Variation in breast cancer risk associated with factors related to pregnancies according to truncating mutation location, in the French National BRCA1 and BRCA2 mutations carrier cohort (GENEPSO)

Introduction Mutations in BRCA1 and BRCA2 confer a high risk of breast cancer (BC), but the magnitude of this risk seems to vary according to the study and various factors. Although controversial, there are data to support the hypothesis of allelic risk heterogeneity. Methods We assessed variation in BC risk according to factors related to pregnancies by location of mutation in the homogeneous risk region of BRCA1 and BRCA2 in 990 women in the French study GENEPSO by using a weighted Cox regression model. Results Our results confirm the existence of the protective effect of an increasing number of full-term pregnancies (FTPs) toward BC among BRCA1 and BRCA2 mutation carriers (≥3 versus 0 FTPs: hazard ratio (HR) = 0.51, 95% confidence interval (CI) = 0.33 to 0.81). Additionally, the HR shows an association between incomplete pregnancies and a higher BC risk, which reached 2.39 (95% CI = 1.28 to 4.45) among women who had at least three incomplete pregnancies when compared with women with zero incomplete pregnancies. This increased risk appeared to be restricted to incomplete pregnancies occurring before the first FTP (HR = 1.77, 95% CI = 1.19 to 2.63). We defined the TMAP score (defined as the Time of Breast Mitotic Activity during Pregnancies) to take into account simultaneously the opposite effect of full-term and interrupted pregnancies. Compared with women with a TMAP score of less than 0.35, an increasing TMAP score was associated with a statistically significant increase in the risk of BC (P trend = 0.02) which reached 1.97 (95% CI = 1.19 to 3.29) for a TMAP score >0.5 (versus TMAP ≤0.35). All these results appeared to be similar in BRCA1 and BRCA2. Nevertheless, our results suggest a variation in BC risk associated with parity according to the location of the mutation in BRCA1. Indeed, parity seems to be associated with a significantly decreased risk of BC only among women with a mutation in the central region of BRCA1 (low-risk region) (≥1 versus 0 FTP: HR = 0.27, 95% CI = 0.13 to 0.55) (Pinteraction <10-3). Conclusions Our findings show that, taking into account environmental and lifestyle modifiers, mutation position might be important for the clinical management of BRCA1 and BRCA2 mutation carriers and could also be helpful in understanding how BRCA1 and BRCA2 genes are involved in BC.


Introduction
Carriers of mutations in the BRCA1 and BRCA2 genes are at very high risk of developing breast cancer (BC) and ovarian cancer. Estimates of the lifetime risk of developing BC for BRCA1 and BRCA2 mutation carriers range from 30% to 80% and from 9% to 84%, respectively [1]. Incomplete penetrance and the range of these risk estimates suggest the existence within families of genetic or shared environmental or lifestyle factors that modify the risk of BC.
Many studies have established that women who had their first full-term pregnancy (FTP) at a young age have a lower risk of BC than nulliparous women or women who had their first FTP when they were older than 30 years of age; additional pregnancies are associated with even lower risks (for example, [2,3]). Long-term breastfeeding is also associated with a decreased risk of BC in the general population [4]. Controversial conclusions have been drawn from studies that have examined the risk of BC associated with incomplete pregnancies. While some older studies found a possible positive association between interrupted pregnancies and BC risk [5][6][7][8][9], the most recent meta-analyses concluded that an increased number of either spontaneous or induced abortions was not associated with an increased BC risk [10][11][12].
Some authors have suggested that the effect of pregnancies in BC development is related to the breast mitotic activity, driven by estrogen and progesterone exposure [24]. This activity appears high during the first three months of pregnancy and is followed by a dramatic decrease and by the differentiation of breast tissue during the last six months [25]. Although lasting and high mitotic activity and incomplete differentiation of breast tissue may have a critical effect on cells with inherited mutations, no study has assessed the effect of breast mitotic activity during pregnancy in BRCA1 and BRCA2 mutation carriers.
