Presence of exon 5-deleted oestrogen receptor in human breast cancer: functional analysis and clinical significance.

A variant form of the human oestrogen receptor (ER) mRNA lacking sequences encoded within exon 5 has been described (Fuqua SAW, Fitzgerald SD, Chamness GC, Tandon AK, McDonnell DP, Nawaz Z, O'Malloy BW, McGuire WL 1991, Cancer Res 51: 105-109). We have examined the expression of the exon 5-deleted ER (HE delta5) mRNA variant in breast biopsies using reverse transcriptase polymerase chain reaction (RT - PCR). HE delta5 mRNA was present in only 13% of non-malignant breast tissues compared with 32% of carcinomas (95% CI, P=0.05). Presence of the HE delta5 mRNA was associated with the presence of immunohistochemically detected ER (P=0.015) and progesterone receptor (PR) (P=0.02). There was a positive correlation between the presence of HE delta5 and disease-free survival (P=0.05), suggesting that the presence of HE delta5 may be an indicator of better prognosis. We have raised a monoclonal antibody specific to the C-terminal amino acids of HE delta5. This antibody recognized the variant but not the wild-type ER protein. We show that HE delta5 protein is present in breast cancer using immunohistochemical techniques. We also analysed trans-activation by HE delta5 in mammalian cells and showed that, in MCF-7 cells, HE delta5 competes with wild-type ER to inhibit ERE-dependent trans-activation. Our results indicate that this variant is unlikely to be responsible for endocrine resistance of breast cancer, but its presence at both the mRNA and protein level suggest that it may, nevertheless, be involved in regulating the expression of oestrogen-responsive genes in breast cancer.

Two-thirds of human breast carcinomas are characterized by the presence of appreciable amounts of oestrogen receptor (ER) protein. A proportion of these tumours also contain progesterone receptor (PR) and it is generally accepted that ER regulates PR gene expression. The presence of ER is correlated with a better prognosis and ER+/PR+ tumours are much more likely to respond to endocrine therapy than ER-/PRtumours. Interestingly, ER-/PR+ tumours are twice as likely to respond as ER+/PRtumours. A significant proportion of ER+/PR+ tumours, however, fail to respond to endocrine therapy and those that do so eventually become resistant to such therapy. The mechanisms leading to endocrine resistance are not yet clear (for reviews see McGuire, 1978;McGuire et al, 1991;Fuqua, 1994;Horwitz, 1994;Sluyser, 1994).
The human oestrogen receptor cDNA  and its gene (Ponglikitmongkol et al, 1988) have been cloned and the molecular mechanisms by which it acts are well understood. Alignment of the predicted ER amino acid sequences from different species shows that it can be divided into six regions A to F on the basis of differing amino acid sequence homology . Functional studies have shown that region C encodes the DNA-binding domain (DBD) and region E contains the hormone-binding domain (HBD) Kumar et al, 1987). Regions A/B and E contain trans-activation functions 1 (AFI) and 2 (AF-2) respectively (Kumar et al, , 1987Webster et al, 1988;Lees et al, 1989;Tora et al, 1989a;Berry et al, 1990). Recent studies indicate that region F plays a role in modulating transcriptional activation by ER (Montano et al, 1995).
There is little evidence for gross rearrangements of the ER gene in breast carcinomas. However, we previously reported the presence of multiple mRNA species in breast carcinomas (Barrett-Lee et al, 1987), and since then several groups have described the presence of mutant or variant forms of ER. Restriction enzyme polymorphisms and point mutations have been described. A point mutation leading to a single amino acid substitution in region B has been implicated in increased incidence of spontaneous abortions (Lehrer et al, 1990(Lehrer et al, , 1992, although the mechanism of action of this mutant is unclear. A recent report describes a point mutation at codon 157 (region B) leading to a premature stop codon (Smith et al, 1994). In breast cancer, however, there have been no reported mutations leading to altered ER protein. A number of recent studies have demonstrated the presence of ER splice variants lacking exons 2, 3, 4, 5 or 7 in breast cancer and/or in breast cancer-derived cell lines. Other variants containing intronic sequences have also been described. Most of these variants would be expected to behave as dominant-negative effectors of wild-type ER (McGuire, 1978;Murphy, 1990;Fuqua et al, 1991;McGuire et al, 1991; Wang and *Present address: Faculty of Allied Health Sciences, Kuwait University, PO Box 31470, Kuwait Miksicek, 1991;Dotzlaw et al, 1992;Fuqua et al, 1992;Fuqua, 1994;Horwitz, 1994;Sluyser, 1994). Fuqua et al (1991) identified an exon 5-deleted variant (HEA5) in ER-/PR+ breast carcinomas that would encode a truncated (40 kDa) polypeptide lacking most of the HBD. This variant activated an oestrogen response element (ERE)-containing reporter gene in a yeast expression system at a low level (10-15%) compared with wild-type ER, in the absence of oestrogen. It has been suggested that this variant may be responsible for resistance to tamoxifen, since it lacks most of the HBD and could be constitutively active.
