Evolutionary links as revealed by the structure of Thermotoga maritima S-adenosylmethionine decarboxylase.

S-adenosylmethionine decarboxylase (AdoMetDC) is a critical regulatory enzyme of the polyamine biosynthetic pathway and belongs to a small class of pyruvoyl-dependent amino acid decarboxylases. Structural elucidation of the prokaryotic AdoMetDC is of substantial interest in order to determine the relationship between the eukaryotic and prokaryotic forms of the enzyme. Although both forms utilize pyruvoyl groups, there is no detectable sequence similarity except at the site of pyruvoyl group formation. The x-ray structure of the Thermatoga maritima AdoMetDC proenzyme reveals a dimeric protein fold that is remarkably similar to the eukaryotic AdoMetDC protomer, suggesting an evolutionary link between the two forms of the enzyme. Three key active site residues (Ser55, His68, and Cys83) involved in substrate binding, catalysis or proenzyme processing that were identified in the human and potato AdoMet-DCs are structurally conserved in the T. maritima AdoMetDC despite very limited primary sequence identity. The role of Ser55, His68, and Cys83 in the self-processing reaction was investigated through site-directed mutagenesis. A homology model for the Escherichia coli AdoMetDC was generated based on the structures of the T. maritima and human AdoMetDCs.

lyzes the removal of the carboxylate group from S-adenosylmethionine (AdoMet) to form S-adenosyl-5Ј-(3-methylthiopropylamine) (dcAdoMet), which is committed to act as the n-propylamine group donor for the synthesis of the polyamines spermine and spermidine from the diamine putrescine. AdoMetDC is a crucial control point within the polyamine pathway and its activity is highly regulated during the cell cycle. These polyamines have been shown to be involved in the initiation and maintenance of proliferative states and are essential for cell growth and differentiation (4,5).
AdoMetDC belongs to a small class of decarboxylating enzymes that use a covalently bound pyruvate as a prosthetic group rather than the cofactor pyridoxal 5Ј-phosphate (PLP) typically employed in amino acid decarboxylation reactions (6,7). All AdoMetDCs currently characterized are pyruvoyl enzymes but they can be divided into two classes. Class 1 enzymes are present in bacteria and Archaea, and class 2 enzymes are present in Eukarya (see Table I for representative members of each class). Formation of the active enzyme in both cases involves a self-maturation process in which the active site pyruvoyl group is generated from an internal serine residue via an autocatalytic post-translational modification. Two non-identical subunits are generated from the proenzyme in this reaction, and the pyruvate is formed at the N terminus of the ␣-subunit, which is derived from the carboxyl end of the proenzyme. The post-translation cleavage follows an unusual pathway, termed nonhydrolytic serinolysis, in which the side chain hydroxyl group of the serine supplies its oxygen atom to form the C terminus of the ␤-chain, while the remainder of the serine residue is converted to ammonia and the pyruvoyl group blocking the N terminus of the ␣-chain. Although all AdoMet-DCs undergo the same self-maturation process, the class 1 and class 2 enzymes have almost no detectable sequence homology, and they do not share similarity to any of the other known pyruvoyl-dependent amino acid decarboxylases.
The class 1 AdoMetDCs can be further divided into two groups. Class 1A AdoMetDCs are found primarily in Gramnegative bacteria as the speD gene product. For example, the Escherichia coli enzyme cleaves to give an 18-kDa ␣-chain and a 12-kDa ␤-chain (8,9). The active form of the enzyme is an (␣␤) 4 tetramer and requires a divalent metal ion, such as Mg 2ϩ , for catalytic activity. Class 1B AdoMetDCs have been identified in some Gram-positive bacteria (10) and Archeabacteria (11). Based on the studies with the proenzymes from Bacillus subtilis (10) and Methanococcus jannaschii (11), the class 1B AdoMetDCs cleave to form ␣and ␤-chains, each with a molecular mass of about 7 kDa. This class forms an (␣␤) 2 dimer and does not require Mg 2ϩ or other activators. Class 1A and 1B enzymes show low levels of sequence similarity. These similarities are evident in the residues surrounding the probable cleavage site and those surrounding a cysteine residue, which was identified as part of the active site from modifications that occur during substrate-mediated inactivation of the E. coli and Salmonella typhimurium enzymes (9,12).