Genotype-phenotype correlations have been found in both BRCA1 and BRCA2 showing heterogeneity in BC risk according to the location of the mutation (for example, [26][27][28][29]). Moreover, inconsistencies in the effect of pregnancy-related factors among BRCA1 and BRCA2 mutation carriers between studies could be explained by an additional heterogeneity due to a differential effect of these factors according to location of the mutation. Thus, we first studied the effect of pregnancy-related factors on the risk of BC for BRCA1 and BRCA2 mutation carriers taken together, and by gene. Then we studied the effect of parity, incomplete pregnancies and breast-feeding for homogeneous regions previously described in our data [30] where a central low BC risk region in BRCA1 and BRCA2 was confirmed [27,28,[31][32][33][34], and a new high-risk region in BRCA2 was described [30].

Data
The GENEPSO study was initiated in 2000 to estimate the risk of breast, ovarian, and other cancers in BRCA1 and BRCA2 mutation carriers and to assess potential risk-modifying factors, either lifestyle or genetic. Subjects were ascertained from the family cancer clinics of the Genetic and Cancer Group of Unicancer. Any woman who was known to carry a deleterious mutation in the BRCA1 or BRCA2 gene was eligible, including those diagnosed with cancer and those currently unaffected. They had to be at least 18 years old, mentally capable of giving informed consent to participate in the study, and had been counseled about their mutation status. The research protocol was approved by the relevant ethics committees, and all participants provided written informed consent.
The study population was based on the women enrolled in the GENEPSO study from 2000 to 2010. A total of 1,337 women (from 987 different families) were recruited, 863 (65%) were BRCA1 mutation carriers and 474 (35%) were BRCA2 mutation carriers. To assess variation in BC risk according to mutation position, a sample with one subject per family was randomly selected to avoid overmatching on the mutation, except for one family where two related women carried two different mutations and thus were considered independent. Additionally, two women were counted twice because they carried two mutations in BRCA1 and BRCA2. Thus, 990 women were considered for assessing risk factor main effects and for the analyses by mutation location.
A standardized questionnaire on reproductive factors and lifestyle factors was administered to the study subjects by mail. The questionnaire collected detailed information on pregnancy history. Subjects who indicated that they had at least one pregnancy were asked to provide, for each pregnancy, the month and year when the pregnancy started or was terminated, its duration, and its outcome (live birth, still birth, miscarriage, induced abortion), and the duration of breastfeeding, if applicable.

Genotyping
The mutation screening strategy was similar for all the clinics, that is, the youngest living affected family member was tested first and, if a BRCA1 or/and BRCA2 mutation was found, affected and unaffected family members were offered testing. Mutations were defined as deleterious when their putative protein products were truncated, that is, nonsense mutations and frameshift mutations (nucleotide insertions or deletions, large gene rearrangements, and splicing defects). Some mutations, without disruption of the reading frame, were considered deleterious when they were classified deleterious by the ENIGMA group (Evidence-based network for the interpretation of germline mutant alleles) [35].
The full coding sequences and the exon-intron junctions of the BRCA1 and BRCA2 genes were screened for variants, based on pre-screening (denaturing gradient gel electrophoresis (DGGE), single strand conformation polymorphism (SSCP), protein truncation assay (PTA), denaturing high performance liquid chromatography (dHPLC), high resolution melting (HRM), or enhanced mismatch mutation analysis (EMMA)) and sequencing. Several large rearrangements were identified by large cDNA sequencing, multiplex ligation-dependent probe amplification (MLPA) [36], quantitative multiplex PCR of short fragments (QMPSF) [37], quantitative PCR (qPCR) [38], qPCR HRM [39], EMMA [40], bar code screening [41] or dedicated array comparative genomic hybridization (CGH) [42]. Mutation description was provided by each French laboratory, coded and standardized according to the international nomenclature [See Additional file 1 for the distribution of mutations in the study].

Statistical methods
The data presented here were analyzed using a modified Cox proportional hazards regression model. Standard Cox regression may lead to biased estimates of the hazard ratio (HR) because the women in this study were taken from high-risk families qualifying for genetic testing. The disease status may, therefore, have affected the likelihood of ascertainment and selection leading to an over-sampling of affected women. To correct for this potential bias, the Cox regression analyses were performed using the weighted regression approach described by Antoniou et al. [43]. Individuals were weighted such that the observed BC incidence rates in the study sample were consistent with established BC risk estimates for BRCA1 and BRCA2 carriers [1]. The affected mutation carriers were underweighted (weights <1) and the unaffected mutation carriers were overweighted (weights >1). The weights were applied to all person-years of each subject in the modified Cox model.
