Circulating insulin-like growth factor-I, insulin-like growth factor binding protein-3 and terminal duct lobular unit involution of the breast: a cross-sectional study of women with benign breast disease

The Breast Radiology Evaluation and Study of Tissues (BREAST) Stamp Project is a cross-sectional molecular epidemiologic study of mammographic density in 465 women ages 40 to 65 years, referred for diagnostic biopsy from 2007 through 2010 at the University of Vermont (UVM) College of Medicine and its affiliated academic medical center, the UVM Medical Center (formerly Fletcher Allen Health Care) [17]. The current analysis included women in the BREAST Stamp Project with available serum (>3.2 mL serum available), who were not using exogenous hormones at the time of biopsy, and who had available quantitative histologic assessment of TDLU involution, resulting in 281 eligible women for IGF-I and IGFBP-3 serum testing (see below). Of the 281 women with IGF-I and IGFBP-3 measurements, we further excluded 47 women with breast cancer and 6 without TDLU data, resulting in an analytic sample of 228 women, including 155 premenopausal and 73 postmenopausal women. For 2 of the 228 women, it was unknown whether they were current users of menopausal hormone therapy. These women were included in the analytic population, given that their exclusion did not alter the study results. Participants provided written informed consent, which included providing access to medical records and mammographic images; Institutional Review Boards at the University of Vermont and the National Cancer Institute approved this study.

Women were referred for image-guided breast biopsy following an abnormal breast imaging examination. Digital mammographic examinations were performed with a density phantom, permitting measurement of percent volumetric mammographic density using single x-ray absorptiometry (SXA) [18]. For the same digital images that were selected for SXA analyses, measures of percent dense area were estimated, as described previously [19, 20], using interactive, computer-assisted thresholding software comparable to other validated methods [21]. Percent volumetric mammographic density as measured using SXA has been found to be accurate to equivalent measures acquired from magnetic resonance imaging (MRI) of the breast [22], and significantly associated with breast cancer risk [20, 23]. Density measures used in this study are from the ipsilateral breast on which the biopsy was performed and TDLU measures assessed.

Demographic and breast cancer risk factor information was collected using a self-administered questionnaire (see sample questionnaire at http://​breastscreening.​cancer.​gov/​ [24]) and a supplementary telephone interview. A woman was considered postmenopausal if menstrual periods had stopped more than 12 months prior to interview, if she had undergone bilateral oophorectomy, or if she had undergone hysterectomy (or gynecologic surgery associated with cessation of menses) and was 55 years of age or older; otherwise, a woman was considered premenopausal. Given the age range of our study population, the premenopausal subgroup likely contains a mixture of both pre- and perimenopausal women. For a subset of premenopausal women with previously measured serum follicle stimulating hormone (FSH) and estradiol levels (n = 105), we identified perimenopausal women (n = 19) as previously described [25]. Body mass index (BMI, kg/m²) was computed from measures of height and weight obtained on the day of the breast biopsy. Whole blood samples were collected prior to breast biopsy using standard techniques, allowed to clot for 30 minutes, and processed at the UVM General Clinical Research Center. Samples were centrifuged at 3,000 rpm for 15 minutes, and the serum and clot fractions were frozen at –80 °C until shipment to SeraCare Life Sciences (Gaithersburg, MD, USA), where they were stored in liquid nitrogen.

IGF-I and IGFBP-3 levels were measured in duplicate in masked serum samples by enzyme-linked immunosorbent assay (ELISA) (Diagnostic Systems Laboratory (Webster, TX, USA)) as described previously [26]. Three quality control (QC) samples (follicular phase, luteal phase and postmenopausal samples) were measured in duplicate in a masked fashion with each of the eight assay batches performed. The mean intra-batch coefficients of variation (CVs) were 2.7 % for IGF-I and 5.1 % for IGFBP-3, and the inter-batch CVs were 10.5 % for IGF-I and 10.5 % for IGFBP-3. Mean IGF-I and IGFBP-3 levels for each subject’s duplicate measurements were used in the analyses. To approximate the circulating free, bioactive levels of IGF-I, the molar ratio of IGF-I to IGFBP-3 was estimated as previously described [27, 28]. Overall and menopausal status-specific tertiles of IGF-I, IGFBP-3 and their molar ratio were calculated and used in subsequent association analyses.

