Relationships between Breast Feeding and Breast Cancer Subtypes: Lessons Learned from Studies in Humans and in Mice

A large body of literature over the years has demonstrated that a number of breast cancer risk factors are related to reproductive and hormonal characteristics, with summaries generally noting that both nulliparity and late age at first full-term pregnancy are associated with increased risk (1), with some reduction with breastfeeding (2). A meta-analysis of data from 47 epidemiologic studies including more than 50,000 women with breast cancer and almost twice as many controls showed that having children was associated with decreased risk of breast cancer, with the decreased risk with parity greatest among women who breastfed (3). RRs for women with one, two, three, four, and five children were 1.0, 0.94, 0.86, 0.84, and 0.73, respectively, for women who never breastfed, compared with risks of 0.97, 0.93, 0.83, 0.73, and 0.64 among women who breastfed.

Even from very early studies, however, it was clear that relationships between reproductive risk factors and breast cancer etiology are complex. For example, there were observations of transient increases in risk following a pregnancy (4–7), and the effects of parity appeared to vary depending upon the age at breast cancer diagnosis. Studies by both Brinton and colleagues (8) and Mayberry and Stoddard-Wright (9) found that for African-American/Black women, parity was protective only for those diagnosed after age 40. In fact, among Black women less than 40 years of age, parity actually increased the risk of breast cancer (8). As reviewed by Pathak (10), it was thought that this dual effect of parity could be related to effects on proliferation of initiated cells (to increase risk), countered by inducing terminal differentiation of terminal endbuds, yielding them less susceptible to DNA damage over time, as shown by Russo and colleagues (11, 12). According to this model, the association of breastfeeding with reduced risk could be attributed to its role in the maximal differentiation of the mammary gland.

Relationships between parity, breastfeeding, and breast cancer risk were put into clearer context with advances in understanding the molecular underpinnings of breast tumors, first with classification of cancers by estrogen receptor (ER) status, and then with gene expression and subsequent IHC profiling to distinguish five distinct subtypes: luminal A (ER⁺ or PR⁺ and HER2⁻), luminal B (ER⁺ or PR⁺ and HER2⁺), HER2 overexpressing, (ER⁻, PR⁻, and HER2⁺), basal-like (ER⁻, PR⁻, HER2⁻, and EGFR⁺ or CK 5/6⁺), or triple negative (ER⁻, PR⁻, and HER2⁻) breast cancer (TNBC), which has poorer prognosis than other subtypes, and unclassified (negative for all five markers; refs. 13, 14). Classification by five or more intrinsic subtypes may have important implications for cancer treatment and its outcomes, but as proposed by Anderson and colleagues, in studies of breast cancer risk, it is likely more appropriate to consider breast cancer as having two distinct etiologic subtypes, which essentially correspond to ER⁺ and ER⁻ disease (15). Of interest, a recent analysis of data from 15 prospective cohort studies addressed the question of a crossover effect of childbirth after pregnancy on risk, with consideration of breastfeeding and according to ER status (16). In that study, the investigators found that risk for ER⁻ breast cancer was highest 2.2 years after birth [HR, 1.77; 95% confidence interval (CI), 1.34–2.33]. Risk decreased to a HR of 1.38 (95% CI, 1.01–1.88) at 34.5 years after birth, but did not crossover to inverse associations. Risk for ER⁻ breast cancer was highest for parous women, regardless of breastfeeding. With longer follow-up, risk with parity was reduced for ER⁺ breast cancer, but remained elevated for ER⁻ disease. This was somewhat consistent with the early studies that observed crossover according to age at diagnosis, because women diagnosed before age 40 are more likely to have ER⁻ breast cancer than older women (17, 18).

