Human retroviral gag- and gag-pol-like proteins interact with the transforming growth factor-beta receptor activin receptor-like kinase 1.

Mutations in activin receptor-like kinase 1 (ALK1), a transforming growth factor (TGF)-beta type I receptor, lead to the vascular disorder hereditary hemorrhagic telangiectasia caused by abnormal vascular remodeling. The underlying molecular cause of this disease is not well understood. Identifying binding partners for ALK1 will help to understand its cellular function. Using the two-hybrid system, we identified an ALK1-binding protein encoded by an ancient retroviral/retrotransposon element integrated as a single copy gene known as PEG10 on human chromosome 7q21. PEG10 contains two overlapping reading frames from which two proteins, PEG10-RF1 and PEG10-RF1/2, are translated by a typical retroviral -1 ribosomal frameshift mechanism. Reverse transcription-PCR and Northern blot analysis showed a broad range of PEG10 expression in different tissues and cell types, i.e. human placenta, brain, kidney, endothelial cells, lymphoblasts, and HepG2 and HEK293 cells. However, endogenous PEG10-RF1 and PEG10-RF1/2 proteins were only detected in HepG2 and HEK293 cells. PEG10-RF1, which is the major PEG10 protein product, represents a gag-like protein, and PEG10-RF1/2 represents a gag-pol-like protein. PEG10-RF1 also interacts with different members of TGF-beta superfamily type I and II receptors. PEG10-RF1 binding to ALK1 is mediated by a 200-amino acid domain with no recognized motif. PEG10-RF1 inhibits ALK1 as well as ALK5 signaling. Co-expression of ALK1 and PEG10-RF1 in different cell types induced morphological changes reminiscent of neuronal cells or sprouting cells. This is the first report of a human retroviral-like protein interacting with members of the TGF-beta receptor family.

This family includes various forms of TGF-␤, bone morphogenetic protein, activin, inhibin, nodal, Mü llerian inhibitor substance, and growth and differentiation factors. All family members regulate different critical aspects of cell proliferation and differentiation, developmental patterning, and morphogenesis (3,4).
The general model of signal transduction for the TGF-␤ superfamily is that signaling is mediated through a heteromeric complex of type I and type II transmembrane serine/ threonine kinase receptors (5,6). Binding of ligand to two type II receptors induces recruitment of two type I receptors. Subsequently, type I receptors become phosphorylated in the cytoplasmic GS domain because of the kinase activity of the constitutively phosphorylated type II receptor. Activated type I receptor then phosphorylates members of the receptor Smads. Upon phosphorylation, receptor Smads form a heterotrimeric complex with Smad4 and translocate to the nucleus where they act as co-transcription factors on specific genes. Accumulating data shows that non-Smad signaling pathways can be activated by TGF-␤ such as the Ras/mitogen-activated protein kinase pathways including the extracellular signal-regulated kinases, c-Jun N-terminal kinases/stress-activated protein kinases, and p38 (7).
Mutations in ALK1 or in endoglin, a TGF-␤ type III receptor, lead to the vascular disorder HHT (8 -11), an autosomal dominant disorder characterized by localized angiodysplasia (12). Data from ALK1 and endoglin knock-out mice show that both genes play crucial roles in early embryonic angiogenesis (13)(14)(15)(16). Ligand binding assays for ALK1 suggested that it could be a type I receptor for TGF-␤ and/or activin, but no TGF-␤ and activin typical cellular responses were observed in initial experiments (1,2). Therefore, ALK1 was classified as an orphan receptor. Subsequently, it was shown that activated ALK1 phosphorylates Smad1 and Smad5 (17,18), suggesting that the ALK1 ligand is a bone morphogenetic protein family member. However, we were able to show that ALK1 is a signaling receptor for TGF-␤1 and -␤3 (19). This was further substantiated by a report that the ALK1-Smad1/5 pathway as well as the ALK5-Smad2/3 pathway are activated in a dose-dependent manner in endothelial cells by TGF-␤1 and TGF-␤3, respectively (20). In addition, we demonstrated that human serum contains a third, so far unknown, ALK1-specific ligand that induces ALK1 signaling (19) and that activated ALK1 inhibits TGF-␤ signaling, consistent with the finding that TGF-␤-inducible genes like plasminogen activator inhibitor-1 (PAI-1) or tissue plasminogen activator (tPA) are up-regulated in ALK1 knock-out mice (16). Interestingly, endoglin also binds TGF-␤1 and -␤3 but not TGF-␤2 and inhibits TGF-␤ signaling, although the mechanism is unclear (21,22).
Although ALK1 was identified in 1993, only recently have several genes been reported to be controlled by ALK1 signaling, i.e. Id1 and Smad6 (20,23,24). No proteins other than endoglin, T␤RII, ActRII, Smad1/Smad5, and the nuclear receptor LXR␤ (25) are known to interact with ALK1. To identify other ALK1 binding partners, we performed a two-hybrid screen with the cytoplasmic part of ALK1. We found a new ALK1-interacting protein named PEG10-RF1. PEG10-RF1 is one of two proteins, PEG10-RF1 and PEG10-RF1/2, encoded by an ancient retroviral/retrotransposon element integrated into the human genome as a single copy gene. We report here the first human gene coding for two proteins caused by Ϫ1 ribosomal frameshifting. In addition, we show that co-expression of ALK1 and PEG10-RF1 induces profound changes in cell morphology that were not observed with other TGF-␤ type I and II receptors.
Antibodies and Cytokines-Proteins tagged with the HA epitope were detected in immunoblots or precipitated with the monoclonal antibody 12CA5 (19). N-terminal His-tagged proteins were detected or precipitated with the Anti-HisG or Anti-HisG-horseradish peroxidase antibodies from Invitrogen. TGF-␤3 was purchased from R & D Systems (Wiesbaden, Germany). Human serum was a kind gift from the Blood Center (Mannheim, Germany).
Yeast Strain, Transformation, and Two-hybrid Library Screen-For the two-hybrid screen and assays based on the Gal4 system, yeast strain PJ69-4A was used (26). PJ69-4A allows protein-protein interaction monitoring because of three separate reporter genes HIS3, ADE2, and lacZ under the control of Gal4-inducible promoters GAL1, GAL2, and GAL7, respectively. This allows highly specific screens and reduces false positive clones/results, especially with the ADE2 selection. PJ69-4A is similar to the commercially available yeast strain AH109 (Clontech) except for the lacZ reporter gene.