Despite a considerable number of studies describing the presence of variant forms of ER mRNA in breast cancer, in meningiomas, the uterus and in cell lines, the presence of variant proteins has not been clearly demonstrated. However, immunoblotting and gel retardation studies have suggested that multiple ER polypeptide species are present in some breast tumours, and there are indications that some of these species exhibit abnormal properties, such as altered subcellular distribution (for reviews and references see Murphy, 1990;Foster et al, 1991;Scott et al, 1991). These forms of the ER protein could arise as a result of mutations in the ER gene or they may be specific exon-deleted variants of the ER.
In this study, we have used semi-quantitative polymerase chain reaction (PCR) to examine the expression of the exon 5-deleted ER mRNA in carcinomas and non-malignant breast biopsies and correlated its presence to clinical features of this patient group to determine whether expression of this variant has a bearing on clinical outcome. To address the question of whether HEA5 protein exists in vivo, we raised a monoclonal antibody that is specific for HEA5 and used this for immunohistochemical studies. We have, furthermore, analysed the action of HEA5 in the mammalian COS-1 and HeLa cell lines and in the human breast cancer cell line MCF-7 using transient transfection assays to determine whether its activity could lead to endocrine resistance. The mean follow-up time was 61.2 months. A total of 23 nonmalignant breast biopsies were also collected and these included three biopsies from breast samples adjacent to cancer but histologically normal, seven biopsies representing benign breast disease and 13 normal breast tissues from reduction mammoplasty specimens. All samples were snap frozen and stored in liquid nitrogen immediately after removal.

Steroid receptor determination
Paraffin sections of all malignant biopsies were analysed for ER and PR content immunohistochemically using the specific monoclonal antibodies, lD5 (Dako Ltd, UK) according to Sannino and Shousha (1994) and PR-ICA (Abbott Laboratories, UK) (Burgess and Shousha, 1993) respectively, except in 15 cases for which sufficient tissue was available only for ER determination.
Isolation of RNA Total cellular RNA was isolated from frozen tissue samples using the guanidinium isothiocyanate method (Sambrook et al, 1989), quantitated spectrophotometrically, analysed on 1% agarose gels in standard Tris acetate-EDTA buffer and stored at -70°C in water.
Reverse transcription and polymerase chain reaction amplification RNA (4 ,ug) was converted into cDNA using MMLV reverse transcriptase (RT), as described previously (Bansal et al, 1995). PCR conditions were optimized by varying the amount of RT product input, the number of cycles and magnesium chloride concentration. The optimal conditions are described: 100 ng of RT product was added to 100 ,l of PCR mixture containing 67 mm Tris-HCl, pH 8.8, 16.6 mm ammonium sulphate, 1.5 mm magnesium chloride, 0.45% Triton X-100, 0.2 mg ml-' gelatin, 200 gM dNTP, 1 unit of Taq polymerase (Peninsula, UK) and 250 ng of each of the two ER primers and the two P-actin primers. The samples were overlaid with mineral oil and subjected to 25 cycles of amplification with denaturation at 94°C for 1 min, annealing at 580C for 1 min, extension at 72°C for 2 min and a final extension at 72°C at the completion of 25 cycles for 10 min.
,-actin cDNA (Bansal et al, 1995). Blots were exposed to Kodak XR-OMAT film using intensifying screens for 1-24 h and band intensities were quantified using a Shimadzu laser densitometer. Band intensities were normalized between blots by inclusion of a reference sample on each blot, which served to correct for variations in amplification, blotting, hybridization and autoradiographic exposure times. To estimate the amount of wild-type ER mRNA, the normalized densitometric value of the 419-bp band was divided by the normalized value of the 0-actin signal for each sample. The amount of variant ER was expressed as a ratio of the densitometric values of the 300-bp variant band to the 419-bp wild-type ER band.

Statistical analyses
Wild-type ER-actin and variantwild-type ER ratios were compared between groups using the non-parametric Mann-Whitney test in which the ratios were compared between two groups (ER, PR, menopausal status), and the Kruskal-Wallis test was employed if more than two groups were being compared (histological type and grade). Spearman rank correlation was used to examine the relationship with age, tumour size and nodal status.
The Kaplan-Meier method was used to construct life-tables and the log rank test was employed to compare life-table curves. Multivariate analysis was carried out using forward stepwise selection with the Cox proportional hazards model.

