Polymorphism at 19q13.41 Predicts Breast Cancer Survival Specifically after Endocrine Therapy

Breast cancer is the most common cancer among women worldwide and is a leading cause of cancer-related deaths (1). Breast cancer can be divided into two major types by the estrogen receptor-α (ER) status; ER-positive breast cancer is driven by the female hormone estrogen, whereas ER-negative breast cancer does not depend on estrogen. Endocrine therapies target the ER-positive type, which accounts for about 70% of all breast cancer (2). Currently available endocrine therapies aim to either selectively block the ER by binding ER (tamoxifen), decrease ovarian estrogen production [ovarian ablation, luteinizing hormone-releasing hormone (LHRH) agonists] or, in postmenopausal women, blocking the conversion of androgen to estrogen in peripheral fat (aromatase inhibitors) or selectively downregulating ER (e.g., fulvestrant; ref. 3). Randomized, controlled trials have demonstrated that breast cancer recurrence and death may be reduced by approximately one third by endocrine adjuvant treatments in patients with ER-positive breast cancer (4, 5). However, approximately 30% of ERα-positive breast cancers do not respond to endocrine therapies (de novo resistance; ref. 6) and in addition, the majority of tumors that initially respond to treatment develop resistance over time (acquired resistance; ref. 7).

There are several potential mechanisms of resistance to endocrine therapy (reviewed in ref. 8) including, for example, as the most important mechanism, loss of expression of ERα (due to an emerging subclone of ER-negative cancer). Beside mechanisms related to ER, resistance to endocrine therapy may also occur due to increased growth factor signaling and dysfunctional metabolism of hormonal agents. As an example, patients carrying inactive alleles of CYP2D6 (∼7%–10% of Caucasian women) fail to convert tamoxifen to its primary active metabolite, endoxifen, and may consequently be less responsive to tamoxifen (9–12). However, the association with endocrine treatment outcome remains currently controversial. Presently, aside from ER status, no unequivocal biomarkers have been identified to determine whether a patient will benefit from adjuvant endocrine therapy.

Germline genetic variations, such as single-nucleotide polymorphisms (SNP), have been assessed as potential predictors of survival for patients with breast cancer in general (13–15) and in different subgroups including those defined by endocrine therapies (16, 17). A majority of these studies have applied a candidate gene approach focusing on SNPs within prespecified genes of interest. These studies have provided some indicative results, but further studies will be needed to validate the findings. Recently, genome-wide association studies (GWASs) have been performed with the aim of identifying genetic variants influencing the outcome of breast cancer (18–20), including our previous studies that identified ARRDC3 locus influencing prognosis in especially early-onset breast cancer (21, 22). Furthermore, a GWAS conducted in subgroup of patients receiving adjuvant tamoxifen therapy in a Japanese population detected significant associations with recurrence-free survival at 10q22 (23).

The availability of GWA data from the patient population in our Helsinki Breast Cancer Study (HEBCS) study together with GWA data from Prospective study of Outcomes in Sporadic versus Hereditary breast cancer (POSH) study enables an agnostic genome-wide approach to identify common genetic variants associated with survival for breast cancer. In this study, we focus on ER-positive breast cancer treated with adjuvant endocrine therapy in order to identify putative genetic markers for endocrine treatment outcome. We implemented a two-stage study design with 1,341 patients with breast cancer from the two abovementioned GWAS in stage I and then a further 5,011 patients from the two in stage II validation datasets.

Samples included in stage I discovery dataset came from participants of the HEBCS and POSH studies. For HEBCS GWS, 805 cases were included. Of these, 563 cases originated from a prospective patient series of unselected, incident patients with breast cancer while 242 cases were obtained from additional familial patient series (24–26). All cases were ascertained at the Helsinki University Hospital (Helsinki, Finland); see Supplementary Methods for the details of the collection. Of the 805 samples, 240 samples were recorded to have received endocrine therapy (Table 1).