Yeast media and plates were prepared as described in the Clontech Yeast Protocols Handbook. Media ingredients were purchased from BIO101 Qbiogen. Yeast were transformed with the LiAc method (Clontech Yeast Protocols Handbook). For the library screen, PJ69-4A was transformed with pGBT9-ALK1/c (see "Expression Constructs and Cloning") as the bait. ALK1-transformed yeast were propagated in large scale and subsequently transformed with a MATCHMAKER placenta cDNA library (Clontech) that had been cloned into the pGAD10 vector with the Gal4 activation domain. Transformed yeast were plated on HisϪ plates. Colonies appearing after 2-5 days were transferred onto AdeϪ plates. Colonies that showed growth on both HisϪ and AdeϪ plates were further tested in the ␤-galactosidase (␤-gal) filter assay. Plasmid DNA from positive yeast clones was isolated and transformed into bacteria by electroporation, and isolated plasmid DNA from transformed bacteria was used for further analyses.
Mutations/amino acid changes in the ALK1 coding sequence were introduced with the QuikChange mutagenesis kit (Stratagene, Amsterdam, The Netherlands). Annotations correspond to the changed amino  acid positions: ALK1 W50C , ALK1 R67Q , ALK1 Q201D (activation mutation),  ALK1 K229R , ALK1⌬ S232 , ALK1 S331W , ALK1 R374W , ALK1 R411Q , ALK1/  c M376 , and ALK1 P424T . For the two-hybrid screen and different two-hybrid assays the following vectors were used: (a) pGBT9 (Clontech) with the Gal4 DNAbinding domain, (b) pGBDU2 (26) with the Gal4 DNA-binding domain, and (c) pGAD424 (Clontech) with the Gal4 activation domain. The coding sequence for the cytoplasmic domain of the different type I and II receptors that were tested with the two-hybrid system was amplified by PCR and fused in frame to the Gal4 DNA-binding domain or activation domain, respectively, of the appropriate vector. ALK1 wild-type (ALK1/c 142-503 ) and ALK1 Q201D were cloned into the EcoRI/SalI restriction sites of pGBT9. ALK4 and ALK5 were cloned into the HindIII/ BamHI restriction sites of pGBT9. ActRII and T␤RII were cloned into the EcoRI/SalI restriction sites of pGBDU2.
For expression in mammalian cells and immunoprecipitation assays, PEG10-RF1 (ORF1) was PCR-amplified and fused to the N-terminal His tag of the pcDNA4/HisMax-TOPO vector (Invitrogen). Also, PEG10-RF1/2 (ORF1 and ORF2) was PCR-amplified from genomic DNA and fused to the N-terminal His tag of the pcDNA4/HisMax-TOPO vector.
For fluorescence microscopy studies, PEG10-RF1 (ORF1) was PCRamplified and fused to the C-terminal EYFP sequence of the pEYFP-N1 vector (Clontech) resulting in PEG10-RF1 YFP . Also, ALK1, ALK5, and T␤RII were PCR-amplified and fused to the C-terminal ECFP sequence of the pECFP-N1 vector (Clontech) resulting in the following constructs: ALK1 CFP , ALK5 CFP , and T␤RII CFP . The correct sequences and reading frames of all of the constructs derived from PCR products were verified by automated DNA sequencing with an ABI 310 sequencer (Applied Biosystems, Darmstadt, Germany).
Northern Blot and RT-PCR-Northern blot hybridization was done with a multiple tissue Northern blot from Clontech at 50°C according to the manufacturer's instructions. PEG10 DNA was randomly labeled with DIG-11-dUTP using the DIG High Prime DNA labeling and detection kit (Roche Applied Science) according to the manufacturer's instructions. Bound DNA was detected by chemiluminescence with the Lumi-Imager F1™ (Roche Applied Science).
PCR was performed as follows: 1 l of cDNA was added to 1ϫ PCR buffer (10 mM Tris-HCl, 1.5 mM MgCl 2, 50 mM KCl, 0.2 mM dNTP), 10 pmol of forward primer, 10 pmol of reverse primer, and 1 unit of Platinum® Taq DNA polymerase (Invitrogen) in a total volume of 20 l. These samples were then held at 95°C for 5 min and cycled 35 times at 95°C for 45 s, 56°C for 45 s, and 72°C for 1 min. This was then held at 72°C for a further 7 min. The PCR products were then run on a 2% agarose gel.
PEG10-RF1-specific Polyclonal Antibodies-Generation of PEG10-RF1-specific polyclonal antibodies was done by EUROGENTEC (Belgium). The rats were immunized with the peptide combination H 2 N-TERRRDELSEEINNC-CONH 2 (amino acids 2-15) and H 2 N-REQV-EPTPEDEDDDIE-CONH 2 (amino acids 45-60) generating antibody SAON1. A second antibody, SAOP2, was generated by immunization with peptide H 2 N-CKASKSSPAGNSPAPL-COOH (amino acids 311-325). The specificity of PEG10-RF1 antibodies was tested in comparison with the preimmune sera of rats that were not able to detect PEG10 proteins. Specificity was further shown in Western blots blocking binding of antibodies to PEG10 with increasing amounts of peptides used for immunization (see Fig. 1 in the supplemental material).
Transient Transfections, Immunoprecipitations, and Western Blots-COS-1 cells were cultured in 6-well plates to 70 -80% confluence. For transfections, 6 l of FuGENE (Roche Applied Science) and 2 g of DNA/well were used according to the manufacturer's instructions. Briefly, the cells were transfected with: (a) 2 g of empty vector, (b) 1 g receptor cDNA or PEG10-RF1 and 1 g vector, or (c) in co-transfections with two different expression constructs, 1 g of each. Two days posttransfection, the cells were lysed in 400 l of lysis buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.5% Nonidet P-40, 50 mM NaF, 1 mM phenylmethylsulfonyl fluoride, and a protease inhibitor mixture (Sigma)) at 4°C for 30 min. The cell lysate was precleared with 50 l of protein A-Sepharose (Amersham Biosciences; dissolved in lysis buffer (v/v)) for 3 h. 300 l of precleared lysate was incubated overnight with 50 l of protein A-Sepharose and the indicated monoclonal antibodies at 4°C. The protein A-Sepharose immunocomplex was then washed three times in 500 l of lysis buffer and resolved in 15 l of Laemmli buffer. 20 l of precleared lysates were separated by SDS-PAGE and blotted onto nitrocellulose or polyvinylidene difluoride membranes (Machery & Nagel; Amersham Biosciences) for immunodetection with the indicated antibody. Immunodetection was performed with the SuperSignal® West Pico chemiluminescent system (Pierce) according to the manufacturer's instructions. For chemiluminescence detection, the Lumi-Imager F1 TM (Roche Applied Science) was used.
