Molecular essence and endocrine responsiveness of estrogen receptor-negative, progesterone receptor-positive, and HER2-negative breast cancer

Cohort 1 was obtained from the database of the Surveillance, Epidemiology, and End Results (SEER) program in the United States. Cohort 2 was retrieved from the Fudan University Shanghai Cancer Center (FDUSCC). Cohort 3 was a publicly available gene expression microarray dataset previously published elsewhere [15]. Cohort 4 was also a publicly available dataset including patients undergoing neoadjuvant chemotherapy [16]. The basic characteristics of the four cohorts are shown in Table 1. The study flowchart diagram is shown in Additional file 1: Figure S1. In addition, we analyzed 64 consecutive cases with the ER−/PgR+/HER2− phenotype from FDUSCC between 2005 and 2011 to validate IHC-based markers of subtype classification (characteristics of the 64 cases are available in Additional file 2: Table S1). The datasets (cohorts 1, 3, and 4) we used in this study are publically available and no permissions were required. The research protocols of cohorts 1, 3, and 4 were determined to be qualified for institutional review board exemption by the Ethical Committee of the Shanghai Cancer Center of Fudan University. The research protocols for cohort 2 and 64 consecutive ER−/PgR+/HER2− cases were reviewed and approved by the Ethical Committee of the Shanghai Cancer Center of Fudan University. All participants provided written informed consents.

For cohort 1, obtained from the SEER database consisting of 18 population-based cancer registries, we selected patients diagnosed with invasive breast cancer between January 1, 2010, and December 31, 2013 (SEER provides HER2 status after 2010). We identified 67,932 HER2-negative patients according to the following criteria: female, surgical treatment (either mastectomy or breast-conservation), AJCC stages I–III, pathologically confirmed invasive ductal carcinoma, unilateral, known ER/PgR/HER2 status, known time of diagnosis, and breast cancer as the first cancer at diagnosis. SEER database does not conduct central review for ER/PgR/HER2. Since we enrolled the cases after 2010, the positivity of ER and PR expression should be according to the ASCO/CAP guideline (≥1 % of tumor cells with nuclear staining) [12]. Data extraction was performed by SEER*Stat software v8.1.5 [17]. The outcome of interest was breast cancer-specific survival (BCSS), which was calculated from the date of diagnosis to the date of breast cancer death. Patients who died of other causes were censored at the date of death.

For cohort 2 from FDUSCC, we included 2,338 consecutive HER2– cases of primary operable invasive breast cancer between January 1, 2008, and December 31, 2011. This is a well-characterized series of patients, whose clinicopathologic and follow-up information were maintained on a prospective basis [18]. Patient treatments were based on St. Gallen consensus [11, 19]. The cut-off for ER/PgR positivity was ≥1 % of tumor cells with nuclear staining [12]. Pathologic HER2 status was defined according to ASCO/CAP guidelines [20]. Re-assessment of ER−/PgR+ cases was carried out routinely. The outcome for this cohort was relapse-free survival (RFS), which was calculated from the date of diagnosis to the date of the first event of local, regional, or distant metastasis of breast cancer.

For cohort 3, retrieved from 36 publicly available breast cancer microarray datasets [15], among the original 5,715 unique breast cancer with expression profiles, 837 cases were identified to be HER2-negative and had information on immunohistochemical ER, PgR, and HER2 status. The normalization of gene expression data was performed by Haibe-Kains et al. [15]. Hybridization probes were mapped to Entrez GeneID as described by Shi et al. [21]. When multiple probes mapped to the same GeneID, the one with the highest variance was used. All untreated patients had surgery, although information was not available for all datasets. The PAM50 classifier was applied to the data to determine the intrinsic subtype of each case as previously described [22]. The survival outcome of interest was RFS.

