Research ArticleCancer

Genomic profiling of ER+ breast cancers after short-term estrogen suppression reveals alterations associated with endocrine resistance

See allHide authors and affiliations

Science Translational Medicine  09 Aug 2017:
Vol. 9, Issue 402, eaai7993
DOI: 10.1126/scitranslmed.aai7993

A patient look at cancer resistance

A variety of drugs that inhibit estrogen signaling are in use for breast cancer, but patients often develop resistance to these treatments. To understand how this resistance develops, Giltnane et al. evaluated 143 patients who were receiving the aromatase inhibitor letrozole to block estrogen signaling before undergoing surgery for breast cancer. By performing genomic analysis on these patients’ tumors, the authors were able to identify not only changes in gene expression and estrogen receptor gene fusions that correlated with resistance to therapy but also potential leads for future treatments that could help overcome this resistance.

Abstract

Inhibition of proliferation in estrogen receptor–positive (ER+) breast cancers after short-term antiestrogen therapy correlates with long-term patient outcome. We profiled 155 ER+/human epidermal growth factor receptor 2–negative (HER2) early breast cancers from 143 patients treated with the aromatase inhibitor letrozole for 10 to 21 days before surgery. Twenty-one percent of tumors remained highly proliferative, suggesting that these tumors harbor alterations associated with intrinsic endocrine therapy resistance. Whole-exome sequencing revealed a correlation between 8p11-12 and 11q13 gene amplifications, including FGFR1 and CCND1, respectively, and high Ki67. We corroborated these findings in a separate cohort of serial pretreatment, postneoadjuvant chemotherapy, and recurrent ER+ tumors. Combined inhibition of FGFR1 and CDK4/6 reversed antiestrogen resistance in ER+ FGFR1/CCND1 coamplified CAMA1 breast cancer cells. RNA sequencing of letrozole-treated tumors revealed the existence of intrachromosomal ESR1 fusion transcripts and increased expression of gene signatures indicative of enhanced E2F-mediated transcription and cell cycle processes in cancers with high Ki67. These data suggest that short-term preoperative estrogen deprivation followed by genomic profiling can be used to identify druggable alterations that may cause intrinsic endocrine therapy resistance.

INTRODUCTION

Estrogen receptor–positive (ER+) breast cancer is the most common clinical subtype of breast cancer, comprising about 80% of patients (1). Adjuvant endocrine therapies, such as selective ER modulators (SERMs; such as tamoxifen), selective ER down-regulators (SERDs; such as fulvestrant), and aromatase inhibitors (AIs; such as letrozole), are approved for adjuvant treatment of women with ER+ breast cancer. Randomized clinical trials have proven the effectiveness of these antiestrogens in preventing disease recurrence (2). However, about 20% of patients diagnosed with operable ER+ tumors will recur during or after adjuvant endocrine therapy. Notably, mortality from these endocrine-resistant tumors accounts for the majority of breast cancer deaths in the United States each year (3). An increasing number of mechanisms of endocrine resistance have been reported (4, 5). To date, the only mechanisms of endocrine therapy resistance that have been observed in the clinic are ERBB2 amplification (6) and mutations in the ligand-binding domain (LBD) of ESR1 (710).

Breast cancer cell proliferation measured by Ki67 immunohistochemistry (IHC) after short-term antiestrogen therapy was first shown to correlate with recurrence-free survival (RFS) in the Immediate Preoperative Anastrozole, Tamoxifen, or Combined with Tamoxifen (IMPACT) neoadjuvant trial (11). In this trial, tertiles of the posttreatment 2-week Ki67 labeling index showed a strong inverse association with RFS and identified a group of patients with a high 5-year RFS rate of ~40%. Further in this trial, the AI anastrozole induced a stronger suppression of Ki67 expression compared to tamoxifen or the combination of these drugs. This difference translated into improved long-term outcome in patients treated with AIs in the large adjuvant ATAC (Arimidex, Tamoxifen, Alone or in Combination) and BIG (Breast International Group) 1-98 trials (12, 13). These data suggest that tumors exhibiting profound inhibition of cellular proliferation by Ki67 are highly hormone-dependent and may identify patients with an excellent prognosis after adjuvant endocrine therapy alone. Conversely, high tumor cell proliferation upon short-term estrogen deprivation can serve as a biomarker to identify patients with antiestrogen-resistant cancers at risk of early recurrence. Thus, we hypothesized that profiling operable ER+ tumors after short-term estrogen suppression with an AI would identify actionable molecular alterations associated with endocrine resistance. These alterations may ultimately serve as therapeutic targets to combat resistance to antiestrogen therapy in ER+ breast cancer.

RESULTS

A subset of ER+ breast cancers remains highly proliferative despite letrozole-induced estrogen deprivation

One hundred fifty-five tumor biopsies were obtained from a population-representative set of 143 patients with stage I to stage III operable ER+/human epidermal growth factor receptor 2–negative (HER2) breast cancer enrolled in a clinical trial of the AI letrozole administered for 10 to 21 days before surgery (Vanderbilt University NCT00651976). Patients provided written informed consent according to a protocol approved by the Vanderbilt-Ingram Cancer Center Institutional Review Board. Intraoperative biopsies or surgical specimens, snap-frozen in liquid nitrogen or formalin-fixed and paraffin-embedded (FFPE), were obtained from each patient’s tumor(s). A diagnostic (pretreatment) FFPE tumor biopsy was obtained for assessment of baseline Ki67. ER, progesterone receptor (PR), and HER2 status at diagnosis were confirmed by IHC or fluorescence in situ hybridization (FISH) as per clinical guidelines (Fig. 1A). Mean patient age was 64 years (range, 45 to 87 years), with tumors distributed among stages I (54%), II (38%), and III (7.25%). A majority of cancers were of low (35%) and intermediate (54%) histological grade; 92% had an ER Allred score of 5 (≥67% ER+ cells) and 46.5% had a PR Allred score of 5 (≥67% PR+ cells). Detailed clinical characteristics are shown in Table 1. A detailed illustration of the number of patients enrolled in the trial, the number of evaluable tissue samples, and the number available for molecular analysis can be found in fig. S1.

Fig. 1. A subset of ER+ breast cancers remains highly proliferative despite letrozole-mediated estrogen deprivation.

