Characterization of the protein dimerization domain responsible for assembly of functional selenodeiodinases.

Thyroid hormone metabolism is catalyzed by a small family of selenoenzymes. Type I deiodinase (D1) is the best characterized family member and is an integral membrane protein composed of two 27-kDa subunits that assemble to a functional holoenzyme after translation. To characterize the protein domain(s) responsible for this post-translational assembly event, we used deletion/truncation analysis coupled with immune depletion assays to map the dimerization domain of D1. The results of our studies show that a highly conserved sequence of 16 amino acids in the C-terminal half of the D1 subunit, -D148FL-YI-EAH-DGW163-, serves as the dimerization domain. Based on the high conservation of this domain, we synthesized a novel bait peptide-green fluorescent protein fusion probe (DDD(GFP)) to examine holoenzyme assembly of other family members. Overexpression of either the DDD(GFP) or an inert D1 subunit (M4) into SeD2 (accession number U53505)-expressing C6 cells specifically led to the loss of >90% of the catalytic activity. Catalytically inactive D2 heterodimers composed of SeD2: DDD(GFP) subunits were rescued by specific immune precipitation with anti-SeD2 IgG, suggesting that SeD2 requires two functional subunits to assemble a catalytically active holoenzyme. These findings identify and characterize the essential dimerization domain responsible for post-translational assembly of selenodeiodinases and show that family members can intermingle through this highly conserved protein domain.

Thyroid hormone is essential for the normal growth and development of vertebrates. Thyroxine (3Ј,5Ј,3,5-tetraiodothyronine) is the major secretory product of the thyroid gland and must be deiodinated to generate the transcriptionally active metabolite, T 3 . 1 Type I iodothyronine deiodinase (D1; EC 3.8.1.4) is the best characterized member of a small family of membrane-bound enzymes that catalyze the production and disposal of T 3 (1)(2)(3)(4). All deiodinase family members are composed of 27-30-kDa polypeptides that contain the novel amino acid, selenocysteine (SeC) that is required for full catalytic activity (3,(5)(6)(7)(8). They also share an N-terminal membranespanning region and a central core of 14 -16 residues surrounding the SeC, but little else is known about the polypeptide. Gel filtration and density gradient centrifugation of detergent-soluble enzyme activity and/or affinity radiolabeled D1 showed that catalytic activity has an M r of ϳ50,000 and that the ϳ27-kDa D1 polypeptide co-migrated with a complex, suggesting that D1 is a dimer of ϳ27-kDa subunits (9 -11). The physiochemical properties of the other selenodeiodinase family members remain to be determined, although a recent report using transiently expressed, epitope-tagged D2 and D3 suggested that these family members also formed dimers, albeit at modest levels (12).
Direct analysis of D1 biosynthesis confirmed that the active enzyme was a homodimer formed by the post-translational assembly of two p27 subunits (9). Whether each subunit was catalytically functional or required post-translational assembly to become catalytically active was resolved by the finding that D1 dimers with only one functional subunit had ϳ50% of the activity of the wild-type enzyme (9,12). Importantly, these hybrid holoenzymes had a molecular mass of 54 kDa and an S 20,w of ϳ3.5 S, identical to that of the native D1 holoenzyme (9). The ability of each D1 subunit to function as an independent catalytic center was recently confirmed by Curcio-Morelli et al. (12). Interestingly, deletion of the membrane anchor located between residues 9 and 37 of p27 resulted in the formation of catalytically active D1 heterodimers composed of a full-length, membrane-bound subunit and the N-terminal truncated partner lacking a transmembrane domain (9), indicating that insertion of both D1 subunits into the membrane was not required.
Whether holoenzyme assembly required covalent or noncovalent protein-protein interactions was recently examined using transient co-expression of epitope-tagged p27 constructs (12). A minor role for intermolecular disulfide bridging between p27 partners (ranging from 1 to 5% of the total) was recently proposed based on banding pattern(s) of epitope-tagged p27 found after SDS-PAGE analysis. Presumably because of sequence diversity among deiodinase family members, dimer complexes formed between different family members were not observed after SDS-PAGE analysis, suggesting that only like family members could associate (12).
