A Novel MAPK–Microrna Signature Is Predictive of Hormone-Therapy Resistance and Poor Outcome in ER-Positive Breast Cancer Philip C

Published OnlineFirst November 4, 2014; DOI: 10.1158/1078-0432.CCR-14-2053

Personalized Medicine and Imaging Clinical Cancer Research A Novel MAPK–microRNA Signature Is Predictive of Hormone-Therapy Resistance and Poor Outcome in ER-Positive Breast Cancer Philip C. Miller1,2, Jennifer Clarke1,2,3,4,5, Tulay Koru-Sengul1,4, Joeli Brinkman1, and Dorraya El-Ashry1,2,3

Abstract

Purpose: Hyperactivation of ERK1/2 MAPK (hMAPK) leads to publicly available datasets. A robust determination of a recur- loss of estrogen receptor (ER) expression and poor outcome in rence signature and a survival signature identified hMAPK– breast cancer. microRNAs (miRNA) play important regulatory miRNAs commonly associated with poor clinical outcome, and roles and serve as biomarkers of disease. Here, we describe specific subsets associated more closely with either disease þ molecular, pathologic, and clinical outcome associations of an recurrence or disease survival, especially among ER cancers hMAPK–miRNA expression signature in breast cancer. of both luminal A and luminal B subtypes. Multivariate anal- Experimental Design: An hMAPK–miRNA signature was iden- yses indicated that these recurrence and survival signatures tified, and associations of this signature with molecular and significantly associated with increased risk of disease-specific þ þ genetic alterations, gene expression, pathologic features, and death and disease recurrence in ER cancer and ER cancers clinical outcomes were determined in primary breast cancers from treated with hormone therapy. training data and validated using independent datasets. Univar- Conclusions: We report an hMAPK–miRNA signature and two iate and multivariate analyses identified subsignatures associated subsignatures derived from it that associate significantly with with increased disease recurrence and poorer disease survival adverse clinical features, poor clinical outcome, and poor þ among ER-positive (ER ) patients, respectively. response to hormone therapy in breast cancer, thus identifying Results: High-hMAPK–miRNA status significantly correlated potential effectors of MAPK signaling, and novel predictive and with ER-negativity, enrichment for basal and HER2-subtypes, prognostic biomarkers or therapeutic targets in breast cancer. Clin and reduced recurrence-free and disease-specificsurvivalin Cancer Res; 21(2); 1–13. 2014 AACR.

Introduction receptors such as erbB family members, EGFR, or HER2 (3–5). Overexpression or amplification of HER2 occurs in approxi- Treatment decisions in breast cancer are informed not only mately 25% of breast cancers, while EGFR overexpression by tumor grade, stage, and lymph node status, but also by occurs in up to 30% of breast cancers. Both are poor prognostic genomic biomarkers, estrogen receptor (ER), and HER2. indicators independent of ER status; EGFR expression has an Increasingly, multigene assays such as Oncotype DX and Mam- inverse correlation with ER and is more common in triple- maprint are being used in prognostic capacities for patients þ negative and basal breast cancers (6–8), while HER2 amplifi- with breast cancer (1, 2). ER tumors have better overall cation tends to correlate with ER-negativity or lower levels of ER prognosis and response to hormonal therapy, while ER expression. Downstream effector pathways activated by erbB tumors are more aggressive, resistant to hormonal therapies, family signaling include the RAS–RAF–MEK–ERK, PI3K–AKT– and frequently display elevated expression of growth factor MTOR, and the MEKK1–SEK1–JNK pathways. Over the past decade, prominent roles for PI3K–AKT–MTOR signaling in the biology of breast cancer have been established, in particular 1Sylvester Comprehensive Cancer Center, University of Miami Miller downstream of HER2 (ERBB2: gene product: HER2 protein School of Medicine, Miami, Florida. 2Braman Family Breast Cancer product) signaling and in tamoxifen resistance mechanisms Institute, University of Miami Miller School of Medicine, Miami, Florida. – 3Department of Medicine, University of Miami Miller School of Med- (9 11). icine, Miami, Florida. 4Department of Epidemiology and Public Health, We have previously examined the role of hyperactivation of University of Miami Miller School of Medicine, Miami, Florida. 5Depart- ERK1/2 MAPK (hMAPK) occurring downstream of EGFR or HER2 ment of Statistics, University of Nebraska-Lincoln, Lincoln, Nebraska. in breast cancer. We established that hMAPK induces loss of ER Note: Supplementary data for this article are available at Clinical Cancer expression, and that inhibition of hMAPK in ER breast cancer cell Research Online (http://clincancerres.aacrjournals.org/). lines and primary cultures from ER breast tumors restores ER Corresponding Author: Dorraya El-Ashry, Department of Medicine, University expression, suggesting that hMAPK plays a direct role in establish- of Miami Miller School of Medicine, 1501 NW 10th Avenue, BRB 716, Miami, FL ing the ER phenotype in breast cancer (12–14). mRNA expres- 33136. Phone: 305-243-4721; Fax: 305-243-6170; E-mail: sion profiling of these hMAPK cell lines defined an "hMAPK gene [email protected] expression profile" that strongly correlates with ER status in doi: 10.1158/1078-0432.CCR-14-2053 clinical samples, establishing a link between hMAPK and the 2014 American Association for Cancer Research. transcriptomes of ER breast tumors (15).

www.aacrjournals.org OF1

Miller et al.

The Cancer Genome Atlas (TCGA; accessed April 2012; ref. 32), Translational Relevance Lyng (GSE37405, accessed May 2012; ref. 33), and the METABRIC Activated signaling through the ERK1/2–MAPK axis is asso- breast cancer dataset (accessed July 2013; refs. 34, 35). Patient ciated with estrogen receptor (ER)–negativity and poor out- populations have been previously described in detail (22, 31–35). come in breast cancer. microRNA (miRNA) profiling is a Enerly and Lyng cohorts have previously been used as validation powerful tool complementing gene expression analysis and or comparison datasets for identification of novel roles of miRNAs provides additional insight into the biodiversity of breast in breast cancer (35–40). cancer, helping to further stratify existing breast cancer classi- fications into more discrete and coherent subtypes. Our study Gene and miRNA expression datasets defines a miRNA signature associated with hyperactive ERK1/ Gene and/or miRNA expression data for Buffa, Enerly, and Lyng 2–MAPK signaling and investigates associations with genetic datasets were retrieved from NCBI GEO. Gene expression, DNA and molecular alterations, pathologic features, and clinical copy number, reverse phase protein array (RPPA) protein expres- outcomes among primary breast tumors. We report that a sion, miRNA expression, and mutation status for breast cancers majority of ER-negative (ER ) tumors possess this miRNA from the TCGA dataset were obtained using the CGDS-R package þ signature, as well as a group of ER tumors that exhibit v 1.1.30. miRNA expression for the METABRIC dataset was molecular and genetic characteristics similar to ER tumors, retrieved from the European Genome-phenome Archive. and which display significantly poorer clinical outcomes, including poorer responses to hormone therapy. This suggests Signature generation and statistical analysis that these miRNAs may be important in facilitating ER-neg- mRNA expression and miRNA expression values were stan- ativity and poor outcomes in breast cancer and may be dardized by scaling expression of an individual mRNA/miRNA predictive of resistance to hormone therapy. within each individual dataset to have mean, 0 and standard deviation, 1. Buffa and TCGA datasets were used as training datasets for generating the hMAPK–miRNA signature. Cancers from each dataset were classified as high-hMAPK–mRNA or low- hMAPK–mRNA by Pearson correlation to an idealized hMAPK– MicroRNAs (miRNA) play critical regulatory roles in a wide mRNA signature as performed by Creighton and colleagues (15). range of biologic and pathologic processes, and miRNA expres- Briefly, the ideal signature genes upregulated in the signature sion profiling reveals substantial differences between cancer and were given an arbitrary value of "1," and genes downregulated in normal tissue, and among different cancer types (16–18). In the the signature an arbitrary value of "1"; Pearson correlation context of breast cancer, miRNA expression has been correlated between the ideal signature and z-score standardized gene with numerous biopathologic features such as tumor grade, ER, expression for breast cancers from the training and validation PR, and HER2 status (19–21), TP53 mutation status, or prolifer- datasets were determined and cancers with positive Pearson ation status (22), and has also been used to classify the breast correlation were classified as "high-hMAPK—mRNA," whereas cancer subtypes identified by gene expression profiling (23). cancers with negative Pearson correlation were classified as "low- miRNA expression studies have demonstrated the utility for hMAPK–mRNA." miRNA profiles as disease classifiers and prognostic tools in breast miRNAs significantly differentially expressed (P 0.05) cancer (24, 25). Functional studies indicate that miRNAs regulate between breast cancers from training datasets classified as important genes and processes germane to breast cancer pathol- hMAPK–mRNA or not-hMAPK–mRNA were identified by the ogy, and are effectors of disease-associated signaling pathways Student t test. P values for expression differences between groups (26–30). were permutation-adjusted by randomly assigning each sample to In the present study, we identify miRNAs associated with a group (creating a random group assignment for each sample) hyperactivation of ERK1/2–MAPK signaling and investigate their and performing the Student t test for between-group differences involvement in facilitating gene expression, adverse clinical fea- for each gene; this was repeated 1,000 times, and the rank of the tures, poor clinical outcome, and poor response to hormone P value for each individual gene from the nonpermuted analysis therapy. Analysis of gene and protein expression, molecular was divided by 1,001, resulting in the permutation-adjusted alterations, and pathway enrichment of gene targets of these P value (e.g., if a nonadjusted P ¼ 0.022 ranked 34th lowest, the hMAPK–miRNAs identifies miRNAs that may potentially act as adjusted P value would be 34/1,001 ¼ 0.0339; ref. 41). This important regulators of MAPK signaling and downstream path- procedure is effectively a Westfall–Young correction for multiple ways, and which may serve as novel biomarkers or therapeutic comparisons (42). Only miRNAs that were commonly differen- targets in the treatment of breast cancer. Multivariate analysis tially expressed between hMAPK–mRNA tumors and not- indicates that discrete signatures made up of subsets of these hMAPK–mRNA tumors from both the Buffa and TCGA training hMAPK–miRNAs are independently associated with increased datasets comprised the total hMAPK–miRNA signature. risk of disease-specific death and disease recurrence, particularly þ þ Correlation to ideal hMAPK–miRNA signature was performed among patients with ER disease and in ER cancers treated with for all available primary breast cancers from both training and hormone therapy. validation datasets, as described for hMAPK–mRNA signature above. Differential mRNA expression between high-hMAPK– Materials and Methods miRNA and low-hMAPK–miRNA cancers from TCGA training Clinical datasets dataset, and likewise from METABRIC validation dataset, was The following publicly available breast cancer datasets were analyzed by the Student t test. used in this study: GSE22220 (Buffa dataset, accessed April 2012; miRNA target prediction was performed using the miRWalk ref. 31), GSE19536 (Enerly dataset, accessed April 2012; ref. 22), database (40), a database that uses 10 different miRNA target

