Expression of Human Endogenous Retrovirus K Envelope Transcripts in Human Breast Cancer¹ Free

HERVs³ are stably inherited sequences thought to have entered the germ line of their host more than a million years ago (1, 2). These elements are widely dispersed throughout the genome and are estimated to comprise >1% of the entire human genome. Most HERVs are defective because of multiple termination codons and deletions (3), but some appear to contain all structural features necessary for viral replication (4). HERVs are grouped into single- and multiple-copy families, usually classified according to the tRNA used for reverse transcription (2). The type K family (HERV-K) is present in an estimated 30–50 copies/human genome and includes some elements with long open reading frames (5, 6). This family of HERVs was originally identified by its homology to the mouse mammary tumor virus and contains members that are transcriptionally active in several human cancer tissues (1, 7, 8) as well as tumor cell lines, notably in the human breast cancer cell line T47D (9, 10). Two general types of HERV-K genomes exist, distinguished by the absence (type 1) or the presence (type 2) of 292 nucleotides at the boundary of the putative pol and env genes (4, 10).

Additional endogenous retroviral sequences that may be transcriptionally active in humans include ERV3 and HERV-E. ERV3 is a single-copy, full-length provirus that contains a nondefective env glycoprotein gene (11) and functional long terminal repeats. ERV3 env mRNAs are expressed in placenta and some tumor cell lines (12, 13, 14, 15). HERV-E is a multiple-copy HERV family that contains long open reading frames in the pol and env regions of the provirus, indicative of the potential for expression (16). Expression of transmembrane env protein from the prototypic HERV-E, HERV-E4-1, has been reported in some cancer cell lines including colon carcinoma, germ cell tumors, and prostate adenocarcinoma (17).

The widespread distribution of multiple endogenous retroviral elements in mammalian genomes suggests that they may perform significant biological roles in the host and thus have been evolutionarily conserved. Several functions for HERVs have been proposed. Reverse transcription and integration of retroviral elements may contribute to the plasticity of the genome, accelerating the evolution of new genes (18) and altering the transcription of existing genes (19). HERV-encoded proteins may also have functions in vivo. In the placenta, the env protein of HERV-W has been shown to mediate the formation of syncytiotrophoblasts, an important step in placental morphogenesis (20). Expressed HERV proteins may also contribute to autoimmune disorders (21), although this remains controversial (22). A causal relationship between endogenous retroviruses and human cancer has been explored, including the recent demonstration of the transforming ability of an HERV-K central open reading frame gene (cORF; Ref. 23). In mice, endogenous retroviral genes are transcriptionally silent in normal tissues but expressed in several well-studied murine tumor models (24, 25). In some of these models, the expressed env protein acts as a tumor antigen capable of inducing both antibody and T-cell responses (26, 27, 28, 29). In humans, tumors of germ cell origin have been reported to express HERV-K transcripts (30, 31, 32), and in seminoma patients, the HERV-K10 env protein is reported to be a tumor antigen eliciting host antibody responses (33).

As part of an ongoing search for tumor antigens important in breast cancer, we analyzed expression of the env region of ERV3, HERV-E4-1, and HERV-K in human breast cancer cell lines and tissues. Here we demonstrate that HERV-K transcripts are specifically and frequently expressed in human breast cancer, and that some of these transcripts contain open reading frames capable of producing env protein.

Human breast cancer cell lines (BT-20, ZR-75-1, MCF-7, SKBr-3, MDA-MB-231, MDA-MB-453, BT-474, and T47D) were obtained from the American Type Culture Collection (Rockville, MD). Normal HME cells were obtained from Clonetics (San Diego, CA). Breast cell lines were cultured in the medium recommended by the American Type Culture Collection. The human teratocarcinoma cell lines Tera I and Tera II were kindly provided by Drs. Gail H. Vance and Virginia C. Thurston (Indiana University School of Medicine) and were cultured in α-MEM (Mediatech, Herndon, VA) with 10% fetal bovine serum (HyClone) and insulin (6 ng/ml; Sigma Chemical Co., St. Louis, MO). T47D cells were treated with 10 nm β-estradiol (Sigma Chemical Co.) and 100 nm progesterone (Sigma Chemical Co.) as described previously (34). Human breast tumor tissue, nonmalignant breast tissue from patients with other breast disorders (primarily fibroadenomas), normal human breast tissue from reduction mammoplasty, and placenta were provided by the Tissue Procurement Shared Facility of the Comprehensive Cancer Center at the University of Alabama at Birmingham, with Institution Review Board approval. Representative samples from breast tumor tissue and adjacent uninvolved tissue were isolated based on gross inspection. Tissue samples were snap-frozen and stored at −70°C until RNA isolation. For in situ hybridization, both snap-frozen and formalin-fixed, paraffin-embedded tissues were used. For Northern blot analysis, RNA from normal breast tissue (pooled from eight individuals) was purchased from Clontech Lab, Inc. (Palo Alto, CA), and RNA from breast cancer tissues (invasive ductal carcinoma) was purchased from BioChain Institute, Inc. (Hayward, CA).