In this study, we mapped the dimerization domain(s) required for deiodinase assembly by exploiting the ability of the native p27 present in LLC-PK1 cells to form dimers with truncation fragments of the immunologically unique, inert rat p27 (M4). We also determined the role of dimerization in the assembly of a catalytically active SeD2 enzyme using C6 cells that constitutively express the 30-kDa SeD2 subunit. The results of our studies show that the highly conserved sequence present in all deiodinase family members from frogs to hu-* The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

EXPERIMENTAL PROCEDURES
Materials-All reagents were of the highest purity commercially available. Restriction endonucleases and DNA-modifying enzymes were purchased from New England Biolabs (Beverly, MA). L-3Ј-125 I-or 5Ј-125 I-labeled rT 3 was prepared by radioiodination of 3,3Ј-diodothyronine using methods described previously (13). Synthetic oligonucleotides were prepared in house or purchased from either Invitrogen or Midland Scientific, Inc. All iodothyronines were of the L-configuration and were purchased from Henning Berlin GmbH. Dulbecco's modified Eagle's medium, antibiotics, Hanks' buffered salt solution, glucose, trypsin, and G418 were obtained from Invitrogen; supplemented bovine calf serum was from HyClone Laboratories (Boulder, CO); acrylamide and N,NЈmethylenebisacrylamide were from U.S. Biochemical Corp.; ammonium persulfate and TEMED were from Bio-Rad; and dithiothreitol was from Calbiochem. Mouse monoclonal anti-GFP primary antibody and rabbit anti-mouse horseradish peroxidase secondary antibody were obtained from Molecular Probes, Inc. (Eugene, OR). Protein A-agarose and Protein G-Sepharose were obtained from RepliGen (Cambridge, MA) and Amersham Biosciences, respectively.
SeD2 cells (C6 astrocytoma cells constitutively expressing an exogenous SeD2) were grown in the presence of 200 g/ml G418 in 25-cm 2 flasks in a humidified atmosphere of 5% CO 2 and 95% air at 37°C as described previously (15).
Site-directed Mutagenesis and Deletion Constructs-Site-directed mutagenesis was used to substitute Ser for the active center SeC to inactivate the catalytic activity of rat p27 as detailed previously (9). M4, the catalytically inert rat p27 subunit (9), was subcloned into the KpnI-XhoI site of the eukaryotic expression vector, pcDNA3, and used as the template for truncation and deletion construct synthesis.
Progressive deletion of 19 -30-residue long segments from both the N and C terminus of p27 was done by oligonucleotide-primed PCR. All primers used to create deletion and truncation mutants of D1 are listed in Table I. All upstream oligonucleotide primers have a 5Ј-end KpnI site for subcloning and an in-frame Met in the Kozak consensus translation initiation site to ensure optimal expression in mammalian cells. A common downstream primer complementary to nucleotides 895-914 (G21 cDNA; GenBank TM accession number X57999) was used for all N-terminal truncation mutants.
For C-terminal deletion mutants, downstream oligonucleotide primers were synthesized that generate D1 subunits terminating at residues 130, 150, 170, 190, 210, and 230, respectively. The 8-residue-long epitope recognized by the anti-rat p27 IgG followed by tandem univer-sal stop codons was appended to all C-terminal deletion constructs to allow for immune depletion by the species-specific p27 antisera as previously described (antibody 3050 (9,14)) (see Table I). A common upstream primer corresponding to nucleotides 7-19 of G21 cDNA and a Kozak consensus initiation site was used. PCR was done using Vent® DNA polymerase (Stratagene, La Jolla, CA) for a total of 25 cycles according to the following temperature profile: 95°C, 1.0 min; 50°C, 1.0 min; 72°C, 2.0 min, with a final 10-min extended incubation at 72°C. PCR products were digested with KpnI, gel-isolated on 1.2% low melt agarose gels, and ligated into the KpnI-SmaI sites of pcDNA3 (Nterminal deletions) or into KpnI-HindIII sites of pcDNA3 (C-terminal deletions). All deletion/truncation mutants were confirmed by DNA cycle sequencing.
Complementary synthetic oligonucleotides corresponding to nucleotides 445-496 of G21 cDNA were used to synthesize the D1 dimerization domain-GFP fusion protein (DDD GFP ). The 72-mer sense oligonucleotide (5Ј-TAGACCACCATGGTCGCTGACTTCCTCATCATTTACAT-TGAAGAAGCTCACGCCACAGATGGATGGGCTCTG-3Ј) and its 72mer complementary oligonucleotide (5Ј-GATCCAGAGCCCATCCATC-TGTGGCGTGAGCTTCTTCAATGTAAATGATGAGGAAGTCAGCGA-CCATGGTGGT-3Ј) were annealed, yielding a double-stranded 72-bp fragment with an N-terminal Met in a Kozak consensus start site, an NheI-compatible 5Ј overhang, and a BglII-compatible 3Ј-overhang. The annealed DDD encoding sequence was appended to the N terminus of GFP by cloning into the NheI BglII sites of pEGFP-N1, generating pDDD GFP . Correct insertion was confirmed by DNA cycle sequencing.
