Using high-accuracy machine-learning algorithms, we characterized normal breast tissue histology and generated quantitative tissue composition data on epithelium, stroma, adipose tissue, fibroglandular tissue, and ESP (a metric of the proportion of fibroglandular tissue that is epithelium relative to stroma) among 4,108 women volunteers participating in the US-based KTB project. By linking to questionnaire-based data, we found several epidemiological risk factors for breast cancer to be associated with tissue composition metrics in the normal breast. In particular, age, menopausal status, parity, breastfeeding, BMI, ethnicity, FHBC, and MHT use demonstrated varying associations with individual tissue composition metrics. We observed a bimodal ESP distribution with respect to age and menopausal status. Menopause, parity, BMI, ethnicity, FHBC, and MHT use were associated with ESP in ways that were consistent with their associations with breast cancer etiology and/or etiological heterogeneity. Collectively, our findings provide novel clues into the potential tissue pathways by which known breast cancer risk factors might influence the risk of breast cancer development.
By examining quantitative tissue composition variations of the normal breast in relation to established breast cancer risk factors, we showed that differences in both epithelial and stromal tissue composition may be critical for breast carcinogenesis. In particular, our findings suggest that imbalance in the rate of stromal and epithelial involution may manifest as changes in ESP, which may represent a feature of the mammary tissue ecosystem that is conducive for carcinogenesis [6]. The bimodal age- and menopause-related peaks in ESP that we found corresponds to the widely reported early-onset (premenopausal) and late-onset (postmenopausal) peaks in breast cancer incidence [20–22]. For most solid cancers, incidence rates increase linearly with age, but the pattern is different for female breast cancer which is characterized by an initial linear increase up to around age 50 years after which the slope changes to a downward trend and then resumes at a slower rate of increase with advancing age [22]. The point at which the incidence curve changes is known as the “Clemmensen’s” hook [23–25], and is characteristic of female breast cancers that occur around the period of life when ovarian function begins to decline until after its cessation at menopause.
Results from epidemiological studies have shown that the bimodal pattern of breast cancer incidence correlates with differences in breast tumor biology [22, 26]. In general, tumors occurring among younger/premenopausal women tend to be more aggressive (typically high grade) and mostly ER- (non-luminal/basal-like) while those occurring among older/postmenopausal women tend to be less aggressive and mostly ER+ (luminal). To date, tissue correlates of this phenomenon have yet to be fully characterized. Our findings of age- and menopausal-related bimodal ESP distribution suggest that the aging process and menopause-related changes in endogenous hormones might impact both epithelial and stromal tissue composition in ways that are consistent with tissue homeostasis and that perturbations at critical points in this process (manifested as variations in ESP) might explain the breast cancer incidence curve, the Clemmensen’s hook, as well as the age-related differences in tumor biology. Further studies integrating serial breast biopsies will be required to characterize longitudinal changes in stromal, epithelial, and adipose tissue in response to age and menopause-related changes in reproductive hormones and related factors.
Epidemiological studies have demonstrated age-specific heterogeneity in breast cancer incidence and molecular subtype with respect to parity, BMI, and ethnicity [27]. For instance, while parity is associated with increases in breast cancer risk among women younger than 30–44 years, parity is associated with decreased risk among older women [27–30]. In this study, we found parity to be strongly associated with higher ESP, with a statistically significant dose-response relationship between increasing NLB and ESP. We did not observe qualitative age interactions between parity and ESP. Instead, our observed bimodal age distribution of ESP was present among parous but not nulliparous women. Although the first ESP peak (~ 30–45 years) among parous women corresponded to the well-documented, parity-related, increased risk of early-onset breast cancer [31, 32], the second ESP peak is not consistent with the documented protective effect of parity among older women [33]. This apparent inconsistency may be due to etiologic heterogeneity of breast cancer with respect to parity/nulliparity [34, 35]. In general, parity is associated with increased risk of basal-like breast cancer, while nulliparity is associated with an increased risk of luminal breast cancer [32, 34, 36]. Although basal-like tumors tend to predominate among younger women, recent data have shown a bimodal age distribution in the incidence of this tumor subtype [37], which is consistent with our observation of a bimodal age distribution of ESP among parous women. In contrast to basal-like tumors, luminal tumors tend to predominate among nulliparous women and at older ages [38–42], which is consistent with our findings of increasing epithelial tissue among older nulliparous women.