ER⁻ and TNBC are more common in Black women than in other U.S. populations; until recently, there had been few explanations for this phenomenon. It was not until there were fairly large studies with sufficient numbers of Black women with ER⁻ breast tumors that associations between risk factors and higher incidence of ER⁻ breast cancer could be clearly elucidated. A landmark article from the Carolina Breast Cancer Study (CBCS) in 2008, which included Black as well as White women with breast cancer and noncancer controls, was one of the first to draw attention to these differences in risk factors by breast cancer subtypes (19). In that study, Millikan and colleagues found that compared with women who were nulliparous, having children was associated with reduced risk of luminal A (ER⁺/PR⁺/HER2⁻) breast cancer, with similar ORs of 0.70 for one, two, and three or more children. ORs remained the same regardless of breastfeeding. However, for basal-like breast cancer, similar to TNBC, there was almost a 2-fold increase in risk with parity [OR for 1–2 children, 1.8 (95% CI, 1.1–1.3) and OR for three or more children, 1.9 (95% CI, 1.1–3.3)]. Importantly, this risk was only observed among women who had not breastfed; increased risk associated with parity was ameliorated if women breastfed [OR for 1–2 children, 1.1 (95% CI, 0.6–1.3) and OR for three or more children, 1.3 (95% CI, 0.7–2.3)].

These results were subsequently replicated in additional studies of breast cancer in Black women. In the Black Women's Health Study (BWHS), risk of ER⁺/PR⁺ breast cancer was reduced with parity, regardless of breastfeeding; however, there was a 50% increase in risk of ER⁻/PR⁻ disease among women who had three or more children and did not breastfeed (OR, 1.5; 95% CI, 1.1–2.2; ref. 20). This increased risk was greatly reduced with breastfeeding (OR, 1.1; 95% CI, 0.8–1.7). Similarly, among Black women in the Women's Circle of Health Study (WCHS), parity reduced the risk of ER⁺ breast cancer, but increased the risk of ER⁻ and TNBC, with ORs similar to those observed in CBCS among women who did not breastfeed (OR, 1.9; 95% CI, 0.99–3.72; ref. 21). Although these findings were consistent, sample sizes were small and some risk estimates were unstable, making it difficult to draw strong conclusions. Thus, the lead investigators of CBCS, BWHS, WCHS, and the Multiethnic Cohort formed the AMBER Consortium to pool data and samples to have adequate statistical power to be able to investigate risk factors for aggressive breast cancer in Black women (17). Analysis in AMBER confirmed and extended results from smaller studies; with data from almost 4,000 Black women with breast cancer and 14,000 controls, results showed that, while having children was associated with reduced risk of ER⁺ breast cancer, it actually increased risk of ER⁻ and TNBC (22). Importantly, breastfeeding appeared to greatly reduce the increased risk of ER⁻ cancer with parity. For example, having four or more children was associated with more than 60% increased risk of TNBC among women who did not breastfeed (OR, 1.68; 95% CI, 1.15–2.44), but no increased risk among women who breastfed (OR, 1.08; 95% CI, 0.65–1.77).

More recent large studies in other populations have also observed reduced parity-associated risk of ER⁻ breast cancer with breastfeeding. For example, analyses from the Nurses' Health Study, including 12,452 women with breast cancer (8,235 ER⁺ and 1,978 ER⁻), showed that having children reduced risk of ER⁺ (HR, 0.82; 95% CI, 0.77–0.98), but not ER⁻ disease (HR, 0.99; 95% CI, 0.94–1.05; ref. 23). Relationships with ER⁺ breast cancer were observed regardless of breastfeeding. Basal-like breast cancer was highest among women with higher parity (>2 children) who never breastfed (HR, 1.43; 95% CI, 0.92–2.23), with no associations for those who did breast feed their infants (HR, 1.05; 95% CI, 0.70–1.57). A pooled analysis of nine cohort studies found that parity was also associated with reduced risk of ER⁺, but increased risk of ER⁻ and TNBC (24). Data were not available for breastfeeding in that pooled analysis, but an earlier review and meta-analysis of 27 studies with 36,881 breast cancer cases found that there were inverse associations with breastfeeding for ER⁻ (OR, 0.90; 95% CI, 0.82–0.99) and TNBC (OR, 0.78; 95% CI, 0.66–0.91), in addition to ER⁺ or PR⁺ subtypes (OR, 0.89; 95% CI, 0.80–0.99; ref. 25). Similar results were observed in a large case–case analysis, with reductions in odds of TNBC with breastfeeding compared with hormone responsive cancers (OR, 0.90; 95% CI, 0.82–0.99; ref. 26), and a large pooled analysis with 558 TNBCs and more than 5,000 controls found that for younger parous women (<50 years), odds of TNBC were 2-fold higher for women who never breastfed (OR, 2.02; 95% CI, 1.12–3.63; ref. 27). Risk estimates were greatest for women with three or more pregnancies and no or fewer months of breastfeeding (OR, 2.56; 95% CI, 1.22–5.35). Similar relationships with TNBC were observed in a pooled analysis, with 82% lower risk for Black women <44 years of age who had breastfed compared with those who had not (28).

