Relation of FGFR2 Genetic Polymorphisms to the Association Between Oral Contraceptive Use and the Risk of Breast Cancer in Chinese Women

Abstract

The fibroblast growth factor receptor 2 gene (FGFR2) has been associated with the risk of breast cancer in multiple ethnic populations, and its effect has been suggested to be hormone-dependent. A large, 2-stage, population-based case-control study was conducted in urban Shanghai, China, during the periods of 1996–1998 and 2002–2005. Exposure and genotyping information from 2,073 patients with breast cancer and 2,084 age-matched population controls was available for evaluation of the interactions between FGFR2 polymorphisms and exogenous estrogen exposure in the development of breast cancer. A logistic regression model was used to compute adjusted odds ratios and 95% confidence intervals. Of 20 genotyped and 25 imputed single nucleotide polymorphisms (SNPs), 22 were significantly associated with breast cancer. Three genotyped SNPs in close linkage disequilibrium, rs2303568, rs3135730, and rs1078806, and an imputed SNP of rs755793 in complete linkage disequilibrium with other 8 SNPs were observed to interact significantly with oral contraceptive (OC) use. The SNP-cancer association was evident only among OC users, and the OC use was only associated with the risk of breast cancer among carriers of these minor alleles at these loci. These findings suggest that genetic variants in FGFR2 may modify the role of OC use in causing breast cancer in Chinese women.

Breast cancer is a hormone-related malignancy. Both meta- (1) and pooled (2, 3) analyses have consistently shown positive associations between oral contraceptive (OC) use or postmenopausal hormone replacement therapy (HRT) and breast cancer risk. A large-scale clinical trial also found that prolonged exposure to exogenous estrogens and progestins in hormone therapy increased a woman's risk of developing breast cancer (4). This evidence, however, was primarily derived from Western populations. Most studies conducted in Asian populations did not find a significant association between OC use and breast cancer risk (5–9). Because of the low frequency of HRT use in Asian countries, previous studies conducted in Asian populations did not have sufficient statistical power to evaluate the effect of HRT on breast cancer risk (8, 10, 11).

It has been increasingly recognized that breast cancer, which is a complex and multifactorial disease, is a result of interplay between different exposures and host susceptibility. Epidemiologic evidence has suggested an interaction of exogenous and endogenous hormones with an individual's genetic susceptibility in breast cancer etiology (12–16).

The fibroblast growth factor receptor 2 gene (FGFR2) is a well-recognized breast cancer-susceptibility gene (17–23). Multiple single nucleotide polymorphisms (SNPs) in this gene have been associated with breast cancer risk across multiple ethnic groups. Although the mechanism of mammary carcinogenesis for these SNPs remains unclear, these SNPs were consistently observed to be more closely associated with estrogen receptor-positive or progesterone receptor-positive cancer than with estrogen receptor-negative or progesterone receptor-negative cancer (17, 18, 20, 22), which suggests that the effect of the FGFR2 gene on breast cancer may be hormone-dependent.

Using the resources from the Shanghai Breast Cancer Study, a large, population-based case-control study of Chinese women in urban Shanghai, China, we evaluated the interaction of exogenous hormone use, primarily OC use, with FGFR2 genetic polymorphisms in the origin of breast cancer.

MATERIALS AND METHODS

Study population

The Shanghai Breast Cancer Study is a 2-phase, population-based case-control study conducted in urban Shanghai, China (20). In the first phase of the Shanghai Breast Cancer Study, which ran from 1996 to 1998, a total of 1,459 of 1,602 women aged 25–65 years who had newly diagnosed breast cancer and 1,556 of 1,724 eligible controls completed in-person interviews. Of these subjects, 1,193 (82%) cases and 1,310 (84%) controls donated a blood sample. The second phase of the Shanghai Breast Cancer Study was conducted from 2002 to 2005 with a protocol similar to that used in the initial phase. A total of 1,989 cases and 1,989 population controls were recruited, yielding response rates of 83.7% and 70.4%, respectively. Of these participants aged 25–70 years, 1,932 (97.1%) cases and 1,857 (93.4%) controls provided a blood sample or an exfoliated buccal cell sample. The study protocols were approved by the institutional review boards of all institutions involved in the study, and written informed consent was obtained from all participants before the interview.