Generation of M4-and DDD GFP -expressing LLC-PK1 and SeD2 Cells-LLC-PK1 cells were grown in growth medium to 60% confluence and transfected with 10 g of either pcDNA3 (empty vector control), pcDNA3-M4 deletion, and truncation construct(s), as indicated in individual figure legends, or pDDD GFP using cationic liposomes (DOTAP®; Roche Applied Science) according to the manufacturer's instructions. Stable transfectants were selected at 1 mg/ml G418, and antibioticresistant cells were expanded without clonal isolation. Cells constitutively expressing M4 (inactivated D1 subunit), M4 deletion and M4 truncation mutants, and DDD GFP were grown in 800 g/ml G418 to maintain expression of the exogenous gene product.
SeD2-expressing C6 cells were grown to 60% confluence in growth medium containing 200 g/ml G418 and transfected with 10 g of either pEGFP (GFP control) or pcDNA3-M4 (D1 subunit) or 10 g of pDDD GFP fusion construct using cationic liposomes (DOTAP®; Roche Applied Science) according to the manufacturer's instructions. Cells expressing the GFP fusion proteins were amplified by selection with 1.5 mg/ml G418. Greater than 95% of the hyperresistant G418-selected SeD2 cells expressed GFP as judged by fluorescence microscopy.
DNA Sequencing-Double-stranded DNA sequencing was done by the dideoxynucleotide method of Sanger et al. (16) using cycle sequencing and iterative primers in the Nucleic Acid Core Facility at the University of Massachusetts Medical School. All sequence information was confirmed by sequencing both strands.
Antibody Depletion of Deiodinase Dimers-Anti-epitope antibodies directed against the C terminus of rat p27 or rat SeD2 were used for pull-down assays. Incubations were done for 90 min at 4°C in a total volume of 100 l that contained 50 -100 units of tDOC-solubilized enzyme (D1 or D2) prepared from purified membranes (9, 10), 100 mM HEPES buffer (pH 8.0), 100 mM NaCl, 0.1 mM phenylmethylsulfonyl fluoride, 5 mM tDOC, 10 l of either immobilized rProtein A TM -agarose beads (Repli-TABLE I Primers used in the preparation of truncation and deletion mutants of rat D1 For X series, the KpnI site is in italic type. For the C series primers, the unique 8-residue-long epitope recognized by the anti-rat p27 antibody (antibody 3050) was added. The Kozak initiator Met is indicated in boldface type.
Gen, Cambridge, MA) or immobilized rProtein G-Sepharose TM beads (Amersham Biosciences), and 1 l of either anti-rat p27 IgG (antibody 3050) or anti-SeD2 IgG (antibody 763). Where indicated, excess C-terminal p27 peptide (10 g/ml, final concentration) was added to block antibody-binding sites prior to the pull-down assay. Immune complexes were removed by centrifugation, and the deiodinase activity remaining in the clarified supernatant was determined as described below. Expression levels of the exogenous M4 deletion and truncation constructs in LLC-PK1 cells were determined by a competitive binding assay as described previously (9). The minimum detection limit of M4 and its derivatives is ϳ3 fmol/assay tube (81 pg of p27/tube), and the maximum binding capacity of the anti-rat p27 antiserum is 22 nmol of M4/ml of antiserum (ϳ600 ng of p27/l) as determined by Scatchard analysis of the binding data.
SeD2-specific immune complexes were removed by centrifugation and released from the beads by heating at 70°C for 5 min in Laemmli sample buffer. Proteins were separated on 12.5% SDS-polyacrylamide gels, transferred to nitrocellulose, and probed with anti-GFP monoclonal antibodies. Specific immune complexes were identified with horseradish peroxidase-conjugated, anti-mouse IgG (rabbit; Molecular Probes) and visualized by chemiluminescence.
