Association of Somatic Driver Alterations With Prognosis in Postmenopausal, Hormone Receptor–Positive, HER2-Negative Early Breast Cancer: A Secondary Analysis of the BIG 1-98 Randomized Clinical Trial

Key Points

Question  What is the clinical relevance of somatic driver alterations in estrogen receptor–positive, HER2-negative early breast cancer in postmenopausal women?

Findings  In a secondary analysis of the Breast International Group 1-98 randomized clinical trial, a sampling plan selected 764 samples for targeted DNA sequencing and after adjustment for clinicopathologic factors, amplifications on 11q13 and 8p11 and an increasing number of driver alterations were associated with a significantly increased risk of distant recurrence. PIK3CA mutations were associated with significantly greater benefit with letrozole compared with tamoxifen therapy.

Meaning  Classification of somatic driver alterations based on DNA analysis provides valuable prognostic information that may aid treatment decision making in the adjuvant setting.

Importance  A range of somatic driver alterations has been described in estrogen receptor–positive, HER2-negative (ER+/HER2−) early breast cancer (BC); however, the clinical relevance is unknown.

Objective  To investigate associations of driver alterations with prognosis and the role of PIK3CA mutations in prediction of benefit associated with endocrine therapy in postmenopausal patients with ER+/HER2− early BC treated with tamoxifen or letrozole.

Design, Setting, and Participants  The Breast International Group (BIG) 1-98 trial randomized 8010 postmenopausal patients with hormone receptor–positive, operable, invasive BC to monotherapy with letrozole, tamoxifen, or a sequential strategy for 5 years. Driver alterations were characterized using next-generation sequencing in primary tumors from a subset of 764 patients from 7329 eligible patients with ER+/HER2− BC, with 841 distant recurrences after a median of 8.1 years of follow-up. To correct for the oversampling of distant recurrences, weighted analysis methods were used. This analysis was conducted from April 4, 2016, to November 30, 2016.

Main Outcomes and Measures  The prevalence of driver alterations, associations with clinicopathologic factors, distant recurrence-free interval, and treatment interactions were analyzed. Multivariable analyses were performed to adjust for clinicopathologic factors.

Results  Of 764 samples, 538 (70.4%), including 140 distant recurrence events, were successfully sequenced. Nineteen driver alterations were observed with 5% or greater frequency, with a mean of 4 alterations (range, 0-15) per tumor. PIK3CA mutations were the most common (49%) and were significantly associated with reduction in the risk for distant recurrence (hazard ratio [HR], 0.57; 95% CI, 0.38-0.85; P = .006). TP53 mutations (HR, 1.92; 95% CI, 1.21-3.04; P = .006), amplifications on 11q13 (HR, 2.14; 95% CI, 1.36-3.37; P = .001) and 8p11 (HR, 3.02; 95% CI, 1.88-4.84; P < .001), and increasing number of driver alterations (HR per additional alteration, 1.18; 95% CI, 1.11-1.25; P < .001) were associated with significantly greater risk. Amplifications on 11q13 and 8p11 remained significant predictors in multivariable analysis, but not PIK3CA and TP53 mutations. Patients with tumors harboring kinase or helical domain PIK3CA mutations derived significantly greater benefit from letrozole over tamoxifen than patients whose tumors did not (P interaction = .002).

Conclusions and Relevance  In ER+/HER2− postmenopausal, early-stage BC, amplifications on 11q13 and 8p11 were significantly associated with increased risk for distant recurrence and PIK3CA mutations were predictive of greater magnitude of benefit from letrozole. With these findings, DNA-based classification may aid adjuvant treatment decision making in this setting.

Trial Registration  ClinicalTrials.gov Identifier: NCT00004205

Large-scale, next-generation sequencing studies have provided valuable insights into the genomic landscape of primary breast cancers (BCs).^1-5 These studies have highlighted the diverse spectrum of driver alterations and mutational processes in estrogen receptor–positive, HER2-negative (ER+/HER2−) disease, the most commonly diagnosed BC subtype. Despite these findings, the clinical implications of driver alterations including the most frequent—PIK3CA (Entrez 5290) mutations—remains to be fully elucidated.

