Background
Hormone receptor-positive breast cancer has served as a prototype for targeted therapy due to the well-established efficacy of estrogen deprivation. Largely because of these approaches, breast cancers are somewhat unique in that recurrences can occur years, sometimes decades following the primary diagnosis [
1‐
4]. Given that the majority of patients receive long-term maintenance regimens of either a selective estrogen receptor modulator (SERM) or aromatase inhibitor (AI), recurrent breast cancers are often classified as estrogen-independent given their ability to thrive in an estrogen-deprived environment. Identifying the biological mediators that allow breast cancer cells to bypass their dependence on estrogen is a crucial step in understanding advanced breast cancer biology and defining novel therapeutic targets.
Defining these molecular processes in patient samples, however, has been challenging because of the logistics in obtaining well-characterized, longitudinally collected biospecimens. Nevertheless, shared features of more advanced breast cancers have emerged, such as relapsed tumors losing expression of ER and over 20% of metastatic ER-positive breast cancers acquiring mutations in
ESR1 that confer ligand-independent signaling [
5‐
7]. Other largely accepted mechanisms of estrogen-independence are bypass activations of mitogenic pathways such as MAPK and PI3K through initiating FGFR, EGFR, and IGF signaling and exploitation of the Rb-CDK-E2F axis [
8‐
12]. Less well validated, more recently discovered mechanisms include
ESR1 fusions and amplifications [
13,
14].
Recent studies analyzing multiple, longitudinally collected, pre- and post-treatment samples have shown clonal evolution and selection in the context of targeted therapies [
15‐
18]. Similar work analyzing hormone receptor-positive breast cancers have largely been restricted to short-term pre-/post-neoadjuvant therapy analyses [
19‐
22]. One of the most comprehensive genomic studies of this type was a multi-platform effort that characterized the clonal architecture of tumors after 4 months of AI therapy [
23]. Although drastic clonal remodeling was observed at the DNA level, few recurrent resistance mechanisms were appreciated. A more recent, large-scale study showed activating
ERBB2 mutations, MAPK activation, and NF1 loss as mechanisms possibly driving endocrine resistance—with some of these alterations being confirmed in subsequent studies [
24‐
27]. The majority of this work has notably been performed on metastatic tissues—whether or not some of these changes occur locally as a result of estrogen independence before distant spread is unknown.
Thus, to better define both DNA and transcriptional changes that occur in long-term estrogen-independent tumors, we undertook a targeted analysis of DNA/RNA alterations in ~ 1400 cancer genes in 12 paired primary and locoregional recurrences from patients with ER-positive breast cancers that were documented as being treated with estrogen-depleting therapy. The median time to recurrence was 3.7 years, with the longest time to recurrence being over 7 years.
Discussion
In this study, a targeted RNA/DNA analysis of approximately 1400 cancer genes in ER-positive primary breast cancers and matched long-term, endocrine therapy-treated local recurrences was performed. We found general conservation of transcriptional and copy number profiles among the majority of samples—suggesting that even after 7 years of dormancy and the selective pressures of therapies, locally recurrent breast cancers generally retain their intrinsic molecular features. An analysis of recurrence-enriched SNVs revealed limited recurrent mutation events, yet notable “n-of-one” mutation selection was observed—such as case ERLR_01 which showed three distinct, recurrence-enriched PIK3CA mutations. The most striking changes in long-term estrogen-deprived tumors, however, were highly recurrent (up to 42%), outlier expression changes. An analysis of tumors with the most recurrent outlier loss, ESR1, revealed concurrent upregulation of genes typically expressed in basal breast cancers, such as PROM1, KLK7, and NDRG1, suggesting a selection of a more basal-like phenotype in endocrine-resistant disease. Our data showing similar CNA profiles argue against the outgrowth of a distinct ER-negative subclone but instead suggest possible epigenetic, transcriptionally driven remodeling under antiestrogen pressures.
