Sequences in the 3”Untranslated Region of the Human Cellular Glutathione Peroxidase Gene Are Necessary and Sufficient for Selenocysteine Incorporation at the UGA Codon*

Glutathione peroxidase (EC 1.1 1.1.9) is one of a unique group of prokaryotic and eukaryotic enzymes that contain the unusual amino acid selenocysteine. The genes for these selenoproteins encode for the atyp-ical amino acid at a TGA codon (UGA in the mRNA transcripts), which normally functions as a termination signal. The present studies analyzed the functional importance of sequences in the coding and 3‘-untrans-lated regions of transcripts of the primary human cel- lular glutathione peroxidase gene (GPXl ) to the insertion of selenocysteine at this UGA codon. Deletions in potential stem-loop or hairpin structures in the coding region did not substantially diminish incorporation of selenocysteine into glutathione peroxidase transiently expressed by the pCMV4 vector in COS-1 cells. How- ever, selenocysteine insertion was completely abolished by deletion of four-nucleotide sequences in the 3”untranslated region from within a conserved “selenocysteine insertion sequence’’ motif also found in the 3”untranslated region of mammalian genes for other selenoproteins. Moreover, in constructs of hybridization of 3 of total cell 10 pg of yeast 6 pl of riboprobe


Sequences in the 3"Untranslated Region of the Human Cellular Glutathione Peroxidase Gene Are Necessary and Sufficient for Selenocysteine Incorporation at the UGA Codon*
(Received for publication, January 6, 1993) Qichang ShenS, Fong-Fong Chug, and Peter E. NewburgerSll

From the $Departments of Pediatrics and of Molecular GeneticslMicrobiology, University of Massachusetts Medical School, Worcester, Massachusetts 01655 and the §Department of Medical Oncology and Therapeutics Research, City of Hope National Medical Center, Dwrte, California 91010
Glutathione peroxidase (EC 1.1 1.1.9) is one of a unique group of prokaryotic and eukaryotic enzymes that contain the unusual amino acid selenocysteine. The genes for these selenoproteins encode for the atypical amino acid at a TGA codon (UGA in the mRNA transcripts), which normally functions as a termination signal. The present studies analyzed the functional importance of sequences in the coding and 3'-untranslated regions of transcripts of the primary human cellular glutathione peroxidase gene (GPXl ) to the insertion of selenocysteine at this UGA codon. Deletions in potential stem-loop or hairpin structures in the coding region did not substantially diminish incorporation of selenocysteine into glutathione peroxidase transiently expressed by the pCMV4 vector in COS-1 cells. However, selenocysteine insertion was completely abolished by deletion of four-nucleotide sequences in the 3"untranslated region from within a conserved "selenocysteine insertion sequence'' motif also found in the 3"untranslated region of mammalian genes for other selenoproteins. Moreover, in constructs fusing the glutathione peroxidase 3'-untranslated region to the coding region of rab5b (an unrelated protein normally without any selenium moiety), the glutathione peroxidase 3'-untranslated region was sufficient to direct the translation of an opal (UGA) mutation as selenocysteine. Thus, our data directly demonstrate the importance of the selenocysteine insertion motif in the glutathione peroxidase gene and specifically show that sequence elements in the 3"untranslated region are both necessary and sufficient for translational insertion of selenocysteine at a UGA codon in eukaryotic mRNA.
Selenoproteins encompass a unique group of prokaryotic and eukaryotic polypeptides incorporating the unusual amino acid selenocysteine. This small class of proteins includes several enzymes, such as bacterial formate dehydrogenases (1,2) and the mammalian glutathione peroxidase (GPx)' family (3)(4)(5)(6) and type I iodothyronine 5"deiodinase (7), all of which contain a single selenocysteine moiety within the active site. Another common mammalian species, selenoprotein P, contains seven to ten selenocysteine moieties and may function as a transport protein or antioxidant (8). All of these proteins incorporate the selenocysteine cotranslationally at a UGA codon (9), formerly known only as a stop codon, and utilize a unique selenocysteine-charged tRNA containing the appropriate UCA anticodon (10,11).