In our research in AMBER, in addition to findings regarding parity and breastfeeding, we also found differential relationships by ER status with age at menarche (29). Earlier age at menarche was associated with increased risk of ER⁺ breast cancer only among parous women, and the risk was greatest for those with the longest period of time between menarche and first full-term birth. However, early menarche increased risk of ER⁻ disease regardless of whether or not women were parous. These findings support the paradigms of etiologic heterogeneity discussed above in relation to age at first full-term pregnancy. Relationships of early menarche with ER⁺ breast cancer only among women with children and dependent upon time between menarche and childbirth is consistent with the Russo model of susceptibility of the breast to damage before pregnancy-induced differentiation of terminal end buds in the ducts. However, for ER⁻ disease, it appears that menarche alone may be a critical event for the development of ER⁻ breast cancer versus the importance of time between menarche and reproduction for ER⁺ disease.

There is reason to consider how events in early years, such as menarche, parity, and breastfeeding, could affect later development of ER⁻ versus ER⁺ breast cancer. One potential mechanism is possibly through effects on progenitor cells in the mammary gland. There is substantive evidence that most breast tumors, both luminal (ER⁺) and basal-like (ER⁻), arise from luminal progenitor cells (reviewed in ref. 30). It has been proposed that cancer risk factors may alter the number and/or properties of progenitor cells (31), and that some cell populations may be exquisitely sensitive to damage during puberty. It is clear from the data from survivors of atomic bombings in Japan in World War II that exposure to ionizing radiation during adolescence, not adulthood, led to subsequent breast cancer, similar to breast cancer following radiation to the chest for Hodgkin disease in young adults (32, 33). Importantly, breast cancers arising from childhood treatment with radiation tend to be ER⁻ and/or TNBC and also to have more aggressive characteristics (34). Investigating mechanisms behind this phenomenon, Tang and colleagues showed with computational modeling and mouse studies that irradiation during puberty increases stem cell self-renewal and increases susceptibility to developing ER⁻ breast cancer (35). Hormonal exposures at a young age may similarly affect progenitor cells and play a role in the etiology of ER⁻ breast cancer. However, factors that influence the genesis of distinct ER subtypes from this common pool of progenitors are not well-defined. Similarly, knowledge of how breastfeeding ameliorates the observed increased risk of ER⁻ breast cancer is in its infancy.

Although parity has an overall protective effect against breast cancer in later life, studies have shown that childbirth at any age bestows an increased risk of cancer in the first decade postpartum, as discussed above. Moreover, these postpartum breast cancers are often more aggressive with poorer outcomes (36, 37). Following lactation, the mammary gland undergoes a massive remodeling process including programmed cell death of most epithelial cells, regeneration of adipose tissue, and infiltration of multiple immune cell types (38, 39). Through a number of elegant studies over the last 2 decades, Borges and colleagues have dissected the role that involution plays in the generation of more aggressive breast cancers within the decade following childbirth (reviewed in ref. 40). These include increased inflammation and lymphangiogenesis and modification of the extracellular matrix, all characteristics of “wound healing.” In addition, they provided evidence that cells from very early-stage tumors cells can be released into the stroma during involution where they have access to vasculature (41, 42), and also demonstrated that breast involution is accompanied by remodeling in the liver, making this organ more amenable to cancer cell seeding and overt metastasis. Gene expression studies of mouse mammary glands undergoing involution have shown differential expression consistent with many of the observed biochemical changes and tissue remodeling (39, 43, 44). Thus far, however, these gene expression profiles show only moderate correlation with any particular aggressive subtype of cancer, specifically with inflammatory breast cancer (45).