Nearly all recurrences are more similar transcriptionally to their matched primaries than to other, long-term estrogen-deprived tumors—reinforcing the notion that advanced cancers generally retain their core transcriptional programming, even after nearly a decade of dormancy [
26‐
29]. Furthermore, amplifications and deletions of recurrences are markedly similar to primaries, supporting recent evidence from breast cancer single-cell sequencing that structural variation is likely an early event and many CNAs, even in metachronous therapy-resistant tumors, may be shared by the majority of subclones [
57]. An important exception to this conservation was ERLR_03_R1, a recurrence with a completely unique transcriptional and copy number profile than its matched primary. Evidence has emerged of so-called collision tumors, whereby two synchronous, distinct cancers can merge anatomically and only under the selective pressures of therapy or through deep sequencing, their individuality can be unmasked [
23,
58]. Indeed, this “recurrence” switched to ER-negative/HER2-positive from ER-positive/HER2-negative clinically and thus could represent a different cancer than the primary—although the level of shared SNVs suggests some degree of clonal relatedness.
Limited shared, non-silent SNVs were discovered in these specimens, with
AKAP9 and
KMT2C being the only two genes that harbored recurrence-enriched mutations in greater than one case. These mutations are not in a conserved functional domain nor in a hotspot location, making it difficult to assess their pathogenic roles.
AKAP9 and
KMT2C also encode relatively large gene products (3911 and 4911 amino acids, respectively) which may increase the likelihood of obtaining a passenger mutation by chance. Nevertheless,
KMT2C and other lysine methyltransferases have been implicated in breast cancer pathology, argued as potential drivers in large-scale sequencing studies of primary tumors and
KMT2C mutations specifically may confer hormone therapy resistance in breast cancer models [
59‐
61]. Case ERLR_20 harbored an enriched nonsense mutation in
ARID1A.
ARID1A alterations are associated with more unfavorable tumor features in breast cancer and have recently been shown to determine luminal identity and therapy response in ER-positive tumors—consistent with the more basal-like transcriptional features we observe with
ESR1-depleted recurrences [
62‐
65]. A single recurrent cancer (ERLR_01_R1) showed enrichment of three somatic hotspot
PIK3CA mutations (E542K, Q546K, E726K), suggesting strong MAPK signaling selection within that particular tumor and coincident with recent reports of multiple mutations occurring in individual cancer genes in advanced cancers [
66]. SNVs within genes that act as corepressors and coactivators, some with direct influences on estrogen receptor-mediated transcription, were found to be enriched in recurrences—such as
NCOA1,
NCOR2,
FRYL, and
CREBBP—along with transcription factors including
PAX5,
FOXO1, and
TP53. Notably, we did not observe any
ESR1 mutations unlike other studies on locoregional recurrences [
67]—likely due to our small sample size. Interestingly, this study reported lower frequency of
ESR1 mutations in locoregional recurrences versus advanced metastases at an AF > 1% and recent data has emerged regarding a pro-metastatic phenotype of
ESR1 variants [
68]—suggesting locoregional recurrences may have a lower frequency of
ESR1 variants versus distant disease. We also observed a positive correlation between the frequency of acquired, non-silent SNVs and disease-free survival—validating the concept that surviving cancer cells after initial therapy acquire potentially pathogenic mutations as they lay dormant and undetectable over time.
Given the heterogeneity of clinical specimens makes it difficult to rely on typically used differential expression workflows—since resistant mechanisms of individual tumors may be distinct—we undertook an analysis of patient-specific outlier expression gains and losses to identify more extreme transcriptional reprogramming events within individual cases that may be driving estrogen independence. Surprisingly, unlike SNVs, recurrent outlier transcriptional gains and losses were quite common
. Particularly compelling outlier events included recurrent gains within shared pathway members, such as near mutually exclusive upregulations of
NTRK3 (
n = 5 [42%]) and
NTRK2 (
n = 4 [33%]). Notably, activation of
NTRK’s mediates downstream signaling pathways typically associated with breast carcinomas, including PI3K and MAPK, and small molecule inhibitors of this family are showing promising results in recent solid tumor trials [
69]. Other notable pathway member changes included loss of Wnt antagonists
SFRP2 (
n = 3 [25%]) and
SFRP4 (
n = 4 [33%]).
