Background
Transcription factors are the integrators of multiple signaling pathways, converting internal and external stimuli into changes in gene expression. Through this role, the evolutionarily conserved E26 transforming sequence (ETS) transcription factor family controls fundamental cellular processes such as proliferation, differentiation, and apoptosis [
1]. The 28 members of the human ETS family are characterized by an ETS DNA-binding domain that recognizes a core GGAA/T motif. Additional specificity of ETS domain binding is conferred by the amino acids surrounding the key residues, as well as by posttranslational modifications and interactions with other proteins [
2,
3]. Given the vital cellular processes regulated by ETS transcription factors, it is not surprising that they have also been identified as significant contributors to tumorigenesis [
4].
E74-like factor 5 (ELF5) is an epithelial-specific member of the ETS transcription factor family [
5,
6]. In addition to the ETS domain, the full-length ELF5 protein contains an N-terminal Pointed (PNT) domain (83 amino acids) that is similar to the evolutionarily conserved sterile alpha motif (SAM) domain. In humans, the SMART database [
7] identifies 96 SAM/PNT domain-containing proteins, 11 of which are ETS family members. SAM domains have diverse functions, including protein–protein interactions, polymerization, kinase docking, RNA binding, and lipid molecule interactions [
8‐
11]. The ELF5 PNT domain has been shown to have strong transactivation activity [
12]; however, the mechanisms underlying this activity (for example, protein–protein interactions or posttranslational modifications) are unknown.
A critical function of ELF5 is the regulation of cell fate, beginning with specification of the trophectoderm in the blastocyst [
13]. Correct spatial and temporal ELF5 expression is also important for normal development of the embryonic lung [
14]. In the mammary gland, prolactin- and progesterone-driven ELF5 expression during pregnancy directs the development of the luminal progenitor cells into estrogen receptor-α (ER)- and progesterone receptor (PR)-negative milk-producing cells [
15]. In normal human tissues, ELF5 is reported to be expressed in the kidney, prostate, lung, mammary gland, salivary gland, placenta, and stomach [
5,
6,
16].
More recently, there has been increasing interest in the role of ELF5 in cancer. ETS factors are frequently deregulated in cancer through diverse mechanisms, including gene fusions, alterations in localization and/or activity, amplifications, increased expression, and (less commonly described) decreased expression [
4]. ELF5 was originally described as a tumor suppressor [
5]; however, the role of this protein in cancer is complex and context-dependent. In prostate cancer, for example, ELF5 has been shown to inhibit transforming growth factor (TGF)-β-driven epithelial–mesenchymal transition by blocking phosphorylation of the TGF-β effector protein SMAD3 [
17]. Conversely,
ELF5 mRNA has been shown to be upregulated in a cell line model of prostate cancer progression involving acquisition of androgen independence [
18]. Bladder and kidney carcinoma have been associated with loss of ELF5 expression at the protein and RNA levels [
19,
20], whereas in endometrial carcinoma
ELF5 upregulation is associated with higher disease stage [
21].
ELF5 gene rearrangements have been described in several lung cancer cell lines [
5], and the authors of a recent case study described a
ZFPM2-ELF5 fusion gene in multicystic mesothelioma [
22]; however, gene fusions do not appear to be a major mechanism for deregulation of ELF5, in contrast to other ETS factors, such as
TMPRSS2-ERG
/ETV1 fusions in prostate cancer [
23].
The breast is the most well-studied context for the role of ELF5 in cancer, with microarrays showing increased expression in basal-like subtypes and decreased expression in luminal A/B and Erb-b2 receptor tyrosine kinase 2 (HER2)-overexpressing subtypes [
24,
25], suggesting subtype-specific effects. Transient ELF5 expression in cell line models reduced proliferation, invasion, ER -driven transcription and epithelial–mesenchymal transition [
25,
26]. However, sustained increased ELF5 expression in some contexts is associated with disease progression, such as in endocrine-resistant breast cancers, reliant on elevated ELF5 for growth in cell line models, and the basal-like subtype of breast cancer [
25]. This illustrates the complexity and contextual dependence of transcriptional regulation.