Women with a diagnosis of early-stage ER+/HER2− BC typically receive adjuvant systemic therapy consisting of endocrine therapy, often preceded by chemotherapy. In addition to markers of anatomic tumor burden,⁶ a number of gene expression–based markers exist to aid in estimation of prognosis and decision making regarding adjuvant chemotherapy.^7-9 These markers are based on continuous gene expression variables and the characterization of phenotype resulting from underlying genetic and epigenetic processes. A common theme of gene expression signatures is the observation that higher rates of proliferation portend poorer prognosis.¹⁰

For adjuvant endocrine therapy, the Breast International Group (BIG) 1-98 study was a seminal, phase 3 clinical trial that evaluated the superiority of 5 years of letrozole over tamoxifen in postmenopausal women with early-stage, hormone receptor–positive BC. At a median of 8.1 years’ follow-up, 5 years of letrozole treatment demonstrated a significant reduction in the risk of death compared with tamoxifen alone, establishing aromatase inhibition as the standard of care for postmenopausal women with hormone receptor–positive BC.^11-13 Other endocrine therapy options now also exist with the type, sequencing, duration, and possible future incorporation of cyclin-dependent kinase 4/6 inhibitors as considerations in therapeutic decision making.¹⁴ Clinically useful biomarkers to aid in individualizing these options are lacking.

PIK3CA mutations are particularly enriched in luminal BCs,¹ associated with improved outcomes and sensitivity to endocrine therapy in ER+ BC,¹⁵ and have been shown to be associated with the initiation of ER+ tumors.¹⁶ Therefore, owing to its association with estrogen signaling, we hypothesized that letrozole, being the more potent antiestrogen, would have higher clinical benefit than tamoxifen in patients with tumors harboring PIK3CA mutations.

In this study, we aimed to characterize the clinical relevance of oncogenic drivers in postmenopausal patients with early-stage ER+/HER2− BC enrolled in the BIG 1-98 study. Our primary objective was to investigate the prognostic associations of driver alterations with distant recurrence-free interval. As a secondary objective, we evaluated whether PIK3CA mutations could be predictive of a greater magnitude of benefit with letrozole over tamoxifen and hence could have implications for selection of adjuvant endocrine treatment.

BIG 1-98 was a multicenter, randomized, double-blind, phase 3 trial in which 8010 postmenopausal patients with hormone receptor–positive, operable, invasive BC were randomly assigned to monotherapy with letrozole (2.5 mg orally daily) or tamoxifen (20 mg orally daily) for 5 years, a sequential strategy of letrozole for 2 years followed by tamoxifen for 3 years, or the reverse sequence. After a median follow-up of 8.1 years, 1022 distant recurrences were observed. Details of the study have been reported previously.^11-13,17 All patients provided written informed consent. Ethics committees and relevant health authorities approved the protocol. Tumor samples were collected retrospectively in accordance with institutional guidelines and national laws. The International Breast Cancer Study Group Biological Project Working Group approved this investigation.

Library preparation, hybridization-based capture, and next-generation sequencing were performed using Foundation Medicine’s T5-targeted panel of 287 known cancer genes in a Clinical Laboratory Improvement Amendments–certified and College of American Pathologists–accredited laboratory using a previously published method (eTable 1 in the Supplement).¹⁸ Samples with a minimum median exon depth of coverage of 150 × were eligible for analysis.

After filtering for known germline variants in publicly available databases, only alterations predicted with the accredited variant annotation pipeline were used in these analyses.¹⁸ These alterations were annotated as known, likely, ambiguous, or unknown. To reduce false-positives, only alterations annotated as known or likely to be pathogenic were included for subsequent analyses as driver alterations. All sequencing and annotation were done without knowledge of patients’ treatment assignments or outcomes.

