Genomic Aberrations and Late Recurrence in Postmenopausal Women with Hormone Receptor–positive Early Breast Cancer: Results from the SOLE Trial

Women with hormone receptor–positive early breast cancer have a persistent risk of relapse even after 5 years of adjuvant therapies (i.e., late recurrence; ref. 1). Extended endocrine treatment to 10 years has been proposed to reduce the risk of late recurrence in postmenopausal women with breast cancer (2). The overall benefit of this approach is still controversial and the decision about therapy extension is based on personalized risk and tolerability evaluation (3).

The Study of Letrozole Extension (SOLE) is a randomized, phase III trial conducted by the International Breast Cancer Study Group (IBCSG) evaluating the effect of extended intermittent letrozole as compared with continuous letrozole on late-breast cancer events in postmenopausal women with node-positive, hormone receptor–positive, early breast cancer and no recurrence after 4–6 years of adjuvant endocrine therapy (4). SOLE showed the intermittent administration of extended letrozole is safe and feasible and may represent a valid alternative approach in this group of women. Given the debated benefit of extended endocrine therapy and the availability of alternative options, there is an urgent need to identify clinically useful biomarkers of late recurrence.

During the last decades, high-throughput sequencing technologies have unraveled the heterogeneity of the molecular drivers of breast cancers and identified a few recurrent genetic aberrations and novel potential therapeutic targets (5, 6). Gene expression signatures have been demonstrated as clinically useful for risk stratification in hormone receptor–positive early breast cancers (7–9). However, little is known about the molecular underpinnings of late recurrence.

We investigated the genomic alterations in hormone receptor–positive early breast cancer from postmenopausal women enrolled in the SOLE trial. We aimed (i) to identify prognostic molecular biomarkers associated with the late-breast cancer recurrence and (ii) to assess potentially actionable aberrations that may represent therapeutic targets for breast cancer treatment.

From November 2007 to October 2012, 4,884 patients were enrolled in the SOLE trial with 4,851 patients in the intention-to-treat (ITT) population. All patients had lymph node–positive, hormone receptor–positive operable breast cancer and they were free of breast cancer after 4–6 years of previous adjuvant endocrine therapy. Patients were randomized to receive letrozole either continuously or intermittently during 5 years. Central pathology review of primary tumors was performed and formalin-fixed, paraffin-embedded (FFPE) material was available for 3,162 patients from the ITT population (Fig. 1).

Because of the overall low recurrence rates in hormone receptor–positive early breast cancer, a case-cohort–like sampling design was used to define the study cohort (10). The 598 selected patients included all 214 patients who experienced a breast cancer event and a stratified random sample of 384 patients who did not at the time of sampling; there were five stratification factors for sampling (treatment assignment, number of positive lymph nodes, tumor subtype according to 14th St. Gallen Consensus (11), prior adjuvant endocrine therapy, and small vs. large centers; Fig. 1).

The study was conducted in accordance with the 1964 Helsinki Declaration and later amendments. Local ethics committees approved the trial protocol. All patients provided written informed consent for the use of their tumor tissue for unspecified research purposes. The project was approved by the IBCSG Biological Protocol Working Group.

Hematoxylin and eosin slides of 598 patients’ tumor samples were retrospectively reviewed blindly to clinical information to evaluate the sample adequacy and the percentage of tumor cell content. Four 10-μm-thick sections from representative FFPE histologic blocks were subjected to manual macrodissection with a sterile scalpel before nucleic acid isolation to enrich tumor cell content (range, 20%–90%). DNA and RNA were extracted using either the Ambion RecoverAll Multi-sample DNA Workflow (Thermo Fisher Scientific) or the AllPrep DNA/RNA Mini Kit (Qiagen), according to the manufacturer's guidelines. After exclusion of quantitatively and qualitatively inadequate cases, 499 samples were available for targeted next-generation sequencing (NGS).