As discussed above, parity has long been known to reduce lifetime risk of breast cancer (46, 47) as well as enhance the efficiency of milk production in second pregnancies (48, 49). These observations suggest the existence of a long-term memory effect induced in mammary tissue by pregnancy. As proposed by Russo and Russo, the most likely mechanism for this memory effect is epigenetic reprograming in the mammary gland as a result of pregnancy hormone–induced differentiation and tissue remodeling (50, 51). For example, DNA methylation in the mouse mammary gland is affected by the milieu of hormonal changes that occur during puberty, development, pregnancy, and lactation (52–54). Using a targeted bisulfite sequencing approach and DNA from total mammary gland, Katz and colleagues identified hundreds of genes that were differentially methylated between parous and nulliparous mice, some of which persisted over time (52). Importantly, this study identified parity-associated increased methylation at the insulin-like growth factor 1 receptor (Igf1r) and other members of the IGF signaling pathway, supporting the notion that suppression of this pathway is related to the parity-associated protective effect against breast cancer. Using reduced representation bisulfite sequencing and flow-sorted mammary epithelial cells, Huh and colleagues showed that pregnancy had the most prominent effect on cell populations enriched for mammary epithelial stem cells (MaSC) and luminal progenitors, and that many of these methylation changes were associated with downregulation of genes and pathways important in stem cell function (e.g., Hedgehog and TGFβ), a finding also possibly relevant to pregnancy-induced reduction in breast cancer risk (53). By carrying out whole-genome bisulfite sequencing of flow-sorted mammary epithelial cells, dos Santos and colleagues demonstrated that pregnancy results in long-term methylation changes, some of which affect genes important in mammary gland development, lactation, and involution, suggesting that a first pregnancy “primes” the mammary gland for a more rapid response to subsequent pregnancies (54). Changes in DNA methylation accompany and are critical to most differentiation processes (55). Hence, the vast majority of these methylation changes probably reflect normal differentiation-associated modifications in the epigenetic state of mammary epithelial cells and/or differences in the relative proportions of distinct cell populations brought about by the pregnancy and lactation cycles. However, none of these studies evaluated the effect of suckling on these methylation changes, because all animals were allowed to nurse normally.

One possible mechanism whereby reproductive events could influence whether luminal progenitor cells give rise to ER⁻ versus ER⁺ breast tumors is through errors in the establishment and maintenance of differentiation-associated DNA methylation states. We performed a two-stage genome-wide methylation study using the Illumina Infinium Human Methylation 450K Platform, first in a small analysis using fresh frozen samples from Black and White women (56), and then in formalin-fixed, paraffin-embedded tissues from 733 patients (57). In both studies, hierarchical clustering separated groups by ER status. The number of differentially methylated loci (DML) between Whites and Blacks was almost 2-fold higher in ER⁻ breast tumors compared with ER⁺, and there were more DMLs by ER status among Blacks. One of the top DMLs by ER subtype in both studies, which was highly correlated with gene expression, was FOXA1. This pioneer transcription factor, important in mammary gland development, induces expression of a multitude of downstream genes that promote the luminal phenotype in luminal progenitors, while also directly repressing the basal phenotype (58). In our data, FOXA1 was hypermethylated and significantly downregulated in ER⁻ tumors, with the highest methylation and lowest expression in ER⁻ breast cancers from Black women. Moreover, the degree of methylation at FOXA1 in ER⁻ tumors was positively associated with the number of children born to these women; importantly, this increase in methylation at FOXA1 was significantly reduced in women who had breastfed (57). Consistent with methylation results, we found that FOXA1 protein expression in breast tumors was lowest among parous Black women who had not breastfed, with associations most pronounced for ER⁻ breast cancer (59). In addition, in two studies of breast tissue from women undergoing reduction mammoplasty, FOXA1 was hypermethylated in normal breast tissue in parous compared with nulliparous women (60, 61), suggesting that the observations made in breast tumors (56, 57) may be derived from changes in preneoplastic cells. It is interesting that BRCA1-deficient breast cancers, most of which are of the ER⁻ basal-like phenotype, are also thought to derive from an expanded population of aberrant luminal progenitors (62). Moreover, wild-type BRCA1 protein has been shown to positively regulate FOXA1 by preventing its DNA methylation and silencing (63). As illustrated in Fig. 1, it is therefore, plausible that BRCA1-mutated breast tumors are ER⁻ because mutation/inactivation of BRCA1 leads to methylation-induced silencing of FOXA1 in luminal progenitors, and that parity-associated methylation of FOXA1 has similar consequences.