A critical question in the interpretation of this "extended genetic code" is how the ribosomal translation assembly can discriminate the special UGA codon in the open reading frame of a selenoprotein mRNA from the termination UGA codon in other mRNA species. Since the first identification of the UGA codon, speculation has focussed on the possibility that one or more consensus sequences would function as a signal for insertion of selenocysteine, most likely by formation of specific mRNA secondary structures. Two major models have emerged.
In E. coli formate dehydrogenase mRNA, translation of the UGA codon as selenocysteine depends upon a 40-base stemloop structure immediately downstream from the UGA and containing critical bases within the loop, 20 bases downstream from the UGA (2,9). Very similar stem-loop structures were identified in the human GPx sequence, as well as in other bacterial selenoenzymes (2).
In contrast, in rat 5"deiodinase mRNA, recognition of UGA as a selenocysteine, rather than a termination, codon depends upon a 200-nucleotide "selenocysteine insertion sequence" in the 3"untranslated region (3'-UTR) (12). Sequence analysis of this region predicted a large stem-loop structure with unpaired UGAU and AAA sequences in the stem and loop, respectively. The sequence motif was conserved in the rat and human 5"deiodinase 3'-UTR. However, these studies utilized relatively large deletions and did not specifically test the importance of the three-and four-nucleotide sequence elements in the 3'-UTR or of potentially important sequence elements in the coding region. The rat GPx 3'-UTR, although different in primary sequence, also contains a potentially similar stem-loop secondary structure with the unpaired AAA and UGAU sequence pair in an analogous position (12); but the presence of this element has been detected only by sequence comparison and its functional role has never before been tested directly in GPx translation.
The present studies analyzed the functional importance of potential selenocysteine incorporation sequences in the open reading frame and the 3'-UTR to the translation of mRNA transcripts from the primary human cellular GPx gene, GPXl. Our data demonstrate that segments of the 3'-UTR, specifi-cally the conserved AAA and UGAU sequences within the selenocysteine insertion sequence motif, are essential for selenocysteine insertion. Moreover, the GPx 3'-UTR alone is sufficient to signal the translation, as selenocysteine, of an opal mutation (UGA) in the open reading frame of an unrelated non-selenoprotein.

EXPERIMENTAL PROCEDURES
Construction of GPx and rab5b Subclones-GPx subclone GPxR in the vector pBluescript KS (Stratagene) was used as a common template for constructing all GPx deletion subclones. It was derived by inversion of the orientation of a GPXl cDNA (13) in the same vector. DNA sequencing of this clone (using standard dideoxy sequencing techniques with a "Sequenase" kit (United States Biochemical Corp.)) showed one additional GCG codon immediately upstream of the previously reported codon 11 (GCC) (4, 14) (GenBank accession numbers YO0369 and M21304) and a codon 92 CTG as we have reported (14), instead of the CAG observed by Mullenbach et al. (4). The former insertion is a polymorphism we have observed in other normal GPXl sequences.' Unless otherwise indicated, GPx deletion subclones were constructed by overlap extension polymerase chain reaction (PCR) (15), using a Perkin-Elmer Cetus thermal cycler and reagents. This PCR method required two flanking primers defining the size of the final product and two mutually complementary primers directing the desired mutation in the target sequence. The sequences of the flanking primers and of one of each pair of complementary mutagenesis primers are listed in Table I. The final PCR products were inserted back into pBluescript KS, the sequences confirmed, and then the mutant GPx sequence inserted into the eukaryotic expression vector pCMV4 (16) for transfection into COS-1 cells as described below.