Preparation of antigen and immunizations
A peptide with the sequence GTRENV, corresponding to the predicted C-terminal amino acid sequence of the HEA5 polypeptide, was synthesized on a Wang resin (Calbiochem, UK) using an Abimed AMS 422 Multiple Peptide Synthesizer and Fmoc method (Atherton and Sheppard, 1985). Before completion of the synthesis, a cysteine was added to the N terminus to facilitate subsequent coupling to a carrier protein. The peptide was cleaved from the resin according to King et al (1990); reverse-phase high-performance liquid chromatography (HPLC) and mass spectrometry were performed to check purity. A sample of 10 mg of the peptide was coupled to 10 mg of purified protein derivative of tuberculin (Central Veterinary Laboratories, UK) (Morrison et al, 1987) using the heterobifunctional agent mal-sac-HNSA (Bachem Feienchemikalen AG, Switzerland) (Aldwin and Nitecki, 1987). The final conjugate was diluted to 20 ml with sterile physiological saline.
Female Balb/c mice were primed subcutaneously with one dose of BCG vaccine (Glaxo, UK) (Lachmann et al, 1986). Mice were then immunized three times subcutaneously at monthly intervals with 200 gl of an emulsion of peptide/tuberculin conjugate (50 jig of peptide per immunization). The sera were screened as described below. The highest responding animal was selected and boosted on three consecutive days before fusion with an intravenous injection of 200-300 gl of peptide/tuberculin conjugate.

Enzyme-linked immunosorbent assay (ELISA) screening
The peptide GTRENV was synthesized, then biotinylated using a long-chain biotin ester (NHS-LC-Biotin; Calbiochem, UK). Dry peptidyl resin (0.1 g) was suspended in 1 ml of N, N-dimethylformamide (DMF; Rathbum Chemical Co., UK) and 35 mg of long-chain biotin ester together with 13.5 mg of 1-hydroxybenzotriazole (Sigma, UK) added. The mixture was incubated ovemight at room temperature and washed thoroughly with DMF and ether and dried. Biotinylation was checked using a ninhydrin test (Dupont). The peptide was cleaved from the resin as above. The polyclonal antisera from the mice were assayed by ELISA (Harlow and Lane, 1988) using streptavidin microtitre plates (Actiplate S plates; Bioproducts, UK) according to manufacturers' protocols, except that a Tris buffer (0.15 M sodium chloride/25 mm Tris-HCl, pH7.2) was used instead of phosphate-buffered saline (PBS).

Preparation of hybridomas and ascites
The spleen was removed aseptically and the splenocytes fused with Sp2/o myeloma cells (Celltech, UK) as described previously (Galfre and Milstein, 1987). After 7 days, the hybridoma supernatants were screened by ELISA using streptavidin plates as described above. The strongly reacting antibodies were identified and a selection of hybridomas transferred for expansion. Supematants from the latter were titrated by ELISA and the strongly reacting supematants were tested on frozen breast tissue sections by immunoblotting and by Westem blotting (as below). The most promising hybridoma was recloned in methyl cellulose (McCullough and Spier, 1990) and the supematants were re-evaluated for antibody specificity by ELISA, immunoblotting, gel shift and immunohistochemistry. Subclass and antibody concentrations were determined according to manufacturer's protocols (Binding Site, UK). The hybridoma cell line was expanded to approximately 2 x 10 cells. The cells were suspended in 10 ml of IMDM (Gibco, UK). Female Balb/c mice, previously primed with an intraperitoneal injection of 0.3 ml of Freund's incomplete adjuvant (Gibco, UK), were then injected intraperitoneally with 0.5 ml of the suspension. The mice were sacrificed and the ascitic fluid drained 5-7 days later. The ascitic fluid was purified using a protein A column (Bioprocessing UK).

Transfection assays
The mammalian expression vector pSG5 was used for expression of wild-type ER (HEGO) and HE15 (amino acids 1-281 of human ER) and have been described previously (Tora et al, 1989a,b). HEA5 was constructed by site-directed mutagenesis of HEGO (Tora et al, 1989b), using an oligonucleotide with the sequence 5'-AAGAG-GGTGCCAGGAACCAGGGAAAATG-3'. Positive clones were identified by loss of XbaI and NcoI restriction sites and confirmed using dideoxy sequencing (Sambrook et al, 1989). The reporter plasmid, 17M-ERE-globin-CAT, and the expression vectors, GAL-ER(HBD), have been described previously (Webster et al, 1988).