Data collection

Study participants were interviewed in person by trained medical professionals who used a structured questionnaire. Detailed information on demographic factors, menstrual and reproductive history, use of OCs and postmenopausal HRT, prior disease history, physical activity level, tobacco and alcohol use, diet, weight history, and family history of cancer was collected for all participants. Subjects who had ever taken any OCs were asked about age at first use, duration of use, age at last use, and the name of the contraceptive used for each episode. Information on the name of the hormones used was collected for the HRT users. Body weight, height, and waist and hip circumferences were measured according to a standardized protocol at the time of interview. Menopause was defined as the cessation of the menstrual period for ≥12 months before diagnosis for cases and before interview for controls, excluding those lapses caused by pregnancy, breastfeeding, or estrogen hormone use. Body mass index, calculated as weight in kilograms divided by height in meters squared, was calculated using measured anthropometric data.

SNP genotyping

Genomic DNA was extracted from buffy coat fractions using the QIAamp DNA minikit (Qiagen Inc, Valencia, California) following the manufacturer's protocol. Genotyping was conducted mainly using Affymetrix Genome-Wide Human SNP array 6.0 at the Vanderbilt Microarray shared resource following Affymetrix's protocol, as described previously (20). Briefly, genomic DNA (250 ng) at 50 ng/μL was digested with either Sty I or Nsp restriction enzyme and then ligated to Nsp or Sty I adaptor/primers using T4 ligase. After ligation, a generic primer that recognizes the adaptor sequence was used to amply adapt ligated DNA fragments. The products of all 4 polymerase chain reactions were purified with magnetic beads, quantitated with a spectrophotometer, and analyzed on a 2% agarose gel. Successful polymerase chain reaction products were fragmented, and the fragmented products were again analyzed on a 4% agarose gel. Finally, the prepared samples were hybridized to Affymetrix SNP Array 6.0, stained and washed, and subsequently scanned according to Affymetrix's protocol. The Birdseed v2 algorithm (http://www.broad.mit.edu/mpg/birdsuite/) was used to call genotypes. A series of stringent quality-control procedures, such as genomic DNA quantity assessment, cross-genotyping platform validation, cross-genotyping batches validation, individual exclusion criteria, and marker exclusion criteria, were applied to assure the data quality, as described previously (20).

Only study participants (2,073 cases and 2,084 controls) who donated a blood sample to the study were included in the genome-wide association study scan. Of the 42 SNPs genotyped on Affymetrix 6.0 SNP chips that covered about 120 kilobases (kb) of the FGFR2 gene plus the 10 kb region of its 5′ upstream and the 5 kb region of its 3′ downstream, 20 had a minor allele frequency (MAF) >0.05 in the 2,084 controls. Information on these SNPs formed the basis of the current analysis.

Statistical analysis

Imputation analyses of all known SNPs in FGFR2 were conducted using HapMap phased haplotype data of the Japanese in Tokyo Study and the Chinese Han in Beijing Study samples as the reference panel (National Center for Biotechnology Information human genome building 36). A total of 93 SNPs were imputed based on the genotyped data on the Affymetrix SNP Array 6.0 by using the Markov Chain-based Haplotyper program (24). During the imputation, an information score was generated for each imputed SNP genotype. Values <0.90 were considered to be the result of poor quality imputation. A total of 25 SNPs, which are not on chip, had a MAF >0.05 and an average imputation quality score >0.90. All genotyped and imputed SNPs in this analysis are listed in Web Table 1 (available at http://aje.oxfordjournals.org/).

Chi-squared statistics and t tests were used to evaluate case-control differences in the distribution of risk factors and genotypes of the FGFR2 gene. Logistic regression models were used to estimate odds ratios and 95% confidence intervals. The association of genetic polymorphisms with breast cancer was evaluated in dominant, additive, and recessive models. The interactive effects between a risk factor and various genotypes were evaluated by introducing the products of genotypes (in a dominant, recessive, or additive model) and the hormone-use variables (either dichotomous variables of never/ever using OC or HRT or 2 dummy variables describing nonusers of OC (short-time user of OC and long-time user of OC grouped by the median time of OC use in controls, which was 18 months)) in the logistic model along with the main effect terms. A likelihood ratio test was conducted by comparing the model including only the main effects with that including both the main effects and the interaction terms to derive the P value for the multiplicative interaction test. Linkage disequilibrium (LD) between polymorphisms was assessed by using HaploView software (25). Haplotype blocks in FGFR2 were defined using the data from this population with the methods of Gabriel et al. (26). Selected SNPs in each haplotype block were used to reconstruct haplotypes using HAPSTAT software (University of North Carolina at Chapel Hill, Chapel Hill, North Carolina). Haplotype effects and haplotype-OC interactions were estimated using HAPSTAT software in dominant, recessive, and additive models (27). All statistical tests were based on 2-tailed probability.