OF2 Clin Cancer Res; 21(2) January 15, 2015 Clinical Cancer Research

hMAPK–miRNA Signature and Hormone Resistance

A TCGA Buffa C 0.8 0.8 0.7 0.7 0.6 Buffa all patients: disease recurrence 0.6 0.5

0.5 1.0 0.4 0.4 0.3

0.3 0.80.60.40.2 0.2 0.2 0.1 0.1 Proportion of subgroup 0 - 0 - ER+ ER ER+ ER B P = 0.214

0.9 1 % free Event n Events (n) Enerly METABRIC 0.8 107 29 0.7 0.8 100 20 0.6 0.6 0.0 0.5 0.4 0 2010 30 605040 0.4 0.3 Time to event (months) 0.2 0.2 0.1 Proportion of subgroup 0 0 ER+ ER- ER+ ER- High-hMAPK-mRNA

- - - - High hMAPK mRNA Low hMAPK mRNA Low-hMAPK-mRNA

D Enerly: all patients Enerly: all patients METABRIC: all patients Disease recurrence Disease survival Disease survival 1.0 1.00.80.60.40.20.0 1.00.80.60.40.20.0 0.8 0.60.40.2

P = 0.12 P = 0.156 P = 1.36e−06

n Events (n) n Events (n) n Events (n) %Event free %Event 51 16 56 13 605 116 43 9 42 6 681 68 0.0

6050403020100 20100 4030 50 60 0 2010 30 40 6050 Time to event (months)

Figure 1. Breakdown of ER status of primary breast cancers with paired mRNA and miRNA expression data from the two training and two validation datasets classified according to the hMAPK–mRNA signature established by Creighton et al. A, association of hMAPK–mRNA status with ER status in the TCGA and Buffa training datasets. B, association of hMAPK–mRNA status with ER status in the Enerly and METABRIC validation datasets. C and D, the Kaplan–Meier analysis of clinical outcome in patients with tumors classified according to hMAPK–mRNA signature. C, recurrence-free survival in Buffa dataset. D, recurrence-free survival (left), disease-specific survival in Enerly dataset (middle), and disease-specific survival in METABRIC dataset (right). Bar graphs: white, not hMAPK-mRNA; gray, hMAPK–mRNA; the Kaplan–Meier curves: dashed, low-hMAPK–mRNA; solid, high-hMAPK–mRNA. P values from the log-rank tests are indicated. prediction programs to identify putative miRNA targets. Genes relationships that may differ between cancers classified as high- were considered putative targets of hMAPK–miRNAs if they were hMAPK–miRNA and those classified as low-hMAPK–miRNA. predicted to be a target by three or more of the following Standard statistical tests were performed, as described in Supple- prediction programs: DIANA-mt, miRanda, miRDB, miRWalk, mentary Materials and Methods. All statistical analyses were RNAhybrid, PICTAR4, PICTAR5, PITA, RNA22, or TargetScan. performed using R statistical software v 2.15.2 (43). Microsoft Pearson correlation of hMAPK–miRNA target expression to Excel 2007 and R statistical software v 2.15.2 were used for figure hMAPK–miRNA expression was performed to identify regulatory generation. All t tests performed were two-sided.

www.aacrjournals.org Clin Cancer Res; 21(2) January 15, 2015 OF3

Miller et al.

A High-hMAPK–miRNA B High-hMAPK–miRNA AMP 0.5 UP 0.7 UP DOWN DOWN 0.4 0.6 HOMDEL RPPA-overexpressed 0.5 0.3 MUT 0.4 RPPA-underexpressed 0.2 0.3 0.2 0.1 0.1 0.0 0.0 0.5 0.7 Low-hMAPK–miRNA 0.4 0.6 Low-hMAPK–miRNA 0.5 0.3 0.4 0.2 0.3 Proportion of samples with alterations 0.1 Proportion of samples with alterations 0.2 0.1 0.0 0.0 SRF ATF4 ETV1 ETV5 ELK4 ESR1 TOB1 RKIP RAF1 BRAF SOS1 CREB1 KRAS GRB2 PDCD4 NRAS EGFR ERBB2 ERBB3 MAPK1 MAPK3 CDKN1B RPS6KA1 RPS6KA5 RPS6KA4 MAP2K1 MAP2K2 ANGPLT4

C Targets of hMAPK-upregulated miRNAs 20 − 2 − 4 ESR1 CDKN1B GATA3 SMAD3 P = 9.44e-24 P = 0.000934 P = 2.15e-19 P = 3.95e-07 High-hMAPK-microRNA - Standardized gene expression t = 11.8 t = -3.34 t = -10.2 t = -5.21

− 6 Low-hMAPK-microRNA D Targets of hMAPK-downregulated miRNAs ER positive 420 ER negative

ER status unknown − 2 CCNB1 SNAI1 CDH3 NOTCH1 P = 2.1e-07 P = 6.33e-14 P = 3.33e-13 P = 8.02e-13

Standardized gene expression t = 5.33 t = 7.93 t = 7.68 t = 7.58 − 4

Figure 2. A, expression of genes downregulated (ESR1, CDKN1B, TOB1, PDCD4, and RPS6KA5) and upregulated (ETV5, RPS6KA4, CREB1, SRF, ANGPLT4, ETV1, ELK4, ATF4, and RPS6KA1) [underlined and bolded genes reported to be downregulated or upregulated in hMAPK mRNA signature (15)] by activated MAPK signaling was examined in primary tumors from the TCGA dataset that were classiﬁed as high-hMAPK–miRNA (top) or low-hMAPK–miRNA (bottom) by our hMAPK–miRNA signature. Proportions of tumors with a given alteration 1.5-fold are reported. Blue, gene expression downregulation; and green, gene expression upregulation. Alterations in copy number, mutational status, gene expression, and protein expression of numerous genes in the EGFR–ERK signaling pathway found in cancers form the TCGA dataset classiﬁed by our hMAPK–miRNA signature. (Continued on the following page.)

