Background

The RANK-axis consists of three tumor necrosis superfamily (TNF) members; receptor activator of nuclear factor kappa-B (RANK), its ligand (RANKL), and osteoprotegerin (OPG). Binding of RANKL to RANK promotes cell proliferation and, in experimental models, promotes primary mammary tumorigenesis and mammary stem cell expansion [1,2,3,4]. RANK-signaling is a mediator of progesterone-signaling and overexpression of RANK in mouse-mammary-tumor-virus models of hormone responsive breast cancer shows increased rates of hyperplasia and tumor development [1, 5]. OPG is the decoy receptor for RANKL and can downregulate RANK-signaling. OPG additionally serves as a decoy receptor for TNF related apoptosis inducing ligand (TRAIL) and may downregulate TRAIL-signaling, a process promoting cell death, especially in estrogen receptor (ER) negative breast cancer cells [6].

There has been increasing interest in the RANK-axis with respect to breast cancer risk and prognosis given the availability of a RANKL inhibitor, denosumab, which has been shown to reduce skeletal-related events in breast cancer patients with bone metastases [7] and may improve disease-free survival in postmenopausal breast cancer patients with ER and progesterone receptor (PR) positive disease [8]. We and others have recently shown that both sRANKL (soluble homotrimeric isoform of RANKL) and OPG concentrations in circulation may influence risk of breast cancer in humans [9,10,11,12,13]. Following our earlier investigations on pre-diagnosis sRANKL and OPG and breast cancer risk, the aim of this study was to investigate associations between pre-diagnosis concentrations of sRANKL and OPG and risk of death after a breast cancer diagnosis. Given the results reported to date, we hypothesized (1) higher sRANKL and lower OPG would be associated with higher risk of breast cancer-associated death among women with ER+ breast cancer; and, (2) a positive association between OPG and risk of breast cancer-associated death among women with ER- disease.

This study provides the first data on circulating RANK-axis members and breast cancer-specific mortality risk and the first data on differences in mortality risk by tumor hormone receptor status.

Methods

Study population: European Prospective Investigation into Cancer and Nutrition

The European Prospective Investigation into Cancer and Nutrition (EPIC) recruited more than 520,000 participants (367,993 women), aged predominantly 35–75 years, between 1992 and 2000 in ten European countries (Denmark, France, Germany, Greece, Italy, the Netherlands, Norway, Sweden, Spain, and the United Kingdom). Detailed dietary, reproductive, lifestyle, anthropometric, and medical history data were collected using standardized methods [14]. Incident cancer cases were identified through cancer registries in most countries; France, Germany, Greece, and the Naples (Italy) center conducted follow-up through review of health insurance records, contact with cancer and pathology registries, and/or direct contact with cohort members. Mortality data were obtained via active follow-up with participants and their next of kin in Germany and Greece, and via national and regional mortality registries in the remaining countries [16].

Blood sample collection

A total of 64% (n = 235,607) of women provided a blood sample at baseline. Blood samples were collected according to standardized protocols. As independent studies on breast cancer were conducted by the Swedish centers, participants from these centers were not included in the current study. For all countries included in this study, except Denmark, half of the aliquots were stored locally and the other half centrally at the International Agency for Research on Cancer (IARC). The samples used in this study were stored at IARC under liquid nitrogen at − 196 °C, or locally at − 150 °C for Danish participants.

The EPIC study protocol was approved by ethical committees of all participating centers and all participants gave written informed consent. The protocol for the current study was approved by the ethical committees of the International Agency for Research on Cancer (IARC; project no. 12–42) and the University of Heidelberg (project no. S311/2014).

Study design

The breast cancer cases in this study were part of a case-control study nested within the EPIC cohort. The study design and methods have been described previously [9, 12, 15]. Briefly, women diagnosed with a first invasive breast cancer between blood collection (ranging from 1992 to 2000 between centers) and completion of last follow-up for breast cancer incidence at the time the case-control study was initiated (ranging from 2003 to 2006 between centers) were included (Fig. 1a). The majority of EPIC participants were followed-up for breast cancer incidence and subsequent mortality via national or regional registries [16], and available information on tumor characteristics (e.g., hormone receptor subtype and stage at diagnosis) was collected where available. End of follow-up for mortality was defined as date of last complete follow-up for vital status, death, or emigration, and ranged from 2009 to 2015 between centers.