The POSH GWS consisted of 574 participants from the POSH study (27). Prospective early onset breast cancer cases were included in the POSH study, with participants diagnosed with invasive breast cancer of ages 40 years or younger. Details of the patient selection are provided in the Supplementary Methods. POSH GWS included 155 patients who had received endocrine therapy (Table 1). All participants of both studies provided written informed consent before participating in the study.

A further 1,415 patients with breast cancer from the POSH study (27) unselected for any differential survival were included in the stage II validation dataset. POSH validation included 1,027 patients who had received endocrine therapy (Table 1).

As an additional independent validation dataset in stage II, we used a series of 3,596 patients from the prospectively randomized SUCCESS-A trial. Details of the collection are provided in the Supplementary Methods. A total of 2,458 cases of the 3,596 cases had received endocrine treatment.

The age and tumor characteristics of study participants from HEBCS GWS, POSH GWS, POSH validation, and SUCCESS-A are presented in Table 1. The flow of samples through the various stages of the study has been summarized in Supplementary Fig. S1.

Genotyping of the Helsinki samples was conducted using the Illumina 550 platform and POSH GWS using the Illumina 660-Quad SNP array as previously described (21, 28). To ensure the harmonization of genotype calling between HEBCS and POSH GWS, the HEBCS GWS intensity files were processed with Illumina's Genome Studio software to call genotypes consistently with the POSH genotypes using a GenCall threshold of 0.15. Rare SNPs were excluded from analysis based on a minor allele frequency (MAF) cutoff of 0.01, a genotyping call rate <95% and Hardy–Weinberg equilibrium (HWE) P value <0.0001. The detailed description of harmonized quality control is in (21).

For replication genotyping, we selected the top four associations from the genome-wide meta-analysis of HEBCS and POSH GWS in the ER-positive endocrine treatment subgroup that fulfilled the following criteria: most significant independent associations with meta-P < 1.0 × 10⁻⁴ within the ER-positive endocrine treatment subgroup, showing significant (P_{heterogeneity} < 0.01) heterogeneity by endocrine treatment evaluating ER-positive endocrine-treated subgroup and ER-positive subgroup not treated with endocrine therapy, and having significant interaction with endocrine treatment (likelihood ratio test P value per allele < 1.0 × 10⁻³) in a pooled dataset of HEBCS and POSH GWS ER-positive cases. These four SNPs were genotyped in the 1,415 additional young onset cases from the POSH stage II validation study. SNPs were genotyped by KBiosciences using the KASPar chemistry, which is a competitive allele-specific PCR SNP genotyping system. The SUCCESS-A GWAS was genotyped on the Illumina HumanOmniExpress-12v1 G FFPE array.

The imputation of genome-wide SNP information in HEBCS and POSH GWS was performed on the basis of 1000 Genomes Project phase I and release version 3 European reference haplotypes. Quality control measures applied to imputed data included excluding SNPs with HWE P value < 1 × 10⁻⁶, MAF < 5%, and imputed genotype call rate ≤90% and individuals call rate ≤90%. The detailed description of imputation is in ref. (22).

See Supplementary Methods for a detailed description of the statistical analyses. In stage I, the Cox proportional hazards models were used to derive hazard ratios (HR) for breast cancer–specific mortality in association with each SNP. Follow-up time was calculated from the date of diagnosis to the date of last follow-up or breast cancer–related death and right-censored at 10 years.

In stage II, the follow-up time was calculated from the date of diagnosis to the date of last follow-up or breast cancer–related death for POSH validation dataset. For SUCCESS-A, the follow-up time was calculated from the date of diagnosis to the date of last follow-up or death from any cause, due to lack of cause-of-death information. The meta-analysis in stage I as well as the meta-analysis of stage I and II was performed with R package MetABEL (29). The Cox proportional hazard models were performed with R package GenABEL (29).