COS-1, HeLa and MCF-7 cells were maintained as described previously (All et al, 1993a), split into 9-cm plates in Dulbecco's modified Eagle medium (DMEM)-phenol red with 5% double charcoal-stripped FCS and transfected using the calcium phosphate technique (Tora et al, 1989b). COS-1 and HeLa cells were transfected with 2 jig of 17M-ERE-globin-CAT along with 0.5 jg of the 0-galactosidase reference plasmid, pCHI 10 (Pharmacia, UK), 0.5 jig of pSG5, HEGO, HE15 or HEA5 expression plasmids, together with Bluescribe M13+ DNA (BSM+; Stratagene, UK) as carrier DNA to make a total of 20 jig of DNA. MCF-7 cells were transfected with 2.0 jig of 17M-ERE-globin-CAT and 4 jg of pCH1O0. Varying amounts of HEAS DNA together with pSG5, to a total of 5 jig, and 9 jig of BSM+ to a total of 20 jig of DNA, were used. Oestrogen, hydroxytamoxifen or ICI 164, 384 were added as appropriate, cells were harvested and CAT assays were performed as described (Ali et al, 1993a).
COS-l extracts (10 jig) were resolved on 10% sodium dodecyl sulphate polyacrylamide gel electrophoresis (SDS-PAGE) and immunoblotting was performed essentially as described (Ali et al, 1993b). The ER monoclonal antibody, BlO (Ali et al, 1993b), was used at 0.5 jig ml-' and aHEA5 was used at 2 jg ml-' for immunoblotting. Alkaline phosphatase-labelled rabbit anti-mouse IgG was used as the second antibody and visualisation was carried out using BCIP and NBT substrates as indicated in the manufacturer's protocol (Promega, UK).
Gel shift assays were performed using 5 jig of the COS-1 extracts as described (Ali et al, 1993b). For 'supershifts' 1-2 jg of B 10, F3 and xHEA5 were added to the gel shift mix.
For immunocytochemistry, COS-1 cells were grown on glass coverslips in 9-cm dishes. Transfections were performed as above. Cells were fixed in 3.7% formaldehyde-PBS for 10 min, washed with PBS for 5 min, placed in methanol for 3 min at -20°C, cold acetone for 1 min at -20°C and washed twice with PBS for 5 min. The coverslips were incubated in 10% goat serum in PBS for 15 min to block non-specific binding. Primary antibody (at 0.5 jig ml-') was added and incubation carried out for 1 h at room temperature followed by two washes in PBS for 5 min. The second antibody (goat anti-mouse IgG; Sigma, UK) was used at 1:25 dilution in PBS for 20 min at room temperature, followed by two washes in PBS. Horseradish peroxidase-labelled mouse anti-goat IgG (Promega, UK) was used in the third incubation at room temperature for 20 min at a dilution of 1:50 in PBS. The coverslips were washed twice in PBS and DAB substrate/reagent (Abbott, UK) was added for 10 min. The coverslips were washed in gently running water for 5 min, counterstained in 1% Harris's haemotoxylin for 5 min and placed twice for 2 min in 95% ethanol, 100% ethanol, then in xylene and mounted on slides coated with a drop of DPX mountant.

PCR analysis of the exon 5 variant in tumour tissue and in normal breast
Oligonucleotides encoding sequences lying within exons 3/4 and 6 of the human ER gene (see Figure IA) (Ponglikitmongkol et al, 1988) were used to amplify total RNA isolated from 154 breast cancers and 23 non-malignant breast tissues. PCR conditions were optimized such that measurements were made during the linear phase of amplification for both ER and 3-actin (data not shown).
In cases in which high input led to product saturation, measurements were repeated using diluted sample. Figure 1B illustrates some of the results. A predominant band corresponding to wildtype ER (419 bp) was seen in PCR products from non-malignant breast samples (lanes 5 and 6). Identity was confirmed by digestion of the product with HindIII, which produced two fragments of about 160 bp and 270 bp as predicted from the hER cDNA sequence ) (data not shown). The PCR products derived from cancer tissue RNA also showed this 419-bp fragment but in about a third of the cases we also observed a smaller 300-bp fragment (Figure 1, lanes 3 and 4), corresponding to the fragment lacking exon 5 sequences described previously . Cloning and sequencing of the two PCR products confirmed the identities of the 419-bp and 300-bp products as wild-type ER and the exon 5-deleted variant respectively (data not shown).