Total RNA from cell lines and breast tissue samples was isolated using RNA STAT-60 reagent (Tel-Test, Inc., Friendswood, TX) following the protocol provided by the manufacturer. RNA was treated with DNase (RNase-free; 1–2 units of DNase I/μg of DNA; Ambion, Austin, TX) at 37°C for 30 min to remove contaminating DNA, followed by heating at 75°C for 5 min to destroy residual DNase activity.

Oligonucleotide primers derived from the env sequences encoding the putative env surface protein of ERV3 (nucleotides 786-2530; Ref. 11), HERV-E4–1 (nucleotides 6211–7559; Ref. 16), and HERV-K (nucleotides 6471–7575; Ref. 4) were used to amplify cDNA prepared from human tissues and cell lines (Fig. 1). Primer sequences were as follows: ERV3 sense, 5′-ACACTACGTGTCGGGGAACATCATG; ERV3 antisense, 5′-ACCAACCTCTGAAAAGGGAATCTGG; HERV-E4-1 sense, 5′-CTGGTCCACGCACGGCCGAAGCATG; HERV-E4-1 antisense, 5′-AAAAGGACGACTTAATAGAGCCAAT; HERV-K sense, 5′-AGAAAAGGGCCTCCA CGGAGATG; HERV-K antisense, 5′-ACTGCAATTAAAGTAAAAATGAA. The HERV-K primers are 100% homologous to the following published HERV-K type 1 env sequences: K10 (GenBank accession no. M14123.1), K101 (GenBank accession no. AF16409.1), K102 (GenBank accession no. AF164610.1), K103 (GenBank accession no. AF164611.1), chromosome 5 (GenBank accession no. AC016577.4), and chromosome 19 (GenBank accession no. AC008996.5). These primers are expected to amplify both unspliced and spliced env transcripts. All HERV oligonucleotides were synthesized by Life Technologies, Inc. (Grand Island, NY). Control primers to amplify human β-actin were from Stratagene (La Jolla, CA).

Isolated RNA was incubated at 65°C for 10 min, followed by incubation on ice for 2 min prior to reverse transcription. Total RNA from each sample was reverse transcribed using cDNA synthesis beads (Amersham Pharmacia Biotech, Inc., Piscataway, NJ) as per the manufacturer’s directions. The reverse transcribed samples were amplified in a volume of 50 μl containing 3.3 μl of the first-strand cDNA synthesis mixture (corresponding to 1 μg of input RNA), 5 μl of 10× PCR buffer (Qiagen, Valencia, CA), 0.5 μl of Ampli-Taq DNA polymerase (2.5 units; Qiagen), and various sense and antisense oligonucleotide primer pairs at 50 pmol each. Each sample was analyzed in parallel with human β-actin primers. To assure that observations were not attributable to DNA contamination, all RNA samples were treated with DNase before cDNA synthesis. Additionally, 1 μg of RNA from each sample without reverse transcription was PCR amplified to control for genomic DNA contamination. PCR reactions were initially denatured at 94°C for 4 min, followed by 30 cycles of denaturation (94°C for 1 min), annealing (55°C for 1 min), and extension (72°C for 1 min). Amplified products were analyzed on a 1% agarose gel.