NGS analysis was retrospectively performed using the Ion Torrent Oncomine Focus Assay (Thermo Fisher Scientific). This assay enables to target 52 cancer-related or actionable genes, including 35 genes with hotspot single-nucleotide variants (SNV) or small insertions or deletions (InDel; SNV analysis from DNA; AKT1, ALK, AR, BRAF, CDK4, CTNNB1, DDR2, EGFR, ERBB2, ERBB3, ERBB4, ESR1, FGFR2, FGFR3, GNA11, GNAQ, HRAS, IDH1, IDH2, JAK1, JAK2, JAK3, KIT, KRAS, MAP2K1, MAP2K2, MET, MTOR, NRAS, PDGFRA, PIK3CA, RAF1, RET, ROS1, and SMO), 19 genes with focal copy-number gains (CNG analysis from DNA; ALK, AR, BRAF, CCND1, CDK4, CDK6, EGFR, ERBB2, FGFR1, FGFR2, FGFR3, FGFR4, KIT, KRAS, MET, MYC, MYCN, PDGFRA, and PIK3CA), and 23 genes with fusion drivers (RNA analysis; ABL1, AKT3, ALK, AXL, BRAF, EGFR, ERBB2, ERG, ETV1, ETV4, ETV5, FGFR1, FGFR2, FGFR3, MET, NTRK1, NTRK2, NTRK3, PDGFRA, PPARG, RAF1, RET, and ROS1). Genomic DNA extracted from the 499 adequate tumor samples was subjected to DNA analysis of SNVs and CNGs. A total of 140 samples were submitted to both DNA and RNA analyses.

Targeted NGS was performed following the manufacturer's instructions. Briefly, 10 ng of extracted genomic DNA was used for library preparation and templates were prepared using either Ion OneTouch 2 System or the Ion Chef Machine (Thermo Fisher Scientific). Sequencing of templates was performed on Personal Genome Machine Sequencer (Thermo Fisher Scientific) and data were analyzed using the Ion Reporter Analysis software. Samples with a mean coverage <100× were excluded from the analysis. To minimize false-positive results obtained with targeted NGS performed on DNA extracted from old FFPE samples, we retained SNVs and InDels with a variant allele frequency (VAF) ≥10% and an allele coverage of at least 250 reads or with at least 25 variant-containing reads [VAF × reads ≥25 (12, 13)]. We further excluded CNG calls in samples with a median of the absolute values of all pairwise differences above 0.5. CNGs were considered as dichotomous variables (presence vs. absence) according to CNG algorithms in Ion Reporter software. After quality metrics filtering, DNA NGS data about SNVs and CNGs were available for 403 and 350 samples, respectively. As expected, RNA degradation from old FFPE tissue samples was higher than DNA. For this reason, RNA analysis was performed only in a subgroup of samples (n = 140) and data were available for 67 samples successfully sequenced.

Variants were annotated as either pathogenic/likely pathogenic or neutral/unknown according to available online cancer gene mutation databases, including Catalogue of Somatic Mutations in Cancer (14), cBioPortal for Cancer Genomics (15, 16), and ClinVar–NCBI–NIH (17). For unknown variants, annotation was further performed using a combination of “in silico” predictors, including FATHMM, FATHMM-MKL, PROVEAN, SIFT, POLYPHEN, MUTATION ASSESSOR, and MUTATION TASTER, available at VarSome website (18). Unknown variants were considered likely pathogenic when they were defined as damaging by four of seven predictors.

FGFR1 CNG was further investigated using fluorescence in situ hybridization (FISH) with ZytoLight FGFR1/CEN8 Dual Color Probe (ZytoVision). Samples were prepared according to the manufacturer's instruction. FISH signals were evaluated as previously reported (19).

The endpoints included breast cancer–free interval (BCFI), defined as time from randomization to the first indication of invasive breast cancer recurrence (local, regional, or distant) or invasive contralateral breast cancer; and distant recurrence-free interval (DRFI), defined as time from randomization to breast cancer recurrence at a distant site. In the absence of an event, endpoints were censored at date of last follow-up or date of death without an endpoint event. Median follow-up was calculated using reverse Kaplan–Meier estimate of the survival function.