It is becoming increasingly recognized that almost all multiexon genes undergo alternative transcription (such as alternative transcription start or termination sites) and/or alternative exon splicing, increasing diversity of protein structure and function [
27]. Alternative transcription events are also commonly deregulated in cancer, contributing to tumor initiation and progression but also providing potential cancer-specific therapeutic targets. Importantly, different isoforms produced by the same gene may have very different functions. One striking example is vascular endothelial growth factor, which produces both proangiogenic and antiangiogenic isoforms [
28]. Early studies described tissue-specific differences in
ELF5 transcript isoform expression [
6], but recent studies have not distinguished between isoforms or have used a single isoform for overexpression studies.
This study represents the first comprehensive analysis of ELF5 expression at the isoform level, using RNA-sequencing (RNA-seq) data from The Cancer Genome Atlas (TCGA) for 6757 normal tissue and cancer samples. The functional effects of ELF5 isoform expression in breast cancer were also investigated using inducible cell line models and a 116-gene quantitative polymerase chain reaction (qPCR) panel, leading to unique insights into the transcriptional functions of ELF5 and in particular the role of the PNT domain.
Methods
RNA-sequencing analysis
RNA-Seq version 2 data for initial primary tumors and solid tissue normal samples (where
n ≥ 3) were downloaded from TCGA data portal (
https://tcga-data.nci.nih.gov/tcga/) [
29‐
43], with institutional human research ethics committee exemption. Samples with available RNA-Seq version 2 data (August 2013 for breast and April 2014 for all other cancer types) were included. The RNA-Seq version 2 TCGA pipeline for preprocessing of publicly available data used MapSplice [
44] for alignment and RSEM [
45] for quantitation. Non-normalized gene and isoform data were downloaded from TCGA as RSEM expected (“raw”) counts, unadjusted for transcript length, and scaled estimates, adjusted for transcript length. Scaled estimates were multiplied by 10
6 to obtain transcripts per million (TPM) values. Normalized gene and isoform data were downloaded from TCGA as quantile normalized RSEM expected counts (unadjusted for transcript length), with the upper quartile set at 1000 for gene data and 300 for isoform data.
A summary of all TCGA samples used in the analysis is shown in Table
1. For breast cancer samples, PAM50 (Predication Analysis of Microarrays 50-gene classifier) status was used to generate a subtyped cohort of 515 patients and 59 matched normal samples [
29,
46]. Six additional normal samples, matching to tumors in the initial cohort, were included in differential expression analyses.
Table 1
Summary of all TCGA RNA-sequencing samples used in analysis
Bladder | Bladder urothelial carcinoma | BLCA | 19 | 241 |
Breast | Breast invasive carcinoma | BRCA | 59b
| 515 |
| Luminal A | 229 |
| Luminal B | 126 |
| HER2 | 57 |
| Basal-like | 96 |
| Normal-like | 7 |
Cervix | Cervical squamous cell carcinoma and endocervical adenocarcinoma | CESC | 3 | 185 |
Colon | Colon adenocarcinoma | COAD | 41 | 261 |
Head/neck (including mouth and throat) | Head and neck squamous cell carcinoma | HNSC | 43 | 497 |
Kidney | Chromophobe | KICH | 25 | 66 |
| Clear cell carcinoma | KIRC | 72 | 518 |
| Papillary cell carcinoma | KIRP | 30 | 172 |
Liver | Hepatocellular carcinoma | LIHC | 50 | 191 |
Lung | Lung adenocarcinoma | LUAD | 58 | 488 |
| Lung squamous cell carcinoma | LUSC | 50 | 490 |
Pancreas | Pancreatic adenocarcinoma | PAAD | 3 | 85 |
Prostate | Prostate adenocarcinoma | PRAD | 50 | 297 |
Rectum | Rectum adenocarcinoma | READ | 9 | 91 |
Thyroid | Thyroid carcinoma | THCA | 59 | 498 |
Uterus | Uterine corpus endometrial carcinoma | UCEC | 24 | 158 |
| Uterine carcinosarcoma | UCS | NAc
| 57 |
Adrenal gland | Adrenocortical carcinoma | ACC | NA | 79 |
Hematological | Diffuse large B-cell lymphoma | DLBC | NA | 28 |
| Acute myeloid leukemia | LAML | | 173 |
Brain | Glioblastoma multiforme | GBM | NA | 156 |
| Lower grade glioma | LGG | | 463 |
Ovary | Ovarian serous cystadenocarcinoma | OV | NA | 262 |
Skin | Cutaneous melanoma | SKCM | NA | 82 |
Bone/connective tissue/soft tissue | Sarcoma | SARC | NA | 103 |
Limma voom [
47] was used for differential expression analysis of gene-level RNA-seq data, with inputs as non-normalized gene data (RSEM expected counts). Filtering was applied to remove genes with low expression, keeping genes with counts >1 in at least
n samples (where
n = number of samples in smallest group of replicates). The trimmed mean of M-values normalization method [
48] was applied, followed by differential expression analysis using Limma voom. All fold change (FC) and false discovery rate (FDR) values reported were generated by Limma voom analyses. Venn diagrams were created using online software (
http://bioinformatics.psb.ugent.be/webtools/Venn/), and clustered heat maps were created using the R package gplots [
49]. As a comparison, differential expression analysis was also carried out using edgeR [
50‐
54] (see Additional file
1: Methods).