Fig. 1
figure 1

Study design and data collection. a Data were collected at three timepoints. Baseline blood samples were collected between 1992 and 2000. Breast cancer cases were diagnosed between baseline and initiation of the case-control study in 2006, and tumor characteristics at the time of breast cancer diagnosis were collected. Participants were followed-up for mortality from the time of breast cancer diagnosis to 2015. b 2020 breast cancer cases were initially selected for the case-control study. Of these, 1970 cases had complete information on sRANKL concentrations and follow-up for mortality, and 2006 had complete information on OPG concentrations and follow-up

Full size image

All cases with available serum sample and information on ER status of the tumor were eligible for the nested case-control study (Fig. 1b). From 2004, all postmenopausal ER- breast cancer cases were included, with one ER+ case randomly selected for every ER- case (matched on center). A total of 2020 breast cancer cases were initially available for the current analyses however, seven had no follow-up information after their breast cancer diagnosis and were excluded.

Laboratory analyses

Pre-diagnosis sRANKL and OPG concentrations were analyzed at the Laboratory of the Division of Cancer Epidemiology at the German Cancer Research Center (DKFZ). Free serum sRANKL was quantified using an enzyme-linked immunosorbent assay (Biomedica, Austria), total serum OPG using an electrochemiluminescence assay (MesoScale Diagnostics, USA). All batches included the same serum quality control samples in duplicate to monitor inter-batch variation. Measurements and standard curves were done on a Victor system using Workout 2.5 software (Perkin Elmer) for sRANKL. The Quickplex SQ 120 Reader and Workbench 4.0.12 software (MesoScale Diagnostics) were used to measure OPG and create standard curves.

Of the 2013 breast cancer cases with complete follow-up data, four were missing OPG and 44 were missing sRANKL concentrations (38 cases, equipment failure and insufficient volume to re-assay; Fig. 1b). 152 cases (7.5%) with sRANKL values below the lower limit of detection (LLOD, 0.01 pmol/L) were set to half of the LLOD.

Inter-batch coefficients of variation were 1.2% for sRANKL and 16.6% for OPG. Intra-batch coefficients of variation were 14.4% for sRANKL, and 15.3% for OPG. Within-person reproducibility of sRANKL and OPG over one and 14 years observed in our study have been published previously [9, 12], Spearman correlation coefficients were r = 0.85 and r = 0.75 for OPG and r = 0.60 and r = 0.38 for sRANKL over one and 14 years, respectively.

Statistical analyses

Pre-diagnosis sRANKL and OPG concentrations were log2 transformed to normalize the distributions, and to allow estimation of the effect of a doubling in concentrations. The ratio was calculated by dividing sRANKL concentrations by OPG concentrations; the ratio was then log2 transformed.

Outliers were evaluated using the extreme studentized deviate test [17]; three participants with outlying OPG concentrations (two with low, one with high OPG concentrations) were excluded from analyses (Fig. 1b). The final study population included 2006 breast cancer cases with OPG, 1970 with sRANKL, and 1965 cases with both.

We used Cox proportional hazards regression to estimate hazard ratios (HR) and 95% confidence intervals (CI) for risk of breast cancer-specific and all-cause mortality, using time since diagnosis as the time scale. sRANKL, OPG, and the sRANKL/OPG ratio were modeled as quintiles; tests for trend were calculated using continuous (log2) variables. The proportional hazards assumption was assessed using Schoenfeld residuals [18]. Based on our previous work on breast cancer risk showing significant heterogeneity by hormone receptor status [9, 12], and the oversampling of ER- cases after 2004, we decided a priori to evaluate associations for mortality both overall and by ER status. Similarly, confounders were selected a priori. Multivariable models were adjusted for body mass index (BMI; continuous), age at blood collection (continuous), age group at menarche (≤12, 13 and missing (as very few cases were missing information), 14, ≥15 years), age group at menopause (premenopausal, ≤48, 49–51, ≥52 years, missing), age group at first full term pregnancy (nulliparous, < 25, ≥25 years and missing), and breast cancer stage (localized, non-localized (including regional, distant, and unspecified metastatic sites), missing). Models were stratified by age at diagnosis (5-year age groups) and tumor ER status (negative or positive, in models among the whole population), as these variables violated the proportional hazards assumption. Additional adjustment for use of oral contraceptives or postmenopausal hormones at blood collection did not impact results (log2 HR < 10% change). Pre-diagnosis concentrations of sRANKL and OPG were weakly inversely correlated, with Spearman correlations of r = −0.25 in premenopausal and r = −0.32 in postmenopausal women.