(A) Schema of clinical trial of 143 patients with ER+/HER2 breast cancer treated for 10 to 21 days with letrozole. Arrows indicate general time points at which a biopsy was taken or surgery was performed. (B) Heat map displaying pre- and post-letrozole treatment IHC (by AQUA) scores for Ki67, ER, and PR in tumor specimens stratified by Ki67 response to letrozole. Molecular subtype, recurrence score by IHC4, and histologic type are also noted. (C) Paired pre- and post-letrozole treatment tumor specimens from the trial were stratified into sensitive, intermediate, or resistant response categories based on posttreatment Ki67 scores. BrCa, breast cancer; pts, patients; QD, once daily; ln, natural log; NST, invasive carcinoma of no special type; ILC, invasive lobular carcinoma.

Table 1. Baseline clinical characteristics of the 143 study patients (n = 155 tumors).

n, number of patients or number of tumors for which data were available.

View this table:

We measured ER, PR, and Ki67 expression in pre- and post-letrozole tumor samples by automated quantitative immunofluorescence (AQUA) (14, 15). This method uses multiplexed immunofluorescence to delineate invasive cancer cells with tagged antibodies against pan-cytokeratin and nuclei with 4′,6-diamidino-2-phenylindole staining in 10 to 25 regions of interest on whole-tumor sections identified by a breast pathologist, scoring about 10,000 tumor nuclei (7). Because all tumors were clinically HER2 (not amplified), this allowed for allocation of molecular subtype (luminal A or B) by the AQUA-IHC4 algorithm (Fig. 1B) (14).

Tumor response to letrozole was categorized by tertiles of the calculated Ki67 labeling index in the posttreatment (mastectomy) sample according to those used in the IMPACT study (11). The tertiles were as follows: sensitive [Ki67 ln (natural log) ≤ 1.0; 0 to 2.7% Ki67+ tumor cells or labeling index], intermediate (Ki67 ln = 1.1 to 1.9; 2.8 to 7.3%), or resistant (Ki67 ln ≥ 2.0; ≥7.4%). Ki67 was evaluable in 140 FFPE tumors. A representative tumor in which the Ki67 was reduced from ln ≥ 2.0 (37%) to ln ≤ 1.0 (0%) is shown in fig. S2A. Twenty-one percent of tumors were resistant and retained high Ki67 despite estrogen deprivation (mean Ki67 index, 24.5%), 56% were categorized as sensitive (mean Ki67 index, 0.8%) and considered to have achieved complete cell cycle arrest, and the remaining tumors were deemed intermediate (mean Ki67 index, 4.7%) (Fig. 1C and Table 2). Several tumors with high proliferation at baseline (ln ≥ 2.0 or Ki67 index ≥ 7.4%) exhibited marked reductions in Ki67 upon estrogen deprivation with letrozole. The mean number of days of letrozole therapy for sensitive, intermediate, and resistant patients was not significantly different: sensitive, 15 ± 4 days; intermediate, 14 ± 4 days; resistant, 15 ± 4 days.

Table 2. Categorization of study samples by post-letrozole Ki67 proliferative index.

View this table:

Consistent with inhibition of ER signaling by letrozole, there was a significant overall reduction in posttreatment expression of PR, a well-established target of ERα-mediated transcription (16), compared to pretreatment expression as measured by AQUA (P < 0.0001, paired t test; fig. S2B). Endocrine resistance, defined by the posttreatment Ki67 tertiles, was independent of the percentage of ER+ cells as scored by a breast pathologist (M.G.K.) (fig. S2C). With AQUA-based scoring, however, ERα positivity was significantly lower in the resistant tumors (P < 0.01; fig. S2D). Tumors with available baseline biopsies were categorized as luminal A or B based on pretreatment AQUA-IHC4 (Fig. 1B); 70 of 134 (52%) were luminal A, and 64 of 134 (48%) were luminal B. Notably, 23 of 64 (36%) luminal B tumors exhibited complete cell cycle arrest upon treatment, suggesting that they were highly hormone-dependent despite unfavorable characteristics by AQUA-IHC4 profiling.

Whole-exome sequencing identifies copy number alterations associated with resistance to estrogen deprivation

To uncover single nucleotide variations (SNVs), insertions/deletions (indels), and copy number alterations (CNAs), we performed successful whole-exome sequencing (WES) on DNA extracted from 54 posttreatment tumors and matched normal/blood and 5 unpaired tumor samples (see fig. S1). Somatic mutations were detected in 54 tumor-normal pairs, and copy number calls were made in 59 tumors. WES mean target coverage was 90×. Figure 2 shows an overview of the recurrent somatic mutations (Fig. 2A) and copy number changes (Fig. 2B) detected by CoMut and GISTIC analyses, respectively. A summary of mutations per megabase (mean mutation rate, 1.503 × 10−6; SD = 1.400 × 10−6) showed one outlier patient with a mutation rate higher than 2 SDs from the mean (mean + 2 SD = 4.303 × 10−6). This tumor genome contained a loss-of-function mutation in RAD52, which is required for the repair of double-stranded DNA breaks (17). Consistent with other breast cancer genomic studies, PIK3CA mutations were the most common, with 27 mutations present in 24 of 54 tumors (44%; n = 19 exon 20, n = 4 exon 9, n = 4 other). TP53, CDH1 (E-cadherin), and TBX3 mutations were also common, present in 20% or more of samples. Similarly, we found alterations in other genes known to be recurrent in ER+ breast cancer, such as GATA3 (5 of 54, 9%) and SF3B1 (6 of 54, 11%) (18), and frequent CNAs [CNAs/copy number variations (CNVs)] at the 1q32 (19 of 59, 32%), 8p11-12 (11 of 59, 19%), and 11q13 amplicons (9 of 59, 15%) (Fig. 2B and fig. S3) (19, 20). Other notable alterations detected were one ESR1 amplification, one potentially new ERα mutation (L429V) present in the LBD, and four AKT1 missense mutations (4 of 54, 7%). Overall, the genomic profile of the letrozole-treated tumors in our cohort was similar to the genomics of treatment-naïve breast tumors in The Cancer Genome Atlas (TCGA) (1).

Fig. 2. WES identifies CNAs associated with endocrine therapy resistance.

