The relationship between terminal duct lobular unit features and mammographic density among Chinese breast cancer patients

Abstract

Extensive mammographic density (MD), a well-established breast cancer risk factor, is a radiological representation of stromal and epithelial breast tissue content. In studies conducted predominantly among Caucasian women, histologic measures of reduced terminal duct lobular unit (TDLU) involution have been correlated with extensive MD, but independently associated with breast cancer risk. We therefore examined associations between TDLU measures and MD among Chinese women, a low-risk population but with high prevalence of dense breasts. Diagnostic pre-treatment digital mammograms were obtained from 144 breast cancer cases at a tertiary hospital in Beijing and scored using the Breast Imaging Reporting and Data System (BI-RADS) density classification. TDLU features were assessed using three standardized measures (count/100mm², span [µm], and acini count/TDLU) in benign tissues. Associations between each of TDLU measures and MD were examined using generalized linear models for TDLU count and span and polytomous logistic regression for acini count with adjustment for potential confounders stratified by age. Among women ≥50 years, 63% had dense breasts; cases with dense breast (BI-RADS, c-d) had greater TDLU count (21.1 [SE=2.70] vs 9.0 [SE=1.83]; P=0.0004), longer span (480.6µm [SE=24.6] vs 393.8µm [SE=31.8]; P=0.03), and greater acini count (OR_trend=16.1; 95%CI=4.08–63.1; P_trend<0.0001) compared to those with non-dense breasts (BI-RADS, a-b). Among women <50 years, 91% had dense breasts, precluding our ability to detect associations. Our findings are consistent with previously reported associations between extensive MD and reduced TDLU involution, supporting the hypothesis that breast cancer risk associated with extensive MD may be related to the amount of at-risk epithelium.

Introduction

Mammographic density (MD) is a radiological representation of stromal and epithelial breast tissue content and extensive MD is a well-established risk factor for breast cancer.^{1, 2} Several studies have provided insight into the underlying mechanism of the MD and breast cancer risk association by examining histologic features of breast tissues.^3–6 Terminal duct lobular unit (TDLU) involution is a histological measure of breast tissue characterized by a reduction in the epithelial component with aging.⁷ Less extent of involution, characterized by greater number and size of observed TDLUs, has been associated with higher breast cancer risk among women with benign breast diseases.^8–10 Previous studies conducted predominantly among white women using diagnostic biopsy specimens for benign breast diseases have demonstrated that reduced TDLU involution was associated with higher MD ^3–5: women with no or partial involution were more likely to have higher percent MD or dense parenchymal patterns than those with complete involution ³. This association was subsequently replicated using standardized quantitative measures of TDLU involution ⁴. Contrary to these reports, a recent analysis of breast cancer cases within the Multiethnic Cohort, which was highly enriched for women of Japanese ancestry (43%) and included participants who were older (mean age 59.7 years) and thinner than in other studies showed the opposite direction of the association (i.e., greater TDLU involution associated with higher dense area) ⁶. The explanation of this finding is unresolved but may suggest potential racial variation in the TDLU-MD relationship.

Although these measures from the histologic (i.e., TDLU involution) and the radiologic (i.e., MD) assessments of breast tissues are correlated, increased TDLU involution has been associated with lower breast cancer risk independently of the MD^{9, 11}, suggesting that the relationship between TDLU involution and MD, and their associations with breast cancer risk are more complex. To date, most studies of MD in relation to TDLU features, including the one in Multiethnic Cohort⁶, have been conducted among Western populations. Therefore, we examined the relationship between MD and TDLU features in non-neoplastic breast tissues from breast cancer patients in China, where breast cancer incidence rates are historically low and the prevalence of dense breasts and other breast cancer risk factors are substantially different from high-risk populations in Western countries^{12, 13} .