For HEBCS GWS, POSH GWS, and POSH validation, we had cause-of-death information that enabled us to evaluate the breast cancer–specific survival. For SUCCESS-A, the only outcome information was overall survival (endpoint: all-cause mortality) and progression-free survival (endpoints: local or metastatic recurrence or death). In order to assess differences in survival when using different endpoints, we further conducted a sensitivity analysis. In the sensitivity analysis, we analyzed the survival across all the four studies using a common endpoint; either 10-year overall survival or 5-year progression-free survival.

In order to test for interaction between endocrine treatment and a given SNP of interest, SNP genotype data were fitted into two multivariate Cox proportional hazards models including also clinically relevant covariates: one with both endocrine treatment and the SNP represented as individual covariates, and one that included an interaction term between the two. A likelihood ratio test between models was then conducted to examine whether the interaction model is a better fit for the prognostic data. The interaction tests, specifying breast cancer–related death as the endpoint, were conducted in a pooled dataset of ER-positive cases only and were stratified by study.

In order to analyze the correlation between the loci of interest and gene expression, we used the breast cancer sample data generated by the METABRIC project (30, 31). The expression data were obtained from the European Genome-Phenome Archive, which is hosted by the European Bioinformatics Institute, under accession number EGAS00000000083. See Supplementary Methods for the details of the data preparation for eQTL analysis. The analysis was conducted with R-package Matrix eQTL (32) using linear regression and ANOVA models. In addition, we used online results of the peripheral blood eQTL meta-analysis (33) and lymphoblastoid exon expression QTL in Geuvadis project (34).

In order to investigate whether the loci of interest harbor known or predicted regulatory elements, we explored the ENCODE data using HaploReg2 (35) and RegulomeDB (36). To assess gene expression–based survival, we used BreastMark that integrates gene expression and survival data from 26 datasets on 12 different microarray platforms corresponding to ∼17,000 genes in up to 4,738 samples (37). The genes that were identified by eQTL analysis were analyzed at the protein level by exploring the protein–protein interaction network with STRING program (38).

We performed a fixed-effects meta-analysis to combine HR estimates from HEBCS and POSH stage I GWS studies including 805 and 536 study subjects. In the two datasets altogether, 486,478 SNPs were common and passed the QC process. The meta-analysis was performed including all cases and in the subgroup of endocrine-treated patients; combining anti-estrogen, aromatase inhibitor, and LHRH agonist treatments totaling 240 endocrine-treated patients in HEBCS GWS and 155 in POSH GWS. After LD-pruning, the top four associations (rs8113308, rs4082843, rs4767413, and rs11085098 in chromosomes 19, 4, 12, and 19, respectively; Supplementary Table S1) were selected from the genome-wide meta-analysis of HEBCS and POSH GWS in the ER-positive endocrine treatment subgroup for genotyping in the stage II POSH validation samples. The selected SNPs fulfilled the following criteria: the most significant independent associations with meta-P < 1.0 × 10⁻⁴ within the ER-positive endocrine treatment subgroup, showing significant (P_{heterogeneity} < 0.01) survival heterogeneity by endocrine treatment and having significant interaction with endocrine treatment (likelihood ratio test P value per allele < 1.0 × 10⁻³) in a pooled dataset of HEBCS and POSH GWS ER-positive cases. All the SNPs were identified under an additive inheritance model.

Of the four SNPs that were formally tested for replication, all were successfully genotyped and two SNPs demonstrated nominal replication signals in the same direction as in the stage I (HEBCS and POSH GWS) patients. As an additional independent validation dataset in stage II, we used SUCCESS-A. Because SUCCESS-A was genotyped in a different version of Illumina genotyping chip, there were no exact SNP matches for two of the SNPs. For SNP rs8113308, we used the genotype information of a tag SNP rs8108525 (r² = 0.81). For SNP rs4082843, no tag SNP could be found with r² > 0.80. For remaining SNPs, an exact SNP match was present in SUCCESS-A genotyping data.