The amount of ER mRNA was determined as a ratio relative to the P-actin levels. None of the tumours were altogether negative for ER transcripts, including those which were ER negative immunohistochemically. Levels of ER-p-actin varied between samples and, in order to determine whether ER transcript levels were proportional to ER protein levels, we compared ER-1-actin ratios with ER status, as determined immunohistochemically. There was a good correlation using the log rank test at 95% CI, with P=0.005 ( -300bp Figure 1 The exon 5-deleted oestrogen receptor variant in breast cancer. A shows a schematic of the oestrogen receptor mRNA with the positions of exons 1-8 marked. Also shown is the position of the primers used to amplify a region including exon 5. The primer sequences are given in Materials and methods. The predicted amino acid sequence is shown for wild-type ER and the variant (HEA5) together with the position of regions A-F. The five amino acids at the C-terminus of HEA5 that arise owing to a frameshift are shown in the single letter amino acid code. The asterisk denotes the stop codon leading to premature termination and a truncated polypeptide. B shows the PCR products obtained for two breast carcinomas that contain HEA5 and two that do not and two normal breast samples. The upper band is the 41 9-bp product obtained from wild-type mRNA and the lower band (lanes 3 and 4) is the 300-bp variant band. The PCR was performed using 100 ng of RT cDNA and 25 cycles of amplification, as described in Materials and methods ( Table 2). The amount of the HEA5 product relative to wild-type ER within each sample ranged from 0% to 30% (data not shown).  (Tables 4 and 5), compared with other clinical parameters. Using univariate analysis, disease-free and overall survival were not found to be correlated with the age of the patient at the time of diagnosis, or with the menopausal status. There was no correlation with histological features of the cancers, but patients with ER-positive cancers did significantly better than those with ER-negative cancers for both disease-free and overall survival. PR+ patients also showed better overall survival. As expected, clinical and pathological tumour size and nodal status were related to prognosis (Table 4). Interestingly, patients whose tumours contained HEA5 appeared to have marginally better disease-free survival than those who did not have any HEA5 (P=0.05). The presence of HEA5 was also associated with increased overall survival, but this failed to reach statistical significance (P=0.09). Multivariate survival analysis showed that ER and tumour size were significant independent predictors of disease-free and overall survival; clinical nodal status was found to be significant for disease-free but not overall survival; HEA5 was not found to be a statistically significant independent predictor of either disease-free or overall survival. Figure 2 shows survival curves for HEA5-positive (n=50) and HEA5-negative (n=104) patients, illustrating a clear trend towards a survival advantage for HEA5-positive patients for both disease-free and overall survival.
Preparation of exon 5-deleted ER-specific monoclonal antibodies Deletion of sequences encoded within exon 5 of the human oestrogen receptor gene leads to the splicing of exon 4 to exon 6, resulting in a frameshift and the introduction of five new amino acids followed by a stop codon ( Figure IA). Translation of the   immunoblotting and immunocytochemistry of HEA5-transfected COS-1 cells. A single hybridoma was selected, based on the intensity and specificity of the reactions obtained in these tests (data not shown; and see below) and identified as IgG2b. It was further amplified, culture supernatants were collected and ascites fluid was produced, and is hereafter referred to as aHEA5.
Detection of HEA5 protein by the aHEA5 monoclonal antibody in transiently tranfected cells COS-I cells were transiently transfected with the plasmid pSG5 or with pSG5 containing either HEGO or HEA5. Cell extracts were prepared and the xHEA5 monoclonal antibody was tested for specificity by immunoblotting. The monoclonal antibody, B10, raised against aminoacids 150-165 (region B) of the human oestrogen receptor (Ali et al, 1993b) was used for comparison with aHEA5. Figure 3A shows that B 10 recognized both the wild-type ER (HEGO, lane 2) and a band at about 40 kDa, consistent with the size expected for the HEA5 polypeptide (lane 3). aHEA5 did not recognize HEGO but did react with the 40-kDa HEA5 polypeptide (compare lanes 2 and 3 with lanes 5 and 6). The DNA binding of ER to an oestrogen response element (ERE) can be determined in vitro using the gel shift assay by incubating cell extracts containing ER with radioactively labelled ERE-containing oligonucleotides and then resolving the ER-ERE complex using non-denaturing polyacrylamide gel electrophoresis (Ali et al, 1993b). This assay also showed that aHEA5 recognizes which displayed a prominent retarded band (lane 5). HEA5-transfected COS-1 cell extracts also gave a retarded band that migrated slightly faster than the HEGO complex (compare lanes 5 and 6). Note that the HEA5-ERE complex was much weaker than the HEGO-ERE complex, despite similar levels of each protein being present in the extracts as judged by immunoblotting (data not shown; and see Figure 3A, compare lanes 2 and 3).
British Journal of Cancer (1997) 75 (8) COS-1 cells were grown on coverslips, transfected with pSG5, HEGO or HEAM using the calcium phosphate method, fixed and immunostained using mouse IgG, B10 and acEA5 antibodies. Figure 3C shows that no staining was observed with pSG5-trans--fected cells. Both HEGOand HEA5-transfected COS-1 cells showed a proportion of cells with strong nuclear staining. Staining with aHEA5 gave nuclear staining only in COS-1 cells transfected + with HEA5.

Expression of HEA5 protein In breast carcinomas
The above transfection studies showed that aHEA5 can specifically recognize the HEA5 polypeptide in immunoblotting, gel shift and immunostaining assays. We next examined breast cancer samples for the presence of HEA5 protein. Twenty samples were chosen on the basis of high and low levels of HEA5 mRNA, as determined by PCR (above). Extracts were prepared from tissue stored in liquid nitrogen by homogenization, followed by freeze-thawing (-80°C/0°C) and centrifugation at 15 000 g for removal of insoluble material. The lysates were analysed by immunoblotting and gel shift assays. We found no evidence for the presence of the HEA5 polypeptide (data not shown).