RESULTS

As shown in Table 1, HRT use was uncommon in this study population. Fewer than 5% of postmenopausal controls had ever used HRT. On the other hand, 22.0% controls in the stage I study and 18.5% controls in the stage II study had ever used OCs.

Table 1.

Comparison of Demographic Characteristics in Breast Cancer Cases and Controls, Shanghai Breast Cancer Study, 1996–1998 and 2002–2005

Characteristic	Stage I Study					Stage II Study
	Cases (n = 1,104)		Controls (n = 1,109)		P Value	Cases (n = 969)		Controls (n = 975)		P Valuea
	%	Mean (SD)	%	Mean (SD)		Mean (SD)	%	Mean (SD)	%
Age, years		47.6 (8.0)		47.7 (8.3)	0.75		51.4 (8.3)		51.4 (8.3)	0.91
Educational level
Elementary school or less	12.4		15.2			7.1		11.8
Middle school	44.3		42.7			42.9		42.0
High school	31.3		31.9			37.7		36.9
College or more	12.1		10.3		0.17	12.3		9.3		0.0016
Breast cancer in first-degree relatives	3.4		2.6		0.31	5.7		3.4		0.02
Breast fibroadenoma	9.5		4.6		<0.0001	10.0		6.5		0.0051
Age at menarche, years		14.5 (1.6)		14.8 (1.7)	0.0002		14.5 (1.7)		14.7 (1.8)	0.0037
Ever had a livebirth	95.1		96.2		0.20	95.2		96.1		0.30
No. of livebirths		1.5 (0.8)		1.5 (0.9)	0.15		1.3 (0.7)		1.4 (0.8)	0.005
Age at first livebirth, years		26.8 (4.1)		26.3 (3.8)	0.003		26.1 (3.7)		25.7 (3.8)	0.03
Postmenopausal	32.8		36.6		0.06	45.5		47.1		0.49
Age at menopause, years		48.1 (4.7)		47.4 (5.0)	0.04		48.7 (4.4)		48.0 (4.5)	0.03
Oral contraceptive use	22.1		22.2		0.95	18.1		18.5		0.82
Duration of oral contraceptive use		8.6 (31.4)		9.4 (32.9)	0.52		8.3 (31.6)		7.5 (28.1)	0.56
Hormone replacement therapy use among postmenopausal women	3.1		5.0		0.18	8.4		4.6		0.02
Physical activity in past 10 years	19.2		26.3		<0.0001	31.8		34.7		0.18
Body mass indexb		23.6 (3.4)		23.3 (3.4)	0.02		24.0 (3.3)		23.4 (3.3)	<0.0001

^aFrom chi-squared tests (categorical variables) or t tests (continuous variables).

^bWeight in kilograms divided by height in meters squared.

In both stages of the study, the cases and controls were similar with respect to age, ever having had a livebirth, menopausal status, OC use, and duration of OC use. Compared with controls, cases were more likely to have a family history of cancer, an earlier age at menarche, a later age at menopause, an older age at first livebirth, a higher body mass index, and a higher waist-to-hip ratio in the 2 stages. In stage I of the study, cases were less likely to have participated in physical activity than were controls. In stage II of the study, cases also differed significantly from controls in educational level, family history of breast cancer, number of livebirths, and postmenopausal HRT use (Table 1).

The risk factors for breast cancer in this population included early age at menarche, postmenopausal status, late age at menopause and at first livebirth, fewer livebirths, low physical activity, high body mass index, high waist-to-hip ratio, and a family history of breast cancer (28–30). Table 2 presents the odds ratios and 95% confidence intervals for OC use and HRT use in the development of breast cancer among subjects with FGFR2 genotyping data. Overall, no significant association was observed for use of OCs, type of OC used, duration of OC use, or period of OC use. HRT was not significantly associated with breast cancer risk among postmenopausal women, either. The null results were consistently observed in stages I and II of the study (data not shown).

Table 2.