(A) Tile plot of variants identified in significantly mutated genes detected by WES in 54 tumor samples. Samples are listed by response category (13 resistant, 8 intermediate, 30 sensitive, and 3 unknown). Genes were considered significantly mutated if their associated q value was ≥0.1 [−log10 (q value) ≥1.0, delineated by the solid red line in the histogram on the right]. (B) Heat map showing log2 copy number ratios for genomic regions with recurrent gains (red) or losses (green) by GISTIC. Available for CNV analysis were 12 resistant, 8 intermediate, 35 sensitive, and 4 unknown tumors.

Table S1 includes an analysis of all recurrent SNVs and CNVs with respect to categorical response to letrozole across all tumors. None of the recurrent somatic mutations (>5%), including PIK3CA and TP53, correlated with response or resistance to letrozole. Recurrent amplifications correlating with response included 17q21-23, 11q13.3, and 8p11.23. The 8p11-12 and 11q13 amplicons have been previously associated with endocrine resistance (19) and include the histone methyltransferase WHSC1L1, the receptor tyrosine kinase FGFR1, the cell cycle regulator CCND1 (encodes cyclin D1), and the fibroblast growth factor receptor (FGFR) ligands FGF3, FGF4, and FGF19, among others (fig. S3). The 17q21-23 amplicon has not been previously associated with endocrine therapy resistance; it includes S6K and BRCA1. Neither of these two genes was amplified in letrozole-resistant tumors.

To confirm the association of 8p11 and 11q13 amplification with lack of response to estrogen deprivation, focusing on actionable somatic alterations in such loci, we performed FISH using clinically validated probes for FGFR1 and CCND1, respectively. FFPE tumors from the initial WES study and additional tumors from the clinical trial were interrogated (Fig. 3A). FGFR1 was amplified in 12 of 72 (17%) FGFR1-evaluable tumors, and CCND1 was amplified in 18 of 68 (26%) CCND1-evaluable tumors (fig. S4, A and B, respectively). FGFR1 and CCND1 amplification were both evaluable in a total of 67 tumors (Fig. 3B and Table 3). We observed a statistically significant correlation between CCND1 amplification by FISH and copy number by WES (P < 0.0001; fig. S4C) and between FISH-quantified CCND1 amplification and CCND1 transcript expression by RNA sequencing (RNAseq) (P = 0.0001; fig. S4D). FGFR1 amplification by FISH statistically correlated with WES-derived log2 copy number values (P = 0.0279; fig. S4E) but did not correlate with transcript expression (P = 0.1036; fig. S4F). Next, we compared FISH and tumor response data and found that 7 of 20 (35%) endocrine therapy–resistant cancers were FGFR1-amplified, whereas only 3 of 35 (9%) sensitive tumors harbored FGFR1 amplification (Bonferroni-adjusted P = 0.05, Fisher’s exact test). Nine of 20 (45%) letrozole-resistant tumors were CCND1-amplified, whereas only 7 of 35 (20%) sensitive tumors harbored CCND1 amplification (Bonferroni-adjusted P = 0.13, Fisher’s exact test). Seven of 67 tumors (10%) harbored coamplification of FGFR1 and CCND1; 6 of those 7 (86%) were endocrine therapy–resistant (Fig. 3B and Table 3). Overall, a higher proportion (≥40%) of ER+ tumors exhibiting FGFR1 and/or CCND1 amplification was categorically resistant, maintaining elevated proliferation despite estrogen deprivation (Pearson χ2 = 12.67, Bonferroni-adjusted P = 0.06). We observed numerically higher baseline Ki67 in FGFR1/CCND1 coamplified tumors compared to nonamplified or singly amplified tumors, but these data were not statistically significant (fig. S4G). A larger sample size with adequate power would be needed to confirm this association.

Fig. 3. CCND1 and FGFR1 amplifications detected by FISH are associated with resistance to estrogen deprivation.

(A) Representative FISH images from our cohort displaying samples with CCND1 amplification (patient 7629), FGFR1 amplification (patient 7670), CCND1 and FGFR1 coamplification (patient 7657), and a patient negative for CCND1 and FGFR1 amplification (patient 1213). Magnification = ×100 for each image, representative scale bar shown in FGFR1/7629. (B) Graphical summary of CCND1 amplification, FGFR1 amplification, and FGFR1/CCND1 coamplification across letrozole-sensitive, letrozole-intermediate, and letrozole-resistant patients as per their posttreatment Ki67 categorization. Numbers for this analysis are shown in Table 3.

Table 3. Summary of FGFR1 and CCND1 FISH analysis in trial tumors.

Bold text highlights resistant samples amplified for FGFR1 and/or CCND1.

View this table:

Gene expression analysis reveals that E2F, cell cycle, and T cell activation correlate with resistance to estrogen deprivation

To identify transcriptomic alterations, we performed RNAseq on 56 posttreatment tumors (47 fresh-frozen and 9 FFPE). RNAseq libraries were prepared in two batches: The first contained 41 fresh-frozen tumors with RNA prepared using standard, unstranded protocols after polyA purification, and the second batch contained a mix of frozen (6) and FFPE (9) tumors and was prepared using RNA ACCESS. Each batch was analyzed separately to avoid batch effects, with the first cohort serving as a discovery cohort and the second batch serving as a validation cohort. This approach allowed us to observe that the first two principal components separated tumors by response to estrogen deprivation [Fig. 4A (frozen-derived specimens) and fig. S5A (frozen- and FFPE-derived specimens)]. To eliminate the possibility that genes associated with proliferation and, therefore, elevated Ki67 are responsible for the observed separation of resistant tumors, we reanalyzed the tumors after removal of proliferation-associated genes [a list of removed genes can be found in table S2, compiled from (21)]. Still, tumors separated by response to estrogen deprivation [Fig. 4B (frozen-derived specimens) and fig. S5B (frozen- and FFPE-derived specimens)]. We performed differential gene expression analysis to identify profiles associated with endocrine resistance. Previous work in our laboratory showed that E2F activation is associated with acquired estrogen independence and identified a gene expression signature of E2F activation that correlates with poor tumor response to AIs in patients (22). Here, this signature clearly segregated tumors by response to aromatase inhibition, with the majority of sensitive tumors exhibiting a low E2F activation gene expression signature (fig. S5C). Again, because it may be expected that an E2F/cell cycle gene expression signature would be up-regulated in tumors that have high Ki67, we performed the differential expression analysis again after removal of proliferation-associated genes (21). Upon doing so, endocrine therapy–resistant tumors stratified based on enrichment of genes that harbor SH2 and SH3 adaptor protein domains (Fig. 4C).