Stable cell line generation
ELF5 isoforms 1, 2, and 3 were tagged with C-terminal V5 (and short linker sequence), cloned into the pHUSH-ProEx vector [
55], and used as a retrovirus. T47D-EcoR and MDA-MB-231-EcoR cells stably expressing ecotropic receptor were infected with pHUSH-
ELF5 retrovirus and selected using puromycin. To generate clonal cell lines, stable cell line pools were plated at low density in 96-well plates.
Cell lines and treatments
All cell lines were obtained from the American Type Culture Collection (Manassas, VA, USA) and were maintained in RPMI medium supplemented with insulin and 10 % tetracycline-free fetal bovine serum (Clontech Laboratories, Mountain View, CA, USA). Puromycin was added at a concentration of 1 μg/ml. Doxycycline (Dox) was added at a concentration of 0.1 μg/ml daily to induce protein expression.
Cell number assay
Cell number was quantified using a spectrophotometric assay. Cells were incubated with 16 % trichloroacetic acid and stained with 10 % Diff-Quik II solution (Lab Aids, Narrabeen, Australia). 10 % acetic acid was added to dried plates, and 100 μl of solution from each well was added to a 96-well plate, which was read at 595 nm. Absorbance readings were transformed to natural logarithms, and values from three wells (single experiment) were averaged for each time point. The minus Dox and plus Dox slopes for each cell line were compared using linear regression analysis.
Western blot analysis
Protein was prepared in NuPAGE Sample Buffer and Reducing Agent (Life Technologies, Carlsbad, CA, USA) using 10 μg (estrogen-related blots), 65 μg (V5 blot, T47D-ELF5-isoform 2-V5) or 25 μg (V5 blots, all other lines) per lane. Samples were separated on precast 15-well 4–12 % Bis-Tris (estrogen-related blots) or 10-well 10 % Bis-Tris (V5 blots) polyacrylamide gels (Life Technologies), transferred to polyvinylidene fluoride membrane, blocked in 5 % skim milk, and incubated overnight at 4 °C in primary antibody. Secondary horseradish peroxidase–conjugated antibody was added 1:2000 in 5 % skim milk (anti-mouse, NA931V, anti-rabbit, NA934V; GE Healthcare Life Sciences, Little Chalfont, UK). Proteins were detected using enhanced chemiluminescence solution (Western Lightning Plus; PerkinElmer, Waltham, MA, USA) and x-ray film (Fujifilm, Tokyo, Japan). Primary antibodies used were anti-V5 (sc-58052, 1:500–1:1000; Santa Cruz Biotechnology, Santa Cruz, CA, USA), anti-transducin-like enhancer of split 1 (anti-TLE1) (ab183742, 1:1000; Abcam, Cambridge, UK), anti-ERα (sc-8005, 1:1000; Santa Cruz Biotechnology), anti-Forkhead box A1 (anti-FOXA1) (sc-101058, 1:1000, Santa Cruz Biotechnology), and anti-β-actin (AC-15, 1:20,000; Sigma-Aldrich, St. Louis, MO, USA).
Transient retroviral infection
ELF5 isoform 3 was tagged with C-terminal hemagglutinin (HA), cloned into the pQCXIH vector (Clontech) and used as a retrovirus. MDA-MB-231-EcoR-pHUSH-ELF5-isoform 2-V5 Clone 7 cells were infected with ELF5-isoform 3-HA/empty vector retrovirus diluted 1:4. No pQCXIH selection pressure was applied.