Immunofluorescence and Fluorescence Microscopy-To assess the endogenous cellular localization of PEG10 proteins, HepG2 cells were seeded on coverslips. After 1 day, the cells were washed twice with PBS and fixed for 10 min in 4% paraformaldehyde. Subsequently, the cells were washed for 5 min in quenching solution (50 mM Tris-HCl, pH 8.0, 100 mM NaCl) and then permeabilized for 20 min in 0.2% Triton X-100/PBS. After permeabilization, the cells were washed two times with PBS. For immunofluorescence detection, the PEG10-RF1-specific antibody SAON1, as well as the SAON1 preimmune serum, were used in a 1:200 dilution in 1% bovine serum albumin/PBS and the TRITClabeled secondary goat anti-rat antibody (Dianova) was used in a 1:400 dilution in 1% bovine serum albumin/PBS. To block unspecific bindings, the cells were incubated for 1 h in 1% bovine serum albumin/PBS, washed once with PBS, and then incubated at room temperature in a wet chamber with the SAON1 antibody or the SAON1 preimmune serum. After 1 h, the cells were washed five times for 5 min with PBS and then incubated with the secondary antibody for 1 h in a wet chamber at room temperature. The cells were washed again five times for 5 min with PBS, and after a final wash with distilled water, the coverslips were mounted with Mowiol onto microscope slides.
For further fluorescence microscopy studies, the PEG10-RF1 YFP fusion protein construct was transiently transfected into COS-1, CHO-K1, L6E9, and HepG2 cells alone or in various combinations with different type I and type II receptors. The cells were transfected with the indicated constructs using FuGENE 6 (Roche Applied Science) according to the manufacturer's instructions. In brief, approximately 70% confluent cells were transfected, and the expression of fluorescence proteins was examined in a time range of 1-4 days with a fluorescence microscope. For fluorescence imaging of living cells, the images were taken with a Hamamatsu C4880 -80 CCD camera with OpenLab 2.2.5 software (Improvision, Tü bingen, Germany) and a Zeiss Axiovert 25 microscope with 40ϫ magnification. For pictures of fixed cells, an Axiovert 200M microscope (Zeiss, Jena, Germany) and the AxioCam MRM (Zeiss) were used supported by the AxioVision 3.1 software (Zeiss).
Luciferase Reporter Assay-The luciferase assay was done as described previously (19). In brief, R-1B cells were transiently transfected with the luciferase transcriptional reporter pGL3-(SBE) 4 (a Smad response reporter) (29) and empty vector DNA or the indicated receptor cDNA. The pSV-␤-galactosidase vector (pSV-␤) (Promega) was always included to normalize the luciferase activity for the different transfec-tions. After a 16-h incubation, the cells were washed with PBS and lysed for 15 min in 80 l of lysis buffer. 50 l were used to measure the luciferase activity using the luciferase assay system (Promega) and the Lumi-Imager F1 TM (Roche Applied Science). The remaining 30 l of lysates were used to measure the amount of ␤-galactosidase. Luciferase activity was corrected for ␤-galactosidase.

RESULTS
A New Intracellular ALK1-binding Protein-To identify new intracellular interacting partners for ALK1, an improved Gal4 yeast two-hybrid system was used (26). A human placenta cDNA library was screened with the entire ALK1 cytoplasmic domain (ALK1/c) yielding one clone that was able to grow on HisϪ and AdeϪ selection plates and turned blue in the ␤-gal filter assay after about 2 h compared with 1 h for the p53/large T-antigen (pAV3/pTD) positive control.
The ϳ6-kb insert sequence of the candidate plasmid was used in a GenBank TM BLASTn search revealing 100% identity to cDNAs deposited as KIAA1051 (GenBank TM accession number AB028974) and HB-1 (GenBank TM accession number AF216076) of almost 6.2 kb length, and mRNA 23915 (Gen-Bank TM accession number AF038197) of 1.5 kb length corresponding to the 3Ј-end of KIAA1051 and HB-1. All entries described a cDNA for a protein with unknown function. Our clone was ϳ100 bp shorter at its 5Ј-end than the published sequences. Analysis of the KIAA1051 sequence for an ORF revealed two ORFs, one of ϳ1 kb (ORF1) and one of 1.1 kb (ORF2), a 200-bp long 5Ј-untranslated region, and a large 3Јuntranslated region of about 4 kb. ORF2 overlaps with ORF1 by 60 bp. In our two-hybrid clone the third amino acid of ORF1 was fused to the Gal4 DNA-binding domain, whereas the second ORF was out of frame by a Ϫ1 nucleotide shift. This suggested that ORF1 was responsible for the ALK1 interaction in the two-hybrid assay. Over the course of our analyses a paternally expressed imprinted gene PEG10 was reported, localized to chromosome 7q21 and identical to KIAA1051/HB-1 (30).
ORF1 was cloned into a mammalian expression vector with an N-terminal His tag, referred to as PEG10-RF1 (PEG10reading frame 1). PEG10-RF1 codes for a 325-amino acid-long protein of about 50 kDa, although the predicted molecular mass is 38 kDa, suggesting protein modifications. PEG10-RF1 and ALK1 were co-transfected into COS-1 cells to assay for interaction by co-immunoprecipitation analysis. Immunoprecipitation for ALK1 co-precipitated PEG10-RF1 and vice versa (Fig. 1A), demonstrating that ORF1 codes for the polypeptide that mediated ALK1 binding in the two-hybrid screen.
Northern Blot and Computational Analyses of PEG10 -Analysis for PEG10 with a multiple tissue Northern blot showed a ϳ6.5-kb transcript with its highest expression in placenta and brain, as well as lower expression in spleen, kidney, and weak expression in thymus and lung (Fig. 1B). A search of the SAGE data base suggests a ubiquitous expression with varying tissuespecific expression levels. Intriguingly, the deposited PEG10/ KIAA1051/HB-1 sequence is one continuous genomic sequence on chromosome 7q21 showing no exon/intron structure.