"Epitope tagging" of GPx was performed (as diagrammed in Fig.  1) by replacing the first 12 nucleotides (nt) of the open reading frame of GPx with a 30-nt sequence encoding an ATG start codon followed by 27 bases encoding a nine amino acid epitope of human influenza hemagglutinin protein (17). The two oligonucleotides listed in Table   I were annealed and then the resulting short double-stranded fragment inserted into GPx native or mutant subclones in pBluescript KS and/or pCMV4 via the ClaI and NheI restriction sites. In this process, amino acids 2-4 were deleted, so the epitope-tagged "GPxEPI" possessed a net increase of 6 amino acid residues more than wild type GPx. Although rabbit antiserum against this epitope was available, its binding to tagged GPx molecule was much lower than that of the antisera against GPx peptide sequences (data not shown), so the latter was still used to detect the tagged GPx molecule.
Gpx subclones with partial or complete deletion of the 3'-UTR ' Q. Shen  Subclone UTR-D2 was constructed by excision of the AvrII-XhoI fragment and re-ligation of the remaining large fragment in the GPx 3'-UTR sequence from GPxEPI-containing pBluescript KS with the plasmid XhoI site eliminated. The subclone UTR-Dl was obtained by inserting a GPxEPI containing fragment with a sticky ClaI end and an end-filled XhoI end, excised from the construct GPxEPI in pBluescript KS, into the expression vector pCMV4 via the ClaI and SmaI polylinker restriction sites. The overlap extension PCR method was also used to construct mutant and fusion subclones of the rab5b gene, which encodes a member of Ras-related GTPase superfamily (18). The plasmid pMT2, carrying a 1.6-kilobase pair rab5b cDNA clone, was obtained as a kind gift from D. B. Wilson. Construct rab5b(opal)GPx3'-UTR contained a fusion product of the rab5b coding region with an opal (UGA) mutation at codon 63, fused with the GPx 3'-UTR sequence. The oligonucleotide sequence of the flanking and mutagenesis primers are listed in Table I. The 3'-PCR flanking primer sequence resulted in the removal of the native rab5b TGA termination codon, and substitution of the last three codons of the GPx open reading frame, including its TAG stop codon. The resultant rab5b(opal) mutant was inserted into pBluescript KS containing the entire GPx 3'-UTR sequence derived from the ClaI-AurII double digestion of the native GPxR clone in pBluescript KS. The gene fusion product was inserted into pCMV4 as above. The same strategy was also used to construct rab5b(WT)GPx3'-UTR, except in this case, conventional PCR was applied using only the flanking primers, and the fusion product (WT, i.e. wild type without the opal mutation) was inserted into pCMV4. The construct rab5b(opal), which contains the coding region opal the approximately 900-nt NheI-EcoRI fragment of rab5b 3'-UTR mutation but the native rab5b 3'-UTR, was constructed by fusion of sequence with the rab5b(opal)GPx3'-UTR subclone, from which the GPx 3'-UTR had been deleted as an AurII-EcoRI fragment. The resulting rab5b(opal) sequence was then inserted into pCMV4 as above.
Transfection, Labeling, and Lysis of COS-I Cells-COS-1 cells were transfected for transient expression of GPx or rab5b subclones by modified calcium phosphate-mediated or electroporation methods (19) and then cultured in Dulbecco's modified Eagle's medium SUPplemented with 10% fetal bovine serum, 5 ng/ml sodium selenite, and 1 X penicillin-streptomycin-fungizone (GIBCO/BRL). All experiments were performed two to four times.
As a control for transfection efficiency, COS-1 cells were cotransfected with 2 pg of plasmid pXGH5 included in a human growth hormone transient expression assay system supplied by Nichols Institute. Human growth hormone secreted into the medium was de-tected by radioimmunoassay using the Crystal Multidetector Radioimmunoassay System (United Technologies Packard).
For '%e labeling, 10 pCi of '%e as selenous acid diluted in nitric acid, with an original specific activity of 750-1000 Ci/g (from the University of Missouri Research Reactor Facility), was added to the transfected cells in each plate and the cell incubation at 37 "C continued for an additional 2 h.