The analysis population consisted of all the patients with a breast cancer event and a sampled subset of patients without an event, who had their DNA samples successfully analyzed. Using weighted analysis methods (Horvitz–Thompson methods), contributions to survival estimators and tests were weighted proportional to the inverses of the sampling fractions and special methods were used to compute variances (10). Weighted Kaplan–Meier estimates of BCFI and DRFI distributions were calculated. Weighted Cox modeling was used to estimate HRs with 95% confidence intervals (CIs) and log-rank test statistics with one degree of freedom were used for comparing SNV versus wild-type or CNG versus no CNG; the modeling had stratification factors of prior adjuvant endocrine therapy and treatment assignment. For the most frequent genomic aberration, a stratified multivariable weighted Cox model was adjusted for clinicopathologic factors, including age at randomization, number of positive lymph nodes, tumor size (≤2 vs. >2 cm), grade, and Ki-67. Correlations of recurrent SNVs/CNGs with clinicopathologic factors were analyzed using a χ² test for categorical variables and a t test for continuous variables.

The statistical power was assessed for the BCFI endpoint on the basis of the patients with available SNV (n = 403) and CNG (n = 350) data and an adjusted SE accounting for the sampling design. For 403 patients with available SNV data, there was 90% power to detect HR = 2.0 (SNV vs. no SNV; assuming 40% mutation prevalence), with 0.05 two-sided significance level. For 350 patients with available CNG data, there was 85% power to detect HR = 3.2 (assuming 8% prevalence).

All analyses were prespecified and followed REMARK recommendations (20). Two-sided P values were reported for all analyses. No multiple comparison adjustments were implemented. A P value less than 0.05 was considered statistically significant. R software version 3.3.1 (https://www.r-project.org/) was used for all the statistical analyses.

Data about mutations (SNVs) were available for 403 patients whose samples were successfully sequenced and met quality metric criteria, including 148 patients with breast cancer recurrences (BCFIs) during a median follow-up time of 5.2 (95% CI, 5.0–5.6) years. Gene focal CNG analysis was successfully performed for 350 patients, of which 134 experienced breast cancer events (BCFIs) during a median follow-up time of 5.2 (95% CI, 5.1,5.9) years. Unweighted distributions of clinicopathologic characteristics of patients in respective cohorts were consistent with those of patients enrolled in the SOLE ITT population, when considering the case-cohort–like sampling design (Table 1; Supplementary Table S1).

A total of 294 (73%) samples harbored at least one genomic alteration in the targeted genes. Overall, a heterogeneous repertoire of mutations was detected, with few recurrent genomic aberrations. A total of 214 (53%) of 403 SNV samples harbored at least one pathogenic mutation. All genes, but one, had a mutation frequency ≤5% (Supplementary Table S2). PIK3CA represented the only highly recurrently mutated gene. A total of 194 pathogenic mutations of PIK3CA were detected in 169 samples (42%; weighted rate, 43%). Most PIK3CA pathogenic mutations affected the PI3K/PI4K (95/194, 49%) and the PIK helical (63/194, 32%) domains. A total of 126 (36%) of 350 CNG samples harbored at least one CNG. CCND1, ERBB2, FGFR1, and MYC were recurrently affected by gains that were detected in 54 (15%; weighted rate, 14%), 36 (10%; weighted rate, 11%), 29 (8%; weighted rate, 5%), and 28 (8%; weighted rate, 10%) samples, respectively (Supplementary Table S3). A total of 142 samples displayed cooccurrent pathogenic alterations. The most frequent coexistences included the cooccurrences of PIK3CA mutations and ERBB2 (n = 11) or CCND1 (n = 14) copy-number gains and concurrent CCND1 and FGFR1 copy-number gains (n = 8; Fig. 2). No fusion genes were identified in the subset of samples (n = 67) successfully subjected to RNA analysis (data not shown).