Photomicroscopy-Cells expressing the DDD GFP fusion protein were seeded onto poly-D-lysine (10 g/ml)-coated glass coverslips (22 ϫ 22 mm) and grown to ϳ80% confluence. Cells were then fixed with 4% paraformaldehyde (w/v) in phosphate-buffered saline, and the excess cross-linker was quenched with 100 mM glycine. The distribution of GFP fusion protein in LLC-PK1 and SeD2 cells was then determined by fluorescence microscopy. Digital images were captured using a Spot CCD camera (Meyer Instruments, Inc., Houston, TX) and processed using Adobe Photoshop software. Photomicrographs shown are representative of 20 -30 independent fields.
Analytical Procedures-D1 activity was determined in cell lysates by measuring the release of radioiodide from 10 M [ 125 I]rT 3 (100 cpm/ pmol) at 20 mM dithiothreitol (unless otherwise indicated) in a total volume of 100 l. D1 reactions were done, in triplicate, in universal deiodinase buffer (100 mM boric acid, 100 mM HEPES, 100 mM glycine) (pH 5.5), 1 mM EDTA, with 10 -25 g of cell protein and incubated at 37°C for 10 -30 min. Product formation was measured as described previously (13), and the data are expressed as units/mg protein where 1 unit equals 1 pmol of I Ϫ released per min for D1 and 1 fmol of I Ϫ released per min for D2.
D2 activity was determined at substrate concentrations of 2 nM [ 125 I]rT 3 (500 cpm/fmol), 20 mM dithiothreitol in the presence of 1 mM 6-n-propyl-2-thiouracil as described previously (17,18). All experiments were performed at least three times. Statistical analysis was done by Student's t test.

RESULTS
The ability of an overexpressed catalytically inert D1 subunit to partner with Ͼ95% of the native p27 subunit(s) in LLC-PK1 cells (9) allowed us to use changes in catalytic activity as a measure of protein-protein interaction(s). Preliminary attempts to recover deiodinating activity from immune precipitated p27 wt :M4 heterodimers were unsuccessful due, in large part, to the Ͼ80% reduction in enzyme activity due to the nonspecific binding of the iodothyronine substrate to the immunoaffinity beads (data not shown). Since these nonspecific interactions are partially reversed under acidic conditions, we re-evaluated the pH dependence of the D1 reaction in an attempt to find conditions that would allow detection of immune precipitated enzyme. Shown in Fig. 1 is the pH profile of D1 activity present in detergent-solubilized COS7 cell lysates expressing a catalytically active D1 (G21 (5)). Contrary to traditional D1 assay conditions, deiodination showed a maximum at pH ϳ5.5, identical to that of the pK a of the reactive SeC nucleophile and ϳ2 orders of magnitude lower than that used to measure D1 activity (19 -21). At optimal pH, the catalytic potential of D1 was 2-3-fold greater than that measured at pH 7.0. All subsequent D1 assays were done at pH 5.5.
To confirm immune precipitation, D1 heterodimers with catalytic activity, cell membranes from LLC-PK1 cells constitutively expressing M4 or vector controls were solubilized with 5 mM tDOC, and 2.5 mg of solubilized protein was incubated with 10 g of anti-rat p27 IgG (antibody 3050; the IgG does not recognize the endogenous LLC-PK1 p27). Immune complexes were collected with MagPrep® Protein A beads, washed free of unbound proteins, and then resuspended in 100 l of Universal Deiodinase Buffer. As shown in Table II, the parent cell lysates with M4:p27 wt dimers had ϳ50% of the activity of the control cells (9), and more than 95% of the activity lost by immune depletion was recovered in the immune pellets. These data confirm the nearly complete formation of D1 heterodimers composed of the inert M4 subunit coupled to the catalytically active wild-type p27 donated by the LLC-PK1 cell.

D1 Dimerization Domain Mapping by Deletion and Truncation Mutants of M4 Expressed in LLC-PK1
Cells-To define the D1 dimerization domain, we prepared a series of N-terminal deletion (X series) and C-terminal truncation (C series) mutations of M4. X and C series mutants were transfected into LLC-PK1 cells, and constitutively expressing cell lines were isolated by G418 selection. All X-series M4 mutants had a Kozak consensus translation initiation site to ensure adequate expression, and all C-series mutants had the 8-amino acid epitope recognized by the anti-rat p27 IgG at the C terminus. Expression levels of the exogenous X and C series protein(s) in the G418-selected LLC-PK1 cells were determined by competitive displacement analysis and revealed that the cell content of the different truncated/deleted M4s varied from 0.9 Ϯ 0.1 to 1.8 Ϯ 0.3 pmol/mg cell protein, yielding molar ratios of X or C series M4 to p27 wt that ranged from ϳ5 (X133, C230) to 10 (X42, C150), values similar to those reported previously (9).
