypes.
ESR1 variation has indeed been associated with a host of other outcomes such as coronary heart disease [
30], height [
31], bone mineral density [
32,
33], multiple sclerosis [
34], cognitive impairment [
35], and age at menarche [
36]. However, despite the diversity among these outcomes there is reasonable evidence of etiological roles for estrogen. Our comparisons of characteristics among participants and non-participants do not point to any plausible mechanism for selection bias as an explanation to the observed association.
Acknowledgements
The authors are indebted to all the women who have contributed to this study by answering the questionnaire and donating biological samples. Dr Anders Lindgren performed the pathological reviews. Anna Christensson, Boel Bissmarck and Anders Holmberg provided invaluable assistance in collecting samples from the participants. Maria Branting, Anders Westermark, Erika Svensson, Kristina Larsson, and Birgitta Sundelin provided excellent technical support. We also wish to express our gratitude towards primary health care centers and pathology departments all over Sweden for their unselfish cooperation. Hans-Olov Adami contributed with skills, experience and enthusiasm to all stages of the project. This study was supported by the National Institutes of Health, grant number 5 RO1 CA 77973-03, by the Swedish Cancer Society, and by the K&A Wallenberg foundation (Wallenberg Consortium North). K. Humphreys was supported by a grant from the Wallenberg Consortium North.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
SW planned and coordinated the collecting of biological samples, participated in the genotyping, performed most of the biostatistical analyses, interpreted results, and wrote the manuscript. LL performed the bulk of the genotyping and took part in interpreting results and writing the manuscript. KH oversaw all and performed some of the biostatistical analyses, interpreted the results. CM, IP, JB and EW planned the study and interpreted the results. HM, ACS, AK, UL, MLF planned, coordinated and oversaw all DNA extraction and genotyping. FS performed genotyping. All authors critically read and took part in finalizing the manuscript.