Circulating 27-hydroxycholesterol and breast cancer tissue expression of CYP27A1, CYP7B1, LXR-β, and ERβ: results from the EPIC-Heidelberg cohort

The EPIC-Heidelberg cohort includes 25,540 study participants (13,611 women) recruited between 1994 and 1998 with between ages 35 and 65 years. Blood samples, anthropometric, reproductive, and epidemiologic factors including diet and lifestyle behavior were collected at baseline and described previously [12, 13]. Blood samples are stored under liquid nitrogen (− 196 °C) at the German Cancer Research Center (DKFZ, Heidelberg, Germany).

All incident cases of invasive breast cancer were identified and verified by study physicians during prospective follow-up (through January 2013), by active follow-up with study participants or derived from hospital records, or cancer and mortality registries. Each participant case was free of any previous invasive cancer (with the exception of non-melanoma skin cancer) or in situ breast cancer history prior to invasive breast cancer diagnosis, and had a blood sample available collected at baseline. Overall, 530 cases met these criteria, including 287 (54.1%) for which tumor blocks were available and who were included on a tissue microarray (TMA).

Up to two matched controls per case were selected from female participants in EPIC-Heidelberg with a blood sample collected at baseline and who remained cancer free (with the exception of non-melanoma skin cancer) to the time of diagnosis of their matched case. Controls were matched to cases based on age (± 3 months), menopausal status, exogenous hormones use, fasting status (< 3; 3–6; > 6 h), and time of the day (±1 h) at blood donation. Premenopausal women were also matched for phase of menstrual cycle (early follicular, late follicular, peri-ovulatory, early luteal, mid luteal, late luteal). The 287 cases included in this study were matched to 563 controls (11 cases were matched to one control only); however, all 1036 controls from the parent study were included in the current study, with statistical analyses adjusting for the matching factors. All participants provided informed consent and the study was approved by the ethics commission of the Heidelberg University Medical Faculty (reference numbers: 13/94, S-551/2014).

Tumor characteristics such as clinically measured estrogen (ER) and progesterone (PR) receptor status, breast cancer stage, grade, and morphology were extracted from clinical records by trained study physicians. Baseline information on reproductive and lifestyle factors was obtained from standardized self-administered questionnaires and computer-assisted interview at recruitment. Ever full-term pregnancy (FTP; no, yes), number of FTPs (0, 1, 2, ≥ 3), use of oral contraceptives (OC) or postmenopausal hormone therapy (HT), menopausal status (premenopausal, perimenopausal, postmenopausal (includes surgical menopause, n = 6)), ever breast-feeding (no, yes), smoking status (never, former, current), and lifetime alcohol use at recruitment (never users, former users, and lifetime users with an average intake ≥ 6, > 6–12, > 12–24, > 24–60 g/d) were available from the baseline questionnaire. Lifetime alcohol use was assessed based on the alcoholic beverage consumption during the 12 months before recruitment and the consumption reported for ages 20, 30, 40, and 50 years. Anthropometric factors were measured and recorded at baseline by trained personnel [13]. Body mass index (BMI) was categorized according to the World Health Organization definition (kg/m²: thin < 18.5, normal 18.5–25; overweight 25–30; obese ≥ 30) based on height and weight. The two lowest categories were combined, as only one participant had BMI < 18.5. The Cambridge Index of physical activity was used; this index combines occupational and recreational physical activity and is categorized into four groups (active, moderately active, moderately inactive, and inactive). Additional details on baseline data collection have been described previously [12, 13].

Collection of tumor tissue has been previously described [14]. For the current study, formalin-fixed paraffin-embedded (FFPE) breast tumor tissue material was available for 287 cases. Overall, cases included on the tissue microarrays (TMAs) were younger at diagnosis (59.1 versus 61.9 years, p < 0.01) and were more likely to have been diagnosed with higher-grade tumor (p = 0.008) and ductal morphology (p < 0.01), compared to cases not included on the TMAs (Additional file 1: Table S1). There were no significant differences regarding reproductive, anthropometric, and lifestyle factors, or for blood biomarkers, including circulating 27HC concentration.

TMA slides were prepared by taking two cores of 1 mm from a representative section of tumor for each case. Immunohistochemistry (IHC) staining was performed on TMA slides at the Tissue Bank of the National Center for Tumor Diseases (NCT), Heidelberg, Germany. A single pathology resident (B.W.) evaluated protein expression of CYP27A1, CYP7B1, LXR-β, and ERβ. Each case had two cores on the TMA; the highest score of the two cores was used to determine staining positivity. Immunostaining was scored either for cytoplasmic (CYP27A1, CYP7B1) or nuclear (LXR-β, ERβ) expression. Staining intensity in tumor cells was scored as 0 (absent), 0.5 (borderline), 1 (weak), 2 (moderate), or 3 (strong). Tumor cells were considered positive for CYP7B1 [15] and LXR-β when more than 10% of the cells showed moderate or strong staining. ERβ was considered positive when moderate or strong staining was observed for more than 50% of tumor cells [16]. For CYP27A1, any moderate to strong tumor cell staining was considered positive [6]. Table 1 shows antibodies and dilutions used and the definitions applied to determine tumor marker positivity. Where a conclusive result for a marker was not available, that marker was categorized as missing (CYP27A1, 4.8%; CYP7B1, 15%). For one slide, CYP7B1 staining failed, resulting in the exclusion of 43 cases for that marker. Additionally, the expression of Ki67 proliferation activity (low, high), Bcl-2 (negative, positive), and p53 (negative, positive) were also assessed in these cases, as previously described [14]. Pathology personnel were blinded to biomarker status.