Cell membranes were then prepared for each X and C series M4 mutant by Percoll gradient centrifugation, and the enzyme activity in detergent extracts of the membrane fractions was measured after immune depletion with anti-rat p27 antisera with or without excess blocking peptide as described under "Experimental Procedures." As illustrated in Fig. 2A, Ͼ95% of the D1 activity in cells expressing the X series of M4 deletion mutants (X19 to X133) was specifically immune-depleted from detergent extracts, just as found previously with the full-length M4 (9). The percentages of D1 heterodimer formation, calculated from the difference in the quantity of residual D1 activity in immune depleted extracts measured in the absence and presence of blocking peptide, are summarized in Fig. 2B. More than half of the p27 could be deleted from the N terminus (up to residue 133, including the transmembrane anchor, the substrate specificity site, and the catalytic center) without altering FIG. 1. Effects of pH on D1 activity of wild type p27 (G21) expressed in COS7 cells. COS7 cells were transfected with rat p27 (G21) and grown for 72 h. Cell lysates were prepared, and D1 activity was determined at the pH indicated using a universal buffer composed of 100 mM boric acid, 100 mM HEPES, and 100 mM glycine. One unit is equivalent to 1 pmol I Ϫ released/min. Data are reported as the means of closely agreeing (Ϯ10) triplicate determinations the binding of the truncated M4 to p27 wt . These data indicate that the dimerization domain resides in the C-terminal 120 residues of p27.
Direct analysis of progressive truncation mutants from the C terminus of p27 revealed that at least 86 residues could be removed from the C terminus of M4 without altering the interaction between the truncated M4 and p27 wt (Fig. 2B). However, loss of residues located between positions 150 and 170 completely eliminated the ability of the truncated M4 mutant to interact with p27 wt . These data indicate that the D1 dimerization domain resides in the C terminus of p27 between residues 150 and 170, a region that contains the most highly conserved element of the deiodinase family from frogs to humans.
Kinetic Analysis of Engineered D1 Heterodimers Lacking a Transmembrane Anchor (X42), a Substrate Specificity Domain (X72), or One Catalytic Center (X133)-Shown in Fig. 3 are representative initial velocity reaction kinetics for the wildtype D1 wt , the full-length M4:p27 wt , and the X42:p27 wt dimers. All hybrid D1 heterodimers showed ping-pong reaction kinetics. Secondary replots yielded the limiting K m and V max values (Table III) and revealed that the loss of the N-terminal membrane anchor (X42), the iodothyronine specificity site (X101), and even the catalytic center (X133) resulted in a 2.5-fold decrease in the limiting K m for the iodothyronine and a 75% fall in the molar activity compared with the native enzyme. Unex-pectedly, loss of the membrane anchor increased the reduced thiol cofactor requirement ϳ4-fold (Table III). These data show that kinetic properties of the D1 reaction are intrinsic to each subunit.
Characterization of the Interaction(s) between p27 wt and the DDD GFP Fusion Protein in LLC-PK1 Cells-Amino acid residues from 147 to 163 of rat p27 (A 147 DFLVIYIEEAHDGW 163 -) are the most highly conserved of all p27 domains across all species and among family members, and our mapping strategy indicated that residues in this region are responsible for assembly of the D1 holoenzyme. To establish that this conserved region was responsible for D1 dimerization, we appended a 19-residue-long peptide containing an initiator Met and residues 147-163 of rat p27 (G21) to the N terminus of the fluorescent reporter protein, GFP, forming a DDD GFP fusion protein. Transient expression of DDD GFP in LLC-PK1 cells resulted in the appearance of multiple punctuate fluorescent signals along the cell periphery and at the borders between adjacent cells (Fig. 4), a pattern identical to that observed previously for the D1 holoenzyme (9,14). Control cells transfected with the wild type GFP reporter showed no specific membrane localization but abundant reporter protein throughout the cell interior and the nucleus (Fig. 4).
