Host, reproductive, and lifestyle factors in relation to quantitative histologic metrics of the normal breast

Participants in this study were women without a personal history of breast cancer who voluntarily donated breast tissues to the Susan G. Komen Tissue Bank (KTB). Details of the KTB project (http://​komentissuebank.​iu.​edu/​) have been described elsewhere [17‐19]. In brief, the KTB is a continuously growing biorepository that collects, stores, and annotates histologically normal breast tissue donated by volunteers. Participants were generally women ≥ 18 years of age at donation with no breast implants and not receiving strong blood thinners or radiation to the chest. About 5382 tissue donations from 4906 women were recorded in the KTB by 2019. For women with multiple donations (n = 476), we used the earliest donation corresponding to the time when questionnaires were administered. Women without hematoxylin and eosin (H&E)-stained images (n = 526) were excluded from the analytical population. In addition, we excluded women above 75 years of age (n = 51) and those without data on age (n = 3), those who were pregnant and/or breastfeeding at the time of tissue collection (n = 59), previously had breast cancer (n = 149), and those without data on menopausal status (n = 10). The final analytical population comprised 4108 women who donated tissues between 2008 and 2019 and for whom we could retrieve the corresponding digitized H&E-stained sections (Additional file 1: Fig. S1). At the time of donation/enrollment, the participants provided written informed consent and were enrolled under a protocol approved by the Indiana University Institutional Review Board and the National Institutes of Health Office of Human Subjects Research (NIH OHSR #4508).

The methodology for exposure assessment within KTB has been described elsewhere [13, 18]. In general, detailed information on sociodemographic, medical, reproductive, menstrual, and lifestyle factors, as well as information on gynecologic surgeries and mammographic screening, were collected by means of self-administered questionnaires. Relevant exposures for this analysis included age (years; < 30, 30–39, 40–49, 50–59, 60–75) at the time of tissue donation, race/ethnicity (Black, White, Asian/Other), age at menarche (years; categorized as ≤ 12, 13, ≥ 14), parity (gravid vs nulligravid), number of live births (0, 1, 2, ≥ 3), age at first full term birth (years; < 25, 25–29, ≥ 30), breastfeeding (ever vs never; duration), body mass index (BMI; < 25, 25–29, ≥ 30 kg/m²;), hormonal birth control use (yes vs no), menopausal status (post- vs. pre-menopausal), bilateral oophorectomy (yes vs none), menopausal hormone therapy (MHT) use (never, former, current) and MHT type, smoking status (never, former, current), alcohol intake, and family history of breast cancer (FHBC) in a first degree relative (present vs absent).

Up to four tissue cores were biopsied from the upper outer quadrant of the right or left breast using a standard 9-gauge (since 2010) or 10-gauge (prior to 2010) needle and one core was fixed in 10% buffered formalin [13, 18]. The formalin-fixed and paraffin-embedded (FFPE) tissue blocks that were prepared from that core were sectioned and stained using H&E staining according to standard laboratory procedures [13, 18]. Archival, digitized, H&E-stained sections were shared with the Molecular and Digital Pathology Laboratory (MDPL) of the Division of Cancer Epidemiology and Genetics (DCEG) at the National Cancer Institute (NCI), USA, for downstream tissue composition analysis (see below).

Digitized H&E-stained slides were archived using the Halo Link digital image repository (Indica Labs, Albuquerque, NM) at the US National Cancer Institute (NCI). Image analysis was performed using the Halo Client computational pathology software (Indica Labs, Albuquerque, NM). A custom-built, random forest, tissue classifier algorithm was trained by two pathologists (MA and MAD) to develop an optimized, 85-datapoint, tissue classifier script. By annotating regions on randomly selected representative images comprised of epithelium, stroma, and adipose tissue, the random forest algorithm was trained to identify, segment, and quantify areas (in mm²) on each slide comprised of epithelium (42-datapoints), stroma (37-datapoints), and adipose tissue (6-datapoints) as shown on Fig. 1 (Red: epithelium; Green: stroma; Yellow: adipose tissue). In previous reproducibility analyses [6], we demonstrated excellent concordance (Spearman’s correlation coefficients ≥ 0.95) between scripts that were independently trained by two pathologists to identify and segment all three tissue types. Training and centralized image analysis were performed masked to all patient characteristics. Percent epithelium, stroma, and adipose tissue were calculated by dividing the absolute value of each histologic metric (in mm²) by the total tissue area (i.e., epithelium + stroma + adipose tissue, mm²) on each slide and multiplying by 100. Percent fibroglandular tissue area was calculated by adding epithelial and stromal area on the slides, dividing by total tissue area, and multiplying by 100. Percent ESP was calculated by dividing the epithelial area by total fibroglandular tissue area and multiplying by 100 as we previously described [6].