In the meta-analysis of stage I and II, the strongest replication and meta-analysis signal was observed at rs8113308 under the additive inheritance model. The minor allele was found to consistently associate with poor survival specifically after adjuvant endocrine therapy among ER-positive patients [HR, 1.69; 95% confidence interval (CI), 1.37–2.07; P = 6.34 × 10⁻⁷]. The most common endocrine treatment regimen in all the four datasets was tamoxifen (Table 1), and a similar effect was found within the tamoxifen-treated subgroup (HR, 1.65; 95% CI, 1.35–2.03; P = 1.44 × 10⁻⁶). However, the minor allele associated with improved breast cancer outcome in ER-negative patients (HR, 0.71; 95% CI, 0.56–0.91; P = 6 × 10⁻³), with a similar trend in ER-positive patients not receiving endocrine therapy (HR, 0.66; 95% CI, 0.40–1.07; P = 9.32 × 10⁻²; Table 2), suggested a treatment-specific effect. The Kaplan–Meier plots of cumulative 10-year survival of rs8113308 genotypes among ER-positive endocrine-treated patients in pooled stage I (HEBCS and POSH GWS), ER-positive nontreated (available only for HEBCS), and ER-negative patients in pooled stage I (HEBCS and POSH GWS) are presented in Fig. 1. The Kaplan–Meier plots separately for all the four studies are presented in Supplementary Figs. S2 and S3.

We further investigated the survival association of rs8113308 in all patients and in phenotype- and treatment-based subgroups separately in each of the four studies (Fig. 2). The association of the SNP in ER-positive patients receiving endocrine therapy was found consistent throughout the four studies (Fig. 2 and Table 2)

On the basis of the sensitivity analysis where we analyzed the survival across all the four studies using also 10-year overall survival or 5-year progression-free survival, very similar association was seen as in the main meta-analysis, regardless of the used endpoint (Supplementary Fig. S4).

In addition to the SNP rs8113308, one further SNP, rs4767413, showed a consistent association across all four studies among ER-positive endocrine-treated patients; however, the result was not significant in stage II studies (POSH validation and SUCCESS-A). SNP rs4767413, located in an intergenic region in chromosome 12, was found to associate with poor survival among ER-positive patients receiving endocrine treatment with HR = 1.39 (95% CI, 1.15–1.67; P = 5.86 × 10⁻⁴), but not among ER-positive patients not treated with endocrine therapy (HR, 0.91; 95% CI, 0.56–1.48). No consistent association was seen among the four studies in ER-negative subgroup. The remaining two SNPs from stage I included in the meta-analysis of stage I and II did not show concordant association in stage II (Table 2).

Given that endocrine treatment is predominantly administered to ER-positive cases, it is possible that an apparent interaction between rs8113308 and endocrine treatment actually indicates an interaction between the SNP and ER status instead of a predictive, treatment-specific effect. In order to address this possibility, the interaction test, specifying breast cancer–related death as the endpoint, was conducted in a pooled dataset of ER-positive cases only, including HEBCS GWS, POSH GWS, and POSH validation datasets—due to lack of cause-of-death information, SUCCESS-A study was not included. In the multivariate Cox's proportional hazards model including the SNP rs8113308 and endocrine treatment separately along with the following covariates: progesterone receptor status, tumor size, lymph node metastasis, distant metastasis at diagnosis, age at diagnosis, and tumor histologic grade, we found both the SNP and the endocrine treatment to be independently prognostic; per allele (HR, 1.27; 95% CI, 1.02–1.59; P = 0.036) and endocrine treatment (HR, 0.59; 95% CI, 0.44–0.80; P = 6.76 × 10⁻⁴). In contrast, when an interaction term (SNP * endocrine treatment) was added to the model, the SNP lost its independent prognostic value and the interaction between SNP rs8113308 and endocrine treatment associated significantly with poor breast cancer survival (HR for per-allele rs8113308:endocrine treatment, 2.16; 95% CI, 1.30–3.60; P = 3.13 × 10⁻³; Table 3). The interaction model is a better fit for the prognostic data than the model without an interaction term (likelihood ratio test P value = 0.0021). When using a codominant model, the interaction remained statistically significant despite the loss of power introduced by the additional genotype covariate (likelihood ratio test P value = 0.0078; Table 3), while the effect size depended on allele dose: the HR for the interaction between endocrine treatment and the heterozygous AG genotype is HR = 1.95; 95% CI, 1.08–3.49, and HR = 7.77; 95% CI, 0.93–64.71 for the interaction between endocrine treatment and rare homozygous GG genotype.