OF4 Clin Cancer Res; 21(2) January 15, 2015 Clinical Cancer Research

hMAPK–miRNA Signature and Hormone Resistance

To generate an hMAPK–miRNA recurrence signature, univari- set: "METABRIC" (validation cohort); datasets previously ate analyses were first performed on the total hMAPK–miRNA described (22, 31, 32, 34, 35); clinical characteristics in Sup- signature to determine whether expression of individual miRNAs plementary Table S1]. Tumors in each of these datasets classi- (median cutoff) was associated with increased rate of disease fied as "high-hMAPK–mRNA" according to our hMAPK–mRNA recurrence in the Buffa training dataset. Breast cancers from the signature (15) exhibited a significant association with ER-neg- Buffa dataset were then partitioned into two groups based on ativity in training and validation datasets (Fig. 1A and B). median expression of every hMAPK–miRNA on an individual hMAPK–mRNA classification associated with a trend toward basis. Using this grouping as a classifier, the Kaplan–Meier sur- increased incidence of disease recurrence in the training dataset vival curves were generated for disease recurrence, and the log- (Fig. 1C), and increased incidence of disease recurrence and rank test P values were determined; 22 miRNAs for which this log- poorer disease-specific survival in the validation datasets (Fig. rank test P value was 0.1 were included in the recurrence 1D). signature. Correlation to an ideal hMAPK-recurrence signature was determined as described above for mRNA signature analysis. Generation and characterization of a breast cancer To generate an hMAPK–miRNA survival signature, primary hMAPK–miRNA signature Breast Cancers from the METABRIC dataset were first randomized We identified 16 commonly underexpressed and 41 commonly into two populations. A leave-one-out analysis was then per- overexpressed miRNAs in tumors classified as hMAPK–mRNA formed, sequentially removing single individual miRNAs from from both training cohorts (P 0.05; see Supplementary Fig. S1 the hMAPK–miRNA signature, and determining whether removal for study overview), establishing a 57-member hMAPK–miRNA of each miRNA improved the ability of the signature to predict signature (Supplementary Table S2). Primary breast cancers from þ poor survival events in ER population at 5 years by the Cox training and validation datasets were classified as "high-hMAPK– proportional hazards analysis, separately for each randomized miRNA" or "low-hMAPK–miRNA" according to this miRNA þ population. The analysis was limited to the ER population signature. To confirm that this hMAPK–miRNA signature identi- because the overwhelming majority of the ER cancers were fies tumors with activated ERK1/2–MAPK signaling, we analyzed already classified as "high hMAPK." miRNAs whose removal the expression of several genes regulated by ERK1/2 in breast improved the model in both randomized populations from the cancer in the TCGA training cohort. Genes known to be down- METABRIC dataset were tabulated. This randomization and leave- regulated by ERK1/2 signaling (ESR1, CDKN1B, TOB1, and one-out analysis was repeated in 1,000 random permutations of PDCD4) are underexpressed in a higher proportion of high- the dataset to ensure robustness and serve as a self-validation. hMAPK–miRNA tumors and, similarly, genes upregulated by miRNAs whose removal improved the model in more than 20% ERK1/2 (ETV5, RPS6KA4, CREB1, SRF, ANGPLT4, ETV1, ELK4, of randomizations were removed from the signature, establishing ATF4, and RPS6KA1) are overexpressed in a higher proportion of a 21-member refined hMAPK–miRNA signature. Correlation to high-hMAPK–miRNA cancers compared with low-hMAPK– an ideal hMAPK–miRNA survival signature was determined as miRNA cancers (Fig. 2A). We also investigated mutational status, described above for mRNA signature analysis. alterations in DNA copy number, and expression of upstream regulators of ERK1/2 signaling (Fig. 2B). Alterations indicative of miRNA expression array and qRT-PCR analysis activation of ERK1/2–MAPK signaling pathways, such as miRNA expression array and qRT-PCR analysis of cell lines were enhanced expression of EGFR, ERBB2, SOS1, NRAS, RAF1, and performed according to standard practices, and are described in MAPK1 and lower expression of ERBB3 and RKIP, are seen in a Supplementary Materials and Methods. higher proportion of high-hMAPK–miRNA cancers compared with low-hMAPK–miRNA cancers. Gene set enrichment analysis of predicted targets of hMAPK–miRNAs identifies substantial Results enrichment for genes involved in the MAPK signaling pathway hMAPK–mRNA signature associates with adverse features of (Supplementary Fig. S2A and S2B), suggesting that a number of breast cancer and poor clinical outcome these hMAPK–miRNAs may not only be regulated by MAPK We previously observed that the hMAPK–mRNA signature is signaling, but also contribute to sustaining hyperactivation of present in a majority of ER breast cancers and a small subset of ERK1/2–MAPK signaling. Analysis of protein expression data þ ER breast cancers (13, 14). We confirmed that the hMAPK– from the TCGA dataset available for breast cancers with miRNA mRNA signature was similarly represented in the primary breast expression data revealed significant differences in expression and tumor datasets with paired mRNA–miRNA expression data phosphorylation of numerous proteins between high-hMAPK– analyzed in this study [TCGA: "TCGA" (training cohort); GEO miRNA and low-hMAPK–miRNA tumors, including proteins GSE22220: "Buffa dataset" (training cohort); GEO GSE19536: involved in or regulated by ERK1/2–MAPK signaling, as well as "Enerly dataset" (validation cohort); METABRIC miRNA data- epithelial–mesenchymal transition (EMT) proteins, protein

(Continued.) B, proportion of tumors classified as high-hMAPK–miRNA with alterations for a given gene (top); proportion of tumors classified as low-hMAPK– miRNA with alterations for a given gene (bottom). Red, copy number amplification; light blue, homozygous deletion; green, gene expression upregulation; dark blue, gene expression downregulation; pink, protein expression increased (RPPA); and gray, protein expression decreased (RPPA). Gene and protein expression increases considered significant if expression z-score > 1.5-fold different from dataset average. Gene set enrichment analysis for predicted and validated gene targets of miRNAs in the hMAPK–miRNA signature reveals significant enrichment of genes that fall into numerous pathways associated with proliferation, signal transduction, survival, differentiation, and cancer. C, expression of genes that are predicted targets of miRNAs overexpressed in the hMAPK–miRNA signature validated to be upregulated by MAPK signaling. D, expression of genes that are predicted targets of miRNAs underexpressed in the hMAPK- miRNA signature validated to be downregulated by MAPK signaling. Yellow boxplot, high-hMAPK–miRNA cancers; blue boxplot, low-hMAPK– þ miRNA cancers; red circle, ER sample; green circle, ER sample; and gray circle, ER status for sample not available.

www.aacrjournals.org Clin Cancer Res; 21(2) January 15, 2015 OF5

Miller et al.