Patient selection is summarized in Figure 1. Patients were ineligible if the tumor was ER-absent or HER2-positive on central pathology assessment, neoadjuvant endocrine or chemotherapy was administered, or follow-up was less than 3 months. Of the 8010 randomized patients in the BIG 1-98 trial, there were 7329 eligible patients with 841 (11.5%) distant recurrences. From the 2706 patients with banked, assessable DNA extracted from formalin-fixed, paraffin-embedded archival tumor samples, 938 patients were selected, using a sampling plan identifying all patients having a distant recurrence and a stratified random sampling of those without recurrence (stratified by randomization option [2 groups vs 4 groups], treatment assignment, nodal status [negative vs positive], and luminal A- or B-like status¹⁹). After further exclusion for inadequate DNA quality or quantity, 764 samples were sent for sequencing. To correct for oversampling of distant recurrences, weighted analysis methods were used. Sampling weights were calculated for 72 classes, defined by the presence or absence of distant recurrence and the 36 strata, as the inverse of the sampling fractions. Analysis was conducted from April 4, 2016, to November 30, 2016.

At the design stage, we estimated that, with 350 (10%) distant recurrence events in the eligible cohort and a sampling fraction of 0.10 for a total sample size of 700 patient samples, there would be 80% power to detect hazard ratios (HRs) for association of driver alterations with distant recurrence-free interval of 1.5 for alteration prevalence of 20%, and HR of 2.0 for alteration prevalence of 5%, with nominal α = .05 tests.²⁰

The primary end point of interest was distant recurrence-free interval, defined as time from randomization to recurrence at a distant site. Patients without distant recurrence were censored at the date of last follow-up or death. Driver alterations were categorized as present or absent (for short variants, as mutated vs wild type; for amplifications, as amplified vs nonamplified). Because oncogenic driver mutations in PIK3CA cluster in distinct functional protein domains,²¹ we annotated PIK3CA mutations by affected protein domain: kinase or helical domain mutations. PIK3CA mutations not affecting kinase or helical protein domains were uncommon, less well described, and annotated as other. Exploratory analyses by previously described PIK3CA mutation hotspots (N345, C420, P539, E542, E545, E546, H1047) were performed.²²

The generalized Horvitz-Thompson weighted method (inverse probability weighting) was applied to all analyses using the sample weights previously described.²³ For clinicopathologic associations, weighted χ² tests were used for categorical variables (age, tumor size, nodal status, and grade) and weighted t tests were used for Ki-67 (percent). Pairwise associations were investigated without sampling weights, characterized by odds ratios and using Fisher exact test to generate P values; multiple testing correction using the Benjamini-Hochberg method was applied using a false discovery rate of less than 0.2. For the remainder of the analyses, we deemed an unadjusted, 2-sided value of P < .05 to be significant.

Weighted Cox proportional hazards regression models adjusted for treatment arm were used to determine prognostic associations. For each driver alteration, models were adjusted for treatment arm and included the following variables: age, tumor size, nodal status, and grade. Treatment-by-PIK3CA genotype interactions were used to assess the outcome of monotherapy in tumors with or without a kinase or helical domain PIK3CA mutation (both wild-type and other PIK3CA mutations). Hazard ratios and 95% CIs were generated using robust SEs, with a Wald test for significance. Weighted Kaplan-Meier curves were generated for visualization purposes. Subpopulation treatment effect pattern plot (STEPP)²⁴ methodology was used in exploratory analyses to illustrate the association between Ki-67 level, PIK3CA mutation status, and disease outcome.

All statistical analyses were prespecified. We used R software, version 3.3.1 (R Foundation) and Stata, version 14 (StataCorp) for the statistical analyses.

Of 764 samples, 538 (70.4%) successfully underwent DNA sequencing and met quality metrics. The median exon depth of targeted genes ranged from 151 × to 1397 × , with an average of 476 × . Patient characteristics are summarized in eTable 2 in the Supplement alongside the eligible patients from the BIG 1-98 trial (n = 7329).