Presumably, nulliparity might increase the risk of luminal breast cancer through an intrinsically epithelial-proliferation pathway while parity may increase risk of basal-like breast cancer via stromal-epithelial crosstalk. The former idea is supported by the strong association between nulliparity and highly proliferating luminal tumors, defined by expression of the proliferation marker Ki67 [43] and buttressed by results from experimental studies showing parity-induced terminal differentiation of luminal epithelial cells, coupled with downregulation of growth factors and upregulation of growth inhibitory signals [44]. Conversely, the stromal microenvironment may play a critical role in the pathogenesis of parity-related breast carcinogenesis. Conceivably, parity may increase risk of aggressive/basal-like breast cancer phenotypes through functional disruptions in stromal-epithelial homeostasis, a notion that is supported by studies showing that stromal remodeling and perturbed immune response mechanisms constitute pathways by which parity influences breast cancer risk [3, 45–47]. In addition to the strong association that we observed between parity and increasing ESP, our observations that the magnitude of this associations increased with increasing stromal (as opposed to adipose tissue) content are equally supportive of the potential role of stromal-epithelial crosstalk in mediating parity-related breast carcinogenesis.
The association of elevated BMI with breast cancer incidence varies by age [27, 48]. Among women younger than 50 years, being overweight or obese is associated with decreased breast cancer risk, but risk increases among these women thereafter. In the current study, we found a strong association between elevated BMI and increasing ESP. Differences in ESP between women with normal versus overweight/obese BMI were highest after 50 years of age, corresponding to the age period during which elevated BMI is associated with increased breast cancer risk. Among women younger than 50 years, however, overweight/obese BMI was associated with slightly lower ESP than normal BMI, consistent with the lower risk of breast cancer among women with overweight/obese than normal BMI below 50 years of age [49]. The relatively higher ESP among normal than overweight/obese women between 30–50 years might be due to the correspondingly lower stromal proportion among women with normal BMI between 30–50 years. On the other hand, the markedly higher ESP among overweight/obese than normal weight women after 50 years of age appears to be driven by a combination of increasing epithelium and decreasing stroma among overweight/obese women after 50 years. These tissue-level observations reflect the complex relationships between hormonal and non-hormonal factors with respect to BMI, aging, and breast cancer risk among pre- and postmenopausal women [50].
In the current study, ESP was higher among Black than White women, and the ESP distribution appeared to vary by age among racial groups. Before age 40 years, ESP was higher among Black than White women, but this declined with advancing age in parallel with increasing ESP among White women leading to a crossover around 55 years, after which ESP levels were higher among White than Black women. It is unclear why ESP levels are higher among younger Black than White women and vice versa among older women but this pattern is reminiscent of the higher rates of early-onset breast cancer among Black than White women and of later-onset breast cancer among White than Black women [51]. Although this analysis was based on self-reported race and ethnicity, our findings are consistent with those from a previous analysis within this population that found TDLU levels to be higher among women of African than European genetic ancestry [52]. Given the link between higher TDLU levels and TNBC [53, 54], our findings with respect to epithelial and ESP differences by race buttress the notion that changes in mammary tissue composition may reflect cumulative exposure to endogenous and exogenous breast cancer risk factors over the lifespan, holding clues into the etiopathogenesis of breast cancer subtypes. The statistically significantly higher prevalence of parity (76% vs 60%) and obesity (69% vs 38%) among Black than White KTB participants may also have contributed to these findings; however, our estimates were mutually adjusted for these factors, and we did not find statistically significant interactions between race and either of these factors with respect to ESP. Accordingly, our observations may possibly reflect underlying racial differences in rates of age-related mammary tissue changes. Further studies of longitudinal tissue collections will be required to conclusively address this question.