SFRP2 is hypermethylated and silenced in a subset of breast cancers [
70] and experiments in model systems have shown cross-talk between ER and Wnt signaling that may mediate endocrine therapy resistance [
71,
72]. Other recurrent gains included
FGFR4 (
n = 4 [33%]),
TERT (
n = 3 [25%]), and
CCNE1 (
n = 3 [25%])
—particularly relevant given the recent success of CDK inhibitors in hormone-positive disease and the burgeoning use of
FGFR inhibitors against solid malignancies as we and others have reported [
31,
73].
The most recurrent outlier expression loss was
ESR1, which was diminished in 42% of long-term estrogen-deprived local recurrences. Interestingly, the loss of
ESR1 for the majority of cases was not associated with a dramatic change in the tumors’ transcriptional profile. To further explore this counterintuitive result, given
ESR1 is a master regulator of transcription and a driver of luminal breast cancers, we identified genes that were consistently altered in
ESR1-depleted recurrences. The most substantial gains in
ESR1-depleted tumors are genes generally expressed in basal breast cancers—such as
NDRG1,
DKK1,
KIT,
KLK7,
PROM1, and
COL9A3—and genes significantly lost in the
ESR1-depleted subset are generally downregulated in basal cancers—
EVLOVL2,
BCL2,
IGF1R,
MYB,
RABEP, and
ATP8A2 (MsigDB: SMID_BREAST_CANCER_BASAL_DN/UP gene lists) [
74]. These results reveal a common, novel, and distinct
ESR1-depleted subtype of advanced breast cancers that acquire basal-like transcriptional reprogramming. Prior studies have hinted that luminal B tumors, which are known to portend worse outcomes, generally have lower expression of
ESR1 and endocrine-resistant tumors have been shown to have decreased
ESR1 expression relative to matched primary tumors [
75]. The mechanisms driving this loss as well as the
ESR1-independent maintenance of a luminal cell-state with basal-like characteristics will be essential to unravel. Interestingly, prior studies have shown that intrinsic molecular subtypes of breast cancers generally remain consistent in recurrent or metastatic tumors, yet here we see a more nuanced gain of basal-like features in luminal tumors [
76‐
78].
The greatest fold-change difference in
ESR1-depleted recurrences was the upregulation of
PROM1.
PROM1 is a marker for tumor-initiating cancer stem cells and plays a key role in determining ER-positive luminal cell fate during differentiation from multipotent stem cells [
56], suggesting long-term endocrine-deprived breast cancer cells may enrich themselves with stem-like progenitors to achieve estrogen independence. Indeed,
PROM1 has been shown to mediate endocrine therapy resistance in breast cancer models through IL6/Notch3 signaling [
79,
80]. Here, we show that a large portion of long-term endocrine-resistant breast cancers may be exploiting this transcriptional reprogramming. Finally,
NDRG1, also significantly upregulated in
ESR1-depleted recurrences and generally expressed in basal cancers, showed differential expression in three distinct LTED cell lines.
NDRG1 is a suspected metastasis suppressor gene. Counterintuitively, we see upregulation of this gene in resistant disease and show increased expression confers worse survival outcomes in ER-positive primary tumors [
81]. Further functional studies assessing the mechanistic and biological consequences of these transcriptional reprogramming events both in locoregional and metastatic disease will be essential.
A pertinent point these results raise is the benefit of integrating longitudinal, targeted RNA sequencing to inform resistance mechanisms and therapeutic targets in breast cancers. In this study, we find limited DNA-level enrichments yet highly recurrent, acquired transcriptional remodeling events from primary to advanced cancers, including a few of which that are immediately targetable such as NTRKs, FGFR4, and CCNE1—although this study was limited by the small number of patient-matched cases and targeted panel of genes. Nonetheless, this work challenges our lack of emphasis on RNA-level changes, particularly those that can be elucidated from longitudinal biopsies, in clinical profiling of tumors and future work should be geared towards deciphering which of these bypass transcriptional programs may be druggable.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.