The reactivity of aHEA5 was evaluated by using frozen sections, since immunohistochemistry offers a more sensitive method of detection than the methods used above. Frozen sections from a cancer expressing high levels of HEM5 mRNA, as determined by PCR (a ratio of 0.25 variant-wild-type ER; see Figure 1B, lane 3), were immunostained for the presence of HEA5 protein. Serial sections were immunostained using the ER monoclonal antibody iD5 (Dako, UK), which was raised against an epitope in the Nterminal portion of the ER protein, and therefore would be 12 expected to recognize both the wild-type ER and the HM5 polypeptides. iD5 exhibited strong nuclear staining confined to British Journal of Cancer (1997) Figure 4A). xHEA5 similarly gave nuclear staining of malignant cells ( Figure 4B), although some cytoplasmic staining was also evident. However, the specificity of aHEA5 was confirmed by immunostaining with mouse IgG and by competing with the immunizing antigen. Immunostaining with mouse immunoglobulins gave no staining ( Figure 4C). Preincubation of xHEA5 with the peptide C-GTRENV, used for immunization, led to loss of staining ( Figure 4D), whereas lD5 staining was unchanged ( Figure 4E), further indicating that the HEA5 staining was specific.
In order to demonstrate further the specificity of our antibody, immunostaining was performed on another cancer positive for ER A and for HEA5 mRNA expression and one that showed ER positivity but HEA5 negativity by PCR. The staining observed for an ER+/PR+ cancer expressing high levels of HEA5 mRNA (variant-WT ratio=0.25; see Figure IB, lane 4) is shown in Figure   4F-H. Strong nuclear staining of the malignant cells was observed with the ER antibody lD5 ( Figure 4F). aHEA5 also showed nuclear staining of some malignant cells, as well as cytoplasmic positivity ( Figure 4G). Case 2 gave positive staining for ER (using ID5; Figure 41) but no nuclear staining was seen with cxHEA5 ( Figure 4J). PCR results showed that this sample was negative for HEA5 mRNA (Figure lB, lane 1). No staining was observed in any of the cases with mouse IgG ( Figure 4H and K).

Transcriptional activation by HEA5 in mammalian cells
We wished to examine HEA5 activity in mammalian cell systems and used a chloramphenicol acetyl transferase (CAT) gene reporter system in COS-1 and HeLa cells and in the human breast cancer-derived cell line MCF-7, which overexpresses ER. HEA5 is truncated at amino acid 365 in ER and would be expected to act through AF-1 in a constitutive manner. We compared the transactivational ability of HEA5 with ER (HEGO) and with the in vitro generated mutant HE15 [containing amino acids 1-281 of ER (Kumar et al, 1987], which activates transcription through AF-1 (Tora et al, 1989a). The constructs used and the extent of the deletions are shown in Figure 5A. Figure 4 Detection of HEA5 using the aHEA5 antibody and comparison with an ER antibody on HEA5-positive and -negative breast cancer sections. Sections of a strongly ER-and HEA5-positive breast cancer sample were immunostained with 1 D5 (A), aHEA5 (B) or mouse igG (C). (D-E) shows the staining obtained after incubation of alHEA5 (D) or 1 D5 (E) with a 1 00-fold excess of the immunizing peptide. Another strongly HEA5-positive (F-H) and a HEA5-negative (I-K) breast cancer sample were immunostained with the ER antibody 1 D5 (F and 1), aHEA5 (G and J) or mouse IgG (H and K). Counterstaining was performed with haematoxylin. Original magnification x 200 Transfection of HEGO into COS-l cells resulted in a tenfold increase in transcription in the presence of oestradiol (E2) over its absence and a 20-fold increase relative to background ( Figure 5B); the trans-activation produced by HEGO in the absence of E2 may be caused by residual oestrogens in the charcoal-stripped medium used (Tora et al., 1989b). HE15 activated the CAT gene about 15% as well as HEGO and showed the same activity in the presence as in the absence of E2 ( Figure SB; and data not shown). HEA5 transactivated to levels similar to those obtained with HE15 ( Figure SB).
