Background
Percent mammographic density (PMD) is one of the strongest known risk factors for breast cancer [
1]. Fibroglandular tissue attenuates X-rays more than does fat [
2], and it appears white (dense) in mammograms, whereas adipose tissue appears dark. PMD, illustrated in Additional file
1: Figure S1a, refers to the area of white tissue divided by the total area of the breast in the image. The dense area and PMD are both associated positively with risk of breast cancer, and PMD is the stronger risk factor [
3]. The nondense area is associated inversely with risk of breast cancer [
3,
4]. The increased risk of breast cancer associated with PMD persists for at least 8–10 years after the date of the mammogram used to assess PMD [
5,
6], and it cannot be explained by the “masking” of cancers by dense breast tissue [
6,
7]. In addition to an increased risk of breast cancer, PMD is also associated with an increased risk of lesions that are thought to be nonobligate precursors of breast cancer [
8].
Average PMD in the population decreases with increasing age [
5]. A cross-sectional study of 11,000 women in 22 countries showed that average PMD declined with increasing age. Decline was present before and after menopause and was most pronounced over the menopausal transition [
9]. Longitudinal data within individuals has shown average reductions in PMD of from 5% [
10] to 8% [
11] respectively over 10 to 5 years.
Similar variations in breast tissue composition can be seen using measures of fat and water obtained by magnetic resonance (MR). Radiologically dense breast tissue and breast water both reflect fibroglandular breast tissue (Additional file
1: Figure S1b).
Antoni et al. showed in a meta-analysis of 19 studies with a total of > 24,000 breast cancer cases [
12] that, relative to women in the lowest density category, women in the highest density category had 3.1-fold (95% CI 2.2–4.2) and 3.2-fold (1.7–5.9) increased risk of estrogen receptor-positive (ER+) and ER− breast cancer, respectively. In case-only analyses, relative risks of breast tumors for ER+ versus ER− were 1.13 (95% CI 0.89–1.42) for medium versus minimal mammographic density (MD). MD remained associated with screen-detected ER+ tumors. In eight contributing studies, the association of MD did not differ by HER2 status. Variations in the distribution by age of ER+ and ER− breast cancer are likely to be influenced by factors other than MD.
Breast cancer risk increases with increasing extent of PMD, and estimates of attributable risk (which assume causality) suggest that 30–50% of breast cancer may be attributed to the most extensive categories of PMD [
5,
6]. Although MD is associated with relative and attributable risks that are large compared with other risk factors for the disease, the accuracy of risk prediction in individuals is modest [
13].
The mechanisms that underlie the association of PMD with risk of breast cancer are not well defined [
14], and the apparent paradox that with increasing age average PMD decreases while breast cancer incidence increases remains unexplained. We have previously proposed [
15] that the radiological features of the breast of PMD provide an index of cumulative exposure to events that influence the incidence of breast cancer, similar to the concept of “breast tissue aging” proposed by Pike et al. [
16]. However, to date, there is only one published study to support this suggestion [
10].
In this paper, we develop a testable biological hypothesis to explain the origins of breast cancer associated with mammographic density. We summarize evidence that PMD reflects the relative quantities of epithelium, stroma, and fat in the breast, and we use a two-stage model of carcinogenesis as a framework to examine how the known biological properties of these tissues may influence the transition of normal breast epithelial cells to malignant cells [
17]. We expect cumulative exposure to these biological factors to contribute to the age-specific incidence of breast cancer, and we examine the relationship between estimated cumulative exposure to PMD (CBD) in the population and the age-specific incidence of breast cancer.
Discussion
The close relationship observed between log CBD and log breast cancer incidence is consistent with the hypothesis of carcinogenesis in which the accumulation of mutations or other molecular changes increases with increasing duration of exposure to PMD rather than with age. Limitations of our data include the cross-sectional observations and the ecological comparison with breast cancer incidence, as well as the small numbers at ages 30–34 and 80+ years. We estimated CBD from cross-sectional rather than longitudinal observations, using film rather than digital mammograms. However, longitudinal assessments of breast density in women aged 40 or older have shown a decline in average PMD with increasing age and menopause that is very similar to the differences seen here [
10]. Further, Maskarinec et al. showed a strong association between cumulative density and age-specific breast cancer incidence in serial mammograms from 607 patients with breast cancer and 667 control subjects in the Hawaii component of the multiethnic cohort, in which the average age at first mammogram was 57 years [
10]. However, the associations of cumulative density and breast cancer incidence with age were not examined [
17].