Materials and Methods

Study population

We utilized data from a previously conducted TDLU evaluation which included 504 invasive breast cancer cases with luminal A and triple-negative tumors treated at the Cancer Hospital, Chinese Academy of Medical Sciences (CHCAMS), Beijing, China, between 2009 and 2012, as previously described in detail.¹⁴ In brief, eligible cases included those who had a confirmed breast cancer diagnosis, complete pathology data, and no neoadjuvant therapies prior to surgeries. The majority of these patients (>90%) were symptomatic and very few were detected through physical-exams or screening. We restricted the analysis to 144 cases who were diagnosed between 2011 and 2012 and whose diagnostic mammograms were available. Demographic and clinical characteristics of patients with and without BI-RADs data were shown in Supplementary Table 1. Breast cancer risk factors were extracted from medical records, which included age at diagnosis, body mass index (BMI), age at menarche, number of children, breastfeeding (yes or no), and first-degree family history of breast cancer. Pathology data including TNM stage, grade, and expression levels of ER, PR, HER2, CK5/6, and EGFR were retrieved from pathology reports. The status of ER, PR, CK5/6, and EGFR expression in tumors was determined by immunohistochemistry (IHC) and >1% staining was considered as positive. HER2 expression was determined by IHC and fluorescence in situ hybridization (FISH), and either IHC 3+ or FISH-positive was defined as HER2 positive. To avoid misclassification, we excluded all HER2 2+ cases without FISH data. Luminal A tumors (n=95) were defined as ER+ and/or PR+, HER2-, and negative for both basal markers and triple-negative as ER-, PR-, and HER2-. The triple-negative group (n=49) was enriched with basal-marker positive tumors (n=39). The project was approved by the CHCAMS Ethics Committee and informed consent was not required for the use of existing pathological materials with no reveal of identifiable patient information. The study was also exempted from review by the Office of Human Subject Research Protections at the National Institutes of Health (NIH) since NIH investigators working with data do not have abilities to identify the subjects from whom the specimens or data originated (Exempt Number: 11751).

Histologic assessment of TDLUs

The methods for histologic review and TDLU annotation were previously described in detail.¹⁴ In brief, tissue sections prepared from grossly benign breast tissue at the time of mastectomy were stained with hematoxylin and eosin and scanned to create digital image files. Cases with low-quality images or benign changes throughout the section (duct dilation, metaplasia, hyperplasia, microscopic evidence of ductal carcinoma in situ, or invasive cancer) were excluded. We used three standardized TDLU measures previously shown to have high intra/intra reader reproducibility ^{4, 10, 15}: TDLU count per unit area (count/100 mm²), TDLU span (measured with an electronic ruler in microns, µm), and acini count per TDLU, as an indication for involution (with higher levels of all 3 measures indicating lower levels of TDLU involution) ¹⁵. Among women with observable normal TDLUs, up to 10 sequential TDLUs were evaluated for acini count/TDLU (categories: 1, 2–10; 2, 11–20; 3, 21–30; 4, 31–50; 5, 51–100; and 6, >100) and TDLU span (median diameter of a TDLU in µm). For each sample, the median values of acini count/TDLU and TDLU span were used as summary measures. All TDLU measurements were performed by a single trained cytotechnologist (R. Cora).

MD assessment

All digital diagnostic mammograms were acquired on GE Senographe DS full-field digital mammography (FFDM) system at CHCAMS. FFDM images were retrospectively evaluated by a board-certified radiologist (EL) using the Breast Imaging Reporting and Data System (BI-RADS) guidelines recommended by the American College of Radiology (5^th edition)¹⁶ and categorized into four levels using the BI-RADS breast composition scoring system (almost entirely fatty (a), scattered areas of fibroglandular density (b), heterogeneously dense (c), and extremely dense (d)). BI-RADS readings for the two breasts within one woman showed very high correlations (>0.95) and we used the maximal values for BI-RADS density categories across the two breasts in subsequent analyses. For quality control assessment, images for all 144 cases were independently read by four other senior breast radiologists with each reading a subset. Average inter-observer agreement based on k-statistics was 0.66 (range of weighted kappa=0.59–0.70) on a four-grade scale and 0.73 (range of kappa=0.61–0.83) on a two-grade scale (a-b or c-d). The reading from EL was used in the subsequent analyses since the consistency was comparable to a previously published study ¹⁷.