Subcellular fractionation of DDD GFP -expressing cell lysates was on iso-osmotic 16% Percoll gradients as detailed previously (14). Shown in Fig. 5 are the distribution profiles of GFP wt and DDD GFP among the separated cell organelles. In control cells expressing GFP alone, the reporter was found both in the cytosol (top two or three fractions) and in the particulate/nuclear fraction. In cells expressing DDD GFP , little, if any, of the fusion protein was present in the cytosol, and ϳ70% of the total cell fluorescence co-migrated with the cellular membranes found in fractions 9 -11 (22). These data show that exogenous DDD GFP fusion protein was tightly associated with the cellular membrane.
We then examined the functional consequences of the DDD GFP on D1 enzyme activity using the immune depletion approach as detailed under "Experimental Procedures." D1 activity was measured in cell lysates and in detergent extracts after removing GFP-containing proteins with anti-GFP antibodies. As found with all M4 mutants that formed a heterodimer with p27 wt (see Fig. 2), the catalytic activity of the p27 wt :DDD GFP heterodimer was ϳ50% of that present in cells expressing GFP alone, and most, if not all (Ͼ95%), of the deiodinating activity present in detergent extracts of DDD GFPexpressing cells was specifically removed by anti-GFP antibodies (Fig. 6). These data confirm that the 17-residue-long DDD peptide contains all of the necessary structural elements required for D1 assembly.
Analysis of the Role of the DDD Domain in the Assembly of a Functional SeD2 Enzyme-Although the physiochemical properties of the SeD2 enzyme have not been determined, a comparison of the DDD domain across family members (Table IV) showed a Ͼ70% identity for this domain and a fixed positional relationship to the active center SeC, raising the possibility that this domain may also be important for assembly of other family members. To determine the role of the DDD domain in SeD2 assembly, we expressed the DDD GFP fusion protein or a full-length M4 D1 subunit retaining the fixed relationship of the DDD domain to the active center in SeD2 cells (15) and examined the consequences on SeD2 activity. As shown in Fig.  7, SeD2 cells constitutively expressing the DDD GFP fusion protein had a Ͼ90% fall in D2 catalytic activity compared with that present in GFP-expressing control cells. Similarly, SeD2 cells expressing the inactive D1 subunit (M4) also had a ϳ80% decrease in SeD2 activity, whereas cells expressing the GFP  reporter alone had SeD2 activity that did not differ from that in untransfected cells (Fig. 7). These data are similar to the effects of the DDD domain on D1 and suggest that the DDD domain, present in all deiodinase isoforms, is required for the assembly of functional enzymes. Importantly, these data show that different family members will partner with each other through the DDD domain to form mixed isoform heterodimers.
Analysis of the subcellular distribution of DDD GFP in C6 control and SeD2 cells is shown in Fig. 8. C6 cells expressing either GFP or the DDD GFP reporter molecule showed abundant fluorescence through the cell interior and nucleus with no specific labeling of cellular membranes. In contrast, SeD2 cells expressing the DDD GFP reporter molecule showed bright membrane-localized fluorescence with minimal nuclear fluorescence; SeD2 cells expressing GFP alone showed no specific membrane staining and appeared to be identical to the C6 control cells. These data suggest that the DDD domain selectively interacts with a membrane-bound protein, presumably the 30-kDa SeD2 subunit constitutively expressed in SeD2 cells.
Since the presence of DDD GFP in SeD2 cells inactivates enzyme activity, we could not use changes in enzyme activity after antibody "pull-down" to examine the binding of DDD to the SeD2 subunit. Rather, we immune precipitated SeD2 com-plexes with anti-SeD2 antibodies and determined the presence of DDD GFP in the specific immune precipitates by immunoblot. As shown in Fig. 9, immunoblots of immune precipitated membrane complexes from untransfected SeD2-and GFP-expressing SeD2 cells showed only minor nonspecific bands. In contrast, immunoreactive GFP was present in anti-SeD2 antibody immune precipitates from the DDD GFP -expressing SeD2 cells. These data together with the subcellular distribution of the DDD GFP reporter molecule and the selective loss of SeD2 activity when DDD GFP is present confirm that the DDD serves to assemble a functional SeD2 enzyme similar to its role in D1 assembly.

DISCUSSION
Thyroid hormone deiodinases are an important class of selenoenzymes that regulate the availability and disposal of thyroid hormone. Assembly of functional enzyme(s) dimers is a key step in the biosynthetic pathway that takes place after the insertion of the selenoprotein monomer into the membrane bilayer (9,12). In this report, we used deletion/truncation analysis, domain expression, and pull-down assays to identify the protein domain(s) responsible for enzyme assembly and show that a 16-residue-long region found 22 residues downstream of the catalytically essential SeC is required for the holoenzyme assembly of at least two family members.