Ultrasound-guided core needle (14-gauge needle) or stereotactic-guided vacuum-assisted (9-gauge needle) breast biopsies were routinely processed as formalin-fixed paraffin-embedded blocks, which were sectioned and stained with hematoxylin and eosin (H&E) for diagnosis. For study purposes, final diagnoses were categorized as non-proliferative benign breast disease, proliferative (ductal hyperplasia; sclerosing adenosis), proliferative with atypia (atypical ductal or lobular hyperplasia), in-situ, or invasive breast carcinoma based on review of microscopic sections and surgical pathology reports. The present analysis was restricted to women whose biopsy diagnoses were benign (Additional file 1: Table S1). Digitized images of sections from the target block were used for analysis of TDLU involution (described below).

H&E-stained tissue sections were digitized at × 20 magnification (Aperio ScanScope CS, Vista, CA, USA), and prepared for web-based viewing and annotation with Digital Image Hub software (SlidePath/Leica, Dublin, Ireland). The lasso tool in Digital Image Hub was used to manually outline and measure total tissue area (mm²) per section. For women who had multiple benign biopsy targets, measures from the first biopsy were used (n = 5). Sections were visually assessed to estimate percentage fat (0, 1, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90, 95 and 100 %). Normal TDLUs per section were enumerated, and up to 10 TDLUs were evaluated for maximum diameter as TDLU span (measured with an electronic ruler in microns) and categories of acini counts per TDLU (categories: 1, ≤10; 2, 11–20; 3, 21–30; 4, 31–50; 5, 51–100; and 6, >100) to provide reliable estimates, as previously described [6]. Median values for each woman were used as summary measures of TDLU span and acini count. Overall and menopausal status-specific tertiles of median TDLU span and median category of acini counts per TDLU were calculated and used in subsequent analyses. A previous study demonstrated high intra-observer agreement (Spearman’s r >0.90) for the study pathologist (MES) for the TDLU measures [6].

Because IGF levels have been shown to differ by menopausal status [10, 29], analyses of associations between circulating IGF-I or IGFBP-3 levels and TDLU measures were done for all women combined and women stratified by menopausal status. Circulating IGF levels and TDLU measures in premenopausal and postmenopausal women were compared using the Kruskal–Wallis rank test. Relationships between circulating IGF levels, TDLU measures and mammographic density were assessed with Spearman rank correlation; Spearman partial rank correlation was also estimated after removing the effects of age and BMI. To account for excess zero TDLU counts in the study population, multivariable zero-inflated Poisson (ZIP) regression [30] was used to estimate relative risks (RRs) and 95 % confidence intervals (CIs) for the relationship between circulating IGF levels (in tertiles) and number of TDLUs (i.e., TDLU count). ZIP regression models included tissue area as an offset to account for differences in biopsy needle gauges. Potential confounders were identified separately for pre- and postmenopausal women for inclusion in the multivariable models. We also explored inclusion of interaction terms in sensitivity analysis. Among all women, interactions between BMI and menopausal status and BMI and percent fat on the tissue slide were statistically significant; however, their inclusion did not alter the observed RRs, and the most parsimoniously adjusted model is therefore presented. Among postmenopausal (but not premenopausal) women, we identified a significant interaction between BMI and percent fat on the tissue slide, which was retained in the final model to improve fit even though its inclusion did not alter results for the main exposures.

Among women with at least one TDLU, ordinal logistic regression models were used to estimate odds ratios (ORs) and 95 % CIs for associations between circulating IGF levels (independent variable, categorized into tertiles) and the outcomes of median TDLU span and median category of acini count per TDLU (categories: 1, ≤10; 2, 11–20; 3, 21–30; 4, 31–50; 5, 51–100; and 6, >100). For these analyses, adjustment factors were identified by stepwise selection with an inclusion/exclusion criterion of P <0.05 and P >0.10, respectively. The variables that were considered included age at biopsy, BMI, ages at menarche, first birth and menopause (among menopausal women), parity and percent fat on the tissue slide. Final models for each analysis are included within the respective table footnotes.

We tested for effect modification of the association between IGF levels and TDLU count by stratifying models on mammographic density (menopausal-specific tertiles of percent volumetric density). Stratified analyses were adjusted for the confounders identified in the main analysis with the exception of the interaction between BMI and percent fat on the tissue slide, as its inclusion yielded similar results, and age at menarche in the stratum of the highest mammographic density tertile among postmenopausal women, in order to improve model convergence. We used the Wald test to compare the estimates from models stratified by tertiles of mammographic density to obtain a P value for interaction (P-int). Sensitivity analyses using percent dense mammographic area yielded similar conclusions (data not shown). In sensitivity analyses, we also restricted analyses of premenopausal women to those with available hormone data, excluding potentially perimenopausal women. In addition, we mutually adjusted models relating IGF-I and IGFBP-3 to TDLU counts for IGFBP-3 and IGF-I, respectively. Statistical analyses were conducted using Stata version 11.2 (Statacorp, College Station, TX, USA), with the exception of the tests for interaction which were conducted in R version 2.14.1 [31].