The three-dimensional structure of the human AdoMetDC demonstrated that its active site is located far from the interface between the two ␣␤ protomers and that it utilizes residues from both the ␣and ␤-chains (18,20). The topology of each ␣␤ protomer showed an internal structural duplication in which residues 4 -164, which contain the pyruvoyl group, and residues 172-329 have similar topologies. The two halves of the human AdoMetDC protomer are similar in size to the class 1 protomer; however, this is the only indication that the two classes of pyruvoyl-dependent AdoMetDC might be structurally homologous.
Here we report the structures of the wild-type proenzyme and S63A mutant of a class 1B AdoMetDC from T. maritima determined to 1.55 and 1.7 Å resolution, respectively, using selenomethionine multiwavelength anomalous diffraction (MAD) phasing methods. The class 1B TmAdoMetDC structure provides striking evidence of an ancient gene duplication event resulting in the class 2 AdoMetDC enzymes. TmAdoMetDC is synthesized as a 14.8 kDa proenzyme, which after processing contains two chains, ␤ and ␣, of molecular mass 7.0 and 7.8 kDa, respectively. The active form of TmAdoMetDC is an (␣␤) 2 dimer. Each protomer contains one active site that occurs at the dimer interface and also requires residues from the adjacent protomer. Several key active site residues involved in substrate binding, catalysis, or proenzyme processing that were identified in the human and potato class 2 AdoMetDCs are readily recognized in the class 1B TmAdoMetDC despite the overall low primary sequence homology. A sequence alignment of EcAdoMetDC with TmAdoMetDC shows 13% identity. Homology modeling was carried out for the core structure of EcAdoMetDC based on the limited sequence homology and a structural superposition of the T. maritima and human AdoMetDCs.

EXPERIMENTAL PROCEDURES
Cloning of T. maritima speD-PCR was performed using T. maritima genomic DNA (purchased from ATCC) as the template with the follow-ing primer pair: TmSpeDF: TAG TAG CAT ATG AAG AGT CTG GGA AGG CAC (inserts an NdeI site at the start of the gene) and TmSpeDR: TAG TAG CTC GAG TCA GAC GGC GGC CTT GTG CGG (inserts an XhoI site after the stop codon of the gene). The amplified PCR product was purified (QIAquick PCR purification kit from Qiagen), and ligated into pET-28a. A representative clone was sequenced and named pTmSpeD. 28.
Mutagenesis of T. maritima speD-Site-directed mutagenesis with the QuikChange kit was performed according to the manufacturer's guidelines. The following complementary primer pair was used: TmS-peDSAF: GGT GGT GAT ATC TGA AGC TCA CCT AAC CAT TCA CAC CTG GCC and TmSpeDSAR: GGC CAG GTG TGA ATG GTT AGG TGA GCT TCA GAT ATC ACC ACC. Clones were screened by restriction digest for the introduction of an AluI site. A representative clone with the correct restriction pattern was sequenced and named pTmSpeD.28 S63A.
Protein Expression and Purification-For the production of native S63A protein, the pTmSpeD.28 S63A plasmid was transformed into the BL21Star(DE3)pRare strain of E. coli (Invitrogen). A 5-ml saturated starter culture was used to inoculate 1 liter of LB supplemented with 15 g/ml chloramphenicol and 30 g/ml kanamycin. The cells were grown at 37°C until they reached an OD 600 of 0.4, at which point the temperature was shifted to 15°C. The cells were induced with 1 mM isopropyl ␤-D-thiogalactoside (IPTG) at an OD 600 of ϳ0.6. After induction for 16 h the cells were spun down at 5,000 rpm for 10 min and stored at Ϫ80°C.