Statistical Analyses

Spearman rank correlation coefficients were used to assess the correlation between TDLU measures (raw TDLU count, median TDLU span, and median category of acini count/TDLU) and risk factors such as age, BMI (kg/m²), age at menarche, and number of children (0 for nulliparous women) (Supplementary Table 2). Associations between MD and risk factors or clinical characteristics were evaluated using Chi-square or Fisher’s exact test wherever appropriate. Local polynomial regression based on LOWESS (locally weighted scatterplot smoothing) was used to estimate and visualize the average TDLU count as a function of age in dense (BI-RADS, c-d) and non-dense (BI-RADS, a-b) breast groups, separately.

To examine the association between TDLU measures and MD, we used generalized linear models for continuous TDLU measures (count and span) and polytomous logistic regression for the ordinal TDLU measure (acini count). Because TDLU count was not normally distributed, we used double square-root transformed TDLU count (normalized TDLU count/100 mm²) to better approximate normal distribution in the further statistical tests. Generalized linear models were conducted using PROC GLM (SAS) to estimate least square means and standard errors (SE) of TDLU count or span (outcome) by BI-RADS density (independent variable, 4 categories in the overall analysis and 2 categories [dense: BI-RADS, c-d vs. non-dense: BI-RADS, a-b] in the age-stratified analysis) with adjusting for potential confounders. The first model included age (5-year frequency) and BMI (<23, 23–24.9, 25+ kg/m²) as covariates, two variables that are known to be strongly associated with both MD and TDLU measures. The second model additionally included age at menarche (<14 (median), ≥14 years), parity (nulliparous, parous), and tumor subtype (luminal A, triple-negative) as covariates based on observed correlations with TDLU measures and/or MD. For TDLU count and span, the least square means and SEs were back transformed to raw values and presented. Polytomous logistic regression was used to examine the association between BI-RADS density (independent variable) and median category of acini count (dependent variable, categorized into tertiles) adjusting for potential confounders. Odds ratios (ORs) and corresponding 95% confidence intervals (CIs) were estimated to compare the 2^nd and 3^rd categories to the 1^st category of acini count. We also examined the trend in association across tertiles of acini count. Since the association between TDLU measures and MD may vary by menopausal status^{4, 15}, all analyses were done for all women combined as well as for women stratified by age (<50 and ≥50 years, a proxy for menopausal status). We also formally tested the interaction of age (5-year interval or <50 and ≥50 years) and MD in relation to TDLU measures by including a product term of age variable and MD in the model. The interaction was not statistically significant and therefore was not included in the final model. Lowess function was conducted using Stata/SE (version 11.2; StataCorp LP, College Station, TX). Graphs for descriptive statistics of TDLU measures were generated using Origin (OriginLab, Northampton, MA). All other analyses were conducted using SAS (version 9.3; SAS Institute Inc., Cary, NC). All statistical tests were two-sided and considered statistically significant at P < 0.05.

Results

Table 1 shows selected characteristics including distribution of TDLU measures and BI-RADS by age group. The mean (SD) age was 42.3 (5.4) for younger cases (<50 years, n=80) and 57.2 (5.3) for older cases (≥ 50 years, n=64). The mean BMI was similar in younger and older women (24.8 vs. 25.0; P=0.76). Compared to younger women, older women were more likely to have later age at menarche (14+: 32.3% vs 13%; P=0.03) and more children (>1: 39.7% vs 21.8%; P=0.04). Older women also tended to have less and smaller TDLUs (P<0.05 for all three TDLU measures) and a lower frequency of dense breasts (BI-RADS, c-d; 62.5% vs. 91.3%; P<0.0001). Supplementary Table 2 shows the correlation between TDLU measures and patient characteristics. All three TDLU measures were strongly correlated with one another (r²= 0.69 to 0.82; P<0.0001) and inversely correlated with age (r²= −0.53 to −0.48; P<0.0001). Similar age associations were found for both non-dense (BI-RADS, a-b) and dense (BI-RADS, c-d) breast groups (Figure 1) and for both luminal A and triple-negative cases (Supplementary Figure 1).

Table 1.