Immunofluorescence
Cells were infected with pQCXIH retrovirus in eight-well Lab-Tek II chamber slides (Thermo Scientific, Waltham, MA, USA) and allowed to recover for 24 h. Dox /vehicle treatment (lasting 24 h) was then commenced. Cells were fixed with 4 % paraformaldehyde diluted in PHEM buffer (60 mM piperazine-N,N′-bis(2-ethanesulfonic acid) (PIPES), 25 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), 1 mM ethylene glycol tetraacetic acid (EGTA), 2 mM MgCl2, pH 6.9), permeabilized with 0.5 % Triton X-100, blocked with 10 % donkey serum/PHEM solution, and incubated overnight at 4 °C in primary antibody. Secondary antibodies were added at 1:200, and coverslips were applied using Duolink In Situ Mounting Medium with 4′,6-diamidino-2-phenylindole (DAPI) (Olink Bioscience, Uppsala, Sweden). Imaging was performed on a Leica DM5500 microscope (Leica Microsystems, Wetzlar, Germany). Antibodies (in 10 % donkey serum/PHEM solution): anti-V5 (sc-58052, 1:200; Santa Cruz Biotechnology), anti-HA (3724, 1:800; Cell Signaling Technology, Danvers, MA, USA), and donkey anti-mouse Alexa Fluor 647 and donkey anti-rabbit Alexa Fluor 555 conjugates (1:200; Molecular Probes/Thermo Fisher Scientific, Eugene, OR, USA).
Quantitative PCR
RNA was extracted using the RNeasy Mini Kit with DNase treatment (Qiagen, Valencia, CA, USA) and quantified using the NanoDrop spectrophotometer (NanoDrop Products, Wilmington, DE, USA). Complementary DNA (cDNA) was made using the High-Capacity cDNA Reverse Transcription Kit (Life Technologies) with ribonuclease inhibitor (Promega, Madison, WI, USA). All qPCRs were run on an ABI 7900 qPCR machine (Applied Biosystems, Foster City, CA, USA), using standard TaqMan cycling conditions or Roche Universal Probe Library (UPL) protocol with two or three technical replicates per sample (see also Additional file
1).
For the clonal cell line time-course qPCR (Fig.
6f), 0.5 μg of RNA per 20 μl of cDNA reaction and ELF5 (Hs01063022_m1) and glyceraldehyde 3-phosphate dehydrogenase (4236317E) assays were used. For the 116-gene panel, cell lines were treated for 48 h with Dox or vehicle. cDNA reactions were scaled to 100 μl and 2.5 μg RNA. Roche UPL assays were designed using the online Roche ProbeFinder software. All assays are detailed in Additional file
2.
Results were analyzed using SDS 2.4 (Life Technologies) and qbase + software (Biogazelle, Gent, Belgium) [
56]. Paired
t tests were used to calculate
p values, comparing -Dox and + Dox samples (three or four pairs per cell line group). Correction for multiple comparisons was performed using the Benjamini-Hochberg procedure, setting the FDR at 0.10 [
57].
Discussion
This study is the first detailed analysis of ELF5 isoform expression and function, extending previous ELF5 Northern blot analysis, immunohistochemistry, and microarray studies [
5,
6,
16,
25] to the isoform level using 6757 sequenced normal and cancer samples. The kidney appears to be unique in being the only tissue examined to express isoform 1 as its dominant isoform, expanding on the initial Northern blot analysis–based descriptions of
ELF5 isoforms [
6]. In breast cancer,
ELF5 alterations were subtype-specific, with the basal subtype demonstrating unique
ELF5 isoform expression changes. Despite differences in protein domains, the in vitro phenotypic and transcriptional effects of increased ELF5 isoform expression were similar. This suggests that ELF5 action is regulated in various tissues by tissue-specific alternative promoter use rather than by differences in the transcriptional activity of the isoforms.
In cancer,
ELF5 expression is frequently altered. The kidney, one of the highest
ELF5-expressing tissues, showed a dramatic decrease in
ELF5 level in cancer. ELF5 has been characterized as a tumor suppressor in the kidney and bladder [
19,
20], and this may restrict kidney carcinomas to non-
ELF5–expressing cells of origin. In other tissues, cancer was associated with an aberrant increase in
ELF5 expression, as seen in the cervix, colon, rectum, and uterus. This may indicate an oncogenic role for ELF5 in these tissues or broader genomic deregulation, such as DNA hypomethylation, a hallmark of the cancer genome [
64]. The mechanisms regulating ELF5 in different tissues and in cancer have not been widely studied; however, in the early embryo and the developing mammary gland, ELF5 regulation of lineage specification is associated with promoter methylation status [
65,
66]. Increased
ELF5 promoter methylation has also been demonstrated in bladder carcinoma [
19]. These studies establish DNA methylation as an important epigenetic mechanism regulating
ELF5 expression, with possible aberrant methylation in cancer.