We assessed the association of SNP rs8113308 and rs4767413 with clinical predictors of breast cancer prognosis in a pooled set of HEBCS and POSH GWS and POSH validation (Supplementary Tables S2 and S3). There were no significant associations between SNP rs8113308 or rs4767413 and clinical features.

We next examined the rs8113308 LD region (r² ≥ 0.2) for stronger associations conducting a meta-analysis of HEBCS and POSH imputed data of 869 SNPs. We identified an association with HR = 3.40 (95% CI, 2.04–5.66; P = 2.64 × 10⁻⁶) for one imputed SNP, rs10410393 (r² = 0.2). However, it did not show concordant direction of association in the SUCCESS-A data. In addition, the MAF for this SNP in European population is 0.036, being very rare. Subsequently, the rs8113308 remained the strongest associated variant in the region (Supplementary Fig. S5). Similarly, we investigated the LD region (r² ≥ 0.2) of SNP rs4767413 and found stronger associations for four imputed SNPs in the HEBCS and POSH datasets, with rs11611797 (r² = 0.87) being strongest (HR, 2.05; 95% CI, 1.76–2.57; P = 1.76 × 10⁻⁶). However, none of the four imputed SNPs were significant in SUCCESS-A data.

In order to test for correlation of the SNP rs8113308 genotypes and RNA expression, we performed an eQTL analysis in 821 ER-positive and 321 ER-negative breast tumors in the METABRIC project (30, 31). We conducted the analysis both in cis and in trans. For cis-eQTL analysis, we included all the genes within 100 kb of the SNP rs8113308; ZNF613, ZNF350, ZNF615, and ZNF649. The re-annotation of Illumina probes by Barbosa-Morais and colleagues (39) classified the probes for ZNF615 and ZNF649 as unreliable and probes for ZNF613 and ZNF350 as reliable; the eQTL was only considered for reliable probes. The appliance of a linear regression model did not reveal any significant cis- or trans-eQTLs. The cis-eQTL analysis using the ANOVA model, allowing genotype to have both additive and dominant effects, indicated an association between rs11881650 (a tag SNP for rs8113308; r² = 0.81) and expression of ZNF350 (P = 3.85 × 10⁻³) in ER-positive breast tumors, whereas no association was seen in ER-negative breast tumors (Supplementary Table S4). The cis-eQTL for ZNF350 remains significant also after Bonferroni adjustment (adjusted P = 9.22 × 10⁻³). The high ZNF350 mRNA expression level was linked to SNP rs11881650 rare homozygote genotype compared with the common homozygote and heterozygote genotypes (P = 0.018 and P = 0.015, respectively; Fig. 3A) in ER-positive breast tumors; no difference was seen among ER-negative tumors. Concordant supportive evidence was seen in peripheral blood tissue where rs7246064 (a tag SNP for rs8113308; r² = 1) and ZNF350 mRNA expression were correlated (P = 3.27 × 10⁻³; Supplementary Table S5), no significant correlation between SNP rs8113308 and ZNF350 mRNA expression was seen in lymphoblastoid cell lines. The trans-eQTL analysis of the METABRIC data using ANOVA model with Bonferroni-adjusted P of < 0.05 and applied to the reliable probes revealed association with expression of the EPS8L1 and ZNF347 genes in 19q13 locus and CYP26A1 in 10q23 in ER-positive breast tumors, with no significant eQTL for these genes among ER-negative tumors (Supplementary Table S6). The eQTL analysis for SNP rs4767413 did not reveal any statistically significant cis- or trans-eQTL correlations.