For %S labeling, the transfected cells in each plate were first incubated 30 min in methionine-and glutamine-free Dulbecco's modified Eagle's medium (GIBCO), supplemented with 10% dialyzed calf serum, 1 X glutamine (GIBCO), and 25 mM HEPES. Then 250 pCi of Express %S protein labeling mix (Du Pont-New England Nuclear), with a specific activity of 1140 Ci/mmol for methionine, was added to the plate, and the cells were incubated at 37 "C for an additional 2 h.
After "%e or 36S labeling, 5 or 1 pl, respectively, of &isopropyl fluorophosphate was added to ice-cooled labeling mixture in the COS-1 cell plates. After 5 min, the mixture was aspirated and 1.5 ml of COS cell lysis buffer (50 mM HEPES, pH 7.8, 1% Triton X-100, 10 mM EDTA, 1 mM phenylmethylsulfonyl chloride) was added to the plate. After 20-min incubation shaking at 4 "C, the lysed cell suspension was transferred into a microcentrifuge tube and spun at 14,000 X g for 10 min to remove the cell debris. SDS was added to the supernatant to a final concentration of 0.5%, and the cell lysate was heated in boiling water for 5 min and then cooled on ice.
Immunoprecipitation and Protein Electrophoresis-Immunoprecipitation utilized two rabbit antisera raised (by Berkeley Antibody Co., Richmond, CA) against synthetic peptide sequences from the GPx polypeptide chain, one from residues 26 to 46 and the other from residue 174 to residue 192. 15 pl of each antiserum, plus 20 pl of protein A-Sepharose CL-4B beads (Sigma) were added to the lysate. The mixture was incubated tumbling at 4 "C overnight. The beads were subsequently spun down, washed twice with washing buffer (50 mM HEPES, pH 7.8, 150 mM NaCl, 1% Triton X-100, 0.5% deoxycholate, 0.1% SDS) and once with 50 mM HEPES at pH 7.8 and then mixed with 30 pl of SDS-gel loading buffer (50 mM Tris-HC1, pH 6.8, 100 mM dithiothreitol, 2% SDS, 0.1% bromphenol blue, 10% glycerol), heated in boiling water for 3 min, and spun in a microcentrifuge. The supernatant was collected for SDS-polyacrylamide gel electrophoresis.
For COS-1 cells transfected with rab5b constructs, the procedure was the same as above except for the use of 0.2% SDS for cell lysis and the addition of 8 pl of affinity-purified rabbit antibody (obtained from D. B. Wilson), raised against a synthetic peptide from the hypervariable domain of rab5b.
RNase Protection Assay-Total cell RNA was isolated by the guanidine HCl method (20). Riboprobes were generated from the T7 promoter by use of an RNA transcription kit (Stratagene) to synthesize a 224-nt 32P-labeled RNA transcript complementary to a 179-nt segment starting at the ClaI site of the 5"untranslated region of the GPxEPI transcript. The template was a ClaI fragment of a construct formed by recircularization of an end-filled SpeI-RsrII large fragment of GPxEPI. RNase protection assays of hybridization mixtures of 3 pg of total cell RNA, 10 pg of yeast tRNA, and 6 pl of riboprobe (400,000 trichloroacetic acid-precipitable cpm/pl) were performed by standard techniques (19).  Fig. 1B) similar to that found in the mRNA of the E. coli formate dehydrogenases and related prokaryotic selenoenzyme genes (2). Another, incorporating the UGAld2 codon at the tip of a "hairpin" (shown in Fig. l C ) , is conserved among several mammalian GPx as well as E. coli formate dehydrogenase mRNA sequences (21). To test the role of each of these potential secondary structures in the direction of selenocysteine incorporation, we constructed a series of five sequential deletions in the GPx open reading frame, designated ORF-Dl through ORF-D5 and shown on  (2). GPx subclone ORF-D5 was a deletion of codon 47, located immediately upstream of the UGA142 codon of the GPx mRNA and forming part of the stem of the alternative putative hairpin loop structure (21). These deletion subclones, carried by the eukaryotic expression vector pCMV4, were individually transfected into COS-1 cells and GPx expression detected by 75Se labeling, immunoprecipitation, SDS-polyacrylamide gel electrophoresis, and autoradiography.

