The two stable TAMr MCF-7 human breast cancer cell lines have been described previously [9, 10]. All MCF-7 isogenic lines were cultured in phenol-red-free RPMI 1640 containing L-glutamine (Invitrogen, UK) supplemented with 5% charcoal-stripped steroid-depleted foetal calf serum (FCS; Harlan SeraLab, UK), 100 U/mL penicillin and 100 U/mL streptomycin and 100 nM 4-hydroxytamoxifen (TAM; Sigma-Aldrich, UK) for 12–24 months. Parental cells [termed wild type (WT) MCF-7] were cultured in the same medium, but with 0.01% (v/v) ethanol vehicle. Experiments were conducted in phenol-red-free RPMI 1640 supplemented with 5% charcoal-stripped steroid-depleted FCS + TAM (0.1–100 nM) or 17β-oestradiol (E2; 0.001-1 nM). Bi-monthly mycoplasma checks were consistently negative and annual short tandem repeat (STR) profiling confirmed cell provenance, both carried out as a service in Leeds. To minimize genetic drift, cell stocks were frozen at low passage and experimental cultures replaced from these stocks every 3–6 months. We also studied LCC1, -2 and -9 breast carcinoma cells, which were derived originally in the Clarke laboratory at Georgetown, Washington DC and cultured as described [7, 8, 22]. LCC1 cells are estrogen-independent and -responsive TAM-sensitive breast cancer cells, which are derived from an estrogen-independent variant of MCF-7 cells (MCF7/MIII) through in vivo selection in oophorectomized nude mice (with circulating estrogen levels similar to a postmenopausal woman), and subsequently re-cultured in vitro to become a stable cell line. LCC2 cells are stable, ER-positive, estrogen-independent, TAM-resistant, and respond to fulvestrant [7]. They were derived from the selection of LCC1 for TAM resistance in vitro. LCC9 cells exhibits cross resistance to TAM and are unresponsive to fulvestrant [7]. They were derived from the selection of LCC1 for fulvestrant resistance in vitro. MCF7:5C and MCF-7:WS8 are oestrogen-independent and dependent, respectively, ER-positive, progesterone receptor (PR)-negative, and TAM-resistant breast cancer cells [23]. They were derived from WT MCF-7 cells following long-term estrogen deprivation (LTED). The main characteristics of the cell lines used are detailed in Table 1.

Cells were seeded at 300, 000 cells per well in 6-well plates (Corning, UK). They were treated the next day with 1 nM E2 for the indicated durations. All experiments were performed in triplicate. Subsequently, cells were harvested in TRIzol (Life Technologies, UK) and total RNA was isolated using the Qiagen RNeasy RNA purification kit (Qiagen, UK). Isolated RNA (1 µg) was then processed on the Affymetrix Clariom S microarray platform (Affymetrix, UK) and used to determine the significance of differential expression of HPGD at basal levels.

Metabolic activity was used as a surrogate for cell proliferation, using the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay [9]. Cells were seeded into 96-well plates. After overnight attachment, TAM (0.1-100 nM), E2 (0.01–1 µM), PGE₂ (1 µM), butaprost or PGE₁-OH (both 0.1–1 µM) or ethanol vehicle was added. At 96 h post-treatment, cells were incubated with 1 mM MTT for 4 h. The purple formazan product was solubilised in propan-1-ol and read at 570 nm using an Opsys MR^TM 15 (Dynex Technologies, UK). Chemicals were purchased from Sigma-Aldrich (UK) or Cayman Chemical (USA).

This was conducted as described previously [24], except that HPGD was detected using a rabbit polyclonal antibody (catalogue number NB200-179, Novus Biological, UK, 1:500). Bands were then visualised with Supersignal West pico ECL reagent (ThermoFisher, UK).

Total RNA was extracted, and reverse transcribed as described previously [24]. Real-time polymerase chain reaction (RT-PCR) for HPGD or prostaglandin E₂ receptor 4 (EP4) mRNA was performed using a SYBR Green-based assay (ThermoFisher, UK) on an ABI 7700 sequence detection system using primers for HPGD (forward, 5’-TAGTTGGATTCACACGCTCAGC-3’; reverse, 5’-AAAGCCTGGACAAATGGCAT-3’) and EP4 (forward, 5’-TCTTACTCATTGCCACCTCCCT-3’; reverse, 5’-CTTGGCTGATATAACTGGTTGACG-3’). The ribosomal protein gene RPLPO was used as the reference gene (forward, 5’-GAAACTCTGCATTCTCGCTTCC-3’; reverse, 5’-GATGCAACAGTTGGGTGCCA-3’). Levels of HPGD or EP4 transcripts were quantified using the 2-ΔCt method [25].