Samples of total RNA (15 μg/lane) were electrophoresed at 5 V/cm on a 1.2% agarose-formaldehyde gel. After electrophoresis, gels were washed and transferred overnight onto a positively charged nylon membrane (Roche Molecular Biochemicals, Indianapolis, IN) using a Transblotter apparatus (Schleicher and Schuell, Keene, NH) and 10× SSPE buffer (1.5 m NaCl, 0.1 m NaH₂PO₄, and 20 mm EDTA, pH 7.4). Membranes were stained with methylene blue solution to visualize 18 S and 28 S bands.

The probe for Northern blot analysis was prepared using an env cDNA derived from patient 1 breast tumor RNA (Fig. 6, Pt1, clone #2), corresponding to the HERV-K102 sequence from nucleotides 6471–7575 (GenBank accession no. AF164610.1). The cDNA was labeled with [³²P]dATP using the random-primed labeling method. Labeled probe was hybridized overnight at 62°C with the membrane, using a high-efficiency hybridization buffer (Molecular Research Center, Cincinnati, OH). Membranes were washed three times at room temperature with prehybridization/wash solution (1× SSC, 1% SDS), followed by three washes at 65°C with the same solution, and exposure to autoradiography film at −70°C for 24 h.

RNA probes were prepared from patient 1, HERV-K env clone #2 (Fig. 6). One μg of the linearized plasmid was used as template. The in vitro transcription and labeling of probe was performed at 37°C for 2 h with a digoxigenin RNA labeling kit (Roche Molecular Biochemicals) using T7 and SP6 RNA polymerases to obtain run-off transcripts of the antisense (complementary to the mRNA) or sense (negative control) probes. Paraffin-embedded breast tissue specimens were cut into serial 5-μm sections, melted, deparaffinized in xylene, rehydrated in ethanol, and then fixed in 4% paraformaldehyde. Snap-frozen breast specimens were cut into 5-μm serial sections and fixed in 4% paraformaldehyde directly. After fixation, tissue sections were treated with proteinase K (20 μg/ml in 50 mm Tris-HCl, 5 mm EDTA) at 37°C for 15 min, washed with PBS, and incubated at room temperature in 0.1 m triethanolamine-HCl plus 0.25% acetic anhydride. One section from each group was pretreated with RNase A (Sigma Chemical Co.) before proteinase K treatment as a control. Hybridization of RNase-treated sections with antisense RNA probe verified that RNA and not genomic DNA was the target of the hybridization. Tissue sections were equilibrated in Quick-Hyb hybridization buffer (Stratagene, La Jolla, CA) for 30 min. Antisense or sense riboprobes were denatured and added with salmon sperm DNA (250 μg/ml; Stratagene) to the tissue sections and incubated at 68°C overnight. Sections were washed twice with 0.2× SSC, 0.1% SDS and then once with STE buffer (0.5 m NaCl, 1 μm EDTA, and 0.02 m Tris) at room temperature. After treating with RNase A (50 μg/ml in STE buffer) at 37°C for 45 min to minimize nonspecific binding, the sections were washed once with STE and once in 0.1× SSC plus 0.1% SDS at 68°C for 15 min. Detection of digoxigenin-labeled nucleic acids was by enzyme immunoassay with a digoxigenin nucleic acid detection kit (Roche Molecular Biochemicals), using nitroblue tetrazolium and 5-bromo-4-chloro-3-indolyl phosphate substrate (Bio-Rad, Hercules, CA), followed by development at room temperature for 1–2 h in the dark. Sections were then rinsed in deionized water and counterstained with methyl green (Zymed, South San Francisco, CA) for 15 min. Adjacent sections were hybridized with sense probe as a negative control or with H&E for histological evaluation.

For construction of GST-fusion protein expression constructs containing partial HERV-K env sequences, the following modified RT-PCR primers were used: sense 5′-CCGGAATTCGTAACACCAGTCACATGGATG (nucleotides 6494), antisense 5′-ATAGTTTAGCGGCCGCTCTTTTGGATCTATTTAAAACACC (amplifies nucleotides 6494–7552; GenBank accession no. M14123.1), with engineered recognition sequences for restriction enzymes (EcoRI and NotI, respectively) underlined. RT-PCR products from patient breast tumor samples were cloned into the pGEX 4T-1 GST gene fusion vector (Pharmacia). Plasmids with predicted env insert size were screened for protein production on a small scale using isopropyl-β-d-thiogalactopyranoside induction. GST fusion proteins were purified by affinity chromatography using glutathione Sepharose 4B (Pharmacia), subjected to 12% SDS-PAGE electrophoresis and transferred to nitrocellulose. Immunoblotting was performed with an anti-GST antibody (1:1000 dilution; Pharmacia). One of the clones that produced fusion proteins of the expected molecular weight was further characterized by nucleotide sequencing using vector-specific primers (Fig. 6, clone #1 from Pt1).