Kruskal–Wallis tests were used to test differences in tissue composition metrics by participant characteristics. The associations of host (age, race/ethnicity, FHBC, menopause), reproductive (age at menarche, pregnant (gravid vs nulligravid), number of live births, age at first full-term birth (AFFB), breastfeeding), and lifestyle (smoking, alcohol intake, BMI, hormonal birth control, MHT use) factors with tissue composition metrics (epithelium, stroma, adipose tissue, fibroglandular tissue, and ESP) were assessed in linear regression models. All tissue composition metrics were square root transformed to better approximate the normal distributions for the linear regression model. Partially adjusted models included age and tissue area and fully adjusted models included all of the variables under consideration. Associations of breastfeeding, number of live births, and AFFB with tissue composition metrics were assessed in models restricted to previously pregnant women. Analyses were performed overall and stratified by menopausal status. In the overall model, bilateral oophorectomy and uterine ablation were included separately to examine their effects on tissue composition metrics. In stratified analyses, individuals who had a bilateral oophorectomy, irrespective of age, or uterine ablation after the age of 55 years, were considered postmenopausal. Locally weighted scatter plot smoothing (Lowess) functions were used to plot the residuals from multivariable linear regression models for each tissue composition metric as a function of age. Lowess plots were constructed overall, and separately for pre- and postmenopausal women, parous and nulliparous women, normal and overweight/obese women, and for Black and White women. For racial/ethnicity comparisons, plots were restricted to comparisons between Black and White women due to the very small numbers of other individual racial and ethnic groups. To explore whether BMI impacted parity and race-related curves, we conducted sensitivity analyses by stratifying Lowess plots for parity and race by BMI categories (i.e., normal versus overweight or obese). In sensitivity analyses, we also assessed whether parity impacted the race-related curves by creating separate plots for nulliparous and parous women. The majority of the risk factors were complete for participants. For those with missing values (Additional file 5: Table S1), however, these were addressed by the inclusion of missing values indicators in the models. In sensitivity analyses, we compared with multiple imputation and found the results to be similar (Additional file 5: Table S2). Further, we removed AFFB, which had the largest number of missing values (48.9%) from our model and compared models with and without AFFB (Additional file 5: Table S3). Although the results were similar, the model containing AAFB explained more variability in ESP than the model without AFFB (0.057 vs 0.045, respectively). Accordingly, AAFB was retained in models. All analyses were performed using R version 4.2 and all p values were two sided. Lowess plots were created using Stata statistical software version 16.1.

The characteristics of study participants are shown in Table 1. On average, participants were 43.8 years at the time of tissue donation (range = 18–75 years). Of the 4108 participants, 2696 (65.6%) were premenopausal while 1412 (34.4%) were postmenopausal. The majority (72%) of the participants were Non-Hispanic White, while ~ 18% were Black or African American and 9% were Asian or belonged to other ethnic groups (including Native Hawaiian/Pacific Islander, Alaskan native, Filipino, Japanese, Mixed race, and others). Most of the participants had a college degree (32%) or graduate/professional degree (25%). Two-thirds of the participants were either overweight or obese (BMI > 25 kg/m²).

The distributions of all the tissue composition metrics varied statistically significantly by age and menopausal status (Table 2). With the exception of adipose tissue, which was higher among older than younger and among postmenopausal- than premenopausal women, the distributions of all other tissue composition metrics were higher among younger than older women and among premenopausal than postmenopausal women. In multivariable linear regression models (Table 3), increasing age remained statistically significantly associated with decreasing epithelium, stroma, and fibroglandular tissue and with increasing adipose tissue, but not with ESP. On the other hand, compared with premenopausal women, postmenopausal women had significantly lower epithelium and ESP (Table 3).

Similar patterns of age- and menopause-related changes in tissue composition metrics as in the regression models were seen in Lowess curves for all the tissue composition metrics, with the exception of ESP. Unlike epithelium, stroma, and fibroglandular tissue that declined with increasing age, ESP showed a bimodal age distribution, increased starting at age 18 and peaked around 40 years of age, decreased from 40 until 55 years of age, and then increased with age thereafter (Fig. 2). Similar patterns of bimodal ESP distributions were seen with respect to menopausal status, with the first peak among premenopausal women occurring around age 30–40 years and a later peak among postmenopausal women occurring around 60–70 years of age. The bimodal age distribution of ESP corresponded to differences in the rates of decline of epithelial and stromal tissues by age. In multivariable linear regression models, stromal decline was ~ 34 times higher than epithelial decline before age 40 years but this slowed to ~ 2 times more between 40 and 55 years and increased again to ~ 10 times more after 55 years of age.