Nineteen driver alterations had a prevalence of 5% or greater, including 9 genes with recurrent mutations and 10 genes with recurrent focal amplifications (Figure 2; eTable 3 in the Supplement). PIK3CA mutations were the most common (49%), followed by CCND1 (Entrez 595) amplifications (17%), and TP53 (Entrez 7157) short-variants (15%). PIK3CA-mutated tumors were commonly coaltered (76%) and exhibited heterogeneous coalteration partners (eFigures 1, 2, and 3 in the Supplement). Coalteration frequencies by PIK3CA genotype were similar, with the exception of amplifications on chromosome 11q13, which were significantly more common in wild-type tumors (eFigure 4 in the Supplement). No identifiable driver alteration was identified in 5.7% of patient samples. There was a mean of 4 driver alterations (range, 0-15; interquartile range, 2-6) per tumor. Driver mutations were uncommon for AKT1 (Entrez 207) (4%), ERBB2 (Entrez 2064) (2%), and ESR1 (Entrez 2099) (0%). There were no recurrent structural rearrangements.

The associations between driver alterations and clinicopathologic variables are summarized (eFigures 5 and 6 and eTable 4 in the Supplement). MAP3K1 (Entrez 4214), SPEN (Entrez 23013), and PTPN11 (Entrez 5781) mutations were significantly associated with larger tumor size. MAP2K4 (Entrez 6416) mutations were significantly associated with age older than 65 years. There were no significant associations of driver alterations with nodal status. PIK3CA and MAP3K1 mutations were significantly associated with lower tumor grade and lower Ki-67 level. Conversely, TP53 mutations and several focal gene amplifications (CCND1, FGFR1 [Entrez 2260], MYC [Entrez 4609], and ZNF703 [Entrez 80139]) were significantly associated with higher tumor grade and higher Ki-67 levels. We observed that PIK3CA-mutated tumors were particularly heterogeneous in terms of coexisting alterations and their Ki-67 levels (eFigure 5 in the Supplement).

To understand biological relationships between driver alterations, we examined for pairwise interactions (eFigure 7 and eTable 5 in the Supplement). TP53 mutations were significantly associated with the coexistence of several gene amplifications (MYC, ERBB2, FGFR1, ZNF703, and ZNF217 [Entrez 7764]). Strong mutual coexistence was observed between several gene amplifications consistent with coamplification of the same amplicon (CCND1 [Entrez 14174], FGF3, FGF4 [Entrez 14175], FGF19 [Entrez 9965], and EMSY [Entrez 56946] on 11q13; FGFR1, ZNF703, and GPR124 [Entrez 25960] on 8p11; and GNAS [Entrez 2778] and ZNF217 on 20q13). These gene amplifications were subsequently combined and analyzed per amplicon for prognostic associations.

We evaluated associations between frequent driver alterations (≥5% frequency) and prognosis in both univariate and multivariable models (Figure 3). In univariate analysis, PIK3CA mutations were significantly associated with reduced risk of distant recurrence (HR, 0.57; 95% CI, 0.38-0.85; P = .006). Analysis of PIK3CA mutations by protein domain vs wild-type PIK3CA demonstrated favorable outcomes for all subgroups (kinase domain, helical domain, and other) (eTable 6 in the Supplement). Conversely TP53 mutations (HR, 1.92; 95% CI, 1.21-3.04; P = .006), 11q13 amplifications (HR, 2.14; 95% CI, 1.36-3.37; P = .001), and 8p11 amplifications (HR, 3.02; 95% CI, 1.88-4.84; P < .001) were associated with significantly increased risk of distant recurrence (eFigure 8 in the Supplement). In multivariable analysis adjusting for age, tumor size, nodal status, and grade, only 11q13 amplifications (HR, 1.72; 95% CI, 1.12-2.64; P = .01) and 8p11 amplifications (HR, 2.00; 95% CI, 1.25-3.21; P = .004) remained statistically significant. Increasing number of somatic driver alterations per tumor was also associated with worse outcomes in both univariate (HR, 1.18; 95% CI, 1.11-1.25; P < .001) and multivariable (HR, 1.11; 95% CI, 1.04-1.19; P = .002) analysis.