Selected characteristics of breast cancer cases in the Cancer Hospital, Chinese Academy of Medical Sciences (CHCAMS), overall and by age (n=144)

	Overall (n=144)		Age<50 (n=80)		Age≥50 (n=64)		Pa
	n	%	n	%	n	%
Age
Mean, SD	48.9	9.2	42.3	5.4	57.2	5.3	-
20–39	20	13.9	20	25.0	-	-
40–49	60	41.7	60	75.0	-	-
50–59	50	34.7	-	-	50	78.1
60–79	14	9.7	-	-	14	21.9
Body mass index
Mean, SD	24.9	3.5	24.8	3.6	25.0	3.5	0.76
<23	45	31.3	29	36.3	16	25.0	0.4
23–24.9	33	22.9	16	20.0	17	26.6
25–29.9	57	39.6	29	36.3	28	43.8
30+	9	6.3	6	7.5	3	4.7
Family history of breast cancer
No	136	94.4	74	92.5	62	96.9	0.30
Yes	8	5.6	6	7.5	2	3.1
Age at menarche
Mean, SD	14.6	1.8	14.2	1.4	15.1	2.1	0.04
<12	41	29.5	25	32.5	16	25.8	0.03
12–13	68	48.9	42	54.6	26	41.9
14+	30	21.6	10	13.0	20	32.3
Missing	5		3		2
Parity/Number of children
Mean, SD	1.37	0.8	1.18	0.6	1.60	0.9	0.01
Nulliparous	6	4.26	5	6.4	1	1.6	0.04
1 Child	93	66.0	56	71.8	37	58.7
More than 1 Child	42	29.8	17	21.8	25	39.7
Missing	3		2		1
Breastfeedingb
No	5	5.1	4	7.0	1	2.4	0.4
Yes	94	95.0	53	93.0	41	97.6
Missing	36		16		20
Tumor subtype
Triple-negative/Core-basal phenotype	49	34.0	27	33.8	22	34.4	0.94
luminal A	95	66.0	53	66.3	42	65.6
Tumor size
≤2cm	89	62.2	50	62.5	39	61.9	0.94
>2cm	54	37.8	30	37.5	24	38.1
Missing	1		0		1
Lymph node invasion
Negative	86	61.4	46	58.2	40	65.6	0.38
Positive	54	38.6	33	41.8	21	34.4
Missing	4		1		3
Grade
Well differentiated	20	15.0	10	13.0	10	17.9	0.72
Moderately differentiated	63	47.4	38	49.4	25	44.6
Poorly differentiated	50	37.6	29	37.7	21	37.5
Missing	11		3		8
TDLU involution measures (median, range)
TDLU count/100mm2	20.4	1.4 to 128.9	29.0	1.4 to 91.4	15.2	2.0 to 128.9	<0.0001
Median span (µm)	534.3	177.0 to 1162.5	614.8	177.0 to 1162.5	410.0	200.0 to 817.5	<0.0001
Median category of acini count/TDLUc
1–1.5	46	31.9	11	13.8	35	54.7	<0.0001
2–2.5	49	34.0	29	36.3	20	31.3
3–5	49	34.0	40	50.0	9	14.1
BI-RADS density
a, almost entirely fatty	9	6.3	1	1.3	8	12.5	<0.0001
b, scattered areas of fibroglandular density	22	15.3	6	7.5	16	25.0
c, heterogeneously dense	66	45.8	36	45.0	30	46.9
d, extremely dense	47	32.6	37	46.3	10	15.6

BI-RADS Breast Imaging Reporting and Data System, BMI body mass index, SD standard deviation, TDLU terminal duct lobular unit

^aT-test or Kruskal-Wallis test for continuous variables; Chi-squared or Fisher’s Exact test for categorical variables

^bParous women only

^cCategories for acini count/TDLU: 1, 2–10; 2, 11–20; 3, 21–30; 4, 31–50; 5, 51–100

Figure 1.

Estimated TDLU counts per 100 mm² by age based on LOWESS (locally weighted scatterplot smoothing) stratified by BI-RADS density (a-b vs c-d) *The normalized values of TDLU count based on double square-root transformation were used for local polynomial regression and back-transformed to the original values for visualization.