Serum 27HC concentrations were quantified using liquid chromatography–mass spectrometry (LC-MS) at Biocrates Life Sciences (Innsbruck, Austria); lipid biomarkers (cholesterol, triglycerides, high-density lipoproteins, and low-density lipoproteins) were quantified by Synlab MVZ Heidelberg GmbH (Eppelheim, Germany); and sex steroids (testosterone, progesterone, estradiol, estrone, and dehydroepiandrosterone sulfate (DHEAS)) were measured in the Division of Cancer Epidemiology labs at the DKFZ. Case and control sets were analyzed together within the analytic batch, with case and control position randomized within the set. Laboratory personnel were blinded to case or control status, and two blinded pooled quality control (QC) samples were included to monitor the assay precision. Details on the measurement of the blood-based biomarkers have been described previously [8].

Blood biomarkers (27HC, lipids, and sex steroids) were log2-transformed to better fit a normal distribution. Fisher’s exact test (categorical variables) and Welch’s t-test (continuous variables) were used to compare tumor characteristics, reproductive, anthropometric, and lifestyle factors, as well as circulating lipid and sex steroid levels by CYP27A1, CYP7B1, LXR-β, and ERβ tumor marker status (positive, negative). Analyses of sex steroid hormones were stratified by menopausal status and by HT use among woman postmenopausal at blood collection. We used menstrual cycle phase-specific residuals for estrone, estradiol, and progesterone to account for within-person variability in these hormones across the menstrual cycle among postmenopaual women. Odds ratios (ORs) and 95% confidence intervals (CIs) were estimated using unconditional logistic regression models, adjusted for the study matching factors, stratified by tumor marker status. Additionally, ORs were adjusted for age at first menstrual period, BMI, age at first full-term pregnancy, and number of FTP since adjustment for additional variables had minimal impact on the ORs in the parent study [8]. A total of 8 women had missing values for both fasting status and time at blood draw (1 case and 7 controls). These women were excluded from the risk analyses; thus, the sample used for these analyses included 286 cases and 1029 controls. Further, analyses were stratified by menopausal status given significant heterogeneity in associations between circulating 27HC and breast cancer risk by menopausal status was previously observed [8]. Heterogeneity by tumor marker status (positive vs. negative) was assessed using polytomous logistic regression models comparing models assuming the same association versus different associations in breast cancer in subgroups [17]; models were compared using a likelihood ratio test. We evaluated heterogeneity by menopausal status at blood collection by including an interaction term and evaluating the Wald test. Statistical analyses were conducted using SAS software, version 9.3 (SAS Institute, Cary, NC, USA). P values are two-sided, and p < 0.05 was considered statistically significant.

A total of 26.4% of the cases were classified as positive for CYP27A1, 35.4% were classified as positive for CYP7B1, 57.1% were classified as positive for LXR-β, and 70.7% were classified as positive for ERβ. A higher proportion of CYP27A1-positive tumors and LXR-β-positive tumors were also ERβ-positive (p < 0.01) (Additional file 2: Table S2), relative to those negative for the respective markers. No other statistically significant associations between the four evaluated markers were observed (p ≥ 0.16).

No significant associations was observed between CYP27A1 or LXR-β expression and case characteristics (Table 2), though the proportion of cases with high expression of p53 was suggestively higher in CYP27A1-positive vs. CYP27A1-negative tumors (88.4% vs. 77.6%, p = 0.05). Differences in tumor characteristics by CYP7B1 and ERβ expression were minimal. A higher proportion of PR-positive (p = 0.04) tumors was observed in CYP7B1-positive tumors, in comparison with CYP7B1-negative tumors (PR-positive: 83.7% vs. 72.0%, respectively). A higher proportion of Bcl2_low expression (66.7% vs. 49.3%, p = 0.01) was observed in ERβ-positive compared to ERβ-negative tumors. No associations were observed between the investigated markers and other breast cancer characteristics (p ≥ 0.06), including age at diagnosis and cancer stage, grade, or morphology, or with any of the investigated epidemiologic factors (p ≥ 0.07, e.g., BMI, number of term pregnancies, menopausal status (Additional file 3: Table S3)). No associations between the tumor markers and serum 27HC were observed (Additional file 4: Table S4). Circulating high-density lipoproteins differed significantly by tumor CYP27A1 expression (p = 0.04; geometric means, CYP27A1-positive 64.3 mg/dl, CYP27A1-negative 68.8 mg/dl) and triglycerides differed significantly by tumor CYP7B1 expression (p = 0.02; geometric means, CYP7B1-positive 106.4 mg/dl, CYP7B1-negative 123.1 mg/dl). No consistent patterns were observed for sex steroid hormones, though higher circulating premenopausal DHEAS was observed among women with CYP27A1-positive compared to CYP27A1-negative tumors (p = 0.04), and higher circulating premenopausal estradiol (p = 0.02) and lower perimenopausal testosterone (p = 0.03) were observed among women with CYP7B1-positive compared to CYP7B1-negative tumors.

We observed no statistically significant heterogeneity in associations between circulating 27HC and breast cancer risk by tumor expression of CYP27A1, CYP7B1, LXR-β, or ERβ (Table 3). Heterogeneity in associations between circulating 27HC and breast cancer risk by menopausal status at blood collection was only observed in tumor not expressing LXR-β (p_het = 0.02). This was driven by a significant positive association observed among perimenopausal women; however, sample size was limited in this subgroup (n = 54 total perimenopausal cases). In postmenopausal women, higher 27HC concentrations were associated with lower risk of ERβ-negative breast cancer (OR_log2:0.31; 95%CI 0.10, 0.92; p_het by menopausal status in ERβ-negative = 0.06), whereas no association was observed among ERβ-positive cases (OR_log2: 0.72 (95% CI 0.36, 1.47); p_het by ERβ status in postmenopausal = 0.12) (n = 146 postmenopausal cases with ERβ data).