To eliminate the technical challenges encountered when multiple, catalytically disabled constructs are transiently coexpressed in host cells (12), we used LLC-PK1 cells that constitutively express a full-length D1 selenoprotein and an exogenous, unique inert rat p27 to examine enzyme assembly. Earlier work done with this approach provided direct evidence that the D1 holoenzyme was a homodimer formed from p27 subunits and that the membrane anchor of one the subunits could be removed without altering dimerization (9). Interestingly, at least half of the p27 polypeptide, up to 133 residues, could be removed from the N terminus without altering the  binding of the truncated D1 subunit to the native p27. Progressive truncation from the C terminus of the rat p27 revealed that residues from position 170 to 254 could also be deleted without altering dimerization but that removal of residues between positions 150 and 170 completely eliminated the binding of the truncated subunit to the native p27, suggesting that the domain responsible for enzyme assembly is located between residues 150 and 170. Comparison of the primary amino acid sequence of this region across family members revealed that a 16-amino acid-long segment located between residues 150 and 170 of p27 is one of the most highly conserved domains in the family. As shown in Table IV, 10 of the 16 residues found between positions 147 and 163 of the rat p27 are identical in all deiodinase family members from fish to humans, and most, if not all, of the variant residues are conservative substitutions. Importantly, the distance between this conserved segment and the active center SeC is fixed at 22 residues in all family members. Because of this high conservation, we generated a domain-specific probe to study enzyme assembly by appending a 17-amino acid-long peptide corresponding to residues 147-163 of p27 to the N terminus of GFP forming the DDD GFP fusion protein.
As expected, overexpression of the soluble DDD GFP fusion protein in LLC-PK1 cells resulted in a ϳ50% reduction in enzyme activity coincident with specific binding of the fluorescent DDD reporter to the native p27. Immune depletion anal-ysis, subcellular fractionation, and photomicroscopy confirmed that this fusion protein was responsible for the decrease in deiodinating activity and was specifically bound to the fulllength p27 in LLC-PK1 cells. These data obtained with the soluble DDD GFP fusion protein are identical to those obtained with the full-length, M4 mutant of rat p27 reported previously (9) and establish that this highly conserved region contains all of the necessary structural requirements for assembly of the D1 holoenzyme.
Based on the finding that DDD GFP fusion protein was a faithful mimic of the protein-protein interaction(s) required for holoenzyme assembly, we turned our attention to the role of this domain in the biosynthesis of catalytically active SeD2. To ensure sufficient quantities of catalytically active SeD2 for study, SeD2 cells (15), C6 astrocytoma cells that constitutively overexpress a functional SeD2, were used as the source of the membrane-bound SeD2 subunits. Unlike our findings for DDD: p27 heterodimers of D1, transient expression of either the DDD GFP fusion or the full-length, M4 mutant of D1 in SeD2 cells completely inactivated SeD2. Since the SeD2:DDD GFP heterodimers were catalytically inactive, we could not evaluate protein-protein interactions by measuring the loss of deiodinating activity by immune depletion. Instead, direct analysis of the binding between the DDD GFP reporter and the 30-kDa SeD2 subunit was done by pull-down assays with anti-SeD2 antibodies and revealed that the complete loss of SeD2 activity was accompanied by the formation of a DDD:SeD2 heterodimer. The role of the DDD domain in SeD2 assembly was confirmed by subcellular fractionation and by co-localization with the full-length SeD2 in the astrocytoma host cells. Control experiments done with the unaltered GFP fluorescent reporter revealed that the assembly of a DDD GFP -SeD2 heterodimer required the N-terminal DDD peptide. These data show that the DDD domain is essential for the post-translational assembly of holoenzymes of at least two deiodinase family members.