GATA3 has also been shown to be pivotal in the differentiation of luminal progenitors to mature luminal cells (64). Together, GATA3 and FOXA1 proteins bind estrogen-responsive target genes to promote ERα function (65). GATA3 has been shown to promote differentiation and suppress metastasis in breast cancer, at least, in part, by inducing expression of miR-29b (66). Recently, in The Cancer Genome Atlas cohort of breast tumors, GATA3 was shown to be downregulated and hypermethylated in ER⁻ breast cancers compared with ER⁺ cases (65), a finding similar to our independent cohort (57). Interestingly, genetic depletion of either Gata3 (67) or Foxa1 (Sribenja and colleagues, unpublished) in the murine mammary gland results in a similar phenotype, skewing epithelial cell types toward luminal progenitors. It will be interesting to determine whether parity-associated methylation at FOXA1 and GATA3 occurs in the same tumor or are mutually exclusive events. Importantly, a recent report by Basree and colleagues demonstrated a similar increase in the proportion of luminal progenitors in glands from mice that had their pups removed prematurely after only 1 week of nursing (termed abrupt involution) as opposed to the typical 3–4 weeks (gradual involution), however, the methylation status of Foxa1 or any other genes was not assessed (68).

One can speculate how epigenetic reprogramming could occur by lack of breastfeeding and result in aberrant DNA methylation of FOXA1 and other genes in breast luminal progenitor cells. The methylation status of FOXA1/Foxa1 in MaSCs, the cells that give rise to mammary luminal progenitor cells, is not known. However, Foxa1 is expressed at very low levels in the basal/MaSC compartment of mammary epithelial cells (53), and similar to Elf5, another gene pivotal to luminal cell differentiation, may be methylated in MaSCs, but is demethylated during the transition to luminal progenitors (69). If an active demethylation mechanism is involved in this process (70), it is unlikely to be 100% efficient, and it is possible that some luminal progenitors are generated with Foxa1 still methylated and resistant to transcriptional activation and luminal differentiation. Expression of Myc protooncogene is essential for the proper function of mammary stem and progenitor cells (71). ElShamy has proposed the “oncogene elimination hypothesis” that posits that fully differentiated mammary gland cells expressing protooncogenes also express immune cell enlisting factors, as well as tumor-specific peptides presented in the context of MHC class, and that these cells would be detected and eliminated by infiltrating immune cells during involution (72). If terminal differentiation is blocked, for example, by methylation-associated silencing of FOXA1 due to abrupt involution brought about by shortened or a lack of breastfeeding, this putative “seek and destroy” mechanism may be impaired, allowing these cells to escape immune surveillance and destruction, and survive as potential precursors to ER⁻ breast cancer.

A second possibility, supported by the Basree and colleagues' study (68), is that lack of, or short-term, breastfeeding results in more rapid involution of the mammary gland with accompanying increased inflammation (73, 74), a condition previously shown to cause aberrant DNA methylation (reviewed in ref. 75); therefore, even if Foxa1 is in an unmethylated state in early luminal progenitor cells, abrupt involution could result in inflammation-induced aberrant methylation potentially leading to its silencing and differentiation arrest of luminal progenitor cells.

A third possibility concerns differences in the length of time that mammary epithelial cells are exposed to breast milk in women who breastfeed opposed to those who do not. Breast milk contains exosomes, “nanosized” membrane-bound vesicles secreted by various cell types that carry several types of macromolecules including proteins, lipids, and RNAs including miRNAs (reviewed in ref. 76). Exosomes have been implicated in several physiologic and pathologic processes and are thought be important for intercellular communication (77). In the mammary gland, exosomes are thought to be involved in regulating lactation and involution (78). DNA methylation is regulated, in part, by fluctuating levels of DNA methyltransferases (DNMT) and proteins that remove methyl moieties such as TET1, TET2, and TET3 (70). Intriguingly, milk exosomes carrying miR-29s and miR-148a negatively regulate DNMT3a/b and DNMT1, respectively (79, 80). Moreover, inhibition of miR-29s in primary bovine mammary epithelial cells resulted in global DNA hypermethylation, as well as increased promoter methylation of several lactation-related genes, including Elf5 (79). Because milk exosomes can be taken up by cultured mammary ductal epithelial cells and remain functional (81), it is conceivable that decreased exposure of mammary epithelial cells to milk exosomes in vivo could result in increased expression of DMNTs and consequent aberrant methylation.