We searched BreastMark database to analyze and visualize survival differences based on mRNA expression differences in public mRNA expression data. In the BreastMark, the only endocrine treatment group available is tamoxifen, and no other information for endocrine treatment is given. ZNF350 showed gene expression-based survival difference in ER-positive tamoxifen-treated patients with high ZNF350 expression associating with poor survival with HR 1.61 (1.14–2.27), log-rank P = 0.006, whereas no survival difference by different ZNF350 expression levels was seen in ER-positive tamoxifen nontreated patients or in ER-negative tamoxifen nontreated patients (Fig. 3B–D). No survival difference by gene expression level was seen for ZNF615, nor for EPS8L1, ZNF347, or CYP26A1. For genes within 100 kb of the SNP rs4767413 (MAP1LC3B2 and MIR4472-2), survival difference by gene expression could not be analyzed because of lack of the probe data for these genes in BreastMark.

In order to assess the functional role of the rs8113308 locus, we explored ENCODE data with designated tools HaploReg2 and RegulomeDB for regulatory elements and protein binding sites residing in the region. The Encode data indicate that the rs11879758, a tag SNP for rs8113308 (r² = 0.85) locates in the site where there are promoter histone marks in nine cell lines and, based on the regulatory chromatin states, one active promoter in mammary epithelial cells with the closest annotated gene being ZNF350. Protein–protein interaction networks were searched by STRING program (Supplementary Fig. S6) demanding a high confidence score for interactions (0.700) derived from experimental studies, databases and text mining and it showed RNF11, ATXN2, BRCA1, and GADD45A proteins interacting with ZNF350.

ER-positive breast cancer is commonly treated with adjuvant endocrine therapies. Adjuvant endocrine treatment has been shown to increase overall survival and in light of recent studies is likely to be recommended for even longer duration in the adjuvant setting (40, 41). However, many patients do not benefit from these therapies and predictors for response or resistance to endocrine treatment are urgently needed. In this study, we report a meta-analysis of two genome-wide studies and two validation datasets for identifying genetic variants associated with breast cancer–related mortality specifically after adjuvant endocrine treatment. In a meta-analysis involving individuals treated with adjuvant endocrine therapy, we identified SNP rs8113308 specifically and significantly predicting outcome after endocrine treatment. We were further able to show that among patients with ER-positive tumors, there is a significant interaction between the rs8113308 and endocrine treatment, indicating a predictive, treatment-specific effect on survival, independent of conventional prognostic markers. In addition, SNP rs4767413 showed a consistent association across all four studies among ER-positive endocrine-treated patients and a significant interaction result with endocrine treatment in stage I. However, the survival association was not significant in stage II studies (POSH validation and SUCCESS-A), and no further supportive evidence could be obtained from the eQTL and gene expression survival analyses. These results thus remain inconclusive and warrant further studies.

Our primary interest in this study was to evaluate the breast cancer–specific survival. In stage I studies (HEBCS GWS and POSH GWS) and stage II POSH validation, the analyses were performed using breast cancer–specific mortality as the endpoint. A randomized clinical trial, SUCCESS-A, with data available via dbGAP was added to further validate our stage I findings. Because the only outcome data available for SUCCESS-A was overall or progression-free survival, we used overall survival as the endpoint for SUCCESS-A in the stage II main meta-analysis but further performed also a sensitivity analysis assessing alternative endpoints throughout all four studies that showed very similar association, regardless of endpoint.