Targeted recurrent genomic aberrations (SNVs/CNGs) with a rate of >5% were detected in PIK3CA, CCND1, ERBB2, FGFR1, and MYC, and the associations with clinicopathologic factors were reported only for those genes. PIK3CA SNVs and MYC CNGs were associated with lower (P = 0.03) and higher (P = 0.004) tumor grade, respectively; tumor-harboring CNGs of CCND1, ERBB2, and MYC displayed a higher Ki-67 proliferation index (P = 0.01, P < 0.001, and P = 0.03, respectively; Supplementary Table S4). There was no association between breast cancer subtypes, as defined by IHC, and any recurrent gene aberrations, except for ERBB2 CNG with HER2-positive tumors (Fig. 2; Supplementary Table S4). As expected, there was a strong correlation between ERBB2 focal CNG identified by NGS analysis and HER2 overexpression/amplification as defined by IHC and/or FISH. Moreover, the presence of FGFR1 CNG was validated with FISH (Supplementary Fig. S1).

Similarly, targeted genes with genomic aberrations rate >5% were included in the analysis of the associations with disease outcomes (Table 2). No associations between the presence of PIK3CA mutation and outcomes were observed in the univariate analyses, considering both the overall mutations and subgroups stratified according to the different domains affected. In an exploratory analysis of patients with PIK3CA mutation, luminal A breast cancers showed a reduced recurrence risk as compared with luminal B tumors (Supplementary Fig. S2), although a test of subtypes-by-PIK3CA mutations interaction (interaction P value > 0.4) was not significant and likely underpowered. No associations between CCND1, ERBB2, and MYC gene CNGs and disease outcomes were observed (Table 2). In univariate analyses, the presence of FGFR1 CNG was associated with increased risks of both breast cancer events (HR, 3.2; 95% CI, 1.48–6.92; P = 0.003) and distant recurrences (HR, 3.5; 95% CI, 1.61–7.75; P = 0.002; Table 2; Fig. 3). The adjusted P value or FDR for FGFR1 CNG was 0.015 (BCFI analysis) or 0.01 (DRFI analysis). Multivariable modeling confirmed FGFR1 CNG as potential risk factor of late recurrences in this population (Table 3; Supplementary Table S5). There was no association between clinical outcomes and the presence of concurrent genomic aberrations or the number of genomic aberrations per sample (data not shown).

We analyzed the genomic aberrations of 52 cancer-related and actionable genes and their association with late recurrence in primary hormone receptor–positive, node-positive early breast cancers from a cohort of postmenopausal women enrolled in the SOLE trial. Our findings showed that patients with tumor harboring focal CNG of the FGFR1 gene had an increased risk of late relapse despite extended therapy.

NGS analysis of the primary tumors revealed a heterogeneous repertoire of genomic aberrations. As previously shown by extensive genomic profiling, breast cancer comprises a spectrum of molecularly different diseases with few recurrent alterations (6, 21). In our series, only five of 52 genes were altered in more than 5% of tumor samples. Considering the available data of whole-exome sequencing of estrogen receptor (ER)-positive, node-positive breast cancer of postmenopausal women from The Cancer Genome Atlas (TCGA) dataset (n = 202), recurrent genomic aberrations were identified in the same genes (21). PIK3CA gene mutations represent the most frequently mutated gene, being detected in 42% of samples as compared with 38% in TCGA dataset. CNGs of CCND1, ERBB2, FGFR1, and MYC were identified in 15%, 10%, 8%, and 8% of our samples, respectively, and 20%, 15%, 15%, and 20% of TCGA cases, respectively. Differences in the clinical characteristics of the patients and in the methods of analysis may account for the differences observed in the mutation/CNG frequencies. However, our data confirmed the important role of these genes as oncogenic drivers of hormone receptor–positive breast cancers.