Several studies have shown that AF-1 and AF-2 of ER are separable and synergise to give maximal trans-activation by ER (Lees et al, 1989;Tora et al, 1989a). The synergistic activity of AF-1 and AF-2 can be examined using constructs in which AF-2 (region E) British Journal of Cancer (1997) 75(8) is fused to the DNA-binding domain of the yeast trans-activator GAL4 (GAL-ER (HBD)), and using HE1S in which AF-2 sequences have been deleted. When co-transfected together with a CAT reporter gene containing a GAL4 DNA-binding element (17M) and an ERE, AF-1 and AF-2 act synergistically (Webster et al, 1988;Tora et al, 1989a). In order to examine the mechanisms of trans-activation by HEAS further and to determine whether it trans-activates ERE-dependent gene expression in a similar way to HE1S, i.e. through AF-1, we analysed its ability to synergise with AF-2 ( Figure 5B). GAL-ER activated the 17M/ERE-G-CAT gene in the presence of E2. Both HE1S and HEA5 synergised with GAL-ER to similar extents. In HeLa cells, AF-1 is unable to activate transcription but can synergise with AF-2 (Kumar et al, 1987;Lees et al, 1989;Tora et al, 1989a;Berry et al, 1990). Both HE1S and HEA5 gave very little trans-activation on their own but synergised with GAL-ER in the presence of E2 ( Figure 5C). These results indicate that HEA5 acts in a similar manner to HE1S, i.e. through AF-1, although as before 4.0 ,ug of HEA5 needed to be transfected compared with 0.5 ,ug of HElS for this level of synergism.
Tamoxifen (and its metabolite, 4-hydroxy-tamoxifen) is a partial antagonist of oestrogen, in that it enables trans-activation through AF-1, while inhibiting the activity of AF-2 (Berry et al, 1990). ER trans-activated about 20% as well in the presence of hydroxytamoxifen (0) as in the presence of E2 in COS-1 cells, whereas little activity was seen in HeLa cells. HE1S and HEA5 also transactivated in COS-1 cells but not in HeLa cells. These results are in broad agreement with those reported previously (Berry et al, 1990). Note also that, in the presence of hydroxy-tamoxifen, little synergistic activity was observed when GAL-ER and HE1S or HEAS were co-transfected. As expected, in the presence of the 'pure' antagonist, ICI 164,384, very little ER activity was observed.
In order to examine the effect of HEAS on trans-activation in the presence of wild-type ER, MCF-7 cells were transfected with HEAS and the CAT reporter gene. We observed a fivefold induction by the endogenous ER, on addition of E2, in cells transfected with the CAT reporter alone. Co-transfection with increasing amounts of HEAS inhibited this trans-activation by up to 60% indicating that HEAS was acting as a 'dominant negative' mutant, presumably by virtue of its lower trans-activational ability compared with the wild-type ER. Note, however, that the inhibition never reached levels obtained with hydroxy-tamoxifen ( Figure SD). An inhibition by HEA5 of the low level trans-activation in the absence of ligand and in the presence of hydroxytamoxifen and ICI 164,384 was also observed and might suggest that, as in HeLa cells, HEAS cannot trans-activate on its own from this promoter.

DISCUSSION
The exon 5 deletion variant was first shown to be present at the mRNA level in ER-/PR+ breast cancers by Fuqua et al (1991), who showed that it could trans-activate constitutively in a yeast expression system. We have analysed 154 breast cancers, chosen without regard to their clinical status, for the presence of HEA5 by PCR. We have shown that the amount of wild-type ER product obtained by this method was strongly correlated with the presence of ER as determined immunohistochemically (at 95% CI, P=0.005). HEA5 was present in about 30% of the cancers but in only 10% of normal breast tissue (P=0.05). In breast cancer, the presence of HEA5 was related to the presence of ER (at 95% CI, P=0.015) and PR (P=0.02), in agreement with other data (Daffada et al, 1995). Prognostic tests using univariate analysis showed that the presence of HEA5 is associated with longer disease-free survival, although the results only reach significance to 95% CI with P=0.05.
Multivariate survival analyses showed that the presence of HEA5 is not an independent prognostic factor. These results were unexpected in view of suggestions that HEA5 could lead to resistance to endocrine therapy, but concur with the results of Daffada et al (1995), who showed similar levels of HEA5 in tamoxifen-resistant cancers and primary controls.
A number of studies have reported the presence of ER variants at the mRNA level but none have provided evidence that these mRNAs are translated. Several studies have found evidence for the presence of ER-like proteins unable to bind DNA or ligand but these could be proteolytic products of wild-type ER protein. We therefore raised a monoclonal antibody that would specifically recognize the HEA5 polypeptide and present the first evidence for the presence of HEA5 protein in breast tissues. In an immunohistochemical study, we have shown that tumours that possess the HEA5 mRNA have immunostainable HEA5 protein. Although we have demonstrated the specificity of our antibody by artificially overexpressing HEA5 in COS-1 cells, we have not yet been able to demonstrate its presence in breast cancer using immunoblotting or gel shift assays, presumably because of low amounts of endogenous HEA5 protein and/or its reduced stability. Further studies are underway to compare the stability of wild-type and exon 5-deleted ER proteins using pulse-chase labelling experiments.