We next investigated whether PIK3CA mutations were associated with a significant treatment interaction. Consistent with our hypothesis, patients with tumors harboring a PIK3CA mutation (kinase or helical, n = 114) derived a greater magnitude of benefit with adjuvant letrozole over tamoxifen (HR, 0.18; 95% CI, 0.06-0.50) than patients whose tumors did not (n = 184) (HR, 1.26; 95% CI, 0.65-2.45; P interaction = .002) (Figure 4; eTable 7 in the Supplement). Further sensitivity analysis by PIK3CA mutation hotspot demonstrated similar results (eTable 8 in the Supplement). Of these patients, 36% were node-positive and 25% had received adjuvant chemotherapy. Patients whose tumors harbored a kinase or helical domain PIK3CA mutation and received letrozole monotherapy had an excellent prognosis, with an estimated 5-year distant recurrence-free proportion of 99% (95% CI, 98%-99%). A statistically significant difference in treatment effect (P interaction = .01) was also observed in the subgroup of no chemotherapy.

Higher Ki-67 levels have been reported to be predictive of increased benefit to adjuvant letrozole.²⁵ As an exploratory analysis, we investigated the association between Ki-67 level and the presence of kinase/helical domain PIK3CA mutations as predictive biomarkers.²⁵ In a Cox proportional hazards multivariable model that included adjustment for Ki-67 level, the treatment interaction with PIK3CA mutation status remained statistically significant (P = .03) (eTable 9 in the Supplement). We further performed a STEPP analysis to examine the association between Ki-67 level as a continuous variable and disease outcome, adjusted for PIK3CA mutation status (eFigure 9 in the Supplement). As illustrated, the increased magnitude of benefit of adjuvant letrozole appeared to be driven by PIK3CA mutation status independent of Ki-67 levels.

In this analysis of primary tumor samples from postmenopausal women with ER+/HER2− BC from the randomized phase 3 BIG 1-98 study, we have described the landscape of driver alterations, their association with clinicopathologic factors, and prognostic associations for distant recurrence-free interval. Furthermore, we report that patients with tumors harboring kinase or helical domain PIK3CA mutations derived a greater magnitude of benefit with letrozole over tamoxifen therapy and had long-term freedom from distant recurrence. To our knowledge, this is the first study to report a DNA-based predictive biomarker in this setting. Given that relapse from ER+/HER2− disease is the major cause of death from BC, our results are important in understanding the biological mechanisms of aggressive ER+/HER2− disease.

Similar to previous studies in heterogeneously selected BCs, we observed a diverse driver alteration landscape.^1,3,4,26PIK3CA mutations were the most common, followed by TP53 mutations, with many other mutations occurring at lesser frequencies. One difference in the BIG 1-98 data set was that recurrent mutations in PTPN11 and SPEN were detected, which have only recently been described in BC,^27-29 representing potential new drug targets as well as novel biology. Amplifications were frequent, with recurrent amplicons on chromosomes 11 (11q13/14), 8 (8p11), and 20 (20q13) detected.³ We observed significant coexistence associations between TP53 mutations and several focal gene amplifications (CCND1, FGFR1, MYC, and ZNF703), higher grade and Ki-67, suggesting that these alterations cooperate synergistically to produce a genomically unstable and aggressive ER+ BC phenotype.^4,26

With respect to risk of distant recurrence, TP53 mutations and amplifications on 11q13 and 8p11 were associated with significantly poorer outcome. This outcome has been shown previously,^3,4,30,31 with higher proliferation rates and promotion of endocrine resistance. Our unique data set lends support to their prognostic implications in early-stage postmenopausal ER+ BC. A novel finding was that progressively poorer outcomes were observed as numbers of driver alterations increased. This has been shown previously in myelodysplasia³³ and suggests that driver alterations have significant implications even if they are subclonal, consistent with an adverse prognostic role described for intratumor heterogeneity³⁴ and the presence of subclonal drivers³⁵ observed in other cancer types.