markers of basal-like breast cancer, and proteins involved in Enerly, 0.02115; METABRIC, <2.2e16; Supplementary Fig. þ mediating tamoxifen resistance (Supplementary Table S3). S7D), and basal and HER2 PAM50 subtypes (P values: Enerly, To study the impact that altering MAPK signaling has on 9.621e16; METABRIC, <2.2e16; Fig. 3A, middle and bottom). miRNA expression in established models of hyperactive MAPK Survival analyses indicate that, among all patients, cancers clas- signaling in breast cancer, we altered MAPK signaling in our stable sified as high-hMAPK–miRNA demonstrated significantly hMAPK cell lines (15) and investigated effects on select miRNAs increased disease recurrence (P value: Enerly, 0.0119; Fig. 3C, from the hMAPK–miRNA signature that had been previously top) and decreased disease-specific survival (P values: Enerly, implicated in breast tumor biology, or that have predicted targets 0.00194; METABRIC, 6.28e06; Fig. 3C, top) at 5 years. Strati- in genes validated by Creighton and colleagues (15) to be regu- fying these analyses by ER status indicates that classification as þ lated by MAPK signaling. In particular, we treated ca-RAF/MCF-7 high-hMAPK–miRNA identifies a population of ER cancers with (which exhibits constitutive activation of MAPK signaling) with a significantly increased disease recurrence (Enerly, P ¼ 0.0059; Fig. MEK inhibitor to abrogate MAPK signaling, and in a complemen- 3C, bottom) and significantly poorer disease-specific survival tary approach, we stimulated MAPK signaling in EGFR/MCF7 cell (Enerly, P ¼ 0.0191; METABRIC, P ¼ 0.0818; Fig. 3C, bottom). line (which overexpresses EGFR and is a ligand-inducible model for hyperactivation of MAPK signaling). Several hMAPK–miRNAs hMAPK–miRNAs associated with recurrence identify an were altered in their expression by abrogating or inducing hMAPK hMAPK–miRNA recurrence signature (Supplementary Fig. S3A and S3B): overexpressed hMAPK–mi- By limiting the signature to miRNAs whose individual expres- RNAs significantly decreased following abrogation of MAPK sion was significantly associated with reduced recurrence-free signaling with U0126 treatment (hsa-miR221 and hsa-miR222) survival on univariate analysis among patients from the Buffa or increased following EGF treatment (hsa-miR378). Likewise, training dataset, we identified a 22-member hMAPK–miRNA underexpressed hMAPK–miRNAs were increased following recurrence signature (Supplementary Table S4). Cancers from the abrogation of MAPK signaling with U0126 treatment (hsa- Buffa dataset classified as high-hMAPK–miRNA by this recurrence let-7a, hsa-miR30a, hsa-miR30a , hsa-miR125a-5p, and hsa- signature demonstrated significantly reduced recurrence-free sur- miR375) or decreased following EGF treatment (hsa-let-7a, vival at 5 years in all patients (P ¼ 5.52e06; Supplementary Fig. þ þ hsa-let-7e, hsa-miR29c, and hsa-miR30c), indicating that these S8) and in ER - and ER -specific cohorts (ER P ¼ 0.00117; ER P miRNAs are regulated by MAPK signaling. We observed that ¼ 0.0177; Supplementary Fig. S8). In the Enerly validation target genes of these validated hMAPK–miRNAs are differen- dataset, classification by this recurrence signature significantly tially expressed between tumors classified as high-hMAPK– associated with reduced recurrence-free survival among all þ miRNA compared with those classified as low-hMAPK–miRNA patients and patients with ER disease (P ¼ 0.015 and 0.0523, in the TCGA training dataset (Fig. 2C and D), and verified these respectively; Fig. 4A). In the Enerly and METABRIC datasets, high- relationships in the METABRIC validation dataset (Supplemen- hMAPK–miRNA status also significantly associated with shorter tary Fig. S4A–S4F). In addition, we observed differential pro- disease-specific survival in all patients (P values: Enerly, tein expression of targets of hMAPK–miRNAs in breast cancers 0.00121; Fig. 4B; METABRIC, 4.49e10; Fig. 4C) and patients þ from the TCGA dataset classified as "high-hMAPK" versus "low- with ER disease (P values: Enerly, 0.0672; Fig. 4B; METABRIC, hMAPK" by our miRNA signature, and inverse relationships 0.00144; Fig. 4C). þ between expression of miRNAs and protein expression of Cancers from the Lyng ER dataset (GSE37405; ref. 33) clas- selected gene targets in the TCGA dataset, suggesting that sified as high-hMAPK–miRNA by this recurrence signature dem- differential expression of these miRNAs may have a regulatory onstrated significantly earlier disease recurrence among breast impact on the protein expression of their predicted gene targets cancers treated with adjuvant tamoxifen monotherapy (P ¼ þ (Supplementary Fig. S5A–S5H). 0.00348; Fig. 4D). ER cancers from the METABRIC dataset receiving hormone therapy, either alone or in conjunction with hMAPK–miRNA signature is associated with adverse clinical chemotherapy or radiotherapy, which were classified as high- features of breast cancer hMAPK by the hMAPK–miRNA recurrence signature demonstrat- Cancers classified as high-hMAPK–miRNA in the TCGA and ed significantly poorer disease-specific survival (P ¼ 0.00301; Fig. Buffa training data were enriched for ER status (TCGA, P ¼ 4D). 4.308e16; Buffa, P ¼ 8.508e14; Supplementary Fig. S6A), high tumor grade (Buffa, P ¼ 7.711e11; Supplementary Fig. S6B), Multivariate analysis of hMAPK–miRNA recurrence signature þ and basal-like or HER2 PAM50 molecular subtypes (TCGA, P < Multivariate analysis of the hMAPK–miRNA recurrence signa- 2.2e16; Fig. 3A, top). High-hMAPK–miRNA cancers also dis- ture in the METABRIC dataset using the Cox proportional hazards played significantly earlier disease recurrence among all patients analyses revealed that among all patients, high-hMAPK–miRNA (Buffa, P ¼ 0.00242; Fig. 3B); segregation of patients by cancer ER status is associated with a significant increase in risk of breast status revealed significant trends for earlier disease recurrence cancer-specific death within 5 years from diagnosis [multi- þ among patients with ER disease (P ¼ 0.0835) and ER disease plicative hazard factor, 1.48; 95% confidence interval (CI), (P ¼ 0.167; Fig. 3B). 1.002–2.175; P ¼ 0.04886; Supplementary Table S5A]. Because In the Enerly and METABRIC validation datasets, the hMAPK– we observed that classification as high-hMAPK–miRNA by the miRNA signature was significantly associated with cancers that are hMAPK–miRNA recurrence signature was significantly associated þ þ ER (P values: Enerly, 2.728e08; METABRIC, <2.2e16; Sup- with poor outcome in ER breast cancers, and ER breast cancers plementary Fig. S7A), higher grade (P values: Enerly, 1.198e09; displayed a large variation in correlation values with this recur- METABRIC, <2.2e-16; Supplementary Fig. S7B), classified by a rence signature ranging from very positive to very negative (in high proliferation metric [described by Enerly and colleagues contrast to ER breast cancers, whose correlations were almost þ (22); P ¼ 2.218e05; Supplementary Fig. S7C], HER2 (P values: entirely positive), we sought to determine whether an incremental

OF6 Clin Cancer Res; 21(2) January 15, 2015 Clinical Cancer Research

hMAPK–miRNA Signature and Hormone Resistance

A 100% Fisher exact test 50% P < 2.2e-16 TCGA 0% High-hMAPK-miRNA LumA LumB HER2 Basal Not classified 100% Low-hMAPK-miRNA

50% Fisher exact test

Enerly P = 9.621e-16 0% LumA LumB HER2 Basal Normal like Not classified Figure 3. 100% Molecular subtyping and the Kaplan– Proportion of subgroup Meier survival analysis of disease 50% Fisher exact test recurrence and survival in patients P < 2.2e-16 with tumors classiﬁed according to METABRIC 0% LumA LumB HER2 Basal Normal like Not classified hMAPK–miRNA signature from + - training (Buffa and TCGA) and Buffa all patients: Buffa ER : Buffa ER : B disease recurrence disease recurrence disease recurrence validation datasets (Enerly and 1.0 1.0 METABRIC). A, association of hMAPK– 1.00.80.60.40.20.0 miRNA status with PAM50 molecular subtypes in TCGA (top), Enerly 0.80.60.4 0.80.60.4 (middle), and METABRIC (bottom) – datasets. B, the Kaplan Meier survival P = 0.00242 P = 0.0835 P = 0.167 analysis of disease recurrence in n Events (n) n Events (n) n Events (n) patients with tumors classiﬁed % Event free 86 29 27 8 59 21 – 0.2 0.2 according to the hMAPK miRNA 121 20 100 16 21 4 signature from training (Buffa 0.0 0.0 dataset); disease recurrence among all 6050403020100 6050403020100 0 10 20 30 605040 patients, patients with ERþ disease, Time to event (months) patients with ER disease (Buffa). C, disease recurrence and disease C Enerly all patients: Enerly all patients: disease METABRIC all patients: survival among all patients (Enerly) disease recurrence survival survival 1.00.80.60.40.2 1.00.80.60.4 and disease survival among all 1.00.80.60.4 patients from METABRIC dataset (top), disease recurrence and disease survival among patients with ERþ disease (Enerly) and disease survival P = 0.0119 P = 0.00194 P = 5.15e-08 among patients from METABRIC þ n Events (n) n Events (n) n Events (n) dataset with ER disease (bottom). 0.2 0.2

% Event free 40 15 44 14 565 117 Bar graphs: white, low-hMAPK– 54 10 54 5 701 67 miRNA; gray, high-hMAPK–miRNA; P 0.0 0.0 x2 0.0 values given are for the test or the 6050403020100 6050403020100 100 20 60504030 Fisher exact test, as indicated. The Enerly ER+ cancers: Enerly ER+ cancers: METABRIC ER+ cancers: Kaplan–Meier curves: dashed, low- Disease recurrence Disease survival Disease survival – – 1.0 1.0

hMAPK miRNA; solid, high-hMAPK 1.0 miRNA; the log-rank test P values are 0.8 0.8 indicated. 0.8 0.6 0.6 0.60.40.2 P = 0.0059 P = 0.0191 P = 0.0818 0.40.20.0 0.4

% Event free n Events (n) n Events (n) n Events (n)

13 6 0.20.0 14 4 300 40 46 7 47 4 692 66 0.0 6050403020100 50403020100 60 0 10 6050403020 Time to event (months) High-hMAPK-miRNA Low-hMAPK-miRNA

change in correlation with the hMAPK–miRNA recurrence signa- Similarly, incremental increased Pearson correlation with the ture (represented as 1% increase in Pearson correlation coeffi- ideal hMAPK–miRNA recurrence signature is also associated cient) associated with change in risk of disease-specific death. In with a significant incremental increase in risk of breast cancer- þ the ER METABRIC cohort, multivariate analysis indicated that specific death within 5 years from diagnosis among patients with þ incremental increase in Pearson correlation with the ideal ER cancer receiving hormone therapy, either alone or in con- hMAPK–miRNA recurrence signature carries a significant corre- junction with chemotherapy or radiotherapy (multiplicative spondingly incremental increase in risk of breast cancer-specific hazard factor, 1.008; 95% CI, 1.002–1.014; P ¼ 0.006285; death within 5 years from diagnosis (multiplicative hazard factor, Supplementary Table S5C). Multivariate analysis of the Lyng þ 1.01; 95% CI, 1.001–1.011; P ¼ 0.0185; Supplementary Table ER cohort indicated a significant association with high- S5B). hMAPK–miRNA status according to the hMAPK–miRNA

www.aacrjournals.org Clin Cancer Res; 21(2) January 15, 2015 OF7

Miller et al.