For production of the S63A TmAdoMetDC mutant protein with selenomethionine (SeMet) incorporated, the methionine auxotrophic strain of E. coli, B834(DE3) (Novagen), was transformed with pTm-SpeD.28 S63A. The growth medium contained M9 salts supplemented with all amino acids (40 g/ml each) except L-methionine, which was replaced with L-SeMet, 0.4% (w/v) glucose, 2 mM MgSO 4 , 25 g/ml FeSO 4 ⅐7H 2 O, 1 mM CaCl 2 , 35 g/ml kanamycin, and a 1% BME vitamin solution (Invitrogen). The cells from the initial 5-ml starter culture were washed with the above medium, and used to start a 1-liter culture. This culture was grown to an OD 600 ϳ0.6, at which point the temperature was lowered to 25°C, and the cells were induced with 500 M IPTG. After induction for 5 h, the cells were spun down and stored at Ϫ80°C.
For production of the TmAdoMetDC proenzyme, pTmSpeD.28 was transformed into the B834(DE3) strain of E. coli (Novagen). The cells were grown using LB media supplemented with 35 g/ml kanomycin. The remainder of the growth protocol was as described for the SeMet S63A protein.
All purification steps were carried out at room temperature. The cells were resuspended in 35 ml of binding buffer (50 mM Tris-HCl, pH 8.0, 10 mM imidazole, 500 mM NaCl) and lysed using a French press. The crude extract was centrifuged, and the resulting supernatant was stirred for 1 h with 1.5 ml of Ni-NTA beads (Novagen) equilibrated with the binding buffer. The resin was then loaded into a polypropylene column and washed with binding buffer (100 ml) followed by 50 ml of wash buffer (50 mM Tris HCl, pH 8.0, 35 mM imidazole, 500 mM NaCl). S63A AdoMetDC was eluted from the column with 15 ml of elution buffer (50 mM Tris HCl, pH 8.0, 150 mM imidazole, 500 mM NaCl). After dialysis with 20 mM Tris-HCl, pH 8.0 and 1 mM dithiothreitol, the protein was concentrated to 6 mg/ml, frozen, and stored at Ϫ80°C. The purification protocol for the wild-type TmAdoMetDC was as described

Structure of T. maritima AdoMetDC
for the S63A protein. The wild-type TmAdoMetDC was found to be greater than 95% unprocessed as determined by Coomassie-stained SDS-PAGE (data not shown).
Processing of TmAdoMetDC-The proenzyme derived from the Tm-speD gene was synthesized in vitro using the Promega TNT system (catalog L4610). Protein was translated from the pTmSpeD.28 or the same plasmid carrying mutants of TmSpeD (S55A, S63A, H68A, or C83A). The TNT reaction was carried out at 30°C for 1 h according to manufacturer's instructions with 1 g of DNA used/50 l of reaction. Processing was then induced by heating at 65°C. Aliquots of 5 l of the TNT reaction product were diluted with 10 l of 0.075 M sodium phosphate buffer, pH 6.8, and the tubes were transferred to a heat block pre-equilibrated to 65°C. At the end of the desired processing time, 15 l of Tricine gel loading buffer (4% SDS, 12% glycerol, 2% ␤-mercaptoethanol, 50 mM Tris, 0.01% Coomassie Brilliant Blue G) was added to each tube to stop the reaction, and the tube was placed on ice. As soon as the tube had cooled, it was placed at Ϫ20°C. When all samples had been obtained, the samples were thawed, boiled 5 min, microcentrifuged briefly, and loaded onto 12% Tricine gels. Gels were run at 45 mA/gel and stopped when the dye front was just at the bottom of the gel. Gels were then fixed in 10% acetic acid, 30% MeOH in H 2 O. Fixative was changed twice at 15-min intervals and then gels were left overnight (12-16 h) in fixative on a gently rocking platform. After fixing, the gels were dried on a gel drier for 2-2.5 h at 80°C and then placed on a phosphorimager screen for visualization. Visualization was carried out on a Molecular Dynamics PhosphorImager 595, and quantitation was done with ImageQuant software. In some experiments, at the end of the 65°C incubation to induce processing, the tube was cooled and 15 l of 1 M hydroxylamine hydrochloride was added. The sample was then incubated at 37°C for the specified time, and the gel-loading buffer was added at the end of the 37°C incubation.
