Background
Over 80 % of breast cancers (BCs) express estrogen receptor alpha (ER) at primary diagnosis. ER is a transcription factor that controls proliferation and survival by binding and activating estrogen response elements (ERE) on target genes controlling proliferation and survival. Current clinical strategies include the use of endocrine agents, which inhibit estrogen (E) signalling either by blocking the conversion of androgens to E in the case of aromatase inhibitors (AIs), by directly antagonizing ER function with agents such as tamoxifen, which compete with E for the ER and results in the recruitment of nuclear corepressors, or by use of drugs such as fulvestrant (ICI182780), which targets ER for proteasomal degradation [
1‐
4].
Despite the efficacy of endocrine therapies, many patients relapse with either
de-novo or acquired resistance [
4‐
6]. Preclinical and clinical data support cross-talk between ER and growth factor receptor pathways, such as IGF1R and ERBB2/HER2 [
7‐
10], which can lead to ligand-independent activation of the ER or can alter the phosphorylation state of nuclear co-activators, thereby changing the balance of ER transcription factors and potentiating transcription [
11]. Despite this knowledge, few clinical trials have shown benefit from the targeting of endocrine resistance using signal transduction or receptor tyrosine kinase inhibitors. One explanation for this is the complexity/heterogeneity of the tumour background and the lack of definitive biomarkers. Data from large studies such as The Cancer Genome Atlas (TCGA) indicate that other than a small number of high-frequency mutations, such as
TP53,
PIK3CA and
GATA3, which have little association with endocrine resistance [
12], primary ER+ BC shows a very low frequency of individual mutations, making targeting difficult. In contrast, expression profiling of primary ER+ BC samples has identified several promising signatures/networks for targeting [
13].
Previously, using global gene expression data from patients treated with neoadjuvant anastrozole, we showed that certain gene signatures such as IGF-1, MAPK and obesity were associated with poor response to therapy [
13]. In order to identify common adaptive pathways associated with acquired resistance to AIs, we derived a panel of cell lines modelling resistance to E-deprivation and analysed these using both high-throughput transcriptomic and proteomic data. The cholesterol biosynthesis pathway was identified as a common adaptive mechanism only in models that retained ER at the point of resistance. Of particular note, the oxysterols 25-hydroxycholesterol (HC) and 27-HC were shown to influence ER transcriptional activity via recruitment to endogenous E-regulated genes. Furthermore,
in-silico interrogation of data from two separate patient cohorts treated with neoadjuvant AIs or adjuvant tamoxifen showed that genes identified within our
in-vitro models encoding enzymes within the cholesterol biosynthesis pathway were associated with poor outcome. Overall, these data provide further links between obesity and BC risk.
Discussion
Despite the success of AIs in treating patients with ER+ BC and reducing their risk of dying from the disease, resistance remains a significant problem [
50]. To study novel mechanisms of resistance to E-deprivation on an AI, we generated a panel of cell line models with varying genetic backgrounds and interrogated alterations in both their global gene transcriptome and proteome. Our main focus was to identify common adaptive mechanisms of resistance. Using pathway analysis, we showed a concordant upregulation of the cholesterol biosynthesis pathway in our LTED models that retained ER expression but not in those that lost expression of the steroid receptor.
Evidence suggests both primary and recurrent tumour cells have a high demand of lipids in order to proliferate and metastasize [
44]. For instance, activation of the PI3K/AKT/mTOR pathway is strongly associated with lipogenesis [
44,
51,
52], as well as accumulation of cholesteryl esters in various cancers via
SREBP1 and
LDLR activation [
41]. In this setting, hyperactivation of PI3K/AKT/mTOR pathway activates SREBP1, thereby potentiating esterification and compartmentalization of cholesterol into lipid droplets, allowing uptake of fatty acids and leading to increased proliferation [
42]. As 40 % of BCs harbour activating mutation/amplification of
PIK3CA or loss of
PTEN, a feature modelled in our ER+ LTED cell lines, we assessed mRNA levels of
SREBP1,
LDLR and
HMGCR. However, expression at the transcript or protein was either undetected or downregulated. Additionally,
ACAT1, which is important for cholesterol esterification and prevents cellular toxicity of excess cholesterol, was less abundant in the LTED at both the mRNA and protein level. Furthermore, the levels of esterified cholesterol were not significantly different between wt and LTED cells (data not shown), suggesting this axis is less important in the ER+ LTED phenotype.
There is compelling evidence for the role of cholesterol metabolites in promoting tumour growth [
45,
47,
53,
54] and acting as endogenous SERMS [
27,
46]. In particular, it has been hypothesized that 27-HC may be the primary biochemical link between lipid metabolism and cancer [
55]. We showed that both 25-HC and 27-HC were elevated in our ER+ LTED models compared with their parental controls. Of note, addition of both oxysterols promoted ER-mediated transcription and enhanced recruitment of ER to EREs on both
TFF1 and
GREB1, two E-regulated genes. Furthermore, this process was antagonized by fulvestrant, suggesting a mechanism dependent on a functional ER-transcription axis. We hypothesized that both 25-HC and 27-HC may substitute for E2 in the LTED setting, a notion supported by our
in-silico analysis, which showed that both oxysterols were capable of binding with the LBD located in the AF2 domain of ER.
Obesity and lipogenesis have been associated with increased BC risk and a worse outcome as a result of increased levels of adipocyte-secreted endocrine factors and insulin-like growth factors (IGFs) [
56,
57]. Previously, we showed that gene signatures modelling both obesity and IGF signalling predicted a poor anti-proliferative effect of anastrozole [
13]. To assess whether these enzymes within the cholesterol biosynthesis pathway had relevance for clinical response or prognosis of ER+ BC treated with endocrine therapy that were upregulated in our ER+ LTED models, we interrogated
in-silico data from patients treated with neoajuvant AI therapy or adjuvant tamoxifen. These analyses revealed that higher expression of four of the nine upregulated genes (
MSMO1,
EBP,
LBR and
SQLE) correlated with poor response to AI therapy judged either by clinical response [
34] or by the validated response biomarker Ki67 [
12,
35], but only
SQLE associated with poor response to endocrine therapy or poor long-term outcome on such therapy or both. Previous studies have shown that
SQLE is amplified in 10 % of BC [
58] and that high expression is associated with poor prognosis in early-stage ER+ BC [
59]. Studies have shown that the 8p11-p12 chromosomal region is a hotspot for genomic aberrations and that co-amplification with the
MYC oncogene (8q24.21) as well as altered transcriptional patterns (hypomethylation) of genes spanning 8q12.1-q24.22, which includes
SQLE, is associated with more aggressive tumours [
60]. Taken together, this would suggest
SQLE is indicative of poor prognosis while the other cholesterol biosynthesis genes associate more strongly and specifically with poor response to AI therapy. Of note, some genes involved in cholesterol biosynthesis are already incorporated in clinically relevant gene signatures. For instance, DHCR7, which was more abundant in our MCF7 LTED cells (1.25-fold,
p < 0.001) but did not meet our stringent selection criteria, forms part of the eight-gene EndoPredict signature [
61].
Acknowledgements
The authors thank The Breast Cancer Now Toby Robins Research Centre for generous funding. They also acknowledge NHS Trust funding to the Royal Marsden Hospital’s NIHR Biomedical Research Centre. The authors thank Dr Faraz Khosravi Mardakheh for providing the OGE system. They also thank Professor Elsa Lundanes and Dr Steven Ray Haakon Wilson for providing the platform for the LC-MS/MS analysis of oxysterols.