Association Between Sex Hormone Use and Breast Cancer Risk Among Chinese Women, Shanghai Breast Cancer Study, 1996–1998 and 2002–2005

Sex Hormone Use	No. ofCases	No. ofControls	Adjusted Odds Ratioa	95% Confidence Interval
Oral contraceptive use
Never	1,654	1,657	1.00
Ever	419	426	0.98	0.83, 1.15
Type of oral contraceptive used
None	1,654	1,657	1.00
Progesterone plus estrogen	328	324	1.05	0.88, 1.26
Progesterone only	42	52	0.85	0.56, 1.30
Estrogen only	24	30	0.82	0.47, 1.41
Other	25	20	1.33	0.73, 2.42
Duration of oral contraceptive use
Never	1,654	1,657	1.00
<18 months	208	225	0.96	0.78, 1.18
≥18 months	211	201	1.11	0.89, 1.37
P for trend			0.52
Period of oral contraceptive use
Never	1,654	1,657	1.00
Recent use (within 10 years)	53	56	0.98	0.66, 1.44
Previous use (>10 years ago)	366	369	0.97	0.82, 1.15
Hormone replacement therapy use (postmenopausal women only)
Never	753	822	1.00
Ever	48	41	1.16	0.75, 1.80

^aAdjusted for age, educational level, age at menarche, menopausal status, breast cancer in first-degree relatives, number of livebirths, and body mass index.

As shown in Web Figure 1, of 20 genotyped SNPs and 25 imputed SNPs, 22 were significantly associated with breast cancer risk under an additive model. These significant markers in FGFR2 fell into 11 major haplotype blocks, as showed in Web Figure 1. Of these SNPs, only associations for rs10736303, rs2981582, and rs2981581 remained significant after adjustment for any other SNPs. The gene-disease association did not differ by the stage of the study.

The associations were significant in both OC users and nonusers for rs10736303, but only among OC users for rs7073360, rs3135749, rs3135747, rs755793, rs3135739, rs3135737, rs3135736, rs12245334, rs9888022, rs2303568, rs3135730, rs9787582, rs1078806, rs2981575, rs2936870, rs17102287, and rs2981582 in multiple comparisons (P < 0.025). The interaction tests were significant for rs7073360, rs3135749, rs3135747, rs755793, rs3135739, rs3135737, rs3135736, rs12245334, rs9888022, rs2303568, rs3135730, and rs1078806 (P for interaction < 0.05) (Web Figure 1). No significant interaction was observed for HRT use with any genotyped SNPs in the FGFR2 gene (data not shown).

In the present study, rs2303568 was in close LD with rs3135730 (r² = 0.86) but in low LD with rs1078806 (r² = 0.01). The 9 imputed SNPs, rs7073360, rs3135749, rs3135747, rs755793, rs3135739, rs3135737, rs3135736, rs12245334, and rs9888022, were in complete LD (r² = 1.0), and thus rs755793 was used as a tag for these SNPs in further analyses.

We further examined the breast cancer risk for OC use according to the number of minor alleles of FGFR2 at rs755793, rs2303568, rs3135730, and rs1078806 (Table 3). Because of the low MAF at rs755793, rs2303568, and rs3135730, we combined the carriers of 1 and 2 minor alleles into 1 group. After adjustment for age and other potentially confounding factors, such as educational level, age at menarche, menopausal status, breast cancer in first-degree relatives, number of livebirths, and body mass index, OC use was observed to increase the risk of breast cancer among minor allele carriers of rs755793 (odds ratio (OR) = 1.83, 95% confidence interval (CI): 1.14, 2.94), rs2303568 (OR = 2.04, 95 % CI: 1.26, 3.29), and rs3135730 (OR = 1.97, 95% CI: 1.25, 3.12). An increased risk of breast cancer was also found for the CC genotype of rs1078806 (OR = 2.11, 95% CI: 1.19, 3.76). The significant positive association between OC use and breast cancer, however, was not observed for the AA genotype at rs755793, the AA genotype at rs2303568, the AA genotype at rs3135730, or the TT or TC genotype at rs1078806. The P values from interaction tests were 0.01 for rs755793, 0.005 for rs2303568, 0.007 for rs3135730, and 0.016 for rs1078806, as shown in Table 3.

Table 3.