For cohort 4, we selected 483 HER2– patients who participated in a prospective Institutional Review Board-approved biomarker discovery study at MD Anderson Cancer Center as published previously elsewhere [16]. The cut-off for ER/PgR positivity was ≥1 % of tumor cells with nuclear staining. All patients received neoadjuvant chemotherapy containing a taxane/anthracycline-based regimen (followed by endocrine therapy if ER+). In our analysis, cases with indeterminate ER and PgR had been excluded, and the outcome for analysis was distant RFS (DRFS). Detailed methods for RNA purification and microarray hybridization have been reported previously [16, 23]. Gene expression profiling with Affymetrix U133 gene chips was performed. Gene expression levels were derived from multiple oligonucleotide probes on the microarray that hybridize to different sequence sites of a gene transcript (probe sets). Gene expression data are available under Gene Expression Omnibus accession number of GSE25066. The PAM50 classifier was applied to determine the subtype of each case [22].

In cohorts 3 and 4, each ER−/PgR+ case was assigned an intrinsic subtype by the PAM50 classifier [22]. The original gene expression profile data were only available in cohort 4. Using these original data [16], we compared the gene expressions of interest between different subtypes of ER−/PgR+. To determine the functional ER pathway, mRNA expression of estrogen-responsive genes, TFF1 (pS2), GREB1, and PDZK1, were measured [24]. Expression of basal-associated cytokeratins (CKs) and EGFR were measured [25, 26]. Moreover, because the claudin-low subgroup is associated with a specific subtype of triple-negative breast cancer, mesenchymal stem-like [27], we also measured the expression of the epithelial-mesenchymal transition-associated gene CDH1 and claudin genes to discriminate mesenchymal stem-like from basal-like [28]. Probe sets used for measurement of mRNA expression are listed in Additional file 3: Table S2. Expression data were normalized with the MAS5 algorithm, the mean was centered to 600 and log2 was transformed as previously described [13].

An endocrine therapy sensitivity score was calculated by the average log2 transformed expression values of ER, PgR, BCL2, and SCUBE2 with following measurement: (0.8*ER + 1.2*PGR + BCL2 + SCUBE2)/4 as previously described in OncotypeDX [29]. This ER group score could predict of response to tamoxifen and a higher score indicates a higher sensitivity to endocrine therapy [30, 31]. For 64 cases (consecutive cases with the ER−/PgR+/HER2− phenotype from FDUSCC) with formalin-fixed paraffin-embedded samples, the method of RNA extraction and real-time PCR is provided in Additional file 4: Supplemental Methods. PCR primers are listed in Additional file 5: Table S3.

IHC was performed in the 64 cases from FDUSCC according to the standard procedure [25]. Staining patterns were as follows: cytoplasmic and/or membranous staining for EGFR and CK5, and cytoplasmic staining for TFF1 (pS2). The cutoff value for positivity for TFF1 was 10 % [32]; CK5 and EGFR scored positive if any (weak or strong) staining was observed as previously described [25]. The antibodies used were reported in our previous study [33].

Comparisons of patient and tumor characteristics were performed using the χ² test or two-sample t-test. Survival curves were constructed using Kaplan–Meier method and tested by log-rank test. Multivariate adjusted hazard ratios (HRs) with 95 % confidence intervals (CIs) were calculated using the Cox proportional hazards model. The Mann–Whitney test was used to test gene expression differences. To analyze the combined results, we employed a two-step approach [34]. At first, the individual participant data from each study were analyzed separately (i.e. to obtain the results of each cohort). Then, the results were synthesized in the second step using a suitable model for meta-analysis of aggregate data. The meta-analysis was conducted in adherence to the standards of quality [35]. To pool the proportions, we used the command “metaprop_one” in Stata. According to a previous study [36], the score methods are recommended for proportion interval estimates and in our study the Wilson score confidence intervals were computed. We also assessed the heterogeneity among cohorts by using Cochran χ² Q statistics and I ² statistics. If P values <0.05 or I ² > 25 % were obtained, we determined that there was a significant heterogeneity [35]. Use of a fixed-effects method (Inverse-variance method) or a random-effects method (DerSimonian and Laird method) was performed according to heterogeneity. When we compared survival estimates of ER−/PgR+ versus ER+/PgR+ and ER−/PgR− versus ER+/PgR+, we used multivariate meta-analysis (command “mvmeta” in Stata). Multivariate meta-analysis has been described previously [37, 38]. The method we used was restricted maximum likelihood and the variance-covariance matrix was defined as “unstructured”. Statistical analyses were performed with Stata v.14.0 and SPSS v.17. Two-sided P <0.05 was considered statistically significant.