A Enerly: all patients Enerly: ER+ cancers Disease recurrence Disease recurrence High-hMAPK-miRNA

Low-hMAPK-miRNA

P = 0.015 P = 0.0523 n Events (n) n Events (n) % Event free 44 15 16 6 50 9 43 7 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 6050403020100 6050403020100 Time to event (months)

B Enerly: all patients Enerly: ER+ cancers Disease survival Disease survival

Figure 4. Five-year outcome analysis of tumors from validation datasets (Enerly, METABRIC, and Lyng) classiﬁed P = 0.00121 P = 0.0672 according to the hMAPK–miRNA – n Events (n) n Events (n) recurrence signature. A, the Kaplan % Event free 47 15 16 4 Meier survival analysis of disease 45 4 recurrence among all patients and 51 4 þ 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 patients with ER disease (Enerly 6050403020100 6050403020100 dataset). B, disease survival among all þ Time to event (months) patients and patients with ER disease (Enerly dataset). C, disease survival METABRIC: all patients METABRIC: ER+ cancers among all patients and patients with C þ Survival Survival ER disease (METABRIC dataset). D, disease recurrence among patients from Lyng dataset treated with tamoxifen monotherapy and patients þ from METABRIC dataset with ER disease who received any hormone P = 4.49e-10 P = 0.00144 therapy. The Kaplan–Meier curves: n Events (n) n Events (n) dashed, low-hMAPK–miRNA; solid, % Event free 586 122 307 47 high-hMAPK–miRNA. The log-rank 700 62 685 59 test P values are indicated. 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0 102030405060 6050403020100 Time to event (months) METABRIC: ER+ cancers with any hormone therapy D Lyng: recurrence Survival

P = 0.00348 P = 0.00301 n Events (n) n Events (n) 73 38 % Event free 220 38 77 20 480 45 0.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 6050403020100 0 102030405060 Time to event (months)

recurrence signature and increased risk of disease recurrence (multiplicative hazard factor, 1.01; 95% CI, 1.0006–1.011; P ¼ within 5 years (multiplicative hazard factor, 2.109; 95% CI, 0.029455; Supplementary Table S6A). 1.1992–3.7085; P ¼ 0.009583; Supplementary Table S5D). In addition, multivariate analysis indicated that increased Pearson Leave-one-out analysis of hMAPK–miRNA signature identifies a correlation with the ideal hMAPK–miRNA recurrence signature subset associated with poor disease-specific survival outcomes þ associates with a significant increase in risk of breast cancer- in ER population þ specific death within 5 years from diagnosis among ER patients We hypothesized that miRNAs from the hMAPK–miRNA sig- from METABRIC dataset stratified by adjuvant therapy received nature may contribute to de novo or acquired resistance to

OF8 Clin Cancer Res; 21(2) January 15, 2015 Clinical Cancer Research

hMAPK–miRNA Signature and Hormone Resistance

hormone therapy, and may be associated with poor survival that associates significantly with the ERK1/2 hMAPK–mRNA þ outcomes in patients with ER disease. To test this hypothesis, expression signature. Classification of primary breast cancers we performed a leave-one-out analysis in the large METABRIC using this hMAPK–miRNA signature indicates significant associ- dataset. Briefly, miRNAs were individually excluded from analysis ation with ER status, high tumor grade, increased proliferation, of the 57-member hMAPK–miRNA signature, and the Cox pro- basal and HER2 molecular subtypes, and poor clinical outcomes. þ portional hazards analysis for hazard of disease-specific death This signature identifies a population of ER primary breast þ event within 5 years among patients with ER disease from the cancers classified as high-hMAPK–miRNA that not only exhibit METABRIC dataset was performed (see Materials and Methods for gene expression patterns similar to ER primary breast cancers, þ additional details). Individual miRNAs whose removal improved but also exhibit significantly poorer clinical outcomes than ER the ability of the hMAPK–miRNA signature to predict poor cancers classified as low-hMAPK–miRNA. These observations þ survival events in the ER population at 5 years by the Cox were validated in independent, publicly available cohorts (Enerly proportional hazards analysis were tabulated (see Materials and and METABRIC). We report a subsignature composed of 22 Methods), and the 21 retained miRNAs make up a "hMAPK– miRNAs, an hMAPK–miRNA recurrence signature, which retains miRNA survival signature" (see Supplementary Table S7). As significant associations with poor clinical outcome among expected, this survival signature has substantially improved the patients with breast cancer and significantly associates with poor ability to predict poor disease survival in the Kaplan–Meier response to hormone therapy in two independent validation survival analysis and the Cox univariate hazard analysis of datasets by multivariate analysis (Lyng and METABRIC datasets). patients in the METABRIC dataset, in both total patient popula- Finally, we report a subsignature of this hMAPK–miRNA signa- tion (P ¼ 4.12e17; univariate HR, 3.7 [2.67–5.13]) and in the ture, consisting of 21 miRNAs, which is significantly associated þ þ ER patient population (P ¼ 1.44e08; univariate HR, 2.87 with poorer disease-specific survival outcome in patients with ER [1.96–4.21]; Fig. 5A) compared with either the total hMAPK– disease; an hMAPK–miRNA survival signature. miRNA signature or the hMAPK–miRNA recurrence signature. The hMAPK–miRNA signature (Supplementary Table S2) con- Breast cancers of the luminal A and luminal B subtypes that were tains numerous miRNAs with previously reported roles in breast classified as "high-hMAPK" by this hMAPK–miRNA survival cancer etiology. Several hMAPK–miRNAs are significantly differ- þ signature exhibited significantly poorer 5-year survival compared entially expressed between basal/HER2 subtype and luminal A þ with those classified as "low-hMAPK" (luminal A: P ¼ 0.0151; type cancers, between ER and ER , and between high-grade univariate HR, 2.28 [1.15–4.52]; luminal B: P ¼ 0.000614; uni- versus low-grade cancers (overexpressed: hsa-miR150, -142-3p, variate HR, 2.48 [1.45–4.25]; Fig. 5A). Breast cancers treated with -142-5p, -148a, -155, and -135b; underexpressed: hsa-miR30a-3-, any hormone therapy that were classified as "high-hMAPK" by this -30a-5p, let-7a, and -342; ref. 35). miR29c, underexpressed in the survival signature were at significantly higher risk for poorer hMAPK–miRNA signature, correlates with expression of GATA3, a disease survival than those classified as "low-hMAPK" (P ¼ luminal identity promoting transcription factor, and the miR29 1.7e06; univariate HR, 2.74 [1.78–4.23]; Fig. 5B). Multivariate miRNA family facilitates GATA3 mediated maintenance of lumi- þ analysis of all breast cancers and ER breast cancers in the nal identity (44). The let-7 family of miRNAs, underexpressed in METABRIC dataset indicated that high-hMAPK status, as deter- the hMAPK–miRNA signature, negatively regulates RAS (45), and mined by our hMAPK–miRNA survival signature, is an indepen- represses self-renewal and tumorigenicity of breast cancer stem dent risk factor for reduced disease survival at 5 years among all cells (30). The EMT-promoting transcription factor SNAI1 has patients (P ¼ 0.0031; multiplicative hazard factor, 1.89 (1.24– been shown to be a direct target of miR30a (46). Overexpression þ 2.89) and especially among patients with ER disease (P ¼ 0.0027; of the hMAPK-downregulated miRNAs miR29c, miR30c, and multiplicative hazard factor, 1.95 (1.26–3.02; Tables 1 and 2). miR342 has been reported to significantly associate with good þ To validate this hMAPK–miRNA ER survival signature, we prognosis (35), and we observe significantly lower expression of þ assessed its association to disease survival in all patients and GATA3 as well as miR29c in ER tumors classified as hMAPK– þ patients with ER disease in the Enerly dataset (P ¼ 0.00689 and miRNA, and higher expression of basal markers and genes asso- 0.0538, respectively; Fig. 5C). We additionally assessed associa- ciated with poor clinical outcome, suggesting poor outcomes tion of this survival signature with disease recurrence in all associated with these tumors may be due to loss of luminal þ patients and patients with ER disease in the Buffa (P ¼ 1e07 identity and acquisition of ER characteristics. A number of and 0.000245, respectively; Fig. 5D) and Enerly datasets (P ¼ miRNAs upregulated in the hMAPK–miRNA signature þ 0.00393 and 0.0167, respectively; Fig. 5D), and in ER patients (miR142-5p/3p, miR146a, miR150, and miR155; ref. 35) are treated with tamoxifen monotherapy from the Lyng dataset (P ¼ associated with markers of lymphocytic infiltrate and may reflect 0.0371; Fig. 5B). This survival signature was vastly superior to the altered host immune or inflammatory responses associated with total hMAPK–miRNA signature, and performed comparably with this hMAPK–miRNA signature. Thus, the downregulation and or better than the hMAPK-recurrence signature in the ability to upregulation of these family members, respectively, in the predict poor outcome in patients classified as "high-hMAPK" hMAPK- miRNA signature may contribute to the breast cancer versus "low-hMAPK," even in the smaller publicly available aggressiveness predicted by this profile. datasets. This strongly suggests that the miRNAs retained in the A number of hMAPK–miRNAs regulate the ER (28–30, 47, 48). survival signature may drive the poor survival outcome associated The hMAPK-upregulated miR221/222 family is overexpressed in with the hMAPK–miRNA signature. ER cancers (49), directly targets ER (49), is upregulated in HER2- amplified versus nonamplified breast cancers (29), and mediates Discussion tamoxifen resistance in MCF-7 cells by repression of CDKN1B, a cell-cycle regulator that is underexpressed in breast cancers clas- Using paired mRNA and miRNA expression data from primary sified as high-hMAPK by our miRNA signature (see Fig. 2; refs. 28, breast tumor datasets (TCGA, Buffa), we report a miRNA signature 29). Six underexpressed miRNAs in the hMAPK- miRNA signature

www.aacrjournals.org Clin Cancer Res; 21(2) January 15, 2015 OF9

Miller et al.