Role
As shown in Fig. 2  respectively) all show high levels of "5Se incorporation into GPx protein. These deletions resulted in a slight, but not substantial, decrease in GPx expression. Repeated experiments (including the creation of identical deletions in the epitope-tagged construct) also showed slightly diminished expression (data not shown). Similarly, as shown in Fig. 2B, deletion ORF-D5 produces little or no diminution of selenocysteine insertion into GPx. Thus, the putative open reading frame mRNA loop structures may slightly modulate GPx expression, but neither one appears absolutely necessary for translation of the UGA14* codon as selenocysteine in human GPx.
Role of the 3"UTR"To test the role of the 3'-UTR in selenocysteine insertion, we constructed GPx subclones containing deletions of various lengths in that region. Epitope tags (17) were incorporated into these subclones in order to improve the resolution of the transiently expressed human GPx products from the COS-1 background. As diagrammed in Fig. l A , we replaced the first four codons of GPx with a 30-nt sequence encoding an ATG start codon and a 9-amino acid epitope of human influenza hemagglutinin protein (17). The unambiguous discrimination of the transiently expressed epitope-tagged GPx was possible, because the tagged GPx migrated slowly enough on SDS-polyacrylamide gel electrophoresis that its band resolved a t a position detectably higher than that of the untagged GPx. This difference of mobility permitted assessment of transient expression of transfected constructs without the need for the substantial overexpression necessary for evaluation of the coding region deletion constructs described above. The epitope sequence was also inserted into the wild type GPx subclone GPxR to yield a new A -

Clal
EcoRl Nhel RsrlI Computer analysis of either the entire GPx mRNA or its 3'-UTR sequence using the FOLD program of the University of Wisconsin Computer Group software (22) revealed a potential secondary structure consisting of a long stem with two small loops (Fig. 4), similar to that found in the 3'-UTR of rat and human 5"deiodinase and rat Gpx genes (12). Moreover, two 4-nt sequences within the loops (UAAA in the first and UGAU in the second, indicated in Fig. 4) were identical to those at the same positions within the selenocysteine insertion sequence motif in the 5"deiodinase gene (12). In order to test whether these two short sequences were necessary for selenocysteine insertion into GPx, we constructed two small deletion mutants, UTR-D4 and UTR-D5 (diagrammed on Fig. l), that specifically eliminated each sequence. Fig. 5 shows that either of these short deletions (lanes 3 and 4 ) completely abolished detectable selenocysteine incorporation into epitope-tagged GPx in transfected COS-1 cells.

An-11 Xhol
In order to rule out the possibility that the deletion mutations affect the level of GPx transcripts, we measured levels of GPx mRNA in transfected COS-1 cells by RNase protection assays. As shown in Fig. 6   cific for the epitope-tagged GPx transcript, indicating multiple sequence differences between the human and COS-1 (monkey) GPx transcripts. COS-1 cells transfected with epitope-tagged wild-type GPx (lane 3) contain the same amount of transcript as those transfected with the UTR-D4 and UTR-D5 deletions (lanes 4 and 5 ) . Deletions in the open reading frame (ORF-Dl, -D2, and -D3) also produced no detectable change in GPx transcript levels (data not shown). Transfection efficiency, assayed by cotransfection with a vector encoding human growth hormone, was also similar from group to  group in these experiments (data not shown).