Although PIK3CA represented the most frequently mutated gene in this study population, we did not identify any association between the presence of PIK3CA mutations and clinical outcomes. Several studies have investigated the prognostic role of PIK3CA in early breast cancer, although with inconsistent results (22–24). In the larger series, the presence of mutations has been associated with better clinical outcomes (22, 25). However, most of these studies have not been specifically focused on the evaluation of the risk of late recurrence, particularly in women with extended treatment. Our population included women with node-positive early breast cancer, but no early recurrence during the 4–6 years of standard adjuvant therapies, and all receiving extended therapy. Moreover, the prognostic role revealed by univariate analyses was less strong or lost in multivariable models given the frequent association between PIK3CA mutations and other favorable clinical and pathologic features (22, 26). Indeed, in our series, the presence of PIK3CA SNVs was associated with a lower tumor grade. Among driver genes affected by copy-number aberrations, only FGFR1 gene was associated with late relapse. Focal CNGs of CCND1, ERBB2, and MYC were identified in tumors with unfavorable clinicopathologic features (i.e., higher Ki-67 proliferation index and/or higher tumor grade). As previously shown, these genomic aberrations characterize aggressive subgroups of genomic unstable hormone receptor–positive breast cancers and they have been associated with poor early prognosis (25, 27–30). Our data showed no association of CCND1, ERBB2, or MYC CNGs with late clinical outcomes suggesting that the same oncogenic drivers may play different roles in the early and late risk of breast cancer relapse.

The association of focal CNG of FGFR1 with worse late clinical outcomes was retained in the multivariable model demonstrating an independent prognostic role of FGFR1 CNG. FGFR1 belongs to the family of FGFR that binds FGF ligands. FGF signaling is involved in fundamental physiologic mechanisms of the cell and aberrant activation of this signaling by means of FGFR gene alterations leads to the activation of several oncogenic mechanisms (31). Amplification of FGFR1 gene (i.e., focal CNG), located in the 8p11-12 chromosomal region, is the most common mechanism of deregulation of FGFR1 function and it has been reported in 5%–27% of breast cancers, particularly in hormone receptor–positive tumors (32, 33). This alteration has been associated with poor prognosis in patients with unselected ER-positive and HER2-positive breast cancer and with the resistance to endocrine therapies and CDK4/CDK6 inhibitors (34–36). Several studies have been focused on the evaluation of FGFR1 aberrations as a potential therapeutic target in breast cancer. Despite the promising preclinical results, clinical trials have shown small benefit from anti-FGFR therapies in patients with advanced tumors and the actionability and the predictive value of the FGFR1 aberration are still controversial (31, 33). However, the role of FGFR1 alterations in late recurrence has been rarely investigated. In an unselected consecutive population of women with ER-positive/HER2-negative early breast cancer, FGFR1 overexpression, but not the gene CNG, was associated with relapse events particularly after 5 years of standard adjuvant therapy (37). Indeed, few data are available about the molecular characteristics of late-recurrent tumor. Multigene expression assays that are currently used for the early-risk stratification in hormone receptor–positive breast cancer have demonstrated a prognostic value for late recurrence. Moreover, an higher expression of estrogen responsive genes has been associated with an increased risk of late relapse (38). Among the integrative breast cancer subtypes, four hormone receptor–positive subgroups with higher risk of late recurrence have been identified. These tumors are characterized by the presence of gene copy-number alterations, and identified women who might benefit from extended treatments (39). In our study population, including postmenopausal women with previous operable early breast cancer receiving extended endocrine therapy, we found that focal CNG of FGFR1 gene may represent a prognostic biomarker of late recurrence despite extended treatment with an aromatase inhibitor.