Odds Ratios for Breast Cancer by Oral Contraceptive Use, According to the Number of Minor Alleles of FGFR2 Single Nucleotide Polymorphisms of rs2303568, rs3135730, and rs1078806, Shanghai Breast Cancer Study, 1996–1998 and 2002–2005

	Number of Minor SNP Alleles												P for Interaction
	0				1a				2
	No. of Cases	No. of Controls	ORb	95% CI	No. of Cases	No. of Controls	ORb	95% CI	No. of Cases	No. of Controls	ORb	95% CI
rs755793
OC use
Never	1,421	1,444	1.00		230	209	1.00
Ever	348	388	0.95	0.80, 1.12	71	37	1.83	1.14, 2.94					0.01
Duration of OC use
<18 months	175	201	0.92	0.74, 1.15	33	23	1.32	0.72, 2.40
≥18 months	173	187	0.98	0.78, 1.24	38	14	2.70	1.38, 5.29					0.011
P for trend			0.69				0.0038
rs2303568
OC use
Never	1,321	1,349	1.00		226	211	1.00
Ever	302	336	0.94	0.78, 1.13	73	37	2.04	1.26, 3.29					0.005
Duration of OC use
<18 months	153	172	0.93	0.73, 1.17	34	21	1.63	0.88, 3.00
≥18 months	149	164	0.96	0.75, 1.23	39	16	2.62	1.36, 5.05					0.011
P for trend			0.59				0.002
rs3135730
OC use
Never	1,318	1,337	1.00		234	225	1.00
Ever	299	332	0.93	0.78, 1.12	76	43	1.97	1.25, 3.12					0.007
Duration of OC use
<18 months	150	169	0.92	0.72, 1.16	37	25	1.61	0.90, 2.87
≥18 months	149	163	0.96	0.75, 1.22	39	18	2.51	1.33, 4.75					0.017
P for trend			0.56				0.002
rs1078806
OC use
Never	785	824	1.00		708	679	1.00		161	154	1.00
Ever	199	232	0.94	0.75, 1.18	172	172	0.99	0.77, 1.26	48	22	2.11	1.19, 3.76	0.016
Duration of OC use
<18 months	100	112	0.96	0.72, 1.29	89	98	0.91	0.67, 1.25	19	15	1.06	0.50, 2.27
≥18 months	99	120	0.92	0.68, 1.24	83	74	1.09	0.77, 1.54	29	7	4.52	1.88, 10.87	0.008
P for trend			0.55				0.86				0.0016

Abbreviations: CI, confidence interval; OC, oral contraceptive; OR, odds ratio.

^a Number of minor alleles ≥1 for rs2303568 and rs3135730.

^b Adjusted for age, educational level, age at menarche, menopausal status, breast cancer in first-degree relatives, number of livebirths, and body mass index.

Interestingly, a significant dose-response relation between duration of OC use and cancer risk was observed among those carrying minor alleles at rs755793, rs2303568, and rs3135730 and among those carrying 2 minor alleles at rs1078806. The associations were significant for the 4 SNPs after correction for multiple comparisons (P < 0.0167). The associations of the cancer with rs755793, rs2303568, rs3135730, and rs1078806 were more pronounced among women who had used OCs for at least 18 months, with P for interaction being 0.011 for rs755793, 0.011 for rs2303568, 0.017 for rs3135730, and 0.008 for rs1078806 (Table 3).

In our population, rs2303568 and rs3135730 were contained in the same block, and rs7073360, rs3135749, rs3135747, rs755793, rs3135739, rs3135737, rs3135736, rs12245334, rs9888022, rs2303568, rs3135730, and rs1078806 fell into a different block (Figure 1). As shown in Table 4, under the dominant model, Hap1 of GG, the haplotype containing the risk alleles rs2303568 and rs3135730, was significantly associated with an increased risk of breast cancer compared with Hap1 of AA (OR = 1.22, 95% CI: 1.02, 1.46), mainly among OC users (OR = 2.02, 95% CI: 1.34, 3.04), particularly in those who had used OCs for ≥18 months (OR = 2.29, 95% CI: 1.27, 4.15). A similar association pattern was observed under additive and recessive models (data not shown). We further constructed a haplotype across rs755793, the tag SNP of 9 imputed SNPs in complete LD, and rs2303568, rs3135730, and rs1078806, 3 significant genotyped SNPs, and we found that the haplotype effect was driven by risk alleles at the 4 polymorphic sites. However, the interaction tests were of borderline significance.