The full-length sequence of PEG10 was determined to be 6253 bp, which differed from the estimated 6.5 kb for the PEG10 transcript based on our Northern blot data. To solve this discrepancy we performed a computational EST walk from the 5Ј-end of PEG10 using the EST data base to extend the 5Ј sequence. Of several ESTs overlapping with PEG10, one (Gen-Bank TM accession number AL589326) showed the longest extension, a 100-bp overlap with an additional 300 bp. Blasting the human genome with the EST sequence confirmed its 7q21 localization (chromosome 7 contig NT029333.1). At the genomic level EST AL589326 is divided into 200 bp directly extending into PEG10 with a remaining 200 bp of ϳ8 kb upstream of the main sequence. This suggests that PEG10 has a noncoding first exon, an 8-kb intron, and a large second exon of 6.3 kb (Fig. 1C). To confirm our computational results, we performed an RT-PCR analysis. RT-PCR with RNA derived from abdominal and breast microvascular endothelial cells and a primer pair flanking the putative intron resulted in an amplification product of the expected size of about 100 bp but gave no PCR product with genomic DNA (Fig. 1D). This indicates that the PEG10 gene has at least one intron at its 5Ј-end and codes for a Ն6.5-kb transcript.
Further blast searches with PEG10 ORF1 and ORF2 polypeptide sequences showed homologies to gag-and pol-like proteins, respectively, of retroviral/retrotransposon elements. The highest homologies, 43 and 47%, respectively, were found to the retrotransposon element sushi of the pufferfish, Fugus rubripe. Further, ORF1 and ORF2 exhibited ϳ85% identity at the nucleotide level to the mouse MyEF3-like protein gene, which is identical with the mouse gene Edr (31,32). ORF2 is ϳ87% identical to the partial cDNA TRT1, a transcript from mink lung epithelial cells that is down-regulated after TGF-␤ treatment (33).
Examination of the putative protein sequences of PEG10 ORF1 and ORF2 with the protein domain program SMART (smart.embl-heidelberg.de/) predicted an N-terminal coiled-coil domain (amino acids 1-50) and a C-terminal zinc finger domain (amino acids 249 -310) for PEG10-RF1 (see Fig. 4A). The zinc finger domain is commonly found in retroviral gag proteins and is characterized by the highly conserved amino acid sequence Cys-Xaa 2 -Cys-Xaa 4 -His-Xaa 4 -Cys (CCHC), flanked by many basic residues (34 -36). The CCHC zinc finger is known to be involved in viral RNA packaging (37). Thus, PEG10-RF1 might be able to bind to RNA, or DNA, because zinc finger motifs are also found in transcription factors.
The PEG10 Gene Utilizes a Typical Retroviral-1 Translational Frameshifting Mechanism to Encode at Least Two Proteins-KIAA1051/PEG10 was considered to be the human or- PEG10-RF1 and HA-tagged ALK1 were co-transfected into COS-1 cells, and interaction was subsequently shown by co-immunoprecipitation (IP) for PEG10-RF1 and ALK1, respectively. B, tissue-specific expression of PEG10 was analyzed by multiple tissue Northern blot showing a ϳ6.5-kb transcript with its highest expression in placenta and brain. C, putative genomic organization of the PEG10 gene on human chromosome 7q21. The 200-bp-long exon 1 is separated by an 8-kb intron from the 6.3-kb exon 2, which contains open reading frames 1 and 2 of PEG10. D, RT-PCR analysis of RNA from microvascular endothelial cells, monocytes (U937), and lymphoblasts, plus controls as indicated. Two different primer pairs were used. Lane 1 shows the PCR product for primer pair one, located 5Ј in exon 1 and 3Ј in exon 2 separated by an ϳ8-kb intron sequence. Lane 2 shows the PCR product for primer pair 2, located in exon 2 of the PEG10-RF1 coding sequence. The amplification product in lane 2 for the U937 cDNA represents genomic DNA contamination. WB, Western blot. thologue of mouse Edr (32) based on gene structure, two overlapping reading frames of gag-and pol-like sequences, and 85% nucleotide identity. In in vitro translation assays it was shown that Edr codes for two polypeptides, one gag-like protein and one gag-pol-like protein, because of Ϫ1 ribosomal frameshifting, which is typically found in viral genomes. The sequence motif most likely responsible for this Ϫ1 slippage is highly conserved between Edr and PEG10/KIAA1051 (32). To determine whether PEG10 uses the same Ϫ1 translational frameshift mechanism as Edr, we cloned the complete reading frame 1 and 2 sequence of PEG10 into a eukaryotic expression vector with an N-terminal His tag referred to as His PEG10-RF1/2. Expression of His PEG10-RF1/2 in COS-1 cells demonstrated that the PEG10 gene indeed codes for two proteins ( Fig.  2A), one the size of PEG10-RF1 and one ϳ100 kDa. The predicted molecular mass for PEG10-RF1/2 is 80 kDa, and the difference between what we observe and the predicted size is probably due to post-translational modifications. In addition to these two proteins, we observed a third protein species of ϳ35 kDa, referred to as PEG10-CNF (PEG10-cleaved N terminus fragment), because it is most likely a cleavage product of PEG10-RF1/2 cleaved at the junction of RF1 and RF2 where there is an aspartyl protease motif. The predicted size of such a cleavage product would match with the detected ϳ35-kDa protein. This motif is highly conserved between PEG10, Edr, and retroviruses and is used for further gag protein processing (38). We were unable to detect any potential second cleavage product because only the N terminus of our construct has a His tag. The estimated amounts of the three protein products visualized were 75-80% for PEG10-RF1 and 10 -15% for PEG10-RF1/2 and PEG10-CNF.
We further tested PEG10-RF1/2 to see whether it is also able to interact with ALK1. As shown in Fig. 2B, ALK1 precipitates PEG10-RF1/2. However, the shorter PEG10-CNF does not bind to ALK1, suggesting that the cleavage might affect an ALK1binding domain or that the binding domain is in the unlabeled cleavage product. For further analyses we concentrated on PEG10-RF1.
Endogenous PEG10-RF1 Is Located in the Cytoplasm-After demonstrating that the retroviral-like PEG10 gene could be transcribed and translated in in vivo transfection assays, we wanted to know whether endogeneous PEG10 proteins could be found in human cells. For this purpose polyclonal antibodies were raised against PEG10-RF1. Antibody SAON1 is directed against the N-terminal part of PEG10-RF1 and antibody SAOP2 against the C-terminal part of PEG10-RF1. Using RT-PCR and Northern blot analyses we had already shown that the PEG10 gene is expressed in different cell types (Fig. 1D). However, when we tested these cells by Western blot analysis, no endogenous PEG10 proteins were detected (data not shown). Thus, we continued to test additional cell lines, including T47D, LCLC, HepG2, HEK293, the endothelial cell line HMEC-1, and primary human umbilical cord endothelial cells. As shown in Fig. 3A, only HepG2 and HEK293 cells were positive for endogenous PEG10-RF1, although T47D and LCLC cells do express the PEG10 gene as demonstrated by RT-PCR (data not shown). HMEC-1 cells and HUVECs do not express PEG10. HepG2 cells were also positive for the PEG10-RF1/2 protein. The endogenous expression of PEG10 in HepG2 cells is in agreement with recent findings (39,40).