This study has some limitations. First, strict thresholds of quality metrics are required to minimize false-positive results when NGS analyses are performed on old FFPE samples (13). For this reason, we obtained data for SNV and CNG for 80% and 70%, respectively, of the available samples. Nevertheless, the sample size was still powered to assess the association between the most frequent and relevant genomic alterations and clinical outcomes. Second, the targeted gene panel used in this analysis did not include some genes frequently altered in breast cancers and previously associated with poor prognosis, such as TP53 (23, 25, 40). Although we could not assess the prognostic role of this gene on late recurrence, the targeted gene panel surveyed in this study was specifically selected to include not only cancer-related, but also potentially actionable genes. Third, the 8p11-12 chromosomal region comprises other cancer-related genes that may be involved in the biological effects and thus, in the clinical role of FGFR1 aberrations (41–43). We cannot exclude that concurrent genomic events not targeted in this study might modulate the role of FGFR1 in breast cancer late recurrence. However, in our NGS assay, the amplicon that spans the 8p11-12 chromosomal region was designed to cover only the FGFR1 gene. Fourth, given that the SOLE population did not included patients with early recurrence, we cannot rule out a prognostic role of FGFR1 CNG in this setting. However, in our population with no early relapse, the presence of FGFR1 CNG in the primary tumor was associated with late-breast cancer events. Finally, given the relatively small number of events (i.e., FGFR1 focal CNG), our study is underpowered to assess the association between FGFR1 aberrations and the treatment arms of the SOLE trial, although in an exploratory analysis, no indication of interaction was seen.

In postmenopausal patients with hormone receptor–positive, node-positive early breast cancer and no recurrence during the first approximately 5 years of adjuvant therapies, the detection of focal CNG of FGFR1 gene may identify a population of women with higher risk of late-breast cancer recurrences despite the extension of endocrine therapy. These women might benefit from an intensive monitoring and different treatment strategies. Further studies are needed to evaluate whether the genomic aberrations of FGFR1 are clonal and still preserved in the recurrent tumors where they might represent potential actionable targets.

E. Guerini-Rocco reports grants from Italian Ministry of Health and nonfinancial support from International Breast Cancer Study Group during the conduct of the study, personal fees and other from Thermo Fisher Scientific, Novartis, AstraZeneca, and Roche, personal fees from MSD Italia, and other from Biocartis and Illumina outside the submitted work. P. Rafaniello Raviele reports grants from Ministry of Health, Italy during the conduct of the study. E. Munzone reports grants from Ministry of Health, Clinical Health Care Research, and Ricerca Finalizzata during the conduct of the study and personal fees from Genomic Health, Pierre Fabre, and Eisai outside the submitted work. G. Jerusalem reports grants, personal fees, and nonfinancial support from Novartis during the conduct of the study, grants, personal fees, and nonfinancial support from Roche, and Pfizer, personal fees and nonfinancial support from Lilly, Amgen, BMS, and AstraZeneca, personal fees from Daiichi Sankyo and AbbVie, and nonfinancial support from Medimmune and MerckKGaA outside the submitted work. A. Gombos reports other from Lilly and Daiichi Sankyo outside the submitted work. P. Karlsson reports grants and nonfinancial support from PFS Genomics and other from Roche outside the submitted work. S. Aebi reports other from Roche, Pfizer, and Pierre Fabre outside the submitted work. J. Chirgwin reports other from International Breast Cancer Study Group during the conduct of the study. A. Thompson reports other from Pfizer outside the submitted work. S. Loibl reports grants and other from AbbVie, Amgen, Celgene, Novartis, Roche, Daiichi Sankyo, AstraZeneca, and Pfizer, other from SeaGen, PriME/Medscape, Lilly, Samsung, BMS, Puma, MSD, Pierre Fabre, and Merck, personal fees from Chugai, and grants from Immunomedics outside the submitted work, as well as has a patent for EP14153692.0 pending. J. Gavila reports grants and personal fees from Novartis and grants from Pfizer and Lilly during the conduct of the study, and grants from Roche and AstraZeneca outside the submitted work. K. Kuroi reports other from Taiho Pharmaceutical Co, Kyowa Hakko Kirin Co, and Eisai Co outside the submitted work. S. O’Reilly reports personal fees from Novartis (makers of letrozole used in SOLE trial) during the conduct of the study and personal fees from Novartis outside the submitted work. A. Di Leo reports personal fees and nonfinancial support from AstraZeneca, Celgene, Pfizer, and Roche, personal fees from Amgen, Athenex, Bayer, Daiichi Sankyo, Eisai, Genentech, Genomic Health, Lilly, Ipsen, Pierre Fabre, and Seattle Genetics, and grants and personal fees from Novartis outside the submitted work. G. Viale reports personal fees from Roche Genentech, MSD Oncology, Bayer, AstraZeneca, Daiichi Sankyo, Dako Agilent, and Menarini outside the submitted work. M.M. Regan reports grants from Novartis during the conduct of the study, and grants from Pfizer, Ipsen, TerSera, Pierre Fabre, Roche, AstraZeneca, and Bayer, grants, personal fees, and nonfinancial support from Bristol-Myers Squibb, personal fees from Tolmar Pharmaceuticals, and other from Ipsen outside the submitted work. M. Colleoni reports grants from Ministry of Health during the conduct of the study and personal fees from Novartis outside the submitted work. No disclosures were reported by the other authors.