Fig. 4. Gene expression analysis reveals multiple pathways that strongly correlate with resistance to estrogen deprivation.

(A) Principal components analysis of gene expression shows that tumors separate by response to estrogen. PC, principal component. (B) This separation is maintained even after removal of proliferation-associated genes. (C) With the removal of proliferation-associated genes, differential gene expression analysis using RNAseq data for response to letrozole shows enrichment for SH2 domain– and SH3 domain–containing genes (highlighted in red text). (D and E) Pathway analyses performed using RNAseq data from letrozole-treated breast cancers reveal an up-regulation of genes involved in the cell cycle (D), particularly in tumors with FGFR1 and CCND1 coamplification, as well as ECM proteins (E). ES, enrichment signature. (F) Pathway analysis after removal of proliferation genes shows up-regulation of secretome-related signaling. (G and H) Analysis of the RB1 loss gene expression signature shows increased expression in the resistant tumors (G) and tumors coamplified for CCND1 and FGFR1 (H). Bars, mean ± SD. P values represent results of a one-way analysis of variance (ANOVA) followed by Tukey’s post hoc test (*P < 0.05 and ***P < 0.001).

Next, pathway analyses were performed on differentially expressed genes using a gene-by-gene linear model with posttreatment Ki67 categorization (sensitive, intermediate, and resistant). After correction for multiple hypothesis testing, 2328 genes (false discovery rate, <0.025) were deemed differentially expressed in letrozole-resistant tumors. Specifically, genes involved in cell cycle and mitosis (Fig. 4D; enrichment score, 6.08), in extracellular matrix (ECM) remodeling and migration (Fig. 4E; enrichment score, 7.56), and in secretome-related signaling (Fig. 4F; enrichment score, 7.91) were differentially enriched in letrozole-resistant versus letrozole-sensitive tumors. Together, these data support the idea that signaling pathways other than cell proliferation are “programmed” differently between endocrine therapy–resistant and endocrine therapy–sensitive tumors.

In our study, FGFR1 and/or CCND1 amplification correlated with a high Ki67 index in letrozole-treated tumors. To identify differentially expressed genes in FGFR1- and CCND1-amplified cancers in an independent data set with a large proportion of early breast cancers, we interrogated the RNAseq data from TCGA ER+/HER2 breast cohort (1). Pathway analysis of genes differentially expressed in CCND1-amplified breast tumors showed robust up-regulation of genes involved in cell cycle and mitosis (fig. S5D; enrichment score, 31.10), whereas in FGFR1-amplified tumors, up-regulation of cell cycle and mitotic genes was modest (fig. S5E; enrichment score, 6.10). Tumors with coamplification of FGFR1 and CCND1 showed additional enrichment of cell cycle genes greater than the enrichment explained by single amplification, consistent with an interaction between FGFR1 and cyclin D1 to drive estrogen-independent proliferation (fig. S5F; enrichment score, 8.12). Finally, genes involved in T cell activation and nuclear factor κB 1 pathway signaling were enriched in FGFR1-only amplified tumors (fig. S5, G and H, respectively).

Gene Set Enrichment Analysis (GSEA) of letrozole-sensitive, letrozole-intermediate, and letrozole-resistant tumors from our trial cohort also demonstrated a strong direct association between resistant tumors and gene signatures up-regulated in high-grade breast cancers, E2F target genes, and genes involved in cell proliferation (fig. S5, I, J, and K, respectively). Expression of genes that were previously identified as part of the estradiol response (23) was also increased, suggesting that the ER was still transcriptionally active despite letrozole-induced estrogen deprivation in these tumors (fig. S5L). Furthermore, and consistent with the differential gene expression analyses above, genes involved in the early activation of T cells were also up-regulated (fig. S5M). Consistent with the GSEAs, the transcription factor and target analysis of the differentially expressed genes by iRegulon (24) showed strong enrichment for genes coregulated by the transcription factors E2F4, FOXM1, and CBFB, all of which were up-regulated in the resistant tumors (fig. S5N). Cumulatively, these data suggest that canonical cell cycle signaling pathways are still active in estrogen-deprived tumors, consistent with their high Ki67 scores despite letrozole therapy.

To better understand the potential functional effect(s) of FGFR1 and/or CCND1 amplification, we examined the expression of a previously published RB1 loss gene signature (25) in our RNAseq data set. We found that the RB1 loss signature score showed a direct correlation with lack of sensitivity to estrogen deprivation as per Ki67 tertiles (Fig. 4G). Furthermore, the signature was also highest in FGFR1/CCND1 coamplified tumors, although there were only two coamplified samples with data available for this analysis (Fig. 4H). Together, these data suggest that letrozole-resistant tumors are commonly associated with FGFR1 ± CCND1 amplifications, allowing tumors to escape canonical proliferation control mechanisms.

RNAseq identifies fusions associated with lack of response to letrozole

In light of recent studies and to make full use of our available RNAseq data, we sought to identify fusion transcripts that may correlate with response to antiestrogen therapy. Fusion transcript identification from frozen samples resulted in 346 putative gene fusions from 50 tumors (~7 fusions per tumor; table S3). We designed primers with universal sequencing tags against the 3′ and 5′ sites of each fusion with a breakpoint mapping to RefSeq exon coding regions (n = 187) (table S4). Quantitative polymerase chain reaction (PCR) was used to amplify breakpoint flanking sequences from tumor complementary DNA (cDNA). Sanger sequencing results from positive quantitative PCR products were mapped to the human RNA reference transcriptome using the National Center for Biotechnology Information’s BLAST (Basic Local Alignment Search Tool) (table S5). Overall, 26 putative fusion transcripts from 15 unique tumors were validated by mapping to open reading frames of the predicted 3′ and 5′ genes (table S5 and fig. S6A). Validated transcripts were more likely to be called by more than one program [13 of 46 duplicate programmatic calls (28%) versus 13 of 300 unique programmatic calls (5%); P < 0.0001, Fisher’s exact test] (table S3, columns H to L).