A METABRIC: all patients METABRIC: ER+ B METABRIC: Any hormone disease survival disease survival therapy—disease survival

HR = 3.7 (2.67–5.13) HR = 2.87 (1.96–4.21) HR = 2.74 (1.78–4.23) P = 4.12e–17 P = 1.44e–08 P = 1.79e–06 % Event free n n Events (n) Events (n) % Event free n Events (n)

6050403020100 6050403020100 0 102030405060 Time to event (months) Time to event (months) METABRIC: Luminal A METABRIC: Luminal B disease survival disease survival Lyng: 5-year recurrence

HR = 2.28 (1.15–4.52) HR = 2.48 (1.45–4.25) P = 0.0151 P = 0.000614 P = 0.0371 % Event free % Event free n Events (n) n Events (n) n Events (n)

6050403020100 0 102030405060 0 605040302010 Time to event (months) Time to event (months)

C Enerly: All patients D Enerly: All patients Buffa: All patients disease disease survival disease recurrence recurrence

P = 0.000689 P = 0.00393 P = 1e–07 % Event free % Event free n Events (n) n Events (n) % Event free n Events (n)

6050403020100 6050403020100 0 605040302010 Time to event (months) Time to event (months) Time to event (months)

Enerly: ER+ disease Enerly: ER+ disease Buffa: ER+ patients disease survival recurrence recurrence

P = 0.0538 P = 0.0167 P = 0.000245 % Event free

n % Event free Events (n) n Events (n) % Event free n Events (n)

6050403020100 6050403020100 0 605040302010 Time to event (months) Time to event (months) Time to event (months)

Figure 5. Five-year outcome analysis of tumors from training (Buffa) and validation datasets (Enerly, METABRIC, and Lyng) classiﬁed according to the hMAPK–miRNA survival signature. A, the Kaplan–Meier survival analysis of disease recurrence in patients with tumors classiﬁed according to hMAPK–miRNA survival signature þ from METABRIC dataset; disease survival among all patients, patients with ER disease (top), disease survival among patients with luminal A and luminal B type breast cancer (bottom). (Continued on the following page.)

OF10 Clin Cancer Res; 21(2) January 15, 2015 Clinical Cancer Research

hMAPK–miRNA Signature and Hormone Resistance

Table 1. Multivariate analysis of hMAPK–miRNA survival signature in METABRIC Table 2. Multivariate analysis of hMAPK–miRNA survival signature in dataset—Cox proportional hazards analysis of all patients METABRIC dataset—Cox proportional hazards analysis of METABRIC ERþ breast All patients: hazard of 5-year death event cancers only, including otherwise identical covariates Covariate HR (95% CI) ERþ cancers: hazard of 5-year death event Positive LN (3þ) 2.97 (1.89–4.69)a Covariate HR (95% CI) Positive LN(2) 2.43 (1.37–4.31)b Positive LN (3þ) 2.81 (1.62–4.86)a Basal subtype 2.03 (0.95–4.31)d Positive LN(2) 2.16 (1.05–4.45)c High-hMAPK 1.89 (1.24–2.89)b High-hMAPK 1.95 (1.26–3.02)b ERBB2-positive 1.79 (1.18–2.73)b Luminal B 1.64 (1.04–2.56)c Luminal B 1.62 (1.04–2.52)c PR-positive 0.59 (0.39–0.87)b c PR-positive 0.67 (0.46–0.99) NOTE: Only covariates with significant hazard ratios are indicated. NOTE: Cancers from METABRIC dataset were classified as high-hMAPK–miRNA Significance codes: 0 "a"; 0.001 "b"; 0.01 "c"; 0.05 "d". or low-hMAPK–miRNA by hMAPK–miRNA survival signature, and this hMAPK– Baseline: low hMAPK, grade 1, stage 0, 0 positive LN, ERBB2-negative, PR- miRNA classification was included in the Cox proportional hazards analysis negative, luminal A. along with standard clinical covariates including tumor grade, stage, HER2 status, progesterone receptor status, lymph node status, and tumor PAM50 molecular subtype. ment of estrogen independence and tamoxifen insensitivity and a b c d Significance codes: 0 " "; 0.001 " "; 0.01 " "; 0.05 " " that these hMAPK–miRNAs may mediate these effects. The Baseline: low hMAPK, grade 1, stage 0, 0 positive LN, ERBB2-negative, PR- þ hMAPK–miRNA recurrence signature and the hMAPK–miRNA negative, ER , luminal A. survival signature identified here significantly associate with increased early risk of recurrence and poor disease survival in þ (hsa-miR125a, -let7e, -30d, -30a-5p, -30a-3p, and -149) are ER patient populations, supporting the notion that these miR- þ underexpressed in treatment-na€ve ER primary cancers that NAs contribute to an ERK1/2–MAPK mechanism of tamoxifen expressed lower ESR1 mRNA levels, higher ERBB2 mRNA levels, resistance. and which had higher proportions of basal and HER2 subtype Multivariate analysis of the breast cancers from METABRIC þ tumors than ER tumors with higher expression of these miRNAs and Lyng datasets indicates that classification as high-hMAPK– (50). Elevated expression of miR30 family members (hsa- miRNA by the hMAPK–miRNA recurrence signature significant- miR30a, -3p, and -30c) is significantly associated with tamoxifen ly contributes to increased risk of breast cancer-specificdeath sensitivity (51), whereas repression of miR375, underexpressed in and disease recurrence within 5 years after diagnosis. Although the hMAPK- miRNA signature, is associated with resistance to we report that the majority of ER breast cancers fall under the þ tamoxifen in an ER breast cancer model (52). Analysis of the classification of high-hMAPK–miRNA, we did not identify ER þ Lyng dataset containing only ER tumors treated with adjuvant status as a significant covariate in our multivariate analysis, tamoxifen monotherapy and the hormone therapy–treated indicating that increased risk associated with hMAPK–miRNA cohort of the METABRIC dataset demonstrates that this status is independent of ER status. We observed a significant þ hMAPK–miRNA signature identifies a subset of ER tumors that increase in hazard associated with incremental increase in exhibit a higher incidence of recurrence and decreased disease- correlation to the idealized hMAPK–miRNA recurrence signa- þ specific death following hormone therapy, suggesting an associ- ture in ER breast cancers. Importantly, the hMAPK–miRNA ation of this hMAPK–miRNA signature with tamoxifen resistance survival signature demonstrated prognostic capability in both and agreeing with previous observations linking activated MAPK luminal A and luminal B subtypes of breast cancer from the signaling and resistance to endocrine therapy (53). The Lyng METABRIC dataset. This indicates that hMAPK–miRNA survival þ dataset consists of only ER cancers treated with adjuvant tamox- signature is not just a surrogate for luminal B breast cancers þ ifen monotherapy, selected such that half of the patients exhibited from the ER group, which often exhibit higher proliferation disease recurrence over the course of 10-year follow-up (33). Lyng rates, more activated growth factor signaling, and less reliance and colleagues (33) were not able to identify a miRNA signature upon estrogen signaling than luminal A breast cancers. In þ þ which consistently predicted disease recurrence among ER breast addition, multivariate analysis of ER breast cancers from the tumors, and suggested that as numerous mechanisms contribute METABRIC and Lyng datasets receiving hormone therapy, to hormone therapy resistance there may not be a single miRNA either alone or in combination with chemotherapy and/or þ profile able to predict response to tamoxifen therapy. In our radiotherapy, revealed that classification of ER breast cancers approach, we identified a miRNA signature indicative of a com- as high-hMAPK–miRNA by the recurrence or survival signature mon biology, that of hMAPK signaling, and queried whether there was significantly associated with increased risk of disease recur- was an association with clinical outcome, rather than taking rence (Lyng) and disease-specific mortality (METABRIC). These patients with discrete clinical outcomes and searching for a results reinforce the observations we made from survival anal- miRNA signature that would predict those outcomes. That our ysis of the hMAPK–miRNA signatures in the other training and hMAPK-recurrence and survival signatures show predictive value validation datasets, and suggest that these hMAPK–miRNA in the Lyng dataset reinforces the idea that activation of MAPK signatures may have predictive value regarding response to þ signaling represents a significant biologic event in the establish- tamoxifen therapy for patients with ER disease.

(Continued.) B, disease recurrence among patients from Lyng dataset treated with tamoxifen monotherapy and patients from METABRIC dataset with ERþ disease þ who received any hormone therapy. C, disease survival among all patients and patients with ER disease (Enerly dataset). D, disease recurrence among all þ patients and patients with ER disease (Enerly dataset and Buffa datasets). The Kaplan–Meier curves: dashed, low-hMAPK–miRNA; solid, high-hMAPK–miRNA. The log-rank test P values are indicated.

www.aacrjournals.org Clin Cancer Res; 21(2) January 15, 2015 OF11

Miller et al.