The GPx 3'-UTR Is Sufficient for Selenocysteine Insertion-Having demonstrated that sequences in the 3'-UTR of GPx mRNA are necessary for translational insertion of selenocysteine, we next investigated whether this 3'-UTR would be sufficient to direct the same process at a UGA codon in an unrelated coding sequence. The chosen target gene, rab5b, encodes a 25-kDa GTP-binding protein, a member of Rasrelated GTPase superfamily (18). This gene was used for three constructs. rab5bfopal) had a codon 63 UGU (cysteine) modified to a UGA (opal) mutant, with the native rab5b 3'-U T R gene fusion construct rab5b(opal)GPx 3'-UTR consisted of the rab5b(opal) coding sequence fused to a 3' portion of GPx cDNA incorporating the last three codons of the GPx coding region, including its stop codon (UAG), and the entire GPx 3'-UTR and rab5bfwt)GPx 3'-UTR was also a rab5b-GPx fusion product but carried the wild type codon 63 rather than the opal mutation. The fusion constructs placed the UGU (cysteine) or UGA (potential selenocysteine) codon the same number of nt upstream from the GPx 3'-UTR as in native GPx transcripts. Fig. 7 presents the results of a representative transient expression experiment with these constructs in COS-1 cells, transfected in these experiments by electroporation. The expression of rab5b was detected by an affinity-purified rabbit antibody against a synthetic peptide sequence, following either 35S (lunes [1][2][3][4] or ''%e (lanes 5-8) radioisotope labeling. COS-1 cells transfected with the vector alone (lanes 1 and 5 ) showed no detectable immunoreactive protein at the appropriate 25-kDa molecular mass for rab5b. All three constructs directed the synthesis of a 35S-labeled polypeptide of approximately 25 kDa, at detectable but widely differing levels; but only rab5b(opal)GPx3'-UTR, the fusion product of the rab5b with the opal mutation coupled to the GPx 3'-UTR, incorporated %e ( l a n e 6). The rab5b(opal) transfectants expressed only a very low level of 35S-labeled protein, probably reflecting the existence of an alternative opal nonsense suppression mechanism (23) in the COS-1 cells. No I5Se was detectable even on very long exposures (data not shown), indicating that no selenocysteine insertion occurred at the UGA codon in the presence of the rab5b 3'-UTR. No truncated polypeptide was detectable on 35S-labeled immunoprecipitates, suggesting that the short polypeptide product was unstable or not immunoreactive with the antiserum. Transfection with the rab5b (wt)GPx3'-UTR fusion construct resulted in expression of immunoreactive protein without any detectable I5Se incorporation, as expected for the wild type rab5b coding region. For these experiments, transfection efficiency was again confirmed by cotransfection with a vector encoding human growth hormone and measurement of secreted growth hormone (data not shown).

DISCUSSION
The present studies analyzed the functional importance of mRNA sequence elements in the coding region and the 3'-UTR of human GPXl transcripts to the translation of the UGA142 as the unusual amino acid selenocysteine, rather than as a termination signal. Two different models have been proposed for the direction of selenocysteine incorporation in GPx, based upon findings in other selenoprotein genes. The GPx mRNA sequence contains a potential stem-loop configuration immediately downstream from UGA'42 and homologous to a structure essential to selenocysteine incorporation in E. coli formate dehydrogenases (2,9). However, the GPx 3'-UTR also contains AAA and UGAU sequence elements in an analogous position to those found within a selenocysteine 35s label W e label  2 and 6), and rab5b(opal) (lanes 3 and 7), rab5b(wt)GPx3'-UTR (fusion product of wild type rab5b coupled to the GPx 3'-UTR) (lanes 4 and 8).
insertion sequence motif in rat type I iodothyronine 5"deiodinase (12). Our data demonstrate that these conserved short sequences in the 3'-UTR, but not the potential stem-loop or hairpin structures in the coding region, are essential for selenocysteine translation in human GPx. Moreover, our data indicate for the first time that the GPx 3'-UTR alone is sufficient to signal the translation, as selenocysteine, of an opal mutation (UGA) in the open reading frame of an unrelated non-selenoprotein, rab5b. As shown in Fig. 4, the two critical short sequences in the selenocysteine insertion motif lie within a potential stem-loop that is structurally conserved in gene sequences for several other eukaryotic selenoproteins (12). These unpaired AAA and UGAU segments had been identified previously as conserved elements by sequence analysis in a study that also demonstrated the importance of the 3'-UTR to selenocysteine incorporation in 5"deiodinase (12). However, those experiments utilized relatively large deletions and rearrangements and did not specifically test the importance of the small threeand four-nucleotide sequence elements. The present studies have demonstrated the importance of the selenocysteine insertion motif in GPx for the first time and have specifically shown the necessity of the AAA and UGAU sequence segments. Further investigation will be necessary to determine whether the surrounding secondary structure or other sequence elements within the ZOO-nt motif are also necessary for its function in GPx or 5"deiodinase. The conservation of the stem-loop structure among different selenoprotein genes in rat and human suggests that some such additional elements are necessary for this unique function of these very short sequences that must commonly occur in the 3'-UTR of other gene transcripts.