Next, we were interested in the intracellular distribution of PEG10-RF1. HepG2 cells were tested by immunostaining with the SAON1 antibody, which revealed a cytoplasmic localization of the endogenous PEG10-RF1 but no nuclear staining (Fig.  3B). The same pattern was also shown by Tsou et al. (40), whereas Okabe et al. (39) report nuclear as well as cytoplasmic staining. To demonstrate that the anti-PEG10 antibody SAON1 is specific, we also performed immunostaining with HMEC-1 cells, which do not express PEG10. HMEC-1 cells showed no staining with SAON1, therefore confirming the specificity of this antibody in PEG10 protein detection (see Fig.  2 in the supplemental material).
Identification of the Reciprocal ALK1/PEG10-RF1-binding Domains-To determine the protein segment in PEG10-RF1 responsible for the interaction with ALK1, we used the twohybrid system and created constructs representing different FIG. 2. The PEG10 gene codes for two proteins by a retroviral ؊1 ribosomal frameshifting mechanism. A, COS-1 cells were transiently transfected with the expression constructs His PEG10-RF1 and His PEG10-RF1/2 or empty vector. Total cell lysates were subjected to SDS-PAGE under reducing conditions, followed by Western blot (WB) with anti-His antibody. Western blot for cells transfected with His PEG10-RF1/2 reveals three proteins recognized by the anti-His antibody. One protein, PEG10-RF1, represents the protein encoded by reading frame 1. One protein of ϳ100 kDa, PEG10-RF1/2, represents the polypeptide encoded by reading frames 1 and 2 together, confirming that PEG10 uses a Ϫ1 ribosomal frameshifting for protein translation. The third protein, PEG10-CNF, is thought to be the cleavage product of an internal aspartyl protease motif of the PEG10-RF1/2 protein. B, ALK1 interacts with PEG10-RF1/2. COS-1 cells were transiently transfected with His PEG10-RF1/2 alone or in combination with ALK1. Total cell lysates were subjected to immunoprecipitation (IP) with anti-HA antibodies. Immunoprecipitates were separated by SDS-PAGE under reducing conditions, followed by Western blot with anti-His antibody, showing that PEG10-RF1/2, but not PEG10-CNF, is precipitated by ALK1. The asterisk marks the unspecific band.
parts of PEG10-RF1 that were tested with ALK1/c. The scheme and the results of our two-hybrid based interaction study are summarized in Fig. 4A. A large domain of 200 amino acids, amino acids 76 -275 (PEG10-RF1 76 -275 ), was identified as being responsible for the PEG10-RF1/ALK1 interaction, named PEG10-RF1/ALK1-interaction region (PAIR). Within this 200amino acid PAIR domain, we could not identify smaller segments mediating ALK1 binding. PEG10-RF1 76 -275 was able to activate all three reporter genes; however, the growth rate and ␤-Gal assay results suggest a weaker ALK1 binding capacity than for the full-length PEG10-RF1. On AdeϪ plates, good colony growth for PEG10-RF1 76 -275 was only seen after 5 days compared with 2 days for PEG10-RF1. To induce the Gal7 promoter above background levels in the ␤-gal assay, PEG10-RF1 76 -275 had to be grown under selection pressure. The blue color change for PEG10-RF1 76 -275 took ϳ4 h, whereas for PEG10-RF1 it took only 2 h. A fragment consisting of amino acid residues 1-103, containing the coiled-coil domain, might also be involved in ALK1 binding because of colony growth on HisϪ plates. However, growth was not observed until 5 days, and no growth at all was seen on AdeϪ plates. The putative His-positive colonies exhibited only background levels of ␤-gal activity; therefore, it is unlikely that these residues play a major role in ALK1 binding.
We cloned the PAIR domain (PEG10-RF1 76 -275 ) into a eukaryotic expression vector with an N-terminal His tag to test the interaction with ALK1 in COS-1 cells. ALK1 immunoprecipitations were unable to co-precipitate PEG10-RF1 76 -275 , and the reciprocal co-precipitation was also negative (data not shown), indicating that unlike in the two-hybrid system, the PAIR domain alone is insufficient to mediate ALK1 binding in cell culture. The interaction domain might require the flanking amino acid sequences to fully obtain a specific conformational state for its strongest interaction with ALK1, and/or further factors might be necessary to stabilize or modulate PEG10-RF1 binding to ALK1 under in vivo conditions.
Next we investigated which ALK1 protein segment mediates the interaction with PEG10-RF1. Again, we used the twohybrid system and created constructs representing different parts of ALK1/c that were tested with PEG10-RF1. The scheme and the results of our two-hybrid based interaction study are summarized in Fig. 4B. We identified two binding regions, amino acid segment 340 -386, ALK1/PEG10-RF1 interaction domain-1 (APID-1) and the C-terminal region from amino acid 434 -503, ALK1/PEG10-RF1-interaction domain-2 (APID-2), that might act in a synergistic fashion. Each region alone was able to interact with PEG10-RF1, although the interaction was weak and was only seen under His selection and not under Ade selection. Interestingly, amino acid fragment 340 -503, which contains both binding domains, was not able to interact with PEG10-RF1, suggesting that the two binding epitopes under this condition are obscured. We also tested the HHT patient mutation M376R (ALK1/c M376R ), which is located in APID-1, in the two-hybrid system. Yeast co-transformed with PEG10-RF1 and ALK1/c M376R were not able to grow on HisϪ or AdeϪ plates (data not shown), demonstrating that this mutation prevents PEG10-RF1 binding and supporting our results that amino acid epitope 340 -386 is involved in PEG10-RF1 binding.
We also investigated whether ALK4, ALK5, T␤RII, and ActRII could bind to PEG10-RF1 using the two-hybrid system. None of the receptor/PEG10-RF1 co-transformed clones were able to activate the ADE2 reporter gene and only T␤RII-transformed yeast showed growth on HisϪ plates. This demonstrates that only ALK1 shows a highly specific interaction with PEG10-RF1. We also examined endoglin for binding to PEG10-RF1. No interaction was detected in co-immunoprecipitation experiments with full-length endoglin protein or in a two-hybrid assay with the endoglin cytoplasmic domain (data not shown).