Non-parametric restricted cubic splines were used to examine possible non-linearity, comparing models with linear and cubic terms to models with only the linear term [19]. There was no evidence of significant deviation from linearity (p >  0.11). We evaluated interaction between pre-diagnosis sRANKL and OPG concentrations and reproductive and lifestyle factors by comparing models with an interaction term to models without, using likelihood ratio tests. We observed significant interaction between OPG and BMI (p ≤ 0.04 in the whole population and in ER+ cases), and thus investigated associations between pre-diagnosis OPG and mortality after a breast cancer diagnosis in stratified models (BMI </≥ 25 kg/m2; i.e. non-overweight/overweight). We observed no significant interaction for the remaining factors (p ≥ 0.05), including menopausal status at blood collection, ages at blood collection, menarche, menopause, and first full term pregnancy, and postmenopausal hormone (PMH) use at blood collection. Similarly, there was no heterogeneity in associations by breast cancer stage at diagnosis (localized vs. non-localized). Information on breast cancer treatment was not available, thus we were not able to evaluate treatment-related factors as covariates or effect modifiers.

In addition to the ratio of pre-diagnosis sRANKL and OPG concentrations we evaluated mutually adjusted models (i.e. sRANKL models additionally adjusted for log2 OPG concentrations and vice versa) and a cross-classification of pre-diagnosis sRANKL and OPG at the median concentration (as OPG may inhibit sRANKL signaling, those with sRANKL concentrations ≤ median and OPG concentrations > median were chosen as the reference group). We further conducted a sensitivity analysis excluding those diagnosed within two years of blood collection to address potential reverse causation.

All statistical tests were two-tailed and considered significant at p < 0.05. Statistical analyses were conducted using SAS 9.4 (SAS Institute Inc., Cary, NC, USA).

Results

Population characteristics

Among 2006 breast cancer cases in whom OPG concentrations were available, the median age at blood collection was 56.6 (range: 26.7, 75.5) years; 1543 (76.9%) of breast cancer cases were postmenopausal and of these, 48.8% were using PMH at blood collection (Table 1). The majority (86%) of cases had at least one full term pregnancy, with a median age at first full term pregnancy of 25 (16.0, 44.0) years. sRANKL and OPG concentrations were measured in blood samples collected a median of 4.7 (0.02, 11.7) years before breast cancer diagnosis. The median age at diagnosis was 60.9 (35.2, 83.6) years and the majority of cases (n = 1620, 80.8%) were diagnosed with ER+ breast cancer. The median time between diagnosis and end of follow-up was 10.9 (0.05, 19.1) years; the median survival time between diagnosis and death was 6.5 (0.1, 18.5) years among those who died of any cause and 5.0 (0.8, 15.2) years among those who died of breast cancer. A total of 421 deaths, including 250 breast cancer deaths, occurred over 21,253 person years of follow-up (Table 1). Compared to those diagnosed with ER- breast cancer, those diagnosed with ER+ disease where slightly older at blood collection and diagnosis, more likely to also have PR+ disease, and had a longer survival time. Though sRANKL concentrations were available for a smaller population than OPG concentrations (n = 1970 cases for sRANKL and n = 1965 for the sRANKL/OPG ratio), population characteristics were very similar (data not shown).

Table 1 Characteristics of the full study population and ER+ and ER- subgroups
Full size table

OPG concentrations

Higher pre-diagnosis OPG concentrations were associated with an increased risk of breast cancer-specific mortality among women with ER+ disease (quintile (q)5 vs. q1 HR 1.77 [CI 1.03, 3.04]; ptrend 0.10) (Table 2). Additional adjustment for pre-diagnosis sRANKL concentrations strengthened this association (q5 vs q1 HR 2.02 [1.15, 3.54]; ptrend 0.07). For all-cause mortality, higher pre-diagnosis concentrations of OPG were associated with a suggestive increased risk of mortality in all cases (q5 vs. q1 HR 1.25 [CI 0.90, 1.73]; ptrend 0.02) and in ER+ cases (q5 vs. q1 HR 1.39 [CI 0.94, 2.05); ptrend 0.02). Though the p-value for linear trend was attenuated, additional adjustment for pre-diagnosis sRANKL concentrations did not substantially impact risk estimates (Table 2). We did not observe heterogeneity by ER status at diagnosis (phet 0.58); however, OPG was not associated with mortality risk in those diagnosed with ER- breast cancer.