As expected, tumor PIK3CA mutations were associated with a better outcome but did not remain significant in the multivariable model owing to associations with good prognosis variables.^15,22,32,36 Our data, however, highlight the diversity in coexisting driver alterations with PIK3CA. This variation may result in a heterogeneous clinical phenotype but also suggests the need for future clinical studies in this setting to take genomic context into consideration.

We also report the novel finding of a predictive role for PIK3CA genotype in identifying patients who derived greater benefit from 5 years of letrozole adjuvant therapy. In patients with tumors harboring kinase/helical PIK3CA mutations, the estimated proportion of patients without a distant recurrence at 5 years for those who received 5 years of letrozole was 99%. This finding is biologically plausible^15,16 and important as PIK3CA mutations are enriched in ER+/HER2− BC. Further studies to establish the exact mechanisms underlying this observation and validate our finding are warranted. If confirmed, PIK3CA genotyping could be easier to implement in the clinic because of the ease of interpreting a binary result (mutated vs not) and reproducibility of a DNA-based test.

Higher Ki-67 levels have previously been reported to be predictive of increased benefit to adjuvant letrozole.²⁵ However, to our knowledge, no prior study has established the interaction between Ki-67 and PIK3CA mutation status on treatment effect comparing letrozole with tamoxifen. The interaction between PIK3CA status and treatment assignment remained statistically significant in a multivariable model, suggesting the genotype had independent predictive value even after adjustment for Ki-67 level. Moreover, using STEPP analysis (eFigure 9 in the Supplement), the magnitude of increased benefit of letrozole at higher Ki-67 levels appeared to be limited to the PIK3CA-mutated subgroup, suggesting that PIK3CA genotype provided additional predictive value above Ki-67 alone.

Although retrospective, our data are strengthened by the use of a clinical trial data set with prospective treatment randomization and robust recurrence data with a median follow-up of 8 years. The strength of biomarker analyses in this context has been previously reported.³⁷

Our study had a limited ability to examine subgroups and infrequent driver alterations. While we used only 20% of the samples from the total trial cohort, the sampling design and weighted analysis is a financially efficient, well-accepted, and methodologically sound approach for a biomarker study (eTable 2 in the Supplement).^23,38 For example, the lobular histologic subtype has also been reported to derive increased benefit from adjuvant letrozole.³⁹ While lobular carcinomas have been recently reported to harbor high frequencies of PIK3CA mutations,⁴⁰ our study was underpowered to confirm that PIK3CA mutations are independently driving this outcome in the lobular carcinomas.

Our study highlights the feasibility of DNA sequencing of formalin-fixed, paraffin-embedded samples,¹⁸ especially as next-generation sequencing technologies become increasingly accessible as costs reduce. We propose that DNA-based biomarkers have advantages compared with gene expression–based prognostic scores^7-9 in that categorical findings are achieved rather than continuous scores, reducing the uncertainty of intermediate values as well as the chance to identify therapeutic targets.

We have described the prognostic and predictive relevance of somatic driver alterations in ER+/HER2− early BC in postmenopausal women from a large, phase 3 adjuvant randomized study. DNA-based classifications have promise to add to current prognostic markers and aid our adjuvant treatment decisions in ER+/HER2− disease.

Corresponding Author: Sherene Loi, MD, PhD, Peter MacCallum Cancer Centre, University of Melbourne, 305 Grattan St, Melbourne, Victoria 3000, Australia (sherene.loi@petermac.org).

Accepted for Publication: March 25, 2018.