Focusing the analysis on the 12 miRNAs in common between hMAPK–miRNAs, and patients then stratified by high hMAPK– the hMAPK-recurrence and hMAPK-survival, miRNA signatures miRNA classification or low hMAPK–miRNA. Such classifica- reduced the ability to prognosticate increased disease recur- tion by this signature could then inform decisions on tamox- rence and decreased disease survival in all patient cohorts ifen treatment alone or more aggressive therapy, such as in compared with the recurrence and survival signatures, but was combination with MEK or other signaling inhibitors, as still more significant than the total 57-member hMAPK miRNA informed by the targets of these miRNAs. Such prospective signature (data not shown). This suggests that, while there are studies could be performed in tandem with current predictive hMAPK–miRNAs common to both the recurrence and the gene expression assays (such as Oncotype DX or Mammaprint) survival signature that may be useful biomarkers or potential to determine the added value of incorporating miRNA analysis þ drivers of general poor outcome (i.e., low expression of with existing mRNA-based assays in predicting response of ER miR29c, miR30 family, high expression of 221/222, etc.), breast cancer to targeted therapy. alterations in particular hMAPK–miRNAs associate specifically fi with either increased disease recurrence or poor disease-speci c Disclosure of Potential Conflicts of Interest survival (i.e., in the survival signature miR22 and miR224 are No potential conflicts of interest were disclosed. overexpressed, whereas let-7a is underexpressed), especially þ among ER breast cancers. Authors' Contributions The data presented here suggest that molecular and behav- Conception and design: P.C. Miller, D. El-Ashry ioral characteristics of tumors with activated growth factor Development of methodology: P.C. Miller, J. Clarke, D. El-Ashry signaling through the ERK1/2–MAPK axis are in part coordi- Acquisition of data (provided animals, acquired and managed patients, nated by aberrant miRNA expression associated with such provided facilities, etc.): P.C. Miller, J. Brinkman activated MAPK signaling. The reported functions of members Analysis and interpretation of data (e.g., statistical analysis, biostatistics, of the hMAPK–miRNA signature, together with the survival and computational analysis): P.C. Miller, J. Clarke, T. Koru-Sengul, D. El-Ashry Writing, review, and/or revision of the manuscript: P.C. Miller, J. Clarke, recurrence data from the Enerly, METABRIC, and Lyng valida- T. Koru-Sengul, D. El-Ashry tion datasets (Figs. 4 and 5; Tables 1 and 2), suggest that the Administrative, technical, or material support (i.e., reporting or organizing miRNAs identified in our hMAPK–miRNA signature contribute data, constructing databases): P.C. Miller, J. Brinkman to endocrine resistance associated with activated ERK1/2– Study supervision: D. El-Ashry MAPK signaling, and raise the provocative possibility that the miRNAs contained within this signature may ultimately prove Acknowledgments to have value as prognostic indicators of clinical outcome and The authors thank Drs. Joyce Slingerland (University of Miami, Miami, FL), þ as predictors of ER breast cancers with de novo endocrine Dan Hayes (University of Michigan, Ann Arbor, MI), James Rae (University of resistance. These findings support further evaluation of the Michigan), Jennifer Richer (University of Colorado, Boulder, CO), and Marc – Lippman (University of Miami) for critical review of the article; Dr. Marc predictive potential of members of this hMAPK miRNA signa- Lippman and members of the El-Ashry and Lippman laboratory group for ture for hormone therapy resistance and suggest that a subset of thoughtful discussion; and The University of Miami Sylvester Comprehensive these miRNAs may also prove to be potential therapeutic Cancer Center Oncogenomics Core Facility for technical assistance and gener- targets. The identification of subsignatures of hMAPK–miRNAs ation of microarray data. that significantly associate with increased disease recurrence and reduced disease-specific survival outcomes, particularly in Grant Support þ patients with ER breast cancer, indicates that stratification of Funding for this research was provided by NIH grant NIH 1R01 CA113674 patients according to expression of these miRNAs may ulti- and by Bankhead Coley Foundation 09BW-04. The costs of publication of this article were defrayed in part by the payment of mately provide important information related to disease prog- page charges. This article must therefore be hereby marked advertisement in nosis and response to therapy; these data indicate that the accordance with 18 U.S.C. Section 1734 solely to indicate this fact. clinical application of these signatures warrants further inves- tigation prospectively in large patient cohorts where patients Received August 7, 2014; revised October 16, 2014; accepted October 30, tumors would be arrayed for miRNA expression of these 21 2014; published OnlineFirst November 4, 2014.

References 1. Buyse M, Loi S, van't Veer L, Viale G, Delorenzi M, Glas AM, et al. Validation 6. Toi M, Osaki A, Yamada H, Toge T. Epidermal growth factor receptor and clinical utility of a 70-gene prognostic signature for women with node- expression as a prognostic indicator in breast cancer. Eur J Cancer 1991; negative breast cancer. J Natl Cancer Inst 2006;98:1183–92. 27:977–80. 2. Paik S, Shak S, Tang G, Kim C, Baker J, Cronin M, et al. A multigene assay to 7. Sainsbury JR, Farndon JR, Needham GK, Malcolm AJ, Harris AL. Epidermal- predict recurrence of tamoxifen-treated, node-negative breast cancer. N growth-factor receptor status as predictor of early recurrence of and death Engl J Med 2004;351:2817–26. from breast cancer. Lancet 1987;1:1398–402. 3. Knight WA, Livingston RB, Gregory EJ, Mcguire WL. Estrogen-receptor as an 8. Rimawi MF, Shetty PB, Weiss HL, Schiff R, Osborne CK, Chamness GC, independent prognostic factor for early recurrence in breast-cancer. Cancer et al. Epidermal growth factor receptor expression in breast cancer asso- Res 1977;37:4669–71. ciation with biologic phenotype and clinical outcomes. Cancer 2010;116: 4. Desombre ER, Thorpe SM, Rose C, Blough RR, Andersen KW, Rasmussen 1234–42. BB, et al. Prognostic usefulness of estrogen-receptor immunocytochemical 9. Loi S, Haibe-Kains B, Majjaj S, Lallemand F, Durbecq V, Larsimont D, et al. assays for human-breast cancer. Cancer Res 1986;46:4256–64. PIK3CA mutations associated with gene signature of low mTORC1 signal- 5. Clark GM, Mcguire WL. Steroid-receptors and other prognostic factors in ing and better outcomes in estrogen receptor-positive breast cancer. Proc primary breast-cancer. Semin Oncol 1988;15:20–5. Natl Acad Sci U S A 2010;107:10208–13.