The normal human breast expresses relatively high levels of
ELF5, with subtype-specific alterations in cancer. High ELF5 has been shown to maintain the ER− basal phenotype, paralleling the normal developmental role of specification of the ER− alveolar lineage [
25]. In all breast cancer subtypes, there was a broader distribution of
ELF5 isoform expression. Increased variability of isoform distribution (“transcriptome instability”) is a known phenomenon and is proposed as a molecular hallmark of cancer [
67,
68]. A recent study identified 244 cancer-associated isoform “switches” involving consistent changes in the most abundant isoform [
69]. An ELF5 isoform switch has not been identified in breast cancer, in keeping with the present study, which showed an inconsistent pattern of isoform expression variation. Although not consistently identified, this does not mean that ELF5 isoform switches do not play an important role in the subset of patients in which they occur.
Other ETS transcription factors have also been shown to be important in breast cancer. Extension of RNA-seq analysis to the entire ETS family revealed a number of cancer-associated expression changes. The ETS family as a whole has previously been studied in breast cancer at the qPCR level in mouse models [
70] and human cell lines [
71], although the present study is the first, to our knowledge, to include examination of the expression of the entire human ETS family in both the normal breast and subtyped breast cancer samples using RNA-seq data. The normal human breast expressed a diverse range of ETS factors. Compared with the normal breast, the basal-like subtype showed a distinct pattern of ETS factor expression changes, with several ETS factors changing in the opposite direction in basal compared with other subtypes.
ELF5 and
SPDEF were the most striking examples of this phenomenon.
SPDEF is also a luminal epithelial lineage-specific transcription factor in the breast and has been shown to promote the survival of ER+ breast cancer cells [
72]. The inverse relationship seen between these two transcription factors in breast cancer is intriguing and may well have a parallel during normal mammary development.
Finally, the phenotypic and transcriptional effects of isoforms 1, 2, and 3 were found to be similar in inducible cell line models. This was unexpected, as the PNT domain in murine ELF5 has previously been shown to have strong transactivation activity [
12]. In many proteins, SAM and/or PNT domains act as protein–protein interaction modules, an important mechanism of biological specificity for ETS factors, which often bind only weakly to DNA in the absence of binding partners or posttranslational modifications [
3,
12]. The importance of the PNT domain is also shown by other ETS family members in which removal of the PNT domain significantly alters protein function. The endogenous ETS1 isoform p27, for example, lacks the PNT and transactivation domains and negatively regulates full-length ETS1 by competing for DNA-binding sites and promoting its translocation from the nucleus to the cytoplasm [
63]. Although this splicing event is similar to those that occur to produce ELF5 isoforms 3 and 4, it appears that ELF5 isoform 3 can alter gene transcription in a very similar way to the full-length isoforms. In addition, there was no subcellular relocation of full-length isoform 2 seen when isoform 3 was coexpressed. Interestingly, however, while exogenous ELF5 localized to the nucleus in this study, cytoplasmic ELF5 staining is seen in some human breast cancer samples and is a predictor of outcome [
73]. This indicates that endogenous ELF5 can localize to the cytoplasm and that this has functional significance in breast cancer. A potential nuclear export sequence exists in the ETS domain of ELF5 (amino acids 165–174) similar to one identified in ELF3 [
74,
75]. It is possible that cytoplasmic relocation of ELF5 is mediated by the relative amounts of isoforms but that this effect is not recapitulated by exogenous expression, particularly in the context of MDA-MB-231 cells, which do not normally express ELF5 and therefore may be lacking essential protein binding partners. Given the importance of context in the function of ETS factors, it is possible that the differential effects of ELF5 isoforms may also require a stimulus (for example, growth factors) or challenge (for example, estrogen deprivation) in order to become apparent, an avenue that was not explored in this study.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
CLP performed RNA-sequencing data analyses, in vitro functional studies, and drafting of the manuscript. DLR assisted with bioinformatics and drafting of the manuscript. HJL, DGO, and SRO assisted with in vitro experiments and revision of the manuscript. CJO conceived of the study and its design and participated in the drafting of the manuscript. All authors read and approved the final manuscript.