RT-PCR products resolved on agarose gels were purified using the QIAquick gel purification kit (Qiagen), subcloned into pCR-II vectors (Invitrogen, Carlsbad, CA), and sequenced using vector-specific reverse and forward primers plus internal HERV-K env-specific primers. Sequence analysis and alignment were carried out with DNAsis for Windows (Hitachi Software, San Francisco, CA) and the GenBank database.

To investigate the expression of HERV elements in breast cancer cell lines, RT-PCR was performed using primers specific for env genes of ERV3, HERV-E4-1, and HERV-K. The HERV primer pairs for RT-PCR allowed detection of transcripts encoding the putative surface portion of each HERV env protein (Fig. 1). Expression of ERV3 and HERV-E4-1 RNA was not detected in breast tumor samples, although amplification of these products from placenta served as a positive control for PCR validation (data not shown). In contrast, HERV-K transcripts were detected in SKBr-3, MCF-7, BT20, ZR-75-1, MDA-MB-231 and T47D, but not in MDA-MB-453, BT-474, and normal HME cells (Fig. 2 A). A single fragment of ∼1100 bp, consistent with the size of a type 1 HERV-K env region, was detected. Because the primers were identical with several known type 1 HERV-K family members but contained 1–4 bp mismatches with known type 2 HERV-K env sequences, amplification of a single fragment was not unexpected. Expression of type 1 env transcripts in T47D cells has been reported previously (10).

These studies were next extended to surgical samples of breast carcinoma and nonmalignant breast tissues. Again, ERV-3 and HERV 4-1 transcripts were not detected in 43 breast tissue samples (20 tumor samples and 23 nonmalignant samples). In contrast, amplification products derived from the HERV-K env region were frequently detected (Fig. 2,B and Table 1). To date, 130 breast tissue samples have been analyzed for expression of HERV-K transcripts by RT-PCR using the HERV env primers (Table 1). HERV-K env RNA was detected in 45% of breast cancer samples, in 18% of breast tissues derived from breast cancer patients but judged to be uninvolved by gross pathological dissection, as well as in 83% of placentas analyzed. In contrast, breast tissues derived from individuals not having breast cancer had no detectable HERV-K env expression (n = 35).

Northern blot analysis was used to further assess expression of HERV-K transcripts (Fig. 3). Using the env region as a probe (99% homologous to HERV-K102), positive radioactive species corresponding to the full-length proviral transcript (∼8.3 kb) and the putative spliced env mRNA transcript (∼3.0 kb) were detected in human breast cancer tissues but not in normal breast tissue (see Fig. 3). As a positive control, the full-length and env transcripts were also detected in the teratocarcinoma cell lines Tera I and Tera II (data not shown).

To demonstrate that HERV-K RNA was transcribed specifically in breast tumor cells, we performed in situ hybridization. Serial tissue sections were prepared and hybridized with an HERV-K env probe. HERV-K RNA was detected by ISH in 7 of 10 RT-PCR-positive breast tumor samples. This difference in the percentage of positive samples may reflect the lower sensitivity of ISH as compared with RT-PCR or may reflect differences in the specific area of each tissue sample used for analysis in the two independent assay types. In the positive tumor samples, the hybridization signal was detected specifically in the tumor cells and not in the surrounding normal cells. More importantly, when so-called “uninvolved” areas of breast cancer patient tissues were examined (i.e., normal breast tissue from breast carcinoma samples), it became apparent that the positive signal detected in some of these samples by RT-PCR (Table 1) was likely attributable to the presence of previously undetected malignant cells (data not shown). Examples of ISH results are presented in Fig. 4, where a strong positive signal is detected in tumor epithelial cells of a tissue sample containing ductal carcinoma in situ (Fig. 4,A), but no signal is detected in the matched, uninvolved epithelial cells of the same tissue sample (Fig. 4,C). Matched tissue sections analyzed with the control sense probe were negative (Fig. 4, B and D). In addition, if the tissue was first treated with RNase, there was no signal using the antisense probe (data not shown). As expected from RT-PCR and Northern blot results, breast tissue from six individuals not having breast cancer had no detectable HERV-K RNA by ISH analyses. Thus, HERV-K transcripts were only detected in tumor cells by this assay.