PEG10-RF1 Inhibits TGF-␤ Receptor Type I Signaling-Previously we had shown that TGF-␤1, TGF-␤3, and an unknown human serum factor induce ALK1 signaling (19). Although ALK1 is a TGF-␤ receptor, it phosphorylates Smad1/Smad5 (17,18). The (SBE) 4 luciferase reporter responds to Smad2/ Smad3 as well as to Smad1/Smad5 (29) and therefore is a suitable reporter for signaling assays with ALK1 as well as ALK5. To study a possible role of PEG10-RF1 in ALK1 signaling, the ALK5-deficient mink lung cell line R1B was co-transfected with the (SBE) 4 luciferase reporter and ALK1 alone or ALK1 plus PEG10-RF1, respectively. To induce ALK1 signaling we used human serum. R1B cells transfected with ALK1 alone showed a clear, increased reporter activity upon human serum induction (Fig.  6A). Nonactivated ALK1-transfected cells showed a higher basal reporter activity than mock transfections or ALK1and PEG10-RF1 co-transfected cells. However, in ALK1 and PEG10-RF1 co-transfections, human serum induced ALK1 signaling was reduced by 50%, suggesting that PEG10-RF1 blocks ALK1 signaling. The same inhibitory activity of PEG10-RF1 was also seen when we tested TGF-␤3-induced ALK5 signaling in R1B cells in the presence of PEG10-RF1 (Fig. 6B). Thus, we conclude that PEG10-RF1 is a general inhibitor of ligandinduced type I receptor signaling.
PEG10-RF1 and ALK1 Co-localize and Induce Morphological Changes in Cells-The interaction of PEG10-RF1 with ALK1, a cell surface transmembrane receptor, suggested a localization of PEG10-RF1 in the cytoplasm. We had already demonstrated this with our immunostaining results for the endogenous PEG10 protein in HepG2 cells. For further cellular localization studies and functional assays, we used the PEG10-RF1 YFP construct and fluorescence microscopy.
COS-1 and CHO-K1 cells transfected with PEG10-RF1 YFP showed good expression of PEG10-RF1 YFP 24 -48 h post-transfection. In both cell types, PEG10-RF1 YFP is restricted to the cytoplasm (Fig. 7), resembling the pattern seen for endogenous PEG10-RF1 in HepG2 cells. It was also noted that PEG10-RF1 YFP preferentially localized to what appear to be cytoskeletal and/or rough endoplasmic reticulum structures.
We then determined the effects of ALK1 and its constitutively active form ALK1 Q201D on PEG10-RF1 YFP cellular localization. Co-transfections of neither ALK1 and PEG10-RF1 YFP (Fig. 7) nor ALK1 Q201D and PEG10-RF1 YFP (data not shown) changed the intracellular localization of PEG10-RF1 YFP . To investigate whether other type I or II receptors effect PEG10-RF1's cellular distribution, the different type I and II receptors were tested with PEG10-RF1 YFP in COS-1 and CHO-K1 cells. Again, PEG10-RF1 was localized solely in the cytoplasm (Figs. 7 and 8A and data not shown). ALK1 CFP , ALK5 CFP , and T␤RII CFP were used for co-localization studies. Co-expression of PEG10-RF1 YFP and ALK1 CFP showed a strong co-localization. The merged photographs of PEG10-RF1 YFP and ALK1 CFP showed co-localization to the cytoskeletal and/or rough endoplasmic reticulum structures as well as to the plasma membrane and in vesicles (Fig. 7). Colocalization of PEG10-RF1 YFP with either ALK5 CFP or T␤RII CFP was more restricted to the cytoskeletal and/or rough endoplasmic reticulum structures.
In CHO-K1 cells expressing PEG10-RF1 YFP alone, we observed a number of cells accumulating PEG10-RF1 YFP around the nucleus after ϳ24 h. Subsequently, these fluorescent cells showed a more rounded shape before detaching, and they were observed floating in the medium. For the first 48 h following transfection, the number of cells expressing PEG10-RF1 YFP alone was similar to cells co-expressing ALK1 and PEG10-RF1 YFP . In contrast to YFP transfected cells, after 4 days only a small number of PEG10-RF1 YFP transfected cells still expressed PEG10-RF1 YFP , possibly because of cells undergoing apoptosis induced by PEG10-RF1. A similar effect of reduced cell numbers expressing PEG10-RF1 YFP was also seen in coexpression studies with the different type I and II receptors, whereas ALK1 and PEG10-RF1 YFP co-transfected cells showed steady and prolonged expression of PEG10-RF1 YFP for at least 4 days.
Co-expression of ALK1 and PEG10-RF1 not only demonstrated the co-localization of the two proteins, but we also observed a dramatic effect on cell morphology. When ALK1 and PEG10-RF1 were co-expressed in either COS-1 or CHO-K1 cells, lamellipodia, filipodia, and dendrite-like changes appeared that were not seen in cells expressing ALK1 or PEG10-RF1 alone (Fig. 7). Some of the cellular extensions were so thin that they were only clearly visible under higher magnification (63ϫ). Many of the cells were not in direct contact along their cell membranes but formed a network in which the lamellipodiae and filipodiae extended to make the contact with neighboring cells. In some cases it appeared as if the ends of the filipodiae were fused with the contacted cell (data not shown), although we were unable to prove this. These changes were less obvious in COS-1 cells than in CHO-K1 cells. COS-1 cells form cellular extensions and protrusions, which we believe are different from what we observed for COS-1 cells co-expressing ALK1 and PEG10-RF1. However, as shown in Fig. 7B, the effects in CHO-K1 cells were profound and reminiscent of neuronal cell networks. We also tested rat myoblast L6E9E cells and HepG2 cells that demonstrated similar morphological changes (data not shown).