Four of these 26 fusions mapped to chromosome 6q25.1, each involving the 5′ end of ESR1 (Fig. 5A). All four ESR1 fusions were present in tumors that did not respond to estrogen deprivation (table S5). We also observed a statistically significant increase in ESR1 coding transcript expression in tumors harboring an ERα fusion over those that did not (P < 0.0001; fig. S6B). Although these genes are on the same chromosome, the fusions are not the result of run-on transcription, because the 5′ end of ESR1 is fused to the 3′ ends of CCDC170, AKAP12, or c6orf211, which are upstream of ESR1 (Fig. 5A). All of these rearrangements fuse the two noncoding 5′ exons of ESR1 to the C-terminal portions of CCDC170, AKAP12, or c6orf211. The ESR1-CCDC170 fusion variants are thought to result in the expression of a truncated form of CCDC170, which reduces sensitivity to tamoxifen (26).

Fig. 5. RNAseq identifies ESR1 fusion transcripts associated with resistance to antiestrogen therapy.

(A) Diagram of chromosome 6 highlighting the 6q25 locus and showing that ESR1 is recurrently rearranged with other genes in 6q25, including AKAP12, c6orf211, and CCDC170. (B) Heat map of ESR1 fusion results from NanoString nCounter analysis. Left panels represent counts resulting from analysis of probes covering each exon of 5′ 6q25 genes; right panels represent counts resulting from probes covering each exon of ESR1 (3′ on 6q25). See tables S6 and S7 for validation results. E2/C2, cells transduced with a construct encoding the ESR1 ex2–CCDC170 ex2 fusion; GFP, cells transduced with a construct encoding green fluorescent protein.

Consistent with other recent reports (26, 27), variants of the ESR1-CCDC170 fusion were also identified in MCF7 and other ER+ breast cancer cell lines by RNAseq (Fig. 5A). We also observed putative ESR1 fusion transcripts in 12 breast cancers from TCGA (table S6). To validate the ESR1 fusions identified by RNAseq in our trial set and to uncover additional ESR1 fusions, we designed NanoString nCounter probes against each exon of ESR1, CCDC170, c6orf211, and AKAP12 on chromosome 6q25.1. For this analysis, we included 40 samples from our RNAseq discovery cohort and an additional 23 tumors from this trial (a total of 63 tumor samples) (Fig. 5B). We also screened several ER+ breast cancer cell lines and select long-term estrogen-deprived (LTED), HER2+, and triple-negative breast cancer cell lines (Fig. 5B). As positive controls, MCF7 and T47D cells were transduced with the ESR1-CCDC170 fusion (specifically, the variant fusing ESR1 exon 2 to CCDC170 exon 2, E2/C2) (Fig. 5B). Reverse transcription PCR (RT-PCR) was used to confirm the presence of any putative ESR1 fusions. The results of the nCounter analysis are summarized in Fig. 5B, and the results of nCounter-identified ESR1 fusion transcript validation are summarized in tables S7 and S8. Of 11 putative ESR1 fusion transcripts identified in six patient samples, 4 fusions were confirmed by PCR (36%). These fusions were the same ones validated after identification by RNAseq (table S5) and were only validated in endocrine-resistant or endocrine-intermediate tumors. Retrospective nCounter analysis revealed that these fusions were detectable above background levels in four of five of matched pretreatment FFPE tumor biopsies, suggesting that these fusions occur de novo (fig. S6C). Of 22 putative ESR1 fusion transcripts identified by nCounter analysis in 11 cell lines (including LTED, ESR1 fusion–transduced, and GFP-transduced cell lines), 16 fusion transcripts were verified by fusion-directed PCR (72%) (tables S7 and S8).

FGFR1 and CCND1 amplification–mediated resistance to estrogen deprivation is therapeutically actionable

The overall purpose of this presurgical study was to identify therapeutically actionable somatic alterations associated with resistance to estrogen deprivation in ER+ breast cancers. WES analysis identified amplification of WHSC1L1, FGFR1, CCND1, FGF3, FGF4, and FGF19 as the top amplifications associated with resistance to letrozole (table S1). WHSC1L1 is a histone methyltransferase that, when amplified, has been proposed as a driver of estrogen-independent ERα activity (28). At this time, however, WHSC1L1 may not be therapeutically actionable. On the other hand, aberrant signaling triggered by FGFR1 and its ligands or the interaction between cyclin D1 (CCND1) and CDK4/6 can be targeted with clinically available small-molecule inhibitors.

Therefore, to document a causal association of FGFR1 and CCND1 amplification/overexpression with therapeutic resistance, we modeled the effects of estrogen deprivation in CAMA1 ER+ breast cancer cells, which harbor coamplification of FGFR1 and CCND1 [cBio Cancer Genomics Portal, Cancer Cell Line Encyclopedia data set (29, 30)]. Upon stimulation with FGF3/19, CAMA1 cells proliferate in the absence of estradiol (Fig. 6A). These data suggest that activated FGFR1 and cyclin D1 can mediate persistent growth of ER+ breast cancer cells in the absence of estradiol, a scenario akin to that of the ER+ tumors treated with letrozole in our clinical trial. Phosphorylation of the tumor suppressor Rb by the cyclin D1–CDK4/6 complex uncouples Rb from E2F transcription factors. As a result, E2Fs induce transcription of genes necessary for the G1-to-S transition (31). To inhibit FGFR1 and/or cyclin D1 overexpression, we used the FGFR tyrosine kinase inhibitor lucitanib (32) and the CDK4/6 inhibitor palbociclib, respectively. Combined inhibition of FGFR1 and CDK4/6 in estrogen-deprived CAMA1 cells with lucitanib and palbociclib potently suppressed estrogen-independent cell growth more effectively than either single agent (Fig. 6A). To further demonstrate the ability of combined blockade of FGFR1 and CDK4/6 activity to overcome resistance to estrogen withdrawal, we calculated the combination index (CI) measuring drug synergy based on the Chou-Talalay method (33), where a CI < 1 reflects drug synergism, CI = 1 reflects an additive drug effect, and CI > 1 represents drug antagonism. The table of CI values (Fig. 6B) for CAMA1 cells treated with increasing doses of palbociclib ± lucitanib clearly demonstrated the synergistic effect of this combination on FGFR1/CCND1 coamplified cells. Lucitanib and/or palbociclib elicited a similar impact on the percentage of CAMA1 cells in the S phase as measured by flow cytometry (Fig. 6C and fig. S7A). Furthermore, only the combination simultaneously reduced FGFR1-mediated ERK1/2 activation and pRb expression in CAMA1 cells, thus restoring Rb activity and cell cycle arrest (Fig. 6D). To complement the studies in CAMA1 cells, we stably transduced ER+ MCF7 breast cancer cells with FGFR1 or cyclin D1 expression vectors (fig. S7B). 2D growth and proliferation assays showed that MCF7 cells overexpressing cyclin D1 were markedly resistant to estrogen deprivation, whereas MCF7 cells transduced with FGFR1 were modestly resistant (fig. S7, C and D). Collectively, these data support the idea that amplification/overexpression of FGFR1 and cyclin D1 at amplicons 8p11-12 and 11q13 promote resistance to estrogen deprivation.