Subjects were followed up from birth and censored at the date of diagnosis, for women who were affected by any cancer, or the date of prophylactic bilateral mastectomy or interview, for unaffected women.
Parity, breast-feeding, incomplete pregnancies, menopausal status and oral contraceptive use changed over time, so it was analyzed as a time-dependent covariate and cumulative over life time. All analyses were stratified by period of birth (before 1940, 1940 to 1949, 1950 to1959, 1960 or later). In addition, because menopausal status, oral contraceptive use and gene may substantially modify the risk of BC and thus be a potential confounder, analyses were adjusted for these factors.
To avoid the potential bias due to BC detected during a pregnancy which may cause a bias either toward or away from the null depending on the effect of pregnancy on the risk of BC, pregnancies were included only if they occurred at least one year before the age at censure. Thus, we excluded ten pregnancies, seven among affected women and three among unaffected women.
To assess the variation of BC risk associated with pregnancies and breast-feeding by location of truncating mutations in BRCA1 and BRCA2, we used regions previously defined as homogeneous in BC risk by Lecarpentier et al. We considered two groups of mutation in BRCA1, those located in LR1 (for 'low-risk region in BRCA1': codons 374 to 1161) and those located outside LR1. In BRCA2, we considered three groups of mutation in BRCA2, those located in LR2 (for 'low-risk region in BRCA2': codons 957 to 1827), located in HR2 (for 'high-risk region in BRCA2': codons 2546 to 2968) and those located outside LR2 and HR2 [30]. Heterogeneity in risk by mutation location was assessed by testing the interaction between mutation location and the risk factor of interest.
All statistical analyses were two-sided and were performed using the STATA statistical package (version 10; Stata Corporation, College Station TX).

Results
Characteristics of the whole cohort and of one-womanper-family cohorts are listed in Table 1. A total of 563 women had been diagnosed with BC at the time of their interview, but only 499 of them were considered as affected in this analysis after censoring. The remaining 838 women were censored at age of diagnosis of ovarian cancer (N = 89), at diagnosis of another cancer (N = 16), at prophylactic bilateral mastectomy (N = 11), or at interview (N = 722). The average age at censoring for the 838 participants without BC was 40.0 years (standard deviation (SD) = 0.4), which is similar to the age at diagnosis of the women with BC (41.0 years, SD = 0.4), although the age at interview was substantially higher for the BC patients, reflecting the pattern of genetic testing among participants. Sampling of one woman per family did not change any characteristic distribution or the average of age at censure (39.8 years, SD = 0.5 and 40.4 years, Year of birth, number of full-term and incomplete pregnancies, breast-feeding, menopausal status, and oral contraceptive use are also described. There was a total of 39,666 person-years of observation. The estimated risks of BC associated with parity, age at first FTP, and history of breast-feeding from the weighted Cox regression analysis are summarized in Table 2, both for the entire sample and for BRCA1 and BRCA2 mutation carriers separately. We also analyzed the parity according to attained age (40 years or younger versus older than 40 years).
Overall, compared with nulliparous women, parous women had a slightly lower but non significant risk of BC (HR = 0.77, 95% CI = 0.53 to 1.13). As the number of FTPs increased there was a statistically significant decrease in the risk of BC (P trend < 10 -3 ). The reduction in risk was estimated with an HR = 0.51 (95% CI = 0.33 to 0.81) for women with at least three FTPs. This association remained significant only for the women who were older than 40 years (for women with at least three FTPs, HR = 0.35, 95% CI = 0.17 to 0.70). Among parous women, age at first FTP seems to be associated with BC risk. Indeed, women who had their first FTP when they were 25 years or older had a lower HR point estimate of BC than women who had their first FTP when they were younger than 20 years (between age 25 and 30 versus before age 20, HR = 0.62, 95% CI = 0.36 to 1.06 and after age 30 versus before age 20, HR = 0.67, 95% CI = 0.36 to 1.23). The reduction in risk associated with parity and age at first FTP was similar for carriers of BRCA1 and BRCA2 mutations. After adjusting for parity, we observed no association between ever having breast-fed and BC risk, either for the entire sample or separately for BRCA1 or BRCA2 mutation carriers. There was also no statistically significant association between duration of breastfeeding and BC risk even for long duration (that is, ≥10 months) (data not shown).