Another potential mechanism by which parity and breastfeeding could be related to development of ER⁻ breast cancer is derived from the identification of pappalysin-1 (PAPPA) as a pregnancy-dependent oncogene (82). Using a transgenic mouse model (MMTV-PAPPA), the authors showed that overexpression of PAPPA during lactation and involution results in decreased IGFBP5a levels and increased IGF signaling in the mammary gland, as well as the occurrence of mammary tumors with a tumor-associated collagen signature (TACS-3). They went on to show that extended lactation (>2 weeks) allowed accumulation of high levels of glycoproteins, stanniocalcin-1 and -2 (STC1 and STC2), which inhibit PAPPA protease activity, resulting in abrogation of the phenotype due to aberrant PAPPA expression. To determine the significance of these finding in humans, Takabatake and colleagues also analyzed expression levels of PAPPA and IGFBP-5 by IHC in breast tumors from 46 premenopausal women (82). The analysis demonstrated expression of PAPPA in tumors from 79% of parous women (n = 28), but only 11% of nulliparous women (n = 18), with a significantly higher proportion of tumors with a TACS-3 signature in the parous group (P < 0.0001). Most relevant was the finding of an increased incidence of TNBC in the parous group (32%) relative to the nulliparous group (5.2%). The effect of breastfeeding in the parous group could not be determined because the duration of breastfeeding was not available.

Interestingly, Atkinson and colleagues (83) reported that normal adjacent tissue from women with TNBC contained a disproportionally higher level of cells with breast cancer stem cell characteristics in both ipsilateral and contralateral breast tissue as determined by both IHC staining and gene expression analysis (83). Importantly, these authors found that patients with these putative stem cells in their normal breast tissue were less likely to have breastfed or had a shorter duration of breastfeeding. Whether these potentially tumorigenic stem cells are related to our proposed differentiation-arrested progenitors remain to be determined.

There is still much to be learned about the role of breastfeeding in reduction of aggressive breast cancer. For example, how much breastfeeding is required to reduce risk of ER⁻ and TNBC? Are 3 months sufficient? 6 months? Is the first or last pregnancy the most important or is it the total months of breastfeeding, what is key for risk reduction? These questions can be addressed in large epidemiologic studies with detailed data on reproductive and breastfeeding histories, and also through studies in the laboratory to understand the effects of these variables on cell populations in the breast. It will be important in future population-based studies to also consider the potential confounding effect of age at menarche on relationships with breastfeeding, as well as potential interactions with other lifestyle factors, such as use of oral contraceptives, alcohol consumption, and body mass index.

Advances in technology may also lead to new avenues of investigation. In the future it is likely that, because of its exceptional resolution, application of new single-cell techniques will shed light on the mechanism(s) by which pregnancy without breastfeeding contributes to the increased occurrence of ER⁻ breast cancer in Black women. For example, single-cell RNA-sequencing should facilitate the identification of specific cell clusters within the larger luminal progenitor fraction of mammary epithelial cells (84) whose presence is associated with not suckling, and whose gene expression signature suggests a differentiation-arrested phenotype that may be predisposed to development of ER⁻ cancer upon additional genetic and epigenetic insult. A better understanding of the pathways involved in making luminal progenitors prone to ER⁻ tumors when transformed, and information on the mechanisms involved in mediating the differences in breast tissue involution and remodeling, will facilitate the identification of potential pharmacologic preventive measures for women who are unable or unwilling to breastfeed, particularly Black women, thereby reducing the prevalence of ER⁻ breast cancer and its inherent mortality.

No potential conflicts of interest were disclosed.

This work was supported by grants from The Breast Cancer Research Foundation (to C.B. Ambrosone), R01CA225947 (to C.B. Ambrosone, M.J. Higgins, and J.R. Palmer), R01CA133264 (to C.B. Ambrosone, M.J. Higgins, and K. Demissie), and P30 CA016056.