Among all women combined, median values of TDLU count tended to increase with increasing BI-RADS density (Figure 2). A similar trend was observed among older but not younger women in age-stratified analyses. Similarly, median values of TDLU span were also greater in cases with dense (BI-RADS, c-d) than non-dense (BI-RADS, a-b) breasts, and the difference was again stronger among older women (Supplementary Figure 2).

Figure 2.

Distribution of TDLU counts per 100 mm² by BI-RADS density, overall and by age

Table 2 shows estimated TDLU count and median TDLU span by BI-RADS density categories with the adjustment for potential confounders. Among all women combined, number of TDLU count per 100 mm² and median span increased with increasing BI-RADS category from entirely fatty breast (BI-RADS, a) to heterogeneously dense (BI-RADS, c) but slightly declined in patients with the extremely dense breast (BI-RADS, d). However, the test for trend was not significant in either age- and BMI-adjusted model or the full model. Among younger women (age<50 years), which only included 7 participants with low MD, TDLU count and span did not vary significantly by MD. In contrast, among older women (age≥ 50 years), women with dense breasts compared to non-dense breast had significantly greater number of TDLUs per 100 mm² (21.1 [SE=2.7] vs 9.0 [SE=1.8]; P=0.0004) and median span (480.6 µm [SE=24.6] vs 393.8 µm [SE=31.8]; P=0.03). The association of acini count with BI-RADS density showed a similar trend (Supplementary Table 3), with women of dense breast associated with greater acini count per TDLU among older women (OR_trend=16.1; 95% CI=4.1–63.1; P_trend<0.0001; comparing higher category to lower category) but not among younger women (OR_trend=1.1; 95% CI=0.2–5.6; P_trend=0.92).

Table 2.

Associations between TDLU measures (count per 100mm² and median span) and BI-RADS density, overall and by age

	n	Meana	SEa	Meanb	SEb
TDLU count/100mm2
Overall
BI-RADS, a	9	14.7	4.2	14.2	4.1
BI-RADS, b	22	17.1	2.9	17.2	2.9
BI-RADS, c	66	25.9	2.2	25.9	2.4
BI-RADS, d	47	21.1	2.4	21.3	2.5
Ptrend		0.17		0.15
Age<50
BI-RADS, a-b	7	32.3	7.7	31.9	7.9
BI-RADS, c-d	73	27.1	2.0	27.3	2.2
P		0.58		0.66
Age≥50
BI-RADS, a-b	24	10.0	1.9	9.0	1.8
BI-RADS, c-d	40	21.1	2.6	21.1	2.7
P		0.001		0.0004
Median TDLU span (µm)
Overall
BI-RADS, a	9	540.1	67.6	531.0	69.0
BI-RADS, b	22	495.9	40.8	499.6	42.0
BI-RADS, c	66	589.2	23.2	592.8	24.8
BI-RADS, d	47	544.9	29.6	548.7	30.1
Ptrend		0.55		0.49
Age<50
BI-RADS, a-b	7	601.4	84.6	588.2	88.5
BI-RADS, c-d	73	642.7	24.9	653.7	27.1
P		0.64		0.49
Age≥50
BI-RADS, a-b	24	399.1	30.3	393.8	31.8
BI-RADS, c-d	40	483.0	23.2	480.6	24.6
P		0.03		0.03

BI-RADS, Breast Imaging Reporting and Data System, BMI, body mass index, SE standard error, TDLU terminal duct lobular unit

^aAdjusted for age (5-year), BMI (<23, 23–24.9, and ≥25 kg/m²)

^bAdjusted for age (5-year), BMI (<23, 23–24.9, and ≥25 kg/m²), parity (nulliparous, parous), age at menarche (<14, ≥14), and tumor subtype (luminal and TN).