Next we tested the different above already mentioned type I and II receptors. None of the receptors when co-expressed with PEG10-RF1 in CHO-K1 cells induced cell morphological changes ( Fig. 8A and data not shown). Furthermore, we wanted to examine what effect ALK1 HHT patient mutations might have on PEG10-RF1 binding as well as on cellular behavior when co-expressed with PEG10-RF1. For this purpose, ALK1 missense mutants were co-expressed with PEG10-RF1 in COS-1 cells. Western blots for wild-type ALK1 always show a higher molecular mass immunostaining of ϳ70 -75 kDa and a lower molecular mass band at 60 -65 kDa. The higher molecular mass molecule represents the ALK1 protein present at the cell surface. This was proven by biotin labeling cell surface proteins of COS-1 cells transfected with ALK1 and subjecting the lysed cells to an anti-ALK1 immunoprecipitation. Precipitated, biotin-labeled ALK1 was visualized by immunoblotting with a streptavidin-conjugated antibody (data not shown). First, Western blot analysis showed that most of the mutant ALK1 proteins are primarily not expressed at the cell surface (Fig. 8B), suggesting that these proteins do not pass the cellular quality control and are retained in the rough endoplasmic reticulum and/or Golgi. Only the ALK1 mutant ⌬S232, an in-frame amino acid deletion right after the ATP-binding domain, showed regular cell surface processing. Some of the mutants, such as the extracellular domain mutant R67Q and the kinase domain mutants S331W and R374W, showed a weak to FIG. 5. PEG10-RF1 interacts with different members of the TGF-␤ receptor family. COS-1 cells were transiently co-transfected with His-tagged PEG10-RF1 and the indicated HA-tagged receptors. PEG10-RF1/receptor interaction was tested for by anti-His immunoprecipitation and subsequent anti-HA Western blot immunodetection for co-precipitated receptors. Two day post-transfection cells were lysed, and the lysates were subjected to immunoprecipitations (IP) with anti-HA antibody. Immunoprecipitates were separated by SDS-PAGE under reducing conditions, and co-precipitated PEG10-RF1 was visualized by Western blot (WB) with anti-His antibody.
very weak presence at the cell surface; the majority of the mutant proteins, however, remained intracellular.
We tested whether the different ALK1 mutant proteins could interact with PEG10-RF1. PEG10-RF1 was able to precipitate the ALK1 proteins with intracellularly located mutations with one exception; the P424T mutant was barely detectable, suggesting a weak interaction with PEG10-RF1. ⌬S232 showed an ALK1 wild-type precipitation profile (Fig. 8B). It is possible that the P424T mutation affects the PEG10-RF1-binding domain at the ALK1 C terminus. A surprising observation was that the K229R mutation in the ATP-binding domain showed a normal ALK1 cell surface expression, but PEG10-RF1 precipitated only the intracellularly located ALK1 KR protein. This would imply that phosphorylation events might also play a role in PEG10-RF1 binding.
Next, we examined whether the ALK1 mutants, when coexpressed with PEG10-RF1 in CHO-K1 cells, also induce changes in cell morphology (data not shown). The only mutant that affected cell morphology like the ALK1 wild-type protein was the ⌬S232 mutation. None of the other mutants, including the K229R mutant, changed the cell shapes. Based on these results, we conclude that the observed morphologic changes only occur because of the interaction of PEG10-RF1 with an ALK1 protein present at the cell surface. DISCUSSION In the present study we set out to identify intracellular partners for ALK1, the HHT-2 gene and TGF-␤ type I receptor. The two proteins identified and their corresponding gene, PEG10, have a number of unusual features. PEG10, also reported as KIAA1051, is an imprinted gene (30) suggesting a role in embryonic development. Furthermore, it has two overlapping open reading frames, ORF1 and ORF2. Overlapping reading frames are characteristic of retroviral genomes but are also found in bacteria and yeast (41,42). Using the nucleotide and protein sequences of PEG10 ORF1 and ORF2 in homology searches, we found the highest homologies with retroviral/ retrotranposon-like gag and pol sequences of the Ty3/Gypsy retrotransposon sushi of the pufferfish, Fugus rubripes, agreeing with a previous report for KIAA1051 (31). This explains the two overlapping ORFs in the 6.3-kb-long second exon, which is a common theme in retroviral/retrotransposon elements (43).
The high homology of PEG10 with two genes in mice and mink, Edr and Trt1, respectively (32,33), suggests that these are orthologues. Interestingly, Trt1 expression is down-regu-lated by TGF-␤1, suggesting an additional link of PEG10 to the TGF-␤ signaling pathway. For Edr, the same retroviral/retrotransposon structure as PEG10 was shown with two overlapping reading frames of gag-and pol-like sequences. Edr utilizes Ϫ1 ribosomal frameshifting to encode a gag-like polypeptide and a gag-pol-like polypeptide. We were able to demonstrate that the same mechanism is used by PEG10, synthesizing two proteins of different sizes. The larger protein is presumably cleaved by internal aspartyl protease activity (Fig. 2), common in retroviruses for gag protein processing (44,45). To our best knowledge, PEG10 is the first example of a human gene coding for at least two proteins not by alternative splicing but by Ϫ1 ribosomal frameshifting.
Edr is located at the proximal region of mouse chromosome 6, which is syntenic with the human PEG10 locus on 7q21. We and others (32) hypothesize that these genes are orthologues, but we predict somewhat differing functions for PEG10 and Edr. First, in the middle of its pol-like sequence, Edr contains a 200-amino acid sequence that is absent in both PEG10 and Trt1, suggesting that this insertion/deletion was acquired after the evolutionary divergence of the mice and the human/mink lineages. Second, Edr is a developmentally regulated gene in mice, expressed between 9.5 and 16.5 days of gestation, and expression in adult animals is restricted to brain and testes. PEG10 is also expressed in embryos, adult brain, and lung (32,46), but in addition, we (Fig. 1B) and others (30) have shown that PEG10 is strongly expressed in placenta, less in kidney and testes, and only weakly expressed in spleen, liver, colon, and small intestine. These data suggest a role for PEG10 in humans that is not shared by Edr in mice, which may obstruct the simple use of Edr knock-out mice to model the role of PEG10 in humans.
PEG10 is a member of a small family of proteins coded for by retroviral elements permanently integrated into mammalian genomes. The only other known examples are Fv1 and Fv4, restricted to a subspecies of Mus musculus (47)(48)(49), and syncytin, only seen in man and monkey (50). Syncytin is the product of the env gene of the endogenous human defective retrovirus HERV-W and is involved in placental morphogenesis. Fv1 codes for a gag-like protein, and Fv4 codes for an env-partial pol-like polypeptide. Both block, or render partial resistance, to retroviral infection. It would be tempting to conclude that the gag-like protein PEG10-RF1 plays a role in retroviral defense like Fv1. For example, Zn 2ϩ -binding motifs FIG. 6. PEG10-RF1 inhibits ALK1 and ALK5 signaling. The effect of PEG10-RF1 on ALK1 and ALK5 signaling was tested in the TGF-␤ receptor type I (ALK5)-deficient RIB mink lung cell line using the luciferase reporter construct (SBE) 4 . (SBE) 4 was co-transfected with the indicated constructs. A, ALK1 signaling activity was measured in the absence or presence of human serum. The cells were incubated overnight in low serum medium (0.2% FCS) with or without (w/o) 15% human serum. B, ALK5 signaling was measured in the absence or presence of TGF-␤3. The cells were incubated overnight in low serum medium (0.2% FCS) with or without (w/o) 2 ng/ml TGF-␤3. All of the transfections were done in triplicate, and the luciferase activities shown are the averages. The standard deviation is indicated, and all experiments were normalized for ␤-gal activity.