The estimated risks of BC associated with incomplete pregnancies, from the weighted Cox regression analysis, are summarized in Table 3. First, HR point estimates suggest an association between incomplete pregnancy (induced abortions and miscarriages considered together) and a higher BC risk in the entire sample (≥1 versus 0: HR = 1.28, 95% CI = 0.98 to 1.67), with a maximum risk among women who had at least three incomplete pregnancies (≥3 versus 0: HR = 2.39, 95% CI = 1.28 to 4.45). HR point estimates seem similar whatever the type of incomplete pregnancy (induced abortions or miscarriages) but were not significant. However, as the number of incomplete pregnancies increased, there was a statistically significant increase in the risk of BC for induced abortions (P trend = 0.02), but not for miscarriages. The maximum risk was observed among women who had at least three induced terminations (≥3 versus 0: HR = 3.84, 95% CI = 1.52 to 9.66). Among women who had induced terminations, an age of 20 years or older at first incomplete pregnancy led to a lower risk of BC than an age younger than 20 years (after age 20 versus before HR = 0.50, 95% CI = 0.28 to 0.90). When we considered this risk with respect to the first FTP, the association previously found persisted, but only before the first FTP (HR = 1.77, 95% CI = 1.19 to 2.63). Interestingly, point estimates associated with having miscarriages in the first three months of pregnancy were similar to those associated with induced termination (HR = 1.35, 95% CI = 0.95 to 1.93 and HR = 1.30, 95% CI = 0.93 to 1.82, respectively). There were no differences when stratified by gene.
To take into account simultaneously the contrary effect on BC risk of FTPs and pregnancies interrupted within the first three months, we determined the TMAP score defined as the Time of breast Mitotic Activity during Pregnancies. The TMAP score is the sum of pregnancies with a duration greater than or equal to three months multiplied by three plus the sum of the duration (in months) of each pregnancy with a duration of less than three months divided by the sum of the duration of each pregnancy whatever the outcome of the pregnancy. The TMAP score is a time-dependent variable.
Compared with women with a TMAP score of less than 0.35, an increasing TMAP score was associated with a statistically significant increase in the risk of BC (P trend = 0.02) and reached 1.97 (95% CI = 1.19-3.29) when the TMAP score was greater than 0.5.
Estimated risks of BC associated with parity, age at first FTP, and history of breast-feeding according to the mutation location were assessed by regions of BRCA1 and BRCA2 previously defined as homogeneous for the risk of BC [30]. Among BRCA2 mutation carriers, no variation of BC risk was found (data not shown). Estimated risks of BC associated with parity and incomplete pregnancy by homogeneous region in BRCA1 are shown in Table 4. Parity seems to be associated with a significantly decreased risk of BC only among women with a mutation in LR1 (≥1 versus 0 FTP: HR = 0.27, 95% CI = 0.13 to 0.55) (P interaction <10 -3 ). Similarly, an increasing number of FTPs was associated with a statistically significant decrease in the risk of BC only in LR1 (HR = 0.20, 95% CI = 0.08 to 0.49 for women with at least three FTPs compared with nulliparous women). This protective effect persists whatever the age (HR = 0.33, 95% CI = 0.16 to 0.68 and HR = 0.21, 95% CI = 0.09 to 0.51 before and after age 40 respectively). The HR associated with breast-feeding did not differ between LR1 (ever versus never: HR = 0.73, 95% CI = 0.32 to 1.66) and outside LR1 (HR = 0.86, 95% CI = 0.58 to 1.27) (data not shown). There was also no significant interaction between incomplete pregnancy or age at first FTP and mutation location (data not shown).   a Not including missing data. b Adjusted for parity (0, 1, 2, ≥3), menopausal status (yes, no), oral contraceptives (never, ever), and gene mutated (BRCA1, BRCA2). c Adjusted for parity (0,1 2, ≥3), no. of incomplete pregnancies (0, 1, 2, ≥3), menopausal status (yes, no), oral contraceptives (never, ever), and gene mutated (BRCA1, BRCA2). d Adjusted for menopausal status (yes, no), oral contraceptives (never, ever), and gene mutated (BRCA1, BRCA2). HR, hazard ratio; CI, confidence interval. a Not including missing data. b Adjusted for menopausal status (yes, no) and oral contraceptives (never, ever). c Adjusted for parity (0, 1, 2, ≥3), menopausal status (yes, no) and oral contraceptives (never, ever). HR, hazard ratio; CI, confidence interval.