We have, furthermore, analysed the trans-activational ability of HEA5 by transient transfection of the ER-negative COS-1 and HeLa cell lines and in the ER-positive MCF-7 cells. We find that HEA5 can constitutively trans-activate ERE-dependent gene expression in mammalian cells in addition to the previously described activity in yeast . HE15, which contains the DBD and AF-1 of the human oestrogen receptor but lacks the HBD/AF-2, has been well characterized (Kumar et al, , 1987. This construct has been extensively used to show that AF-l activity is cell-and promoter-specific (Tora et al, 1989a;Berry et al, 1990;Metzger et al, 1995). Furthermore, the activity of the wild-type ER in the presence of tamoxifen has been correlated to the activity of HE15 by comparing the cell and promoter specificity of trans-activation by hER in the presence of tamoxifen, indicating that tamoxifen inhibits AF-2 but allows activation of AF-l (Berry et al, 1990). The experiments outlined in Figure 5 indicate that HEA5 behaves mechanistically like HE15 (amino acids 1-281 of ER), which only contains AF-1 of ER since: (1) HEA5 can activate a reporter gene in COS-1 cells to similar levels as HE15; but (2) like HE15, HEA5 does not trans-activate in HeLa cells (see Kumar et al, 1987;Tora et al, 1989a;Berry et al, 1990); (3) HEA5 can synergise with AF-2 (in GAL-ER) in COS-1; and (4) despite being inactive on its own in HeLa cells, HE15 is capable of synergising with AF-2 (GAL-ER (HBD)), as is HEA5. Given our findings that HEA5 shows similar cell specificity and synergistic properties to HE15, our results indicate that the constitutive trans-activation by HEA5 results entirely from AF-1 sequences and, since the agonistic activity of tamoxifen is caused by its activation of AF-1, while inhibiting AF-2, it is unlikely that resistance to tamoxifen should arise as a result of the induction of, or increase in levels of, an ER variant with the same trans-activational properties as the full-length ER bound by tamoxifen.
Transfection of MCF-7 cells showed that HEA5 inhibits transactivation by the endogenous receptor, presumably by competitive binding on EREs in responsive genes owing to its lower trans-activational ability relative to the wild-type receptor. In this respect, HEA5, despite its ability to trans-activate constitutively, actually inhibits the action of ER, thereby acting in a manner similar to some anti-oestrogens, and should not be expected to be responsible for resistance to tamoxifen therapy, as has been suggested (see Introduction). The analysis of pre-and post-therapy biopsy samples for changes in the level of the HEA5 variant may shed more light on this issue. Fuqua (1994) reported that transfection of HEA5 into MCF-7 cells makes them resistant to tamoxifen. These results are in apparent contradiction to our results showing that HEA5 inhibits trans-activation by ER, although we have performed these studies using chimeric reporter genes. We found the greatest levels of HEA5 mRNA in PR+ cancers and the lowest in PRcancers, regardless of ER status (data not shown). Similarly a study describing the analysis of 27 breast cancers (Zhang et al, 1993) for the presence of HEA5 reported highest levels of HEA5 in PR+ cancers. In this series only 1/13 ER+/PR+ and 0/3 ER-PR+ cancers failed to express HEA5, whereas 3/7 ER+/PR-and 4/4 ER-/PRcancers did not express HEA5. Another study (Daffada et al, 1995) recently showed correlation of pS2 and PR status and the presence of HEA5. Taken together, these results suggest that these ER-responsive genes might be regulated, at least in part, by HEA5.
The correlation between the presence of HEA5 and the presence of PR and pS2 is in agreement with our results indicating better prognosis for HEA5-positive cancers, as these indicators have themselves been shown to be correlated with a better prognosis (Ravdin et al, 1992;Foekens et al, 1993).
In conclusion, we find that the HEA5 variant is present in a proportion of breast cancers at the mRNA level and that its presence does not correlate with a poor prognosis. Its trans-activational properties suggest that it inhibits trans-activation by wild-type ER, most likely by direct competition. Our results do, however, support the notion suggested by others that HEA5 could play a role in regulating the expression of certain genes. The detection of the HEA5 polypeptide in breast cancers using a specific antibody is further indication that HEA5 plays a role in breast cancer. ADDENDUM Rea and Parker (1996) recently reported the creation of MCF-7 cell lines stably expressing HEA5 and showed that the variant stimulated transcription of a reporter gene in chicken embryonal fibroblasts in the absence of hormone but was only weakly active in MCF-7 cells. They further showed that the growth-stimulatory effects of oestrogen and the growth-inhibitory effects of tamoxifen were not influenced by the presence of HEA5. These findings are in general agreement with our results indicating that HEA5 is not responsible for resistance to endocrine therapy in breast cancer.