OF12 Clin Cancer Res; 21(2) January 15, 2015 Clinical Cancer Research

hMAPK–miRNA Signature and Hormone Resistance

10. Lopez-Knowles E, O'Toole SA, McNeil CM, Millar EK, Qiu MR, Crea P, et al. 32. NIH" NCIa. The Cancer Genome Atlas. 2011 [cited 2011]. Available from: PI3K pathway activation in breast cancer is associated with the basal-like http://cancergenome.nih.gov/. phenotype and cancer-specific mortality. Int J Cancer 2010;126:1121–31. 33. Lyng MB, Laenkholm AV, Sokilde R, Gravgaard KH, Litman T, Ditzel HJ. þ 11. Hernandez-Aya LF, Gonzalez-Angulo AM. Targeting the phosphatidylino- Global MicroRNA expression profiling of high-risk ER breast cancers from sitol 3-kinase signaling pathway in breast cancer. Oncologist 2011;16: patients receiving adjuvant tamoxifen mono-therapy: a DBCG study. PLoS 404–14. ONE 2012;7:e36170. 12. Bayliss J, Hilger A, Vishnu P, Diehl K, El-Ashry D. Reversal of the estrogen 34. Curtis C, Shah SP, Chin SF, Turashvili G, Rueda OM, Dunning MJ, et al. The receptor negative phenotype in breast cancer and restoration of antiestro- genomic and transcriptomic architecture of 2,000 breast tumours reveals gen response. Clin Cancer Res 2007;13:7029–36. novel subgroups. Nature 2012;486:346–52. 13. Holloway JN, Murthy S, El-Ashry D. A cytoplasmic substrate of mitogen- 35. Dvinge H, Git A, Graf S, Salmon-Divon M, Curtis C, Sottoriva A, et al. The activated protein kinase is responsible for estrogen receptor-alpha down- shaping and functional consequences of the microRNA landscape in breast regulation in breast cancer cells: the role of nuclear factor-kappaB. Mol cancer. Nature 2013;497:378–82. Endocrinol 2004;18:1396–410. 36. Niemela M, Kauko O, Sihto H, Mpindi JP, Nicorici D, Pernila P, et al. CIP2A 14. Oh AS, Lorant LA, Holloway JN, Miller DL, Kern FG, El-Ashry D. Hyper- signature reveals the MYC dependency of CIP2A-regulated phenotypes and activation of MAPK induces loss of ERalpha expression in breast cancer its clinical association with breast cancer subtypes. Oncogene 2012;31: cells. Mol Endocrinol 2001;15:1344–59. 4266–78. 15. Creighton CJ, Hilger AM, Murthy S, Rae JM, Chinnaiyan AM, El-Ashry D. 37. Zhong L, Zhu K, Jin N, Wu D, Zhang J, Guo B, et al. A systematic analysis of Activation of mitogen-activated protein kinase in estrogen receptor alpha- miRNA–mRNA paired variations reveals widespread miRNA misregula- positive breast cancer cells in vitro induces an in vivo molecular phenotype tion in breast cancer. BioMed Res Int 2014;2014:291280. of estrogen receptor alpha-negative human breast tumors. Cancer Res 38. Hua L, Li L, Zhou P. Identifying breast cancer subtype related miRNAs from 2006;66:3903–11. two constructed miRNAs interaction networks in silico method. BioMed 16. Calin GA, Liu CG, Sevignani C, Ferracin M, Felli N, Dumitru CD, et al. Res Int 2013;2013:798912. MicroRNA profiling reveals distinct signatures in B cell chronic lympho- 39. Bleckmann A, Leha A, Artmann S, Menck K, Salinas-Riester G, Binder C, cytic leukemias. Proc Natl Acad Sci U S A 2004;101:11755–60. et al. Integrated miRNA and mRNA profiling of tumor-educated macro- 17. Lu J, Getz G, Miska EA, Alvarez-Saavedra E, Lamb J, Peck D, et al. MicroRNA phages identifies prognostic subgroups in estrogen receptor-positive breast expression profiles classify human cancers. Nature 2005;435:834–8. cancer. Mol Oncol 2014;pii:S1574-7891(14)00178-1. 18. Jay C, Nemunaitis J, Chen P, Fulgham P, Tong AW. miRNA profiling for 40. Dweep H, Sticht C, Pandey P, Gretz N. miRWalk–database: prediction of diagnosis and prognosis of human cancer. DNA Cell Biol 2007;26:293– possible miRNA binding sites by "walking" the genes of three genomes. 300. J Biomed Inform 2011;44:839–47. 19. Iorio MV, Ferracin M, Liu CG, Veronese A, Spizzo R, Sabbioni S, et al. 41. Good PI. Resampling methods: a practical guide to data analysis. 3rd ed. MicroRNA gene expression deregulation in human breast cancer. Cancer Boston, MA: Birkh€auser; 2006. Res 2005;65:7065–70. 42. Westfall PH, Young S. Resampling-based multiple testing: examples and 20. Lowery AJ, Miller N, Devaney A, McNeill RE, Davoren PA, Lemetre C, et al. methods for p-value adjustment. New York: Wiley; 1993. MicroRNA signatures predict oestrogen receptor, progesterone receptor and 43. Team RDC. R: a language and environment for statistical computing, HER2/neu receptor status in breast cancer. Breast Cancer Res 2009;11:R27. R Foundation for Statistical Computing. 2.12.2 ed; 2011. 21. Mattie MD, Benz CC, Bowers J, Sensinger K, Wong L, Scott GK, et al. 44. Chou J, Lin JH, Brenot A, Kim JW, Provot S, Werb Z. GATA3 suppresses Optimized high-throughput microRNA expression profiling provides nov- metastasis and modulates the tumour microenvironment by regulating el biomarker assessment of clinical prostate and breast cancer biopsies. Mol microRNA-29b expression. Nat Cell Biol 2013;15:201–13. Cancer 2006;5:24. 45. Johnson SM, Grosshans H, Shingara J, Byrom M, Jarvis R, Cheng A, et al. 22. Enerly E, Steinfeld I, Kleivi K, Leivonen SK, Aure MR, Russnes HG, et al. RAS is regulated by the let-7 microRNA family. Cell 2005;120:635–47. miRNA-mRNA integrated analysis reveals roles for miRNAs in primary 46. Kumarswamy R, Mudduluru G, Ceppi P, Muppala S, Kozlowski M, Nik- breast tumors. PLoS ONE 2011;6:e16915. linski J, et al. MicroRNA-30a inhibits epithelial-to-mesenchymal transition 23. Blenkiron C, Goldstein LD, Thorne NP, Spiteri I, Chin SF, Dunning MJ, by targeting Snai1 and is downregulated in non-small cell lung cancer. Int J et al. MicroRNA expression profiling of human breast cancer identifies new Cancer 2012;130:2044–53. markers of tumor subtype. Genome Biol 2007;8:R214. 47. Cochrane DR, Spoelstra NS, Howe EN, Nordeen SK, Richer JK. MicroRNA- 24. Heneghan HM, Miller N, Lowery AJ, Sweeney KJ, Kerin MJ. MicroRNAs as 200c mitigates invasiveness and restores sensitivity to microtubule-target- novel biomarkers for breast cancer. J Oncol 2009;2009:950201. ing chemotherapeutic agents. Mol Cancer Ther 2009;8:1055–66. 25. Janssen EA, Slewa A, Gudlaugsson E, Jonsdottir K, Skaland I, Soiland H, 48. Yu Z, Wang C, Wang M, Li Z, Casimiro MC, Liu M, et al. A cyclin D1/ et al. Biologic profiling of lymph node negative breast cancers by means of microRNA 17/20 regulatory feedback loop in control of breast cancer cell microRNA expression. Mod Pathol 2010;23:1567–76. proliferation. J Cell Biol 2008;182:509–17. 26. Burroughs AM, Ando Y, de Hoon MJ, Tomaru Y, Suzuki H, Hayashizaki Y, 49. Zhao JJ, Lin J, Yang H, Kong W, He L, Ma X, et al. MicroRNA-221/222 et al. Deep-sequencing of human Argonaute-associated small RNAs pro- negatively regulates estrogen receptor alpha and is associated with tamox- vides insight into miRNA sorting and reveals Argonaute association with ifen resistance in breast cancer. J Biol Chem 2008;283:31079–86. RNA fragments of diverse origin. RNA Biol 2011;8:158–77. 50. Foekens JA, Sieuwerts AM, Smid M, Look MP, de Weerd V, Boersma AW, 27. Cochrane DR, Cittelly DM, Howe EN, Spoelstra NS, McKinsey EL, LaPara K, et al. Four miRNAs associated with aggressiveness of lymph node-negative, et al. MicroRNAs link estrogen receptor alpha status and Dicer levels in estrogen receptor-positive human breast cancer. Proc Natl Acad Sci U S A breast cancer. Horm Cancer 2010;1:306–19. 2008;105:13021–6. 28. Di Leva G, Gasparini P, Piovan C, Ngankeu A, Garofalo M, Taccioli C, et al. 51. Rodriguez-Gonzalez FG, Sieuwerts AM, Smid M, Look MP, Meijer-van MicroRNA cluster 221–222 and estrogen receptor alpha interactions in Gelder ME, de Weerd V, et al. MicroRNA-30c expression level is an breast cancer. J Natl Cancer Inst 2010;102:706–21. independent predictor of clinical benefit of endocrine therapy in advanced 29. Miller TE, Ghoshal K, Ramaswamy B, Roy S, Datta J, Shapiro CL, et al. estrogen receptor positive breast cancer. Breast Cancer Res Treat 2011;127: MicroRNA-221/222 confers tamoxifen resistance in breast cancer by tar- 43–51. geting p27Kip1. J Biol Chem 2008;283:29897–903. 52. Ward A, Balwierz A, Zhang JD, Kublbeck M, Pawitan Y, Hielscher T, et al. Re- 30. Yu F, Yao H, Zhu P, Zhang X, Pan Q, Gong C, et al. let-7 regulates self expression of microRNA-375 reverses both tamoxifen resistance and renewal and tumorigenicity of breast cancer cells. Cell 2007;131:1109–23. accompanying EMT-like properties in breast cancer. Oncogene 2013;32: 31. Buffa FM, Camps C, Winchester L, Snell CE, Gee HE, Sheldon H, et al. 1173–82. microRNA-associated progression pathways and potential therapeutic 53. McGlynn LM, Kirkegaard T, Edwards J, Tovey S, Cameron D, Twelves C, targets identified by integrated mRNA and microRNA expression profiling et al. Ras/Raf-1/MAPK pathway mediates response to tamoxifen but not in breast cancer. Cancer Res 2011;71:5635–45. chemotherapy in breast cancer patients. Clin Cancer Res 2009;15:1487–95.

www.aacrjournals.org Clin Cancer Res; 21(2) January 15, 2015 OF13

A Novel MAPK−microRNA Signature Is Predictive of Hormone-Therapy Resistance and Poor Outcome in ER-Positive Breast Cancer

Philip C. Miller, Jennifer Clarke, Tulay Koru-Sengul, et al.

Clin Cancer Res Published OnlineFirst November 4, 2014.

Updated version Access the most recent version of this article at: doi:10.1158/1078-0432.CCR-14-2053

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2014/11/05/1078-0432.CCR-14-2053.DC1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/early/2015/01/04/1078-0432.CCR-14-2053. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.