Increasing age at menarche was associated with higher stromal and fibroglandular tissue but with lower adipose tissue and lower ESP (Table 2). Compared with women who had never been pregnant, previously pregnant women had statistically significantly higher ESP and this increased with increasing number of live births. Among parous women, those who breastfed had higher epithelium, stroma, fibroglandular tissue, and ESP but lower adipose tissue. In addition, increasing duration of breastfeeding was associated with increasing epithelium, stroma, fibroglandular tissue, and ESP but with decreasing adipose tissue (Table 2). The distributions of tissue composition metrics did not differ by age at first full-term birth.

In multivariable linear regression models (Table 3), parity and increasing number of live births remained statistically significantly associated with higher epithelium and higher ESP, with a strong linear trend in the magnitude of these associations with increasing stroma (Additional file 2: Fig. S2). Breastfeeding remained statistically significantly associated with increasing stromal and fibroglandular tissue and with decreasing adipose tissue and ESP (Table 3). Increasing duration of breastfeeding was associated with increasing stroma and fibroglandular tissue and with decreasing adipose tissue. Although different strata of breastfeeding duration were associated with ESP, there was no statistically significant trend in the association between duration of breastfeeding and ESP. The observed associations of parity, increasing number of live births, and breastfeeding with the individual tissue composition metrics were evident in both pre- (Additional file 5: Table S4) and post- (Additional file 5: Table S5) menopausal women.

Separate Lowess plots for nulliparous and parous women revealed rapid increase in ESP among parous women from 20 years, peaking around 40 years, declining slightly between 40 and 60 years, increasing again after 60 years, and remaining higher for parous than nulliparous women throughout life. Among nulliparous women, epithelium decreased progressively up to around 50 years after which it began to increase, surpassing levels in parous women around 65 years. In contrast, ESP decreased progressively up to 50 years among nulliparous women and remained fairly constant afterward (Fig. 3A).

The distributions of all tissue composition metrics varied by BMI categories (Table 2). Increasing BMI was associated with lower proportions of epithelium, stroma, and fibroglandular tissue but with higher proportions of adipose tissue (Table 2). Further, compared with normal weight women, overweight and obese women had statistically significantly higher levels of ESP. These associations persisted in multivariable linear regression models adjusted for other factors (Table 3). In separate plots for normal versus overweight/obese women (Fig. 3B), epithelial, stromal, and fibroglandular tissue components declined with age, while adipose tissue increased, in both groups. However, while decreases in epithelium (i.e., consistent with lobular involution) continued throughout life among normal weight women, age-related decline in epithelial tissue was not evident among overweight/obese women after 50 years. We also found ESP levels to be slightly higher among normal than overweight/obese women before 50 years of age, with a rapid decline among normal weight women after age 50 (Fig. 3B). Conversely, ESP levels were fairly constant among overweight/obese women before 50 years of age after which a rapid increase was observed causing ESP to be markedly higher among overweight/obese than normal weight women after 50 years (Fig. 3B).

Of the tissue composition metrics, the distributions of stroma and ESP varied statistically significantly by race and ethnicity (Table 2). In general, White women had the highest amount of stromal tissue (median = 18.0%), while Black women had the lowest (median = 15.1%). Conversely, ESP was highest among Black women (median = 6.7%) and lowest among White women (median = 5.7%). The difference in stroma by race/ethnicity was not statistically significant in multivariable models; however, epithelium was statistically significantly higher, while ESP was suggestively higher, among Black than White women (Table 3). In separate Lowess plots for Black and White women, the pattern of age-related decline in ESP differed by race. ESP was higher for Black than White women between 20 and 45 years of age, similar for Black and White women 40–60 years of age, and higher among White than Black women above 60 years of age (Fig. 3C). This pattern of association did not differ by parity status or BMI (Additional file 3: Fig. S3).

The distributions of individual tissue composition metrics varied according to several other factors, including FHBC, average number of alcoholic drinks per week, bilateral oophorectomy (Tables 2 and 3), as well as MHT use among postmenopausal women (Additional file 5: Table S5). While a positive FHBC was positively associated with ESP (Table 3), increasing number of alcoholic drinks per week (Table 3) and current use of combined estrogen and progesterone MHT formulation (Additional file 5: Table S5) were statistically significantly inversely associated with ESP in multivariable models. We did not find statistically significant associations between use of hormonal birth control among premenopausal women and any tissue composition metric.