As a test of the coding potential of the above expressed env sequences, HERV-K env cDNA amplified from selected breast cancer samples was cloned into a prokaryotic GST-fusion protein expression vector. The predicted recombinant fusion proteins would contain M_r ∼26,000 of GST plus M_r ∼41,000 daltons of HERV-K env surface protein. Six independent clones derived from a single breast tumor specimen (Fig. 6, Pt #1) were analyzed. Of these, four clones produced a fusion protein of the expected molecular size (M_r ∼67,000), one produced a fusion protein of lower than predicted size, and one produced protein of the size of GST alone (partial results in Fig. 5). Similar results were obtained using clones derived from four additional breast tumor samples. HERV-K env cDNAs derived from the same tumor sample (Pt 1) were cloned into a eukaryotic expression vector for assay by coupled in vitro transcription-translation with [³⁵S]methionine. Radiolabeled protein of the size expected (M_r ∼40,000) for the targeted HERV-K env product was detected in two additional clones (data not shown; sequence analysis in Fig. 6, Pt1, clones #2 and #3). These results suggest that the amplified HERV-K env regions contained no premature stop codons and provide indirect evidence that such HERV-K env proteins can be stably expressed.

HERV-K env RT-PCR products derived from four individual breast cancer patient tissue samples were cloned, and eight independent isolates were completely sequenced (Fig. 6). All demonstrated >97% sequence homology to the previously reported HERV-K102 env (35), but none were identical to any published sequence. Sequence divergence from K102 was noted at 46 of 1035 nucleotides analyzed (4.4%). Of note, none of the isolated clones were identical to each other, even among clones derived from the same tumor sample, although specific nucleotides sequence divergence from K102 was in many cases shared among the eight clones. Specifically, nucleotide divergence was identical with at least one of the other seven isolates for 13 of 46 nucleotide positions (28%). The remaining 33 nucleotide changes (72%) were unique among the eight cDNA clones. A few cDNA sequences showed highest homology to published HERV-K type 1 genes other than K102 (e.g., Pt1clone #3 in Fig. 6 was most homologous to K101, GenBank accession # M14123), but in no case did the sequence homology exceed 99.5%.

Endogenous retroviruses have several potential functional roles in their host. By analogy with mouse models of cancer, we hypothesized that proteins encoded by HERV env genes may act as tumor antigens. As a first test of this hypothesis, we used RT-PCR to examine expression of the env region of several candidate HERV genes in human breast cancer cell lines and surgical specimens. Whereas no breast tissue expression of ERV3 or HERV-E4-1 was found in our analyses, RT-PCR readily detected HERV-K transcripts in six of eight breast carcinoma cell lines and in 45% of the 55 analyzed breast tumor tissue samples. In contrast, a normal breast epithelial cell culture (HME) and multiple breast tissue samples from individuals not having breast cancer showed no detectable HERV-K env RNA. Northern blot analysis using an env probe derived from a breast tumor sample confirmed expression of HERV-K transcripts in breast cancer. Because a small percentage of samples judged to be nonmalignant by gross pathological inspection had detectable env RNA by the RT-PCR assay (18%), ISH was used to localize expression of HERV-K transcripts to individual cells by histological discrimination between tumor and adjacent uninvolved tissues. Such ISH studies indicated that small numbers of tumor cells present in some matched tissue samples previously identified as “uninvolved” based upon gross inspection could account for positive HERV-K env RT-PCR results for these samples (as in Table 1). Considered together, our RT-PCR and ISH results support the conclusion that expression of HERV-K transcripts is restricted to breast carcinoma and is undetectable in nonmalignant breast epithelial cells.