Fig. 6. FGFR1 and CCND1 amplification–mediated resistance to estrogen deprivation is therapeutically actionable.

(A) FGFR1 and CCND1 coamplified CAMA1 breast cancer cells were seeded in estrogen-free medium supplemented with FGF3/19 (100 ng/ml) and treated with 500 nM of the CDK4/6 inhibitor, palbociclib (Palbo), and/or 1 μM of the FGFR1 inhibitor, lucitanib (Luc). When the untreated control cells reached 50 to 70% confluence (day 12), the cells were fixed and stained with crystal violet, and two-dimensional (2D) growth was quantified. The combination of palbociclib + lucitanib was sufficient to overcome resistance to estrogen withdrawal better than either single agent. ns, not significant. Further highlighting this effect, (B) presents the CI values calculated based on the average of the fold change in 2D growth with increasing combination doses of lucitanib (0 to 1000 nM) and palbociclib (0 to 500 nM) relative to untreated controls from three independent experiments. CI < 1 represents synergism, CI = 1 represents an additive effect, and CI > 1 represents antagonism. (C) S phase analysis of CAMA1 cells treated with FGF3/19 (100 ng/ml), ±1 μM lucitanib, and ±500 nM palbociclib reveals that treatment with palbociclib or the combination of lucitanib + palbociclib can overcome FGF-induced persistence of CAMA1 cells under estrogen-deprived conditions. Individual plots for each treatment condition can be found in fig. S7A. (D) Finally, immunoblot analysis of CAMA1 whole-cell lysates shows that only combined inhibition of FGFR1 and CDK4/6 can simultaneously decrease extracellular signal–regulated kinase 1/2 (ERK1/2) and Rb phosphorylation. P values represent results of two-way Student’s t test. Lucitanib = 1 μM and palbociclib = 500 nM. Comb, combination of lucitanib and palbociclib.

Profiling of serial breast tumor biopsies identifies 8p11-12 and 11q13 amplifications and ESR1 mutations as markers of endocrine therapy resistance

Finally, we performed targeted next-generation sequencing (FoundationOne assay) on tumors from a cohort of seven patients with metastatic ER+ breast cancer. These patients were treated with neoadjuvant chemotherapy followed by adjuvant endocrine therapy, and all recurred with metastatic disease within 5 years (Fig. 7A). For all seven patients, DNA was extracted from a diagnostic tumor biopsy, from the chemotherapy-resistant mastectomy specimen, and from the metastatic biopsy (referred to as pre-Rx, post-Rx, and Metastasis in Fig. 7B).

Fig. 7. Profiling of serial breast tumor biopsies suggests 8p11-12 and 11q13 amplifications and ESR1 mutations as markers of endocrine therapy resistance.

(A) Schema of the treatment course of seven patients with ER+ breast cancer from which pretreatment, postneoadjuvant chemotherapy, and metastatic recurrence biopsies were collected. Dashed lines represent time points at which biopsies were taken. A/C, adriamycin and cyclophosphamide. (B) Diagram of the landscape of genomic alterations for the seven patients as per targeted next-generation profiling by FoundationOne, including time to recurrence and previous treatment. The ESR1 mutations that were identified are specifically named.

Consistent with the above WES analysis, amplification of genes on 8p11-12 and/or 11q13 was frequent in this cohort, present at diagnosis in five of seven (71%) tumor sets. Among these, three of seven (43%) patients harbored FGFR1 amplification in all three metachronous biopsies, and two of seven (29%) exhibited CCND1 amplification in all three biopsies. Additionally, amplification of ZNF217 at 20q13.2, an amplicon associated with poor prognosis (34), was observed de novo in three patients who displayed rapid recurrence on adjuvant therapy. GATA3 truncations were detected in four of seven (57%) tumor sets across all biopsies. In general, most of the identified alterations were present in all three consecutive biopsies. Three tumor sets contained ERα LBD mutations (p.V422del, p.Y537S, and p.D538G), present only in the metastatic recurrence. Consistent with these data, the p.Y537S and p.D538G ERα mutants are rare in early breast cancers but detected in about 30% of ER+ metastatic breast cancers that progress on AIs (710). These mutants are transcriptionally active in the absence of estrogen (7, 9, 10, 35). To date, ERα p.V422del has only been reported in a single case of endometrial cancer in the COSMIC (Catalogue of Somatic Mutations in Cancer) database (36) and, most recently, by Toy et al. (35). Secondary analysis of FoundationOne-profiled breast tumors revealed 10 additional tumors that harbor ERα V422del (table S9). All 10 tumors were noted as metastatic, and the 6 tumors with available clinical data were all ER+/PR+ and HER2, similar to the tumor with ERα V422del in our study. To better understand the potential effects of the V422del mutant on ERα transcriptional activity, we transfected ER MDA-MB-231 and SKBR3 breast cancer cells with expression constructs encoding hemagglutinin-tagged wild-type, V422del, or Y537S ERα; ERα transcriptional activity was then measured with an ERα luciferase reporter. Cells expressing V422del exhibited twofold higher ERα transcriptional activity under estrogen-free conditions compared to cells transfected with wild-type ERα (fig. S8, A and B), as reported by Toy et al. (35).