Discussion

In this cross-sectional study examining the relationship between BI-RADS density and TDLU features using surgically removed benign tissue samples from Chinese breast cancer patients, we found that women with denser breasts were more likely to show greater TDLU count, median TDLU span, and acini count per TDLU, reflecting less extent of involution, particularly among older women. Despite the differences in study populations, tissue type, and approaches to measure MD and TDLU involution, we observed a similar direction for the MD-TDLU association as in two previously published studies ^{3, 4} conducted among Caucasian women, supporting the hypothesis that MD associated breast cancer risk may partially be mediated through the abundance of at-risk epithelium.

In the Mayo Benign Disease Cohort study (Mayo), using biopsy specimens of women with benign breast disease (BBD), Ghosh et al. showed that women with no or partial involution were more likely to have extensive MD than those with complete involution ³. The association was replicated using quantitative MD area measurement in the same study ³. In a recent study using data from the STAMP project, where benign biopsy tissues were annotated with more standardized and quantitative TDLU measures, the relationship of denser breast tissues with lesser degrees of TDLU involution was confirmed ⁴. In contrast, a report from Multiethnic Cohort showed that greater TDLU involution assessed in adjacent non-tumor breast tissue was not related to lower MD but was associated with a higher MD area ⁶. There may be several reasons to explain these conflicting results, such as the heterogeneity in the assessment of TDLU involution (standardized quantitative vs. semi-quantitative methods) and MD (BI-RADS vs. continuous measures), tissue source (women with BBD vs. women with breast cancer) or racial groups (white women in Mayo and STAMP vs. more diverse populations in Multiethnic Cohort [43% Japanese American and 35% Caucasians]). Further, women in the Multiethnic Cohort were older (69% postmenopausal), reported past or current hormone replacement therapy user (65%), and thinner than in other studies. Interestingly, our study design was more similar to that of the Multiethnic Cohort⁶ in that we also used surgically removed benign breast tissues from cancer patients for the evaluation of TDLU features, while the Mayo BBD cohort study ³ and the STAMP project ⁴ used benign tissues from women with BBD.

Consistent with what was reported in the STAMP study ⁴, which used the same three quantitative metrics for TDLU assessment as in the current study, we found that TDLU count had a stronger association with MD compared to TDLU span. However, in contrast to the null association between acini count and percent MD in the STAMP study, we observed significant associations between all three TDLU measures and MD, albeit with wide confidence limits. To tease out whether there is a specific TDLU feature(s) that might show stronger association with MD, we analyzed MD as an outcome variable and all three TDLU measures as explanatory variables with the adjustment of the same set of covariates among older women. We found that the association for acini count remained statistically significant while the associations for the two other measures became non-significant (data not shown), suggesting that the acini count may better reflect the at-risk epithelium and capture the MD association.

In addition, while the STAMP study reported significant association only among premenopausal women,⁴ we observed a significant association only among older women. The lack of association among younger women in our study could be due to the insufficient power, given that the vast majority of younger women in our study had dense breasts (BI-RADS, c-d; 91%, n=73). Alternatively, the association between TDLU involution and MD may vary by race or ethnicity in relation to age interaction since both MD and TDLU features are influenced by common factors such as age, BMI, and hormonal exposures and the impacts of these factors may vary by race/ethnicity. For example, a longitudinal analysis showed that the age-related decline of percent mammographic density was slower among Japanese women compared to Caucasians.¹⁸ In addition, studies conducted among Asian women showed that the association between MD and breast cancer risk was stronger among postmenopausal women than premenopausal women.^19,20

Variations in breast size and tissue composition across different populations may also affect the association between MD and TDLU features differently. Although not always consistent, studies comparing racial differences in MD suggest that Asian women have higher percent MD but smaller absolute dense area/volume and breast size compared to Western women after accounting for well-known MD determinants such as age and BMI.^21–26 It is also notable that intact TDLU structures were persistent even in very old women (up to age 79 years) in our study, whereas a high proportion of subjects showed complete involution among younger women (zero TDLU count: 25%; 40–65 years) in the STAMP study.^{4, 27} In addition, while we found strong correlations of all three TDLU measures (all Spearman’s rho>0.63; Supplementary Table 2), previous studies reported a strong correlation between TDLU size measures (acini count and span) but a weaker correlation of TDLU count with either size measure (all Spearman’s rho<0.18).^{4, 27} Future studies using quantitative density in area and volumetric measures among women from diverse racial background are needed to clarify the racial variation in MD-TDLU relationship.