Table 2 Circulating concentrations of OPG and risk of death following a breast cancer diagnosis, by ER subtype
Full size table

Stratifying by BMI at blood collection (pint ≤ 0.04 in models for OPG among all cases and ER+ cases) showed that pre-diagnosis OPG concentrations were not associated with risk of death after a breast cancer diagnosis in those with a high BMI (≥ 25 kg/m2) at blood collection (Additional file 1: Table S1). Among those with a lower BMI (< 25 kg/m2), high pre-diagnosis concentrations of OPG were strongly associated with an increased risk of both breast cancer-specific and all-cause mortality, especially among those with ER+ disease (e.g. breast cancer mortality in ER+ cases q5 vs q1 HR 2.80 [1.30, 6.02]; ptrend 0.01, additionally adjusted for pre-diagnosis sRANKL concentrations HR 3.19 [1.46, 6.97]; ptrend 0.005).

sRANKL concentrations

Pre-diagnosis sRANKL concentrations were not associated with breast cancer-specific or all-cause mortality (e.g. breast cancer-specific mortality, in the full population: q5 vs q1 HR 0.99 [0.66, 1.47]; ptrend 0.84; Table 3). Results were unchanged by additional adjustment for pre-diagnosis OPG concentrations.

Table 3 Circulating concentrations of sRANKL and risk of death following a breast cancer diagnosis, by ER subtype
Full size table

sRANKL/OPG ratio and cross-classification

A higher ratio between pre-diagnosis sRANKL and OPG concentrations was not associated with breast cancer-specific or overall mortality risk (Additional file 1: Table S2). We observed no associations between cross-classified pre-diagnosis sRANKL/OPG and breast cancer-specific mortality (Additional file 1: Table S3). In line with results for OPG, all-cause mortality risk was lower in ER+ cases with both pre-diagnosis sRANKL and OPG values below the median (HR 0.63 [CI 0.44, 0.92]), relative to those with low sRANKL and high OPG concentrations. Results for breast cancer-specific mortality were of similar magnitude, though not significant (HR 0.69 [0.42–1.15]).

Sensitivity analyses

Information on PR status was available for 1478 (74%) of cases; stratification by both ER and PR status (i.e. ER + PR+ and ER-PR-) did not materially affect results (data not shown). Excluding those diagnosed with breast cancer within two years of blood collection did not impact results for pre-diagnosis sRANKL or the pre-diagnosis sRANKL/OPG ratio, but did attenuate associations between pre-diagnosis OPG concentrations and risk death (e.g. ER+ cases: all-cause mortality q5 vs q1 HR 1.24 [0.79–1.93]; ptrend 0.13 and breast cancer-specific q5 vs q1 HR 1.64 [0.89–3.05]; ptrend 0.23). There was no heterogeneity in associations by breast cancer stage at diagnosis (breast cancer-specific mortality phet ≥ 0.18; all-cause mortality phet ≥ 0.43), and we observed no heterogeneity in associations by menopausal status at blood collection for sRANKL and sRANKL/OPG (phet ≥ 0.44). For OPG, we observed no heterogeneity in associations by menopausal status at blood collection for breast-cancer specific mortality (Phet ≥ 0.14), and suggestive heterogeneity in models for all-cause mortality (phet = 0.05 in the full population). In analyses stratified by menopausal status at blood collection associations between OPG and all-cause mortality among postmenopausal women were similar to those in the full population (e.g. q5 vs. q1 HR 1.35 [CI 0.93, 1.98]; ptrend 0.004). In the smaller group of women premenopausal at blood collection (n = 75 deaths and 60 breast cancer deaths), log2 OPG concentrations were not associated with breast cancer risk.

Discussion

In this large-scale prospective study, high pre-diagnosis concentrations of OPG were associated with an increased risk of death after an ER+ breast cancer diagnosis, especially among women with BMI less than 25 kg/m2. Pre-diagnosis sRANKL concentrations and the sRANKL/OPG ratio were not associated with mortality following a breast cancer diagnosis.

Experimental data in mouse models show RANKL blockade using OPG-Fc reduced formation of breast cancer metastases [20,21,22]. In studies in breast cancer patients, most reported either no or an inverse association between tumor RANKL or OPG expression and risk of death [23,24,25,10]. It would be of interest to further investigate these differences in larger studies that can account for e.g. tumor characteristics. We have recently shown a positive association between OPG concentrations and risk of ER- breast cancer (tertile 3 vs. 1 RR = 1.93 [95% CI 1.24–3.02]; ptrend = 0.03) [9], whereas higher sRANKL concentrations were associated with risk of ER+ disease (quintile 5 vs. 1 RR 1.28 [95%CI 1.01–1.63]; ptrend 0.20) [12]. Results from our current study indicate circulating concentrations of OPG and sRANKL may impact cancer risk and mortality differently, though further studies are required to more fully understand the underlying mechanisms.

Conclusions

Higher pre-diagnosis endogenous concentrations of the decoy receptor for RANKL, OPG, appear to increase risk of death after a breast cancer diagnosis especially in those diagnosed with ER+ disease. Further investigations in well-defined patient cohorts are needed to confirm these results, and to clarify whether circulating OPG may be relevant for breast cancer prognosis.