Expression of endogenous retroviral sequences has been implicated in a variety of human disease states (36). Of particular relevance to the current study, Yin et al. (37) previously described expression of HERV-K sequences in both normal and malignant breast tissue, using a classification system and assay targeting the pol genes of HERV-K subgroups HML 1–6. They reported that, on average, all HML groups were expressed at lower levels in breast tissues as compared with placenta, and furthermore, that HML pol genes were not more highly expressed in malignant as compared with nonmalignant breast tissues. Although the conclusions of Yin et al. (37) may seem in conflict with those from the current study, it should be noted that different retroviral gene products (pol versus env) and different HERV-K subgroups (HML versus K102) were targeted. It might be expected that gene-specific expression for individual HERV-K elements would correlate with overall transcriptional activity of HERV-K loci in any given cell type; however, this presumption is untested in the available literature and remains unexamined for samples used in the current study.

Whereas expression of various HERV mRNA species has been reported previously in different tumor tissues, HERV protein expression has been confirmed in only a minority of cases. Most HERV loci are thought to be defective, and certain HERV-K env sequences, including HERV-K10 and HERV-K107 (both type 1), are reported to contain a premature stop codon (TAA at nucleotide 6920) that should ablate env protein expression. The env cDNA clones sequenced in our study contained no stop codon at this position (codon was CAA instead of TAA for all eight clones), and no stop codons were observed over the entire env region analyzed in the presumed reading frame. Furthermore, our in vitro studies suggest that the HERV-K env cDNA clones from breast cancer tissue were capable of producing stable env protein of the predicted molecular size. Others have also reported expression of HERV-K env protein from cloned env cDNA after transfection into a mouse cell line (38). Thus, the potential for expression of HERV-K env proteins in human breast cancer is supported by independent data.

Sequencing of env cDNAs in the current study was performed in part to determine whether the transcripts were likely to have arisen from a single locus. Within the region of the HERV-K env gene sequence analyzed, the eight completely sequenced cDNAs amplified from breast cancer tissues showed >97% nucleotide homology to HERV-K102. Significant homology with HERV-K101, HERV-K10, and env sequences located on chromosomes 5 and 19 was also noted for individual clones. However, none of the eight sequenced clones were identical to each other or to any published env sequence. Although our sample size is small, it is interesting to note that many of the nucleotide changes were shared among clones isolated from either the same or an unrelated patient tumor specimen. Although it is possible that some of these nucleotide differences may have been generated during the amplification and cloning process, it is unlikely that this would account for all of the differences, particularly those specific changes observed in multiple clones. Also, inherited polymorphisms in HERV-K genes would not explain the cDNA sequence divergence among clones isolated from the same patient tumor sample. Random mutagenesis during the malignant transformation of breast tumor precursors would not seem to explain the identical nucleotide changes observed among clones derived from four individual patients. Because genomic sequences are not yet available for all of the estimated 30 or more copies of HERV-K in the human genome, it is difficult to estimate the potential impact of allelic variation on the current study. Considering the minimal estimated complexity of HERV proviral sequences within the human genome (39), clarification of the precise origins of the expressed env sequences studied here awaits more complete sequence data provided by the Human Genome Project.

Collectively, our results are in agreement with the hypothesis that multiple, related HERV-K gene loci may be transcriptionally activated in breast tumors. Such activation might occur because of similarities in HERV-K long terminal repeats, which might be activated in response to a transcriptional factor found specifically in the altered transcriptional milieu of malignant breast epithelial cells (40, 41). Alternatively, activation may be attributable to hypomethylation of genomic DNA. HERV-K sequences were reported to be hypomethylated in urothelial carcinomas (42), suggesting that a similar mechanism may function in breast carcinoma. Further investigations of the mechanisms regulating expression of the multiple individual HERV loci are needed to define those most likely to be active in human malignancy.

In summary, HERV-K env RNA was expressed in many of the breast cancer tissues and cell lines analyzed, with no expression detected in normal breast tissues, suggesting that env expression might be a tumor marker. These transcripts are likely to originate from multiple HERV-K loci, even within the same tumor. Similar analyses with expanded tumor samples will be needed to confirm these findings, to estimate the number of individual HERV-K loci transcribed in tumors, and to establish the potential prognostic or diagnostic value of env RNA and/or protein expression. Investigation of the mechanism resulting in HERV-K transcriptional activation in tumor cells has the potential to elucidate fundamental cell processes involved in malignant transformation.

We thank Drs. Gail H. Vance and Virginia C. Thurston from Indiana University School of Medicine, Indianapolis, IN, for their generosity in providing the Tera I and Tera II cell lines. We also thank John Dubay for critically reading the manuscript.