Overall, 94 % of subjects were white and 68 % were premenopausal. The median age at biopsy for premenopausal women was 47 years and for postmenopausal women was 57 years (Table 1). Approximately 74 % of premenopausal and 77 % of postmenopausal women in our study population were parous. Nearly half (45 %) of parous postmenopausal women had age at first birth <25 years, whereas only 24 % of parous premenopausal women had a first birth at <25 years of age (P = 0.01, Table 1). As previously reported in this study population [17], volumetric and area percent mammographic density measures were higher among premenopausal as compared with postmenopausal women (Additional file 1: Table S1).

Median levels of IGF-I and the IGF-I:IGFBP-3 molar ratio were higher among premenopausal women as compared with postmenopausal women (P ≤0.001; Table 2), whereas levels of IGFBP-3 were not significantly different between these two groups (P = 0.44). Correlations between TDLU count, TDLU span and acini count were generally similar by menopausal status, such that TDLU span and acini count were correlated (r >0.60; P ≤0.001), whereas TDLU count was not significantly correlated with either acini count or TDLU span (Additional file 2: Table S2). Correlations observed for IGF and TDLU measures with mammographic density among pre- and postmenopausal women were attenuated after removing the effects of age at biopsy and BMI (Table 2).

Among all women combined, IGF-I levels in the highest tertile versus the lowest were associated with higher TDLU counts in fully adjusted models (RR_T3vsT1 = 1.31, 95 % CI = 1.18–1.45; P-trend <0.0001; Table 3). IGFBP-3 levels in the highest tertile versus the lowest were associated with lower TDLU counts (RR_T3vsT1 = 0.88, 95 % CI = 0.80–0.96; P-trend = 0.009). IGF-I:IGFBP-3 ratio levels in the highest tertile versus the lowest were associated with higher TDLU counts (RR_T3vsT1 = 1.84, 95 % CI = 1.65–2.06; P-trend <0.0001; Table 3). Results for postmenopausal women were similar to those for all women, although effects for IGFBP-3 (RR_T3vsT1 = 0.56, 95 % CI = 0.45–0.70; P-trend <0.0001) and the IGF-I:IGFBP-3 ratio (RR_T3vsT1 = 2.78, 95 % CI = 2.14–3.62; P-trend <0.0001; Table 3) were stronger. In contrast, only IGF-I:IGFBP-3 ratio levels in premenopausal women were associated with higher TDLU counts (RR_T3vsT1 = 1.17, 95 % CI = 1.04–1.32; P-trend = 0.01). IGF-I and IGFBP-3 and their ratio were not significantly related to TDLU span or number of acini per TDLU (Additional file 3: Table S3 and Additional file 4: Table S4).

Findings from sensitivity analyses restricted to premenopausal women with available hormone data (n = 86) were consistent with those observed among all premenopausal/perimenopausal women combined, such that those in the highest tertile of IGF-I:IGFBP-3 ratio levels had higher TDLU counts (data not shown). In addition, patterns of association between IGF-I or IGFBP-3 and TDLU counts were similar to and somewhat stronger than those reported in Table 3, following mutual adjustment for IGFBP-3 and IGF-I, respectively (Additional file 5: Table S5).

IGFBP-3 was inversely correlated with mammographic density and the IGF-I:IGFBP-3 ratio was positively correlated with mammographic density, but only significantly so among premenopausal women (r = –0.22 and r = 0.27, respectively; P <0.05; Table 2). Similarly, TDLU count was positively correlated with mammographic density, particularly among premenopausal women (Table 2). These associations were attenuated after adjustment for age and BMI (Table 2). Examination of the relationship between TDLU count and circulating IGF levels by tertiles of mammographic density revealed differences in the strength of the association for women across the density spectrum. As shown in Fig. 1a, among premenopausal women a positive association between the IGF-I:IGFBP-3 ratio and TDLU count was observed for women in the highest tertile of mammographic density (P-int = 0.006). For postmenopausal women (Fig. 1b), mammographic density modified all IGF associations with TDLU count; associations were strongest among those with higher mammographic density (IGF-I, P-int <0.0001; IGFBP-3, P-int = 0.02; and IGF-I:IGFBP-3 ratio, P-int <0.0001). Patterns of association were similar, though stronger, following mutual adjustment for IGFBP-3 and IGF-I (Additional file 6: Figure S1).