Discussion
Our results confirm the existence of a protective effect of an increasing number of FTPs toward BC among BRCA1 and BRCA2 mutation carriers. This risk reduction, however, appeared to be significant only for women older than 40 years. Additionally, we found some evidence of an association between pregnancies interrupted within the first three months (induced or spontaneous) and an increased risk of BC. This increased risk appeared to be restricted to incomplete pregnancies occurring before the first FTP. Whatever the outcome of the pregnancy, the results show that a first pregnancy before age 20 was associated with a higher risk of BC than a first pregnancy occurring later. We defined the TMAP score to take into account simultaneously the contrary effect of full-term and interrupted pregnancies. We found a significant positive association between the TMAP score and BC risk. All these results appeared to be similar in BRCA1 and BRCA2. Nevertheless, our results suggest a variation in BC risk associated with parity according to the location of the mutation in BRCA1.
Our study has several limitations. First, our results are based on retrospective information obtained from women who opted for BRCA1 and BRCA2 mutation screening and genetic testing. One assumption that underlies the method of weighting used in our analyses is that the absolute disease risks are well estimated and ascertainment is not dependent on the covariates of interest [43]. This assumption would be violated if any of the factors related to pregnancies changed the likelihood that women might opt to undergo genetic testing. We are unaware of any study that has assessed whether a woman's uptake of genetic testing differs according to these factors and we cannot assess this potential bias [13].
Second, since our data used prevalent cases with some women being interviewed a long time after their BC diagnosis, we cannot exclude that our findings on parity, breast-feeding and incomplete pregnancies are affected by a potential survival bias. However, we could not detect it in our data by performing extra analyses on subsamples of individuals diagnosed or censured within the five-year period before their interview, with a follow-up being counted only during this five-year period. We did not observe differences in our results using this pseudo-incident cohort.
It is well established that increasing parity and early age at first birth are associated with a lower risk of developing BC in the general population. There is evidence that the protective effect of parity may be restricted to women who are over 40 years old [44][45][46][47]. The relationship between pregnancy and risk of BC in BRCA1 and BRCA2 carriers is less clear in the earliest publications [15,16,20,21,23]. Our results are more in line with more recent studies [13,17,22,48] which found a decreased risk associated with an increasing number of FTPs among BRCA1 and BRCA2 mutation carriers. In agreement with our findings, three of these studies showed a reduced risk of BC only after age 40 years [13,22,48]. Among the studies which assessed the risk of BC associated with the age at first FTP [13,15,17,[19][20][21][22]48] results are inconsistent and only two studies found a reduced risk among BRCA1 or BRCA2 mutation carriers associated with a first FTP after age 20 [13,20]. In contrast with our results, the International BRCA1/2 Carrier Cohort Study (IBCCS) study [13] found a variation in this risk by gene mutated. They found that a first FTP after the age of 30 years was associated with a significant decrease in BC risk in BRCA1 and a significant increase in BRCA2 mutation carriers. Antoniou et al. [48] subsequently carried out a similar analysis on 789 BRCA1/2 mutation carriers from the UK and found that in BRCA2 mutation carriers the risk is higher for those who have their first FTP later, that is, after age 30. We did not find such a variation although our data overlap for about one quarter of our subjects (319 out of 1,337) with those of the IBCCS study.