Our study adds new information on the relationship between TDLU features and MD among Asian women, in which breast composition features are unique, but have not been well studied. However, several limitations should be considered to interpret our results. First, we evaluated MD and TDLU features in benign tissues from breast cancer cases, which may be systematically different from those without cancers. Nevertheless, our observation is largely consistent with what was previously reported based on benign tissues from healthy women, despite the potential changes that may have occurred in adjacent-normal tissue in cancer cases. Second, the selection of patients was based on the availability of data with both TDLU measures and BI-RADS density. Several demographic and tumor characteristics, including age, tumor subtype, and grade, and TDLU measures varied between selected and unselected cases (Supplementary Table 1). In addition, we only included luminal and TN patients with MD data in this analysis and patients seen at this tertiary hospital may not represent the general breast cancer population in China, which limits the generalizability of our findings. Third, we used BI-RADS for MD assessment, which is subjective and is known to vary across radiologists ²⁸. In addition, BI-RADS density is a categorical MD measurement based on a two-dimensional view; quantitative measurements assessing different density in three-dimensional views or via volumetric approaches may provide more accurate estimates for the association between specific TDLU features and MD measures. Fourth, the information on hormone-replacement therapy among postmenopausal women, which is known to influence both TDLU involution and MD, is limited in this study population. However, the prevalence of hormone-replacement therapy use is very low among Chinese women (2.1%)²⁹, which is unlikely to have caused a significant bias in the observed associations. Lastly, our sample size is small, which limited the power to address how breast cancer risk factors or breast tumor subtypes may influence the MD-involution association. In summary, in this study of Chinse breast cancer patients, we found that breast cancer cases with dense breasts were more likely to show greater TDLU count, median span, and acini count per TDLU, particularly among older women. The observed associations between MD and TDLU measures are generally consistent with what has been previously observed among Caucasian populations ^{3, 4}. These results suggest that MD and TDLU features may be influenced by common factors and associated with similar biological processes, but the co-regulation of these two features may vary at different ages according to dynamic changes in the morphometric patterns of breast tissue. Our findings also highlight the importance of investigating the biological mechanisms underlying MD and TDLU features in diverse populations where morphometric features as well as their interactions with genetic and non-genetic risk factors may be different from Western populations.

Supplementary Material

Supp Figures

Supplementary Figure 1. Estimated TDLU counts per 100 mm² by age based on LOWESS (locally weighted scatterplot smoothing) stratified by by BI-RADS density and tumor subtype: BI-RADS a-b; luminal A (n=19); BI-RADS c-d, luminal A (n=76); BI-RADS a-b, triple-negative (n=12); BI-RADS c-d, triple-negative (n=37)

Supplementary Figure 2. Distribution of median span (µm) of 10 TDLUs by BI-RADS density in all cases and by age group

Novelty and Impact:

Most epidemiologic findings on terminal duct lobular unit (TDLU) involution in relation to breast cancer were from studies in Caucasian women. Using the standardized TDLU quantifiers assessed in benign tissues from breast cancer cases accompanied by the Breast Imaging Reporting and Data System density classification data from a tertiary hospital in Beijing, China, we extended the previous finding of the association between dense breast and greater number and size of TDLUs to an Asian population. Our finding supports the hypothesis that the higher amount of ‘at-risk’ epithelium may in part mediate the association between extensive breast density and breast cancer risk.

Abbreviations:

BBD

benign breast disease

BI-RADS

Breast Imaging Reporting and Data System

BMI

body mass index

CK5/6

Cytokeratin 5/6

EGFR

Epidermal Growth Factor Receptor

ER

Estrogen receptor

FISH

fluorescence in situ hybridization

HER2

Human epidermal growth factor receptor-2

IHC

immunohistochemistry

MD

mammographic density

OR

odds ratio

PR

Progesterone receptor

SD

standard deviation

SE

standard error

TDLU

terminal ductal lobular unit