E. Guerini-Rocco: Conceptualization, data curation, investigation, visualization, writing-original draft, writing-review and editing. K.P. Gray: Conceptualization, data curation, formal analysis, investigation, visualization, writing-original draft, writing-review and editing. C. Fumagalli: Data curation, investigation, writing-review and editing. M.R. Reforgiato: Data curation, investigation, writing-review and editing. I. Leone: Investigation, writing-review and editing. P. Rafaniello Raviele: Investigation, writing-review and editing. E. Munzone: Conceptualization, resources, writing-review and editing. R. Kammler: Resources, project administration, writing-review and editing. P. Neven: Resources, writing-review and editing. E. Hitre: Resources, writing-review and editing. G. Jerusalem: Resources, writing-review and editing. E. Simoncini: Resources, writing-review and editing. A. Gombos: Resources, writing-review and editing. I. Deleu: Resources, writing-review and editing. P. Karlsson: Resources, writing-review and editing. S. Aebi: Resources, writing-review and editing. J. Chirgwin: Resources, writing-review and editing. V. Di Lauro: Resources, writing-review and editing. A. Thompson: Resources, writing-review and editing. M.-P. Graas: Resources, writing-review and editing. M. Barber: Resources, writing-review and editing. C. Fontaine: Resources, writing-review and editing. S. Loibl: Resources, writing-review and editing. J. Gavila: Resources, writing-review and editing. K. Kuroi: Resources, writing-review and editing. B. Müller: Resources, writing-review and editing. S. O’Reilly: Resources, writing-review and editing. A. Di Leo: Resources, writing-review and editing. A. Goldhirsch: Supervision, writing-review and editing. G. Viale: Resources, supervision, writing-review and editing. M. Barberis: Conceptualization, resources, supervision, writing-review and editing. M.M. Regan: Conceptualization, resources, formal analysis, supervision, writing-original draft, writing-review and editing. M. Colleoni: Conceptualization, resources, supervision, funding acquisition, writing-original draft, writing-review and editing.

We thank the women who participated in the SOLE trial and consented to use of the tissue for unspecified future studies. We thank physicians, nurses, trial coordinators, and pathologists who participated in the SOLE trial and collected and submitted tissue blocks. SOLE is a Breast International Group (BIG 01-07) and an International Breast Cancer Study Group (IBCSG 35-07) trial. We thank SOLE Steering Committee, IBCSG DSMC and IBCSG Data Management Center, Coordinating Center, Statistical Center for coordination, data management and statistical support, and Central Pathology Office for tumor block collection and processing. We acknowledge the laboratory technicians and biotechnologists of the Division of Pathology (European Institute of Oncology IRCCS, Milan, Italy) for sample processing. This work was partially supported by the Italian Ministry of Health with Ricerca Corrente and 5×1000 funds. The SOLE trial, including prospective tissue collection, was funded by Novartis, the manufacturer of letrozole, and the International Breast Cancer Study Group. This translational research study was funded by the MInistero della Salute (Ministry of Health, Italy), Clinical Health Care Research, and Ricerca Finalizzata (RF-2011-02350542).

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.