The importance of the distance between the UGA codon and the selenocysteine insertion sequence also remains unknown. Our target for induced selenocysteine incorporation, rab5b, is similar in amino acid number to GPx, and the codon mutated to UGA is a similar distance (550 nt) from the 3'-UTR as the UGA'42 in GPx. However, in 5"deiodinase the span is approximately 1200 nt, so the distance between these elements necessary for selenocysteine incorporation is probably not critical.
Notwithstanding the critical role of the 3'-UTR in the direction of selenocysteine incorporation, other sequences upstream from the UGA codon probably play a regulatory role in selenoprotein translation. Recently, Mizutani et al. (24) have suggested that the 5'-UTR of the human GPx mRNA is essential, and areas of the coding region are important, for expression of the gene in transfected COS-7 cells. They found hardly any expression of a construct containing only 10 nt of the 5'-UTR and better expression with the full native 311-nt 5'-UTR than with 257-or 408-nt constructs. However, we obtained excellent expression of GPx with only 5 nt of 5'-UTR, perhaps due to differences in the vector or the host cells. They also reported decreased, but not absent, GPx expression in mutants changed in unspecified ways in the coding region upstream of the UGA'42. Although we did not observe any substantial effect of the upstream deletion ORF-D5, we did not extensively investigate the role of the upstream sequence in selenocysteine incorporation. However, in preliminary experiments, we attempted to use bacterial neomycin phosphotransferase as a tag by inserting a nearly full-length neo' gene between the GPx fifth and sixth codons. Transient expression of this construct resulted primarily in the translation of a neo'-GPx fusion polypeptide truncated at the selenocysteine position, plus some (10-20% production) of a full-length fusion protein incorporating selenocysteine (data not shown). Also, we found a minor, albeit inconsistent, effect of downstream open reading frame deletions on selenocysteine insertion (Fig. 2 A ) . Thus, disruption of the 5'-UTR or of the coding region neighboring the UGA14' may also diminish GPx expression. These effects could be mediated by multiple regulatory pathways, of which diminished translation of the UGA14' codon is only one of many possible mechanisms.
In both prokaryotes and eukaryotes, the expansion of the genetic code to include the unusual amino acid selenocysteine has utilized the opal termination codon UGA. Also necessary to this process has been the evolution of a signal mechanism to indicate whether an individual UGA encodes for selenocysteine or for termination. Our data fit an emerging pattern in which prokaryotes utilize a signal in the coding region and eukaryotes a selenocysteine insertion sequence motif in the 3'-UTR. The eukaryotic translation signal does not appear to be interpretable by prokaryotes, at least for GPXl, which cannot be expressed in E. coli ( 2 5 ) . This difference might relate to a requirement for post-transcriptional regulation by elements within individual coding regions of prokaryotic polycistronic mRNAs. More likely it reflects the evolution, in eukaryotes, of a functional or regulatory role for the 3'-UTR and its polyadenylated tail. The latter plays a role not only in mRNA stability (26) but also in a newly described regulation of maternal transcripts during oocyte development (27). Specific sequences within the 3'-UTR determine the regulation of mRNA stability in response to cytokine stimulation (28) and to iron availability (29). The direction of codon translation in selenoproteins by the selenocysteine incorporation sequence further expands the known repertoire of the 3'-UTR in post-transcriptional gene regulation.