Published Online: June 14, 2018. doi:10.1001/jamaoncol.2018.1778

Author Contributions: Dr Luen and Ms Asher contributed equally to the study and Drs Regan and Loi contributed equally to the study. Drs Luen and Loi had full access to all the data in the study and take responsibility for the integrity of the data and accuracy of the data analysis.

Study concept and design: Luen, Lee, Hackl, Thuerlimann, Colleoni, Regan, Loi.

Acquisition, analysis, or interpretation of data: Luen, Asher, Lee, Savas, Kammler, Dell'Orto, Biasi, Demanse, JeBailey, Dolan, Hackl, Thuerlimann, Viale, Regan, Loi.

Drafting of the manuscript: Luen, Asher, Lee, Regan, Loi.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Luen, Asher, Lee, Demanse, JeBailey, Regan, Loi.

Obtained funding: Hackl, Thuerlimann, Regan, Loi.

Administrative, technical, or material support: Luen, Kammler, Dell'Orto, Biasi, Dolan, Thuerlimann, Viale, Loi.

Study supervision: Lee, Kammler, Thuerlimann, Colleoni, Loi.

Conflict of Interest Disclosures: Mr Demanse and Drs JeBailey, Dolan, and Hackl are employees of Novartis. Dr Colleoni has received honoraria from Novartis and acted as unpaid consultant to Pierre Fabre, Pfizer, OBI Pharma, Puma Biotechnology, Celldex, and AstraZeneca. Dr Regan has received research funding and/or provision of drug supply for clinical trials from Novartis, Pfizer, Ipsen, Merck, Ferring, Celgene, AstraZeneca, Pierre Fabre, and Roche; research funding from Novartis, Veridex, OncoGenex, Bayer, and Bristol-Myers Squibb; and serves in an unpaid consulting or advisory role for Ipsen, Merck, and Bristol-Myers Squibb. Prof Loi receives research funding to her institution from Novartis, BMS, Merck, Roche-Genentech, Puma Biotechnology, and Pfizer, and serves as an uncompensated consultant for Seattle Genetics, Pfizer, Novartis, BMS, Merck, and Roche-Genetech. No other conflicts were reported.

Funding/Support: This translational research was funded in part by Novartis (partial funding of tissue block collection and full funding of next-generation sequencing at Foundation Medicine) and Susan G. Komen for the Cure Promise Grant (nucleic acid extraction, grant KG080081 to Drs Viale and Regan). The Breast International Group (BIG) 1-98 trial was supported by Novartis. The International Breast Cancer Study Group (IBCSG), which coordinated the BIG 1-98 trial, was also supported by The Swedish Cancer Society, The Swedish Research Council, The Cancer Council Australia, Australia and New Zealand Breast Cancer Trials Group, Frontier Science and Technology Research Foundation, Swiss Group for Clinical Cancer Research, Cancer Research Switzerland/Oncosuisse, and the Foundation for Clinical Cancer Research of Eastern Switzerland. Dr Luen is supported by the University of Melbourne. Prof Loi is supported by Breast Cancer Research Foundation of Australia and the John Colebatch Cancer Council Victoria Clinical fellowship.

Role of the Funder/Sponsor: The IBCSG and Novartis collaborated on the design and conduct of the study and the collection and management of the data; the affiliated authors were also involved in the analysis and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. The BIG 1-98 Steering Committee, which includes IBCSG and Novartis representatives, reviewed the manuscript prior to publication. The other funding organizations had no role in this research.

Meeting Presentation: This study was presented at the 2016 Meeting of the San Antonio Breast Cancer Symposium; December 7, 2016; San Antonio, Texas.

Additional Contributions: We are indebted to the women, physicians, nurses, and data managers who participated in the BIG 1-98 clinical trial and the many pathologists who submitted tumor blocks; to the BIG Collaborative Group, BIG secretariat, and BIG 1-98 Steering Committee; to the IBCSG for the design of the trial, coordination, data management, medical review, and statistical support; and to the IBCSG Central Pathology Office for tumor block collection and processing.