FIG. 7. Co-expression of PEG10-RF1 and ALK1 induces cell spreading. Cellular localization was monitored using fluorescence microscopy. In co-transfections, fluorescence for the PEG10-RF1 YFP protein is shown in green, and the fluorescence for the ALK1 CFP , ALK5 CFP , and T␤RII CFP proteins is shown in red (instead of blue) as indicated by the color in the title. A and B, COS-1 cells and Chinese hamster ovary (CHO-K1) cells were transiently transfected with PEG10-RF1 YFP , ALK1 CFP , ALK5 CFP , and T␤RII CFP alone or in combination, as indicated. PEG10-RF1 YFP transfected either alone or in combination with the different receptors clearly shows a cytoplasmic (bright fluorescence) but no nuclear (dark spot) localization. Co-expression of PEG10-RF1 with ALK1, but not with ALK5 or T␤RII, induces cell spreading that resembles neuronal-like cell morphology.
in gag proteins, which are usually part of the nucleo-capsid domain, are known to be involved in the retroviral replication cycle binding viral nucleic acids during the assembly and infection stage (34,37). In addition, the nucleo-capsid domain most often contains the interaction domain responsible for the multimerization of 1500 -2000 gag proteins that make up an individual particle (51). We do not know whether PEG10-RF1 contains an interaction domain, but under nonreducing SDS-PAGE conditions we were able to detect PEG10-RF1-containing oligomers (data not shown), suggesting that PEG10-RF1 can form either homomers or heteromers with other cellular proteins. Thus, PEG10-RF1 might be able to block viral assembly by dominant-negative binding to gag proteins or viral nucleic acids, a hypothesis that requires further investigation.
PEG10 was recently found to be highly expressed in hepatocellular carcinoma and in regenerating mouse liver where it was induced in G 2 /M phase (39,40). More recent data show that PEG10, in concert with four other imprinted genes and three genes involved in chromatin integrity/function, is overexpressed in the perinatal form of biliary atresia (52), another disorder affecting liver cells. It was also reported that PEG10 binds to the apoptosis mediator SIAH and that SIAH-induced cell death is decreased by overexpressing PEG10 (39), hinting at a role in cancer development for PEG10. However, we observed a viability decrease with prolonged PEG10-RF1 expression in CHO-K1 cells not seen in COS-1 cells. These data suggest that PEG10 functions are cell type-and dosage-dependent.
We determined the ALK1 interaction region of PEG10-RF (PAIR1), which we localized to a 200-amino acid region (Fig.  4A). It is possible that this relatively large region is necessary to interact properly with the two PEG10-binding domains of ALK1, APID-1, and APID-2, which are separated by ϳ100 amino acids (Fig. 4B). When we expressed the PAIR fragment in COS-1 cells for co-immunoprecipitation experiments, we did not observe any interaction with ALK1. It is possible that for proper binding to occur, the PAIR domain requires a specific conformational state that is only fully obtained in the context of flanking amino acids. This larger sequence context was provided in our two-hybrid based interaction studies by the Nterminal fused Gal4-DNA-binding domain. This hypothesis is currently under investigation. PEG10-RF1 might be an example of proteins with low complexity, which upon binding to other proteins acquire a specific and functional three-dimensional structure. It is interesting to note that in the two-hybrid assay, the HHT ALK1 mutation M376R, which is located in the highly conserved APID-1 domain, prevented PEG10-RF1 binding, whereas an adjacent HHT mutation, R374W, was able to interact with PEG10-RF1 when tested in COS-1 cells (Fig. 8B). This might suggest that additional proteins are involved in the ALK1/PEG10-RF1 interaction, proteins that would not be present in the two-hybrid assay. The second PEG10-RF1-binding domain, APID-2, contains the NANDOR BOX, a 10-amino acid stretch that is identical among the different human type I receptors (53). Mutation or deletion of this motif abolishes TGF-␤-dependent receptor phosphorylation, receptor downregulation, and transcriptional activity. Thus, it is possible that by binding to this motif, PEG10-RF1 blocks TGF-␤-dependent signaling as we demonstrated for ALK1 and ALK5 (Fig. 6).
Co-expression of ALK1 and PEG10-RF1 induces pronounced morphological changes in cultured cells, whereas co-expression of PEG10-RF1 and other type I or II receptors does not. This appears to be a unique feature of the ALK1 and PEG10-RF1 interaction. Our findings suggest that the ALK1 protein, although thought to be not activated in terms of ligand-induced signaling, influences yet unknown cellular functions. Therefore, the definitions of active and nonactive receptors require reconsideration. The observed morphological changes caused by the interaction of ALK1 with PEG10-RF1 link ALK1 to cell adhesion/migration processes. For these functions, ALK1 needs to be present at the cell surface because the morphological changes were not observed for ALK1 mutants, which were retained in the cell. The involvement of ALK1 in cell spreading seems to be independent of its signaling activity, which we conclude from the following. First, the cells were cultured without TGF-␤ or bone morphogenetic protein, although we cannot exclude some signal induction activity because of FCS in the culture medium. Second, we saw that the ALK1 mutant ⌬S232 is expressed at the cell surface and induced morphological changes, although it is thought to be signaling-inactive. Third, we have shown that PEG10-RF1 inhibits TGF-␤ signaling.
Cell migration and spreading require the involvement of cytoskeletal and focal adhesion proteins. PEG10-RF1 might function as an adaptor/scaffolding protein that connects ALK1 with cytoskeletal or focal adhesion proteins. The possible binding of PEG10-RF1 to other proteins in this context could be mediated by its N-terminal coiled-coil domain. This type of domain is a highly versatile protein folding and oligomerization motif, reported in cytoskeletal and motor proteins, as well as in transcription factors (54).
The fact that many different cell types express the PEG10 gene but only a few cell types also translate the message suggests that a post-transcriptional translation control mechanism exists for PEG10. We are currently investigating whether the 5Ј-untranslated region and/or the very long 3Јuntranslated region is involved in this control process. The proteins encoded by PEG10 suggest that retroviral sequences have not merely accumulated as "junk DNA" in mammalian genomes but have played an active role in mammalian evolution. The integration of retroviral genes into mammalian genomes adds a new dimension to the "one gene one protein" discussion and needs to be taken into account when examining the human genome.