Using the immune depletion strategy modeled after that developed to study D1 biosynthesis, Curcio-Morelli et al. (12) used transient co-transfection of FLAG-tagged deiodinase constructs to confirm that a D1 dimer containing only one functional subunit was catalytically active (9,12). Although more than 95% of D1 was immune precipitated by antibodies directed against a single subunit, they estimated that only 1-5% of the holoenzyme was a homodimer formed by intermolecular disulfide bridges based on SDS-PAGE analysis of incompletely reduced samples. Interestingly, the ϳ50-kDa D1 band was only found after SDS-PAGE of inactive Cys mutants of D1; immune precipitates of radiolabeled wild-type D1 failed to show this additional band (12). Since the SeC 3 Cys mutants of the p27 generate a divicinal thiol (-CXC-motif) in the active center of the mutant(s), a protein motif shown to have a strong propensity to aggregate during SDS-PAGE (23,24), it seems likely that the ϳ50-kDa protein aggregates found after SDS-PAGE resulted from the choice of Cys mutants rather than a specific role for disulfide bridging in the assembly of homodimers.
Whereas the results of our experiments make it unlikely that disulfide bridging participates in the assembly of the D1 homodimer, a role for noncovalent interactions is well documented. First, early work showed that Ն95% of detergentsolubilized D1 catalytic activity and the BrAc[ 125 I]thyroxinelabeled p27 co-migrated with a M r of ϳ50 kDa on a molecular sieve chromatography (11,25). Importantly, the choice of detergent has a profound effect on the physiochemical properties  7. Effects of overexpression of the DDD domain on SeD2 activity. C6 astrocytoma cells constitutively expressing SeD2 were transfected with pEGFP-N1, pM4 GFP , or pDDD GFP , and cells expressing these fusion proteins were amplified by G418 selection. D2 activity in cell lysates was determined as described under "Experimental Procedures." Data represent the mean Ϯ S.E. of three different preparations.
FIG. 8. Representative photomicrographs of C6 and C6 cells constitutively expressing SeD2 transfected with cells expressing either pEGFP-N1 vector or pDDD GFP . Cells were fixed 24 h after transfection, and images were taken as detailed under "Experimental Procedures." of this membrane-bound enzyme and molecular sieve chromatography of D1 activity solubilized with nonionic detergents yielded M r values as high as 400,000 (26,27), presumably due to the size of the enzyme-detergent micelles. This problem can be overcome by the use of ionic detergents such as taurodeoxycholate at or near their CMD, where they contribute negligibly to the molecular size of the D1 as judged by sucrose density gradient centrifugation (10,29). Second, that ability of p27 mutants lacking the membrane anchor to bind to the fulllength p27 indicates that integration into the bilipid layer is not required for deiodinase assembly (9). Finally, the ability of the DDD GFP fusion protein to bind specifically to the full-length p27 confirms that enzyme assembly is a post-translational event that is independent of membrane association (9).
The participation of selected residues located in the DDD domain in the catalytic reaction was recently proposed by Larsen and co-workers (30) based on hydrophobic cluster analysis coupled with position-specific BLAST protocols. Mutation of any one of three residues, Glu 156 , His 158 , or Glu 163 , all located in the DDD domain, impaired catalysis of both D1 and D2 family members in transient expression assays (30). Whether this loss of catalytic function was due to improper enzyme assembly, direct participation in the catalytic reaction, or enhanced enzyme instability could not be evaluated. In light of the role of this domain in enzyme assembly and the improved deiodinase reaction conditions, the role of individual residues in the D1 catalytic reaction, especially in this important functional domain, needs to be re-evaluated.
In summary, assembly of a functional deiodinase holoenzyme requires a small 16-residue-long region that is one of the most highly conserved domains among family members. Identification of the domain responsible for enzyme assembly revealed that deiodinase family members retain the ability to interact with each other to form heterodimers. Whereas the consequence of heterodimer formation between different family members is unknown, it is possible that cells modulate the cellular levels of functional selenodeiodinase enzyme(s) family members by sequestering subunits in a pool of poorly functional heterodimers and thereby regulate production and disposal of bioactive thyroid hormone. FIG. 9. Immune precipitation of SeD2:DDD dimers. Eight 75cm 2 flasks of confluent SeD2 cells were transfected with 20 g of either pEGFP or pDDD GFP and grown for 2 days at 37°C. Cell membranes were isolated as detailed under "Experimental Procedures" and solubilized with 0.5% Triton (v/v). Clarified extracts were incubated with anti-SeD2 antiserum (1:100 dilution), immune complexes were isolated with Protein A-agarose beads, and the immune precipitated proteins were separated on a 12.5% SDS-polyacrylamide gel. Separated proteins were electrotransferred to nitrocellulose membranes, and specifically immune precipitated DDD GFP was identified with a monoclonal anti-GFP IgG and visualized using chemiluminscent reagent. Ig HC, rabbit immunoglobulin heavy chain.