A stab culture of human HPGD cDNA IMAGE clone ID 3638799 was obtained from MRC and HPGD was PCR-amplified from this using the following primers: forward, 5′-CCGGGATCCTGCACCATGCACGTGAAC-3′; reverse, 5′-CCCCAAGCTTTCATTGGGTTTTTGCTTG-3′. Products were cloned into pcDNA3.1(-)myc/his (gift from Dr Thomas Hughes, University of Leeds). Vectors containing HPGD or empty vector controls were transfected into TAMr cells using Lipofectamine 2000 (ThermoFisher, UK). Stable clones were selected and maintained in G418 (ThermoFisher, UK)-containing medium (250 µg/mL).

Both assays were performed using previously protocols published [26]. For PGE₂, serum-free conditioned medium was collected from cells after overnight incubation, and PGE₂ levels were measured using a competitive immunoassay (Amersham Biosciences, UK), and normalised to total cellular protein content as measured by Bradford assay. HPGD activity was measured in lysates of WT MCF-7 and TAMr cells by assessing the transfer of tritium from ³H-PGE₂ to glutamate, catalysed by HPGD and glutamate dehydrogenase [27].

Small-interfering (si)-RNAs targeting EP4 or HPGD (ThermoFisher Scientific, UK) were reverse transfected into cells using Lipofectamine 2000 according to the manufacturer’s instructions. Briefly, MCF-7 cells were mixed with media containing small-interfering RNA (siRNA)/lipid (final siRNA concentration 10 nM) and seeded into 96 well plates at 0.4 x 10⁴ cells/well and allowed to establish overnight before incubation with drug or vehicle control. Response to treatment was assessed by MTT assay. EP4 or HPGD knockdown was assessed by real-time quantitative polymerase chain reaction (qRT-PCR) as described previously [28].

Following ethical approval (06/Q1206/180) we investigated HPGD expression in 350 primary invasive breast cancers, all of which, had been surgically resected and received adjuvant TAM, and were represented on tissue microarrays (TMAs) [24, 29]. Out of the 350 cases, 108 cases experienced a relapse (TAMr) and 242 cases did not [tamoxifen-sensitive (TAMs)]. Mean follow-up was 80 months [range 1–229, standard deviation (SD) 44.2]. Tumour staining was achieved using previously described methods [24, 26].

The relationship between HPGD expression and cancer outcomes was analyzed using Kaplan-Meier Plotter (KMplot, https://kmplot.com/analysis/) [30]. Using the 2017 release of the database, a cohort of ERα-positive breast tumours, previously treated with TAM only and no adjuvant chemotherapy, was analyzed and the association between HPGD expression and overall survival (OS). Patients were dichotomized as high or low HPGD expression using lower tertile as a cut-off. Multivariate analysis was done using in-built software with Ki67, ER and HER2 as covariates.

We applied miRWalk version 3.0 [31] and TargetScan version 7.2 [32] to a previously described Affymetrix miRNA dataset of endocrine sensitive MCF-7:WS8 and resistant MCF-7:5C cells [33] in order to identify miRNA-target interactions sites for major HPGD transcripts using default settings. OncoLnc (http://www.oncolnc.org/) was used to retrieve HPGD and miR-3200-3p expression data from matching breast cancer cases in the TCGA database [34]. Kaplan-Meier Plotter [30] was used to investigate the significance of miR-3200-3p on survival in ERα+ breast cancer treated with any endocrine therapy (not limited to TAM). Cases were dichotomised as high or low miR-3200-3p expression using the median expression value as a cut-off.

For in vitro analyses, one-way ANOVA was performed. The log rank test was used to compare patient survival in the primary breast cancer cohort. Analyses were performed using GraphPad Prism version 7.03 (GraphPad Software, La Jolla California, USA). For Affymetrix analyses, ANOVA P-values were calculated using the Affymetrix Transcriptome Analysis Console (Affymetrix). The time course plots were plotted as fold-change with the baseline value set at 1.

We demonstrated that HPGD was downregulated in two independent TAMr MCF-7 cell lines, which had been cultured continuously in 100 nM TAM for 12 (TAMr 1) and 21 (TAMr 2) months [10], respectively by Western blot (Figure 1a) and real time PCR (Figure 1b) compared with parental MCF-7 cells. Microarray analysis of the independent LCC cell line series and MCF7:5C cells showed that only the oestrogen-independent but oestrogen-responsive TAMs cell line LCC1 showed increased expression of HPGD in response to E2 over time (Figure 1c).