Having a positive FHBC is a strong risk factor for breast cancer development. However, the tissue pathways by which FHBC influences breast cancer risk are yet to be fully defined. Results from a previous study suggested that polygenic risk scores for breast cancer development were associated with TDLU involution [55]. Here, we found positive FHBC to be associated with higher ESP, which is consistent with its association with increased breast cancer risk. We also found varying but less consistent associations between other risk factors and individual tissue composition metrics. Breastfeeding was associated with lower adipose tissue and higher epithelium, stroma, and fibroglandular tissue, as previously reported [11], but these relationships were generally stronger in postmenopausal women, among whom breastfeeding was also statistically significantly associated with lower ESP. This inverse association between breastfeeding and ESP might partly explain its risk-reducing impact, particularly on pregnancy-associated breast cancer [56, 57]. Having had a bilateral oophorectomy was suggestively associated with lower epithelial and fibroglandular tissue, a low-risk tissue phenotype that is consistent with the reduced risk of breast cancer among women who have had a bilateral oophorectomy [58, 59]. Current use of MHT was associated with higher fibroglandular tissue, correspondingly lower adipose tissue, and lower ESP. While the association between MHT use and higher fibroglandular tissue is consistent with its association with higher mammographic density [60], a radiological representation of the amount of fibroglandular tissue in the breast, and elevated breast cancer risk [61, 62], its association with lower ESP is not consistent with its risk increasing role. Data from epidemiological studies suggest that MHT use is associated with elevated risk of hormone receptor-positive/low grade but not receptor-negative/high grade breast cancer [43, 63–67]. The disparate associations between MHT use and distinct tissue phenotypes may, therefore, be another clue into the role of variations in exposure-tissue interactions in the etiopathogenesis of breast cancer subtypes.
This study has several important strengths, including the application of high-accuracy machine-learning algorithms for the detailed and centralized assessment of quantitative tissue composition metrics on digitized, H&E-stained, biopsy specimens from over 4,000 normal breast tissue donors. To the best of our knowledge, this is the largest analysis of its kind to date to investigate associations between several questionnaire-based risk factors and quantitative tissue composition metrics of the normal breast. The large sample size allowed us to conduct analysis overall and stratified by menopausal status and other relevant characteristics. Nevertheless, the current analysis is not without limitations. For instance, we were unable to examine longitudinal changes in tissue composition metrics. Also, we were unable to directly evaluate the potential impact of sociodemographic, environmental, and socioeconomic factors on our BMI and race-related findings, but all our estimates were adjusted for educational level as a surrogate for socioeconomic status. Although this study was based on a population of self-selected volunteers, BCRAT (or Gail) scores of absolute breast cancer risk for participants in this study are normally distributed (Supplementary Fig. 4), as in the general population.
In conclusion, we investigated the relationships of host, lifestyle, and reproductive factors on quantitative tissue composition metrics of the normal breast, including epithelium, stroma, adipose tissue, fibroglandular tissue, and histologic ESP (a metric of the proportion of fibroglandular tissue that is epithelium relative to stroma). We found several established breast cancer risk factors to be associated with individual tissue metrics, including novel observations with respect to ESP. In particular, age, menopausal status, parity, BMI, ethnicity, FHBC, and MHT use demonstrated associations with ESP consistent with their documented associations with incidence of molecular breast cancer subtypes. Overall, our findings provide critical insights into the role of stromal-epithelial interactions in breast cancer etiology, with implications for our understanding of the histogenesis of breast cancer subtypes. Conceivably, variations in tissue composition metrics on biopsy, particularly ESP, could serve as intermediate markers of risk and might be used to inform breast cancer prevention strategies for women.