Figure 1.

Pairwise linkage disequilibrium (D′) between 12 significant single nucleotide polymorphisms at the fibroblast grown factor receptor 2 gene, Shanghai Breast Cancer Study, 1996–1998 and 2002–2005. Diamonds without a number correspond to D′ = 1. The block was defined using the method of Gabriel et al. (27).

Table 4.

Haplotype Analysis of the Association Between FGFR2 Under the Dominant Model and the Risk of Breast Cancer, Shanghai Breast Cancer Study, 1996–1998 and 2002–2005

	All Subjectsa			Non-OC Users			OC Users			Used OC for <18 Months			Used OC for ≥18 Months
	%	ORb	95% CI	%	ORb	95% CI	%	ORb	95% CI	%	ORb	95% CI	%	ORb	95% CI
Hap1
AA	92.5	1.00		92.2	1.00		93.6	1.00		93.3	1.00		93.9	1.00
GG	6.5	1.22	1.02, 1.46	6.7	1.08	0.88, 1.32	5.3	2.02	1.34, 3.04	5.7	1.79	1.01, 3.17	5.0	2.29	1.27, 4.15
P for  interaction				0.07
Hap2
AAAT	62.6	1.00		61.8	1.00		68.4	1.00		64.6	1.00		68.0	1.00
AAAC	28.4	1.13	1.01, 1.27	29.2	1.08	0.95, 1.23	24.4	1.36	1.05, 1.77	27.3	1.20	0.84, 1.72	24.8	1.60	1.11, 2.31
GGGT	4.6	1.31	1.05, 1.64	4.6	1.13	0.88, 1.45	2.7	2.36	1.42, 3.92	3.0	2.85	1.39, 5.85	3.1	2.34	1.15, 4.76
P for  interaction				0.09

Abbreviations: CI, confidence interval; Hap1, constructed haplotypes in the order of rs2303568 and rs3135730; Hap2, constructed haplotypes in order of rs755793, rs2303568, rs3135730, and rs1078806; OC, oral contraceptive; OR, odds ratio.

^aAdditionally adjusted for oral contraceptive use.

^bAdjusted for age, educational level, age at menarche, menopausal status, breast cancer in first-degree relatives, number of live births, and body mass index.

DISCUSSION

A high level of estrogen is a well-established risk factor for breast cancer. The carcinogenicity of the hormone has been attributed to its stimulation of estrogen receptor-mediated transcription, which results in cell proliferation and metabolic activation (31, 32). Both OCs and postmenopausal HRT have been classified as group 1 carcinogens (33, 34). Numerous epidemiologic studies have shown evidence of the adverse effects of OCs and HRT in the development of breast cancer (1–4). However, the results are not entirely consistent, particularly in Asian populations (5–9).

Consistent with the previous studies of Asian populations (5–9), we did not observe a significant overall association between OC use and the risk of breast cancer in a Chinese population. Interestingly, we found that the effect of OC use in the development of breast cancer significantly depended on the genotypes of FGFR2 at rs7073360, rs3135749, rs3135747, rs755793, rs3135739, rs3135737, rs3135736, rs12245334, rs9888022, rs2303568, rs3135730, and rs1078806. The dose-response relation between duration of OC use and cancer risk was also observed in women carrying at least 1 minor allele at rs755793, rs2303568, and rs3135730 and those having 2 minor alleles at rs1078806. To our knowledge, this is the first study to report a modifying effect of FGFR2 on the association between OC use and breast cancer.