DISCUSSION

The purpose of this study was to identify actionable alterations potentially causal to de novo endocrine resistance in ER+ breast cancers. To stratify patient tumor responses to estrogen deprivation with letrozole, we quantified pre- and posttreatment Ki67 and divided the post-Ki67 into tertiles based on the criteria delineated in the IMPACT trial (Fig. 1C and Table 2) (11). A similar approach has been reported by Ellis et al. (18), who labeled AI-sensitive tumors as those with a Ki67 index at or below 10% at surgery after 4 months of neoadjuvant antiestrogen therapy. Resistant tumors, with a Ki67 index of >10% at surgery, were associated with luminal B status and presence of patterns of somatic mutations mapping to cellular signaling pathways. Conversely, mutations in GATA3 and MAP3K1/MAP2K4 were associated with decreased proliferation upon treatment with letrozole. Another approach has been to treat patients with postmenopausal ER+ breast cancer for 2 to 4 weeks with an AI and rebiopsy the tumor at that time. Patients with tumors with a residual Ki67 of ≥10% were categorized as endocrine-resistant and switched to neoadjuvant chemotherapy. The rate of pathological complete responses among these patients was very low (37), suggesting that these tumors were also resistant to chemotherapy and underscoring the need to identify molecular drivers of estrogen-independent progression in ER+ tumors.

For patient tumors with a low baseline (pretreatment) Ki67 index (≤2.7% or ln ≤ 1.0), the post-letrozole 2-week Ki67 value may not be sensitive enough to provide a true reflection of drug action and hormone dependence. In the trial reported here, 6 of 134 (4.4%) tumors had a baseline Ki67 of ≤2.7% (ln ≤ 1.0). Three of those six were PR+ by AQUA, and in two of three, there was a clear reduction in PR upon treatment with letrozole (fig. S2B). In addition, we observed an overall decrease in PR expression after letrozole. These results suggest that the presence of active ERα signaling in tumors with a low pretreatment Ki67 can still be down-regulated by estrogen deprivation (fig. S2). These data also suggest that in low-proliferative, ERα-driven tumors, the baseline Ki67 index and/or the difference between baseline and posttreatment Ki67 index are not optimal biomarkers for the detection of a pharmacodynamic drug effect. Other surrogates of antiestrogen action are needed for these hormone-dependent cancers.

We did not find an association between PIK3CA activating mutations and endocrine resistance in this cohort of patients with newly diagnosed ER+ breast cancer (Fig. 2A and table S1). The clinical activity reported in trials of phosphatidylinositol 3-kinase inhibitors in combination with antiestrogens in patients with advanced ER+ breast cancer progressing on AIs (38, 39) may suggest that these mutations are a marker of acquired, but not de novo, resistance, as has been posited by previous studies [discussed in (40)]. We did, however, identify recurrent copy number amplifications in chromosome amplicons 8p11-12 and 11q13 (Figs. 2B and 3, table S1, and Table 3) (19, 20). We also observed that a large proportion of the resistant tumors harbored coamplification of CCND1 (11q13) and FGFR1 (8p11-12) (Fig. 3B and Table 3). FGFR1 amplification occurs in ~15% of patients with advanced ER+/HER2 breast cancer, in which it is associated with early relapse after adjuvant tamoxifen therapy (41). Simultaneous amplifications of CCND1, FGFR1, FGF3, FGF4, and FGF19 occur in 30 to 40% of breast tumors and are associated with reduced patient survival (19). Whereas CCND1 and FGFR1 singly amplified tumors exhibited increased expression of genes involved in cell cycle progression and mitosis, FGFR1/CCND1 coamplified tumors showed additional enrichment of cell cycle genes above that explained by amplification of either gene alone. This finding suggests a potential interaction between FGFR1 and cyclin D1 to drive estrogen-independent proliferation in coamplified tumors (Fig. 4 and fig. S5). Further strengthening the link between endocrine therapy resistance and FGFR1/CCND1 coamplification, a previously published gene signature associated with RB1 loss (25) scored higher in letrozole-resistant tumors and in coamplified tumors (Fig. 4, G and H). We also showed that cotargeting of FGFR1 and cyclin D1 signaling with the kinase inhibitors lucitanib and palbociclib, respectively, can synergistically inhibit proliferation of FGFR1 and CCND1 coamplified ER+ breast cancer cells under estrogen-deprived conditions (Fig. 6 and fig. S7). Thus, coamplification of FGFR1/CCND1 in ER+/HER2 tumors may help identify candidate patients for enrollment into investigational adjuvant combination therapy studies designed to prevent or delay cancer recurrence.

In line with previously published reports, we identified a potential association of ERα fusions with endocrine-resistant and intermediately resistant tumors (Fig. 5 and table S5) (26). We also observed an increase in total ESR1 coding transcript expression in samples with ERα fusions versus those without (fig. S6B). Veeraraghavan et al. (26) reported that ESR1 transcript fusions result in expression of truncated variants of the ESR1 5′ fusion partner (CCDC170). Overexpression of these CCDC170 truncated variants in nontumorigenic MCF10A mammary epithelial cells resulted in increased colony formation, migration, and invasion. Instead of overexpression of a truncated version of the fusion partner, we hypothesize that by relocating the ESR1 promoter upstream of genes not thought to be relevant for ERα-dependent signaling, this may release possible negative transcriptional regulation of the ESR1 coding exons. Testing of this hypothesis will require additional research beyond the scope of this study.

Finally, ERα LBD mutations have been found predominantly in patients with advanced ER+ breast cancer who have progressed on AIs (7, 9, 10). It is less clear whether they are associated with resistance to SERDs like fulvestrant or SERMs like tamoxifen (42). We found only one case with an ERα mutation, L429V, among newly diagnosed tumors in our trial. Retrospective profiling of metastatic recurrences in patients with ER+/HER2 breast cancer revealed three ERα LBD alterations (Y537S, D538G, and V422del), consistent with other reports that ERα mutations arise in late-stage disease (Fig. 7B). The Y537S and D538G mutants induce ligand-independent ERα transcriptional activity (710), but less is known about the V422del variant. From the FoundationOne profiling data, ERα V422del is clearly recurrent in metastatic ER+ disease (table S9). In addition to and consistent with a recent report (35), we showed that V422del activates estrogen-independent ERα transcription about twofold (fig. S8). Further experimental studies are required to fully understand the relevance of V422del for ERα signaling, disease biology, and response to different endocrine therapies.