Various endocrine therapies work by different mechanisms to antagonize the growth-promoting activity of estrogen and different endocrine therapies are administered to premenopausal and postmenopausal women. In our study, we combined anti-estrogen, aromatase inhibitor, and LHRH agonist treatments in order to gain more statistical power. However, the most common endocrine treatment regimen in all the three datasets was tamoxifen and a similar result was found specifically within the tamoxifen-treated subgroup.

The direction of the association was consistent across the studies and remained statistically significant in the study-stratified pooled analyses even though patients in HEBCS GWS had relatively later onset breast cancer and POSH GWS contains only patients with early-onset breast cancer. Furthermore, the age-adjusted multivariate analysis suggests that the treatment interaction was not affected by the patients' age. In stage I, we could specifically assess also the ER-positive patients not treated with endocrine therapy that allowed us to elucidate whether the association is linked to ER positivity or endocrine treatment. The significant interaction between endocrine treatment and rs8113308 among ER-positive patients provided strong evidence that the identified association links specifically and significantly to endocrine treatment subgroup. On the basis of HapMap, 27% of the population carries at least one allele of this SNP and 3% carry the homozygous genotype. In HEBCS GWS, POSH GWS, and POSH validation, 26%, 22%, and 25% of the cases carried at least one allele (HR, 2.16) and 2.11%, 2.06%, and 2.35% were homozygous carriers, respectively (HR, 7.17). Previously, Kiyotani and colleagues performed a GWAS to assess the genetic factors influencing survival among patients receiving adjuvant tamoxifen therapy in Japanese population. They reported significant association with recurrence-free survival for SNP rs10509373 at 10q22 using 240 patients in GWAS and 105 and 117 patients in the replication phase (23). We did not find any significantly associated SNPs on chromosome 10, but because of the differences in allele frequencies comparison between the two studies is difficult.

The SNP rs8113308 is located on the long arm of the chromosome 19, in intron five of a gene ZNF613 encoding for Zinc Finger Protein 613. The 100-kb flanking region harbors a multitude of zing finger genes. The cis-eQTL analysis indicated an association between rs11881650 (a tag SNP for rs8113308; r² = 0.81) and mRNA levels of ZNF350 (also known as ZBRK1) as well as trans-eQTL associations to EPS8L1, ZNF347, and CYP26A1. However, a tag SNP (r² = 0.85) for rs8113308 locates in an active promoter site that is specifically active in mammary epithelial cells with the closest annotated gene being ZNF350; only ZNF350 showed gene expression-based survival difference among ER-positive tamoxifen-treated patients, with the high ZNF350 expression associating with poor survival. When investigating the correlation of the tag SNP rs11881650 genotypes and RNA expression in ER-positive tumorous breast tissue, we found that the rare allele correlated with increased ZNF350 mRNA expression levels. Together, these data indicate that the rs8113308 rare allele associates with increased expression of ZNF350 and poor survival of patients with breast cancer after endocrine treatment.

In a network analysis with a high confidence score, we saw an interplay with ZNF350, RNF11, GADD45A, and BRCA1, with the two latter interacting with ER (Supplementary Fig. S6). ZNF350 expression is altered in different human carcinomas including breast cancer (42). Zinc Finger Protein 350 can bind GADD45A in a BRCA1-dependent manner (43) as well as via binding sites located in the GADD45A promoter region, suggesting that ZNF350 represses GADD45A expression via multiple binding sites (44). The overexpression of ZNF350 has been shown to cause a decrease in both GADD45A transcripts and protein (44). GADD45A takes part in several cellular processes, including cell-cycle arrest and apoptosis (45). Functional assays on GADD45 family proteins, including GADD45A, have shown that they bind to nuclear hormone receptors including ERα and act as nuclear coactivators (46). Recently, it has also been found that high level of GADD45A protein expression correlates with ER-positivity and low level with ER-negativity in breast cancer (47). Because aberrant expression of cell-cycle regulators has been suggested to contribute to tamoxifen resistance (48), it could be hypothesized that altered expression of GADD45A might have an effect on endocrine treatment response. Moreover, the role of GADD45A as coactivator of ERα could affect the antagonist effect of endocrine treatments that block the ER binding, for example, tamoxifen. The biologic rationale behind the identified association of rs8113308 and breast cancer outcome after endocrine treatment might thus be the altered ZNF350 expression levels and subsequent alterations in the expression and activity of GADD45A and its interacting partners, including ERα, but further studies are required to elucidate the actual mechanism.