The FGFR2 gene has been suggested to predispose people to breast cancer in multiple ethnic populations (17–19), including in our population (20). Importantly, the association between the FGFR2 gene and breast cancer appears to be stronger for estrogen receptor-positive and progesterone receptor-positive tumors than for estrogen receptor-negative or progesterone receptor-negative tumors (35, 36), which suggests a sex hormone-dependent role of the FGFR2 gene in breast cancer. The strongest association found in these previous studies mainly involves SNPs in intron 2 of the gene, which includes highly conserved regions and is dense in transcription factor binding sites of estrogen receptors, octamer-binding protein 1/runt-related transcription factor 2, and CCAAT/enhancer-binding protein beta that may cooperate in increasing gene expression (37, 38). In the present study, however, we did not observe a significant modifying effect of rs10736303, the most significant and potential functional variant of the FGFR2 gene, but we did find a significant interaction between OC use and rs1078806, a variant also located in intron 2 in the same block as rs10736303 and in close LD with the SNP (r² = 0.54, P < 0.01). This result provides some evidence of the hormone-dependent nature of the FGFR2 gene. rs3135730, another significant variant, is also located in intron 2 but is far away from the SNPs of rs1078806 (13.2 kb away) and rs10736303 (8.6 kb away). It is in close LD with rs2303568, a significant locus in intron 4 of the FGFR2 gene, and both are contained in the same block. Although haplotype analysis provides further evidence for the interaction between OC use and the 2 variances, the function of the 2 loci is not clear and the mechanism underlying the results is not understood. In addition, we found that rs7073360, rs3135749, rs3135747, rs755793, rs3135739, rs3135737, rs3135736, rs12245334, and rs9888022 significantly interacted with OC use in breast cancer risk. Of the 9 SNPs in complete LD, rs755793 is an imputed missense SNP located at exon 5. The translation of T→C at codon 1204 of FGFR2 results in a change of Met186→Thr. Given that exon 5 encodes immunoglobulin domain II of FGFR2, 1 of the 2 domains involved in ligand binding (39), it is possible that a Met186→Thr change may lead to an alternation in crystal structures and activity of FGFR2, as did the Pro253→Arg or Ser252→Trp mutation (40), and thus predispose to cancer risk. Haplotype analyses across rs755793, rs2303568, rs3135730, and rs1078806 also provide evidence of the contribution of the genetic variant to cancer risk. However, the rate of the variant is 6.7% of MAF in the Chinese population but does not exist in Europeans, which does not appear to support a causal effect for the SNP in breast cancer etiology because the incidence of breast cancer among Chinese women is much lower than that among white women. Therefore, considering the low MAF of rs755793, rs2303568, and rs3135730, as well the low frequency of the CC genotype at rs1078806 in this population, we could not exclude the possibility of chance.

Postmenopausal HRT use, another source of exogenous estrogen, was reported to have a significant modifying effect on the FGFR2-breast cancer association in 2 recent studies (14, 15). Rebbeck et al. (14) observed a significant interaction between HRT and FGFR2 rs1219648 genotypes in breast cancer risk in white Americans. Prentice et al. (15) found a favorable effect of postmenopausal hormone therapy in the TT genotype for the FGFR2 SNP of rs3750817. In a Japanese population, however, FGFR2 polymorphisms in intron 2 were not observed to interplay with HRT use, although the null interaction may be due to inadequate statistical power (41). These studies have commonly focused on “the top hit” in FGFR2 of the previous genome-wide association studies (17, 18) but have shown different SNPs as having modifying effect on hormonal exposure. These results underscore the complicated mechanisms through which FGFR2 may play a role in the etiology of breast cancer.

In the present study, we did not find a main effect of HRT on breast cancer and did not observe a modifying effect of HRT on the FGFR2-cancer association. This can be mainly attributed to the low frequency of HRT use in our population. Our study does not have the statistical power to evaluate the interaction between HRT use and gene polymorphisms.

Strengths of this study included the large sample size, the population-based study design, the high participation rate, the homogeneous ethnic background, and the extensive coverage of the FGFR2 gene of 120-kb plus the 10-kb region of its 5′ upstream and the 5-kb region of its 3′ downstream. Limitations of this study included incomplete knowledge about the function of the genetic variants of FGFR2 that has limited our ability to better understand the mechanisms through which the variants influence the effect of exogenous hormones on breast cancer risk. In this population, very few OC users took OCs containing estrogen only or progesterone only, and we were therefore unable to evaluate the respective effects of estrogens and progesterone. Further studies are warranted to confirm our findings, and results derived from other populations are needed to better understand the complicated mechanisms underlying the modifying effect of the FGFR2 gene.

In summary, our study provides the first evidence that the FGFR2 genetic polymorphism may modify the association of OC use with breast cancer risk among Chinese women. Our results, if confirmed, may have important implications in personalized prevention of breast cancer.

Glossary

Abbreviations

CI

confidence interval

FGFR

fibroblast growth factor receptor 2

HRT

hormone replacement therapy

LD

linkage disequilibrium

MAF

minor allele frequency

OC

oral contraceptive

OR

odds ratio

SNP

single nucleotide polymorphism