We then explored the effects of introduction of HPGD into MCF-7 TAMr 2 cells. Successful expression of HPGD protein following stable transfection was confirmed by Western blotting (Figure 2a). This restored activity of HPGD in each of the transfected cell lines to levels like WT MCF-7 (Figure 2a and b). This also sensitized these cells to the inhibitory effects of TAM to approximately 60% of the parental MCF-7 cell response (Figure 2c), whilst restoring sensitivity to E2 almost completely (Figure 2d).

Since HPGD is responsible for the biological inactivation of prostaglandins including PGE₂, we tested whether increased PGE₂ signaling might explain TAMr. We first examined PGE₂ levels in serum-free conditioned medium collected from MCF-7 and MCF-7 TAMr 2 cells. TAMr cells released higher amounts of PGE₂ compared with parental MCF-7 cells and this could be reduced to levels approaching that of MCF-7 by HPGD overexpression in these cells (Figure 3a). Next, we tested the effect of adding exogenous PGE₂ (1 µM) on TAM response in two different ERα-positive breast cancer cell lines, MCF-7 and T47D. PGE₂ reduced sensitivity of both cell lines to TAM (Figure 3b). As PGE₂-signalling is mediated by G-protein coupled receptors, we examined if the stimulatory EP2 or EP4 receptors might be responsible for mediating the effects of PGE₂ on TAM sensitivity. We demonstrated that PGE₁-OH (an EP4 agonist) was able to mimic the effects PGE₂ in decreasing TAM sensitivity (Figure 3c; P < 0.015), albeit to a lesser extent than PGE₂. However, butaprost (EP2 agonist) had no effect at equivalent concentrations (Figure 3d). This suggested that the EP4 receptor might be responsible for mediating the effect of PGE₂ on TAM sensitivity. Partial silencing of EP4 in WT MCF-7 cells was achieved using one siRNA (Figure 3e). EP4 silencing using siRNA 1 inhibited PGE₂ induced TAM resistance in these cells (Figure 3f).

We analysed the relationship between HPGD expression and clinical outcome for breast cancer patients using the on-line resource Kmplot [30] and from our own cohort of breast cancer patient’s data in separate survival analyses. In the Kmplot cohort, we found that ER+ patients who received adjuvant TAM only and had ‘low’ expression of HPGD, had reduced OS by univariate analysis [hazard ratio (HR) = 0.28 (0.13–0.59), P = 0.0007] (Figure 4a). This remained significant on multivariate analysis (P = 0.001, Figure 4b). We then conducted a retrospective immunohistochemical study of HPGD expression in our own cohort of breast patients, who all received adjuvant TAM therapy and related HPGD expression to survival. As these TMAs have been used extensively in several other studies [24, 29, 35], the number of viable cores that could be reliably evaluated was only 144 out of 350 cases. The level of HPGD staining was generally weak to moderate (Figure 4c) in positive cases, however cases which completely lacked HPGD were associated with worse OS [HR = 0.3 (0.15–0.69), P = 0.047] (Figure 4d), in line with the in silico analysis (Figure 4a). The small number of TMA cases available precluded multivariate analysis.

With growing evidence that HPGD is inactivated epigenetically, we sought to identify miRNAs that may post-transcriptionally modulate HPGD expression. We interrogated a global miRNA data set established in an independent TAM resistance model MCF-7:5C [33]. These TAMr cells were derived from WT MCF-7 cells (Table 1). In silico analysis revealed numerous miRNAs that were differentially expressed in TAMr MCF-7:5C cells compared with TAMs MCF-7:WS8 cells and that are predicted to interact with the three prime untranslated region (3′-UTR) of the HPGD transcripts (Table 2).

We then tested the association between each of the listed miRNAs and HPGD mRNA levels in 987 breast cancer samples from the TCGA database using OncoLnc [34] where both miRNA and mRNA expression data were available. Only one miRNA, namely miR-3200-3p, had an inverse relationship between its expression levels and HPGD mRNA levels in this dataset (Spearman, ρ = –0.25, P < 0.0001). We used the Kmplot online resource [30] to analyse the association between miR-3200-3p expression and OS in ER+ breast cancer patients treated with any endocrine therapy, irrespective of grade, molecular subtype, and lymph node status. High miR-3200-3p expression was associated with significantly reduced OS (Figure 5a) and a weak negative correlation (R² = 0.25) was observed between miR-3200-3p and HPGD (Figure 5b) using data from TCGA analysed using the OncoLnc platform [34].