To our knowledge, this is the first meta-analysis of two genome-wide studies and two validation sets to assess the genetic factors influencing survival for patients with breast cancer receiving adjuvant endocrine treatment among women of European descent. It should be noted that our findings do not reach genome-wide significance as such, despite our use of GWAS as a starting point. However, the results are supported by the significant statistical interaction between the rs8113308 and endocrine treatment among patients with ER-positive tumors, indicating a predictive, treatment-specific effect on survival, independent of conventional prognostic markers, as well as by consistent in silico functional findings and a biologic rationale. Pending further validation in other large datasets, our findings may potentially influence personalized treatment by identifying patients who would not benefit from endocrine treatment. Further fine mapping studies will help to identify the causative and most significant variants responsible for the observed associations, while functional studies will be necessary to fully elucidate the underlying biologic mechanism.

No potential conflicts of interest were disclosed.

Conception and design: K. Aittomäki, C. Blomqvist, D. Eccles, H. Nevanlinna

Development of methodology: S. Khan, S. Rafiq, D. Eccles

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): W. Tapper, K. Aittomäki, J. Liu, C. Blomqvist, D. Eccles, H. Nevanlinna

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Khan, R. Fagerholm, S. Rafiq, C. Blomqvist, D. Eccles

Writing, review, and/or revision of the manuscript: S. Khan, R. Fagerholm, S. Rafiq, K. Aittomäki, C. Blomqvist, D. Eccles, H. Nevanlinna

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Khan, S. Rafiq, D. Eccles

Study supervision: D. Eccles, H. Nevanlinna

The authors thank Drs. Kirsimari Aaltonen, Dario Greco, Xiaofeng Dai, Päivi Heikkilä, and Karl von Smitten, as well as Tuomas Heikkinen, for their help with the HEBCS patient samples and data, and research nurse Irja Erkkilä for the assistance in the HEBCS data collection and management. For the POSH study, the authors thank Nikki Graham, Sylvia Diaper (University of Southampton DNA bank), and Kathy Potter (Southampton CRUK Centre Tissue Bank). Assistance with phenotype harmonization and genotype cleaning, as well as with general study coordination, was provided by the GARNET Coordinating Center (U01 HG005157). For the GWAS in breast cancer patients from the prospectively randomized SUCCESS-A trial, assistance with data cleaning was provided by the National Center for Biotechnology Information.

This work was supported by the Helsinki University Hospital Research Fund, the Finnish Cancer Society, the Nordic Cancer Union, the Academy of Finland [266528] and the Sigrid Juselius Foundation as well as the Emil Aaltonen Foundation and the Maud Kuistila Foundation (to S. Khan). Funding for the POSH GWAS was provided by Breast Cancer Campaign grants 2010NovPR62 and 2013MayPR044 and the POSH study is supported by Cancer Research Grants C1275/A15956 and C1275/A19187. Genotyping at the National Institute of Singapore was financially supported by the Agency for Science, Technology and Research (A*STAR), Singapore. Funding support for the GWAS in breast cancer patients from the prospectively randomized SUCCESS-A trial was provided through the NHGRI Genomics and Randomized Trials Network [GARNET] (U01 HG005152). Funding support for genotyping, which was performed at the Center for Inherited Disease Research (CIDR) at Johns Hopkins University, was provided by the NIH Genes, Environment and Health Initiative [GEI] (U01 HG005137).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.