CBD may also explain many of the known epidemiological associations with breast cancer risk. As shown above, the estimated size of the susceptible cell population of epithelial cells and epithelial cell proliferation are greatest at early ages and decline with increasing age. The greater amount of fibroglandular tissue, as shown by percent water, present at ages 15–18 may be related to the greater susceptibility of the breast at early ages to the effects of known exposures on risk of breast cancer, including radiation, alcohol, and smoking [
36].
Early menarche is associated with an increased risk of breast cancer in later life [
90] and advances the age at which fibroglandular breast tissue develops. This addition to the time of exposure will influence all estimates of PMD at later ages and will increase CBD. An early pregnancy and early menopause both reduce later risk of breast cancer and PMD [
90]. The reductions in PMD associated with these events will influence all measures of PMD at later ages and reduce average CBD in parous and postmenopausal women, respectively. At least some of the effect of pregnancy in reducing risk of breast cancer has been shown to be mediated by the reduction in PMD associated with pregnancy [
91].
Tamoxifen reduces PMD and risk of breast cancer, and reduction in PMD appears to predict response to adjuvant therapy with tamoxifen [
92]. Progesterone as a postmenopausal replacement therapy has been shown to increase both PMD and breast cancer incidence, and the effect of progesterone on breast cancer incidence has been shown to be mediated through the effect on PMD [
93]. The proliferation of mammary epithelium in response to progesterone is mediated by receptor activator of nuclear factor-κB ligand (RANKL), and increased expression of RANKL has been found to be associated with more extensive PMD in premenopausal women [
94].
The biological hypothesis that we propose from the foregoing considerations is that the transition of breast epithelial cells from normal to malignant cells is completed more frequently in dense breast tissue than in nondense tissue. We propose that this transition is associated with the acquisition of mutations or other molecular changes in breast epithelial cells that increase in frequency with increasing exposure to both the amount and duration of PMD. We propose that the probability of acquiring mutations is influenced by the greater number of epithelial cells and by the several known biological properties of the stromal tissues that are associated with PMD, described in Table
1, by the amounts of such tissues, and by the duration of exposure to these influences. Proteoglycans and MMP-3 in the ECM of radiologically dense breast tissue have already been shown, in the absence of breast cancer, to be similar to those expressed in breast tissue associated with breast cancer.
Additional influences may include the greater number of stromal fibroblasts and associated chemokines associated with PMD that may, in the absence of breast cancer, resemble CAFs. CAFs can be distinguished from normal fibroblasts by markers and functional assays. Among these properties is the production of
TGF-β1, which promotes epithelial-mesenchymal transition and has effects on the microenvironment that promote tumorigenesis and tumor invasion (reviewed in [
45]). Epithelial-mesenchymal transition and other changes in the microenvironment may, in the absence of breast cancer, be more extensive in radiologically dense breast tissue than in nondense tissue [
45].
Conclusions
PMD has reproducibly been shown to be a strong risk factor for breast cancer that may account for a substantial fraction of the disease. The biological basis for this association is currently unknown, however. We have examined potential biological mechanisms for the risk of breast cancer associated with PMD using a two-stage model of carcinogenesis as a framework.
It is understood that it is the biological properties of the breast tissues associated with PMD, not the radiological properties, that are responsible for the association of PMD with risk of breast cancer. PMD is known to be associated with a greater number of epithelial cells, greater glandular area, a greater area of collagen, and a greater number of nonepithelial cells. The known biological properties of these breast tissue components increase the probability of mutation and of transition to malignant cells. The finding that CBD in healthy subjects in the population, estimated from cross-sectional observations in healthy women, was strongly associated with the age-specific incidence of breast cancer in Canada and is consistent with the accumulation of mutations with increasing time of exposure to CBD. This biological model gives rise to a number of testable predictions concerning the properties of breast tissue associated with PMD and suggests that the radiological features of the breast may be useful in the design, sampling, analysis, and interpretation of research on the biology of breast tissues in relation to breast cancer.
Acknowledgements
We acknowledge the contribution of Dr. Alice S. Whittemore, Stanford University, who directed our attention to the Moolgavkar model.