A number of studies have examined the risk of BC associated with interrupted pregnancies, but there has been some controversy in the past. A collaborative reanalysis of data from 53 epidemiological studies, including 83,000 women with BC from 16 countries, described inconsistent findings across studies and difficulties in evaluating these associations. It was concluded that BC risk did not appear to be associated with an increased number of either spontaneous or induced abortions [10]. Similar results were obtained subsequently from a prospective study of young women [12]. However, numerous studies have suggested that interrupted pregnancies may moderately increase the risk of BC [5][6][7][8][9]49]. Few studies have examined this association in BRCA1 and BRCA2 mutation carriers. Two studies concluded that BC risk did not appear to be associated with an increased number of either spontaneous or induced abortions [13,15]. Furthermore, Friedman et al. observed that among BRCA2 mutation carriers, two or more therapeutic abortions resulted in a 64% decrease in BC risk, but not among BRCA1 mutation carriers [14]. In 1995, evidence was found that the relative risk conferred by a family history of BC increased with the number of interrupted pregnancies and that this risk was highest for those who had an interrupted pregnancy before the first FTP [50]. Our findings seem consistent with this study. Although, as in many previous studies (for example [51]), a recall bias where BC cases declared interrupted pregnancies more often than controls, would lead to a BC bias away from the null hypothesis. Indeed, we found an increased BC risk associated with an increasing number of induced abortions. However, this risk appeared to be restricted to pregnancies with induced interruptions before the first FTP. This effect may be because the differentiation of mammary cells which occurs during an FTP [52] prevents the carcinogenic effect of subsequent interrupted pregnancies. In addition, our results indicate that spontaneous abortions occurring in the first three months were associated with an increased risk of BC. The difference in risk according to the pregnancy outcome (interrupted versus full-term) and according to the duration of interrupted pregnancy, whatever the nature of the interruption, and our TMAP scores highlight the importance of the duration of pregnancy as a BC risk factor. This is also illustrated by the findings of Vatten et al. [53] who reported that the shorter the length of gestation, the higher the BC risk, in a cohort of about 695,000 women. This score could be useful for the individual estimation of BC risk.
When stratified by homogeneous regions, our results suggest a variation of the BC risk associated with parity according to mutation location in BRCA1, but not in BRCA2. This is the first time that the effects of pregnancy-related factors according to mutation location have been studied. Although, the significance might occur by chance because of a limited power, parity seems to be associated with a significantly decreased risk of BC among women with a mutation in the LR1 region, but not outside this region. Therefore, pregnancies seem to have the same protective effect in LR1 as in the general population, while outside LR1 parity does not seems to have an effect on BC risk.
Although there is no obvious biological hypothesis to explain this variation, one can expect that BRCA1 acts during pregnancy. Indeed, BRCA1 is also involved in cellular anti-proliferation via inhibition of the transcriptional activity of estrogen receptor α (ERα) [54][55][56]. Interestingly, this mechanism is postulated to occur through a proteinprotein interaction involving domains of BRCA1 corresponding to regions outside of LR1: that is, the N-terminus (amino acids 1-300) and the C-terminal region [54]. In addition, Ma et al. [57] provide evidence for a difference in some hormone-related risk factor profiles between triple negative (TN) and other BC subtypes, especially, in line with a protective effect of parity in all subtypes except in TN. Thus, it would be of interest to study the relation between mutation location and the tumor subtype to determine whether the TN tumors are more often associated with mutations located outside LR1.

Conclusions
This study confirms the existence of a protective effect of FTPs toward BC among BRCA1 and BRCA2 mutation carriers which is restricted to women with mutation in the LR1 region for BRCA1 mutation carriers. We also showed the importance of the duration of pregnancies as a BC risk factor. If our findings are confirmed, taking into account environmental and lifestyle modifiers, mutation position might be important for the clinical management of BRCA1 and BRCA2 mutation carriers and could also be helpful in understanding how BRCA1 and BRCA2 genes are involved in BC.

Additional material
Additional file 1: Distribution of mutations. Distribution of mutations found in the population under study by gene and by type: Truncating mutations (non-sense mutations, frameshift, and all other type of mutations leading to a truncated protein) and 'other type' (missense mutations, in-phase skipping, large rearrangements, partial and entire gene deletions).