In sum, we propose that ER+ tumors that remain highly proliferative upon short-term estrogen deprivation harbor molecular lesions causally associated with antiestrogen resistance. According to our study, the genomic landscape of ER+ tumors treated with short-term letrozole is similar to that of treatment-naïve breast tumors in TCGA. These data suggest that the potential for clonal selection during the 10- to 21-day letrozole time frame is minimal. Furthermore, some of the early alterations identified in this study are retained throughout the natural and treated history of these cancers and may mediate early recurrence. As such, we are confident that this presurgical clinical genomics platform can be used for the unbiased discovery of mechanisms of endocrine resistance that, in turn, can be tested in patients with advanced ER+ cancers of the same genotype. Conversely, tumors that exhibit exquisite inhibition of proliferation under this approach are highly hormone-dependent and may be candidates for adjuvant endocrine therapy alone. Finally, the presurgical approach reported here could be of use in hormone-dependent luminal B tumors that, despite high-risk clinical features at diagnosis, respond well to antiestrogens and may also be candidates for endocrine therapy alone.

MATERIALS AND METHODS

Clinical trial study design and tumor biopsies

To molecularly assess markers of response and nonresponse of early breast cancers to aromatase inhibition, tumor samples were obtained from postmenopausal patients with stage I to stage III operable ER+/HER2 breast cancer enrolled in a clinical trial with the AI letrozole administered for 10 to 21 days before surgery (Vanderbilt University NCT00651976) (43). Figure S1 contains a detailed illustration of the number of patients enrolled on this trial, the number of evaluable tissue samples, and, of those, the number available for the various molecular analyses performed in this study. At the time of this study, 157 patients were enrolled: 14 from Allegheny Cancer Center, 31 from Emory University’s Winship Cancer Institute, 5 from Surgical Associates, 98 from Vanderbilt-Ingram Cancer Center, and 9 from Instituto Nacional de Enfermedades Neoplásicas in Lima, Peru. Each patient provided written informed consent for tumor biopsies according to a protocol approved by the Vanderbilt-Ingram Cancer Center Institutional Review Board. Patients were required to complete a pill diary. Intraoperative or surgical specimen biopsies, both fresh-frozen and FFPE, were obtained from each patient’s tumor(s). A diagnostic (pretreatment) FFPE tumor was obtained for assessment of baseline ER, PR, HER2, and Ki67 by AQUA/IHC4.

Statistical analyses

Appropriate statistical analyses were applied to all comparisons here under the consultation of biostatistical experts (T.P.S., Y.S., and L.D.). Specific tests used are indicated within the manuscript text and/or the figure legend(s) and table(s) they support.

SUPPLEMENTARY MATERIALS

www.sciencesignaling.org/cgi/content/full/9/402/eaai7993/DC1

Materials and Methods

Fig. S1. Flowchart of tumor sample availability for study analyses.

Fig. S2. Immunohistochemical response of ER+ breast tumors to short-term letrozole, assessed by Ki67 and PR expression.

Fig. S3. Log2 copy number ratios on chromosomes 8 and 11 as per WES analysis.

Fig. S4. Individual FISH analysis for FGFR1 or CCND1 amplification and comparison with WES log2 CNV values and with RNAseq transcript expression.

Fig. S5. Principal components analysis, Database for Annotation, Visualization and Integrated Discovery (DAVID), Gene Set Enrichment (GSE), and iRegulon analyses based on RNAseq data from ER+, letrozole-treated breast tumors and TCGA breast tumors.

Fig. S6. RNAseq fusion validation pipeline and comparison of ESR1 transcript expression.

Fig. S7. Pharmacological inhibition of FGFR1 and/or CDK4/6 activity mediated by FGFR1 ± CCND1 amplification/overexpression.

Fig. S8. ERα transcriptional reporter assays assessing the activity of ERα LBD mutations.

Table S1. Statistical correlation of whole exome identified SNVs and CNVs with lack of response to letrozole.

Table S2. List of proliferation-associated genes (Excel file).

Table S3. Three hundred forty-six putative gene fusion transcripts identified by RNAseq in 50 tumors (Excel file).

Table S4. Primer sets for validation of RNAseq-identified putative fusion transcripts (Excel file).

Table S5. Summary and sequences of 26 RNAseq-identified, RT-PCR–validated fusion transcripts (Excel file).

Table S6. ESR1 fusion transcripts identified from TCGA RNAseq data.

Table S7. Validation and diagrams of putative ESR1 fusions in patient and cell line cDNA.

Table S8. Sanger sequences from RT-PCR–validated, NanoString-identified ESR1 fusion transcripts.

Table S9. Clinical characteristics of 10 breast cancer patients profiled by FoundationOne harboring the ESR1 c.1265_1267delTGG/ERα p.V422del alteration.

Table S10. Primers for PCR and direct sequencing of ESR1 fusions identified by NanoString in patient and cell line cDNA.

Table S11. Primer combination to validate NanoString-identified fusions in patient and cell line cDNA.

Table S12. CCND1 and FGFR1 cloning primers.

References (4453)

REFERENCES AND NOTES

Acknowledgments: We thank the patients for their participation in this trial. Funding: This study was funded by NIH Breast SPORE (Specialized Programs of Research Excellence) grant P50 CA098131, Vanderbilt-Ingram Cancer Center support grant P30 CA68485, Susan G. Komen for the Cure Foundation grant SAC100013 (C.L.A.), Vanderbilt Institute for Clinical and Translational Research Intramural Clinical and Translational Science Award VR2998, a grant from the Breast Cancer Research Foundation (C.L.A.), and a grant from Novartis (C.L.A.). T.P.S. was supported by NIH grant K08 CA148912. J.M.B. was supported by NIH/National Cancer Institute (NCI) R00 CA181491-01A1 grant, U.S. Department of Defense (DOD) grant BC131494, and Susan G. Komen Career Catalyst Research Award CCR 299052. S.C. was supported by U.S. DOD Breast Cancer Research Program award W81XWH-14-1-0359 and NCI Cancer Center Support Grant P30 CA08748. Author contributions: Trial design/conception: I.A.M. and C.L.A. Clinical research procedures/regulatory approvals: I.A.M., C.L.A., M.R., H.G., and J.M.G. Experimental study design/conception: J.M.G., K.E.H., T.P.S., J.M.B., and C.L.A. Data acquisition/analysis: All authors. Writing of manuscript: J.M.G., K.E.H., T.P.S., and C.L.A. Competing interests: The authors declare that they have no competing interests. Data and materials availability: RNAseq and WES data generated in this study can be accessed through dbGaP number phs000857.
View Abstract

Navigate This Article