Crystallization-Small initial crystals of the native S63A TmAdoMetDC were obtained using the Crystal Screen TM sparse matrix kit (22). The optimized crystallization conditions were found to be 2.0 -2.4 M ammonium formate, 100 mM HEPES, pH 8.0, using the hanging-drop method at room temperature. Crystals appeared after 2 weeks and reached their maximum size of 0.15 mm ϫ 0.1 mm ϫ 0.1 mm within 6 weeks. The best SeMet-S63A protein crystals grew under similar conditions with an alkaline shift to pH 8.6 and an increase in precipitant concentration to 3.2 M. Under these conditions S63A TmAdoMetDC crystallizes in the trigonal space group R3 with unit cell dimensions of a ϭ 104.97 Å and c ϭ 69.52 Å. Each asymmetric unit contains a complete dimer with a solvent content of 52%.
Crystals of the wild-type TmAdoMetDC proenzyme were obtained under similar conditions to the S63A mutant TmAdoMetDC. The protein concentration was 5 mg/ml, and the optimized well solution contained 3.2 M ammonium formate and 100 mM HEPES, pH 7.8.
For cryoprotection, the crystals were transferred to a stabilization solution similar to the mother liquor, but with the ammonium formate concentration increased by 200 mM. The stabilization solution was gradually replaced with a cryoprotectant solution containing increasing concentrations of glycerol (5% steps until the final concentration of 20% was reached). The crystals were frozen by plunging them into liquid nitrogen and stored for later use.
Data Collection and Processing-The diffraction experiments were performed at beamline 8-BM at the Advanced Photon Source using a Quantum 315 detector in binned format. A three-wavelength data set was collected to 2.1 Å resolution on the SeMet-S63A TmAdoMetDC. Following calibration of the beam using a Se foil, a fluorescence scan was taken on a SeMet-TmAdoMetDC crystal. For data collection, one wavelength was chosen at the maximum of f Ј (edge), another was chosen at the maximum of f Љ (peak) and a high energy remote wavelength was chosen 200 eV above the edge. Data were collected over 120°u sing 15 s for each 1°oscillation with a crystal to detector distance of 280 mm. Bijvoet pairs were measured after each 40°wedge using inverse beam geometry. A native S63A data set was collected over a range of 180°using 40 s for each 1°oscillation at a crystal to detector distance of 220 mm. A wild-type proenzyme data set was collected over a range of 120°using a 17 s exposure time for each 0.5°oscillation at a crystal to detector distance of 200 mm. The HKL2000 suite (23) of programs was used for integration and scaling of all data sets. The data collection statistics are summarized in Table II. Ambiguity in space group assignment between R3 and R32 indicated a potential twinning problem. Cumulative intensity distribution analysis and analysis of second moments indicated twinning for all data sets. The merohedral twin fraction was estimated at 0.25, 0.42, and 0.28, respectively, for the native S63A, SeMet-S63A, and wild-type proenzyme TmAdoMetDC data sets using the twinning server (24).
MAD Phasing-The initial selenium atom positions of the two nonterminal methionine residues were determined utilizing the iterative Patterson search technique as implemented in SOLVE (25) with untreated twinned MAD data to 2.7 Å resolution. The initial phases from SOLVE were improved by density modification using RESOLVE (25) yielding an overall figure of merit of 0.72.
Model Building and Structure Refinement-All model building was performed using the computer graphics program O (26). The two monomers of the S63A dimer were built by tracing a polyalanine model through clear stretches of backbone density. From the initial model an approximate 2-fold non-crystallographic symmetry (NCS) axis was identified. The map was further improved through NCS averaging and density modification with a protein mask that was created from residues 10 -115 and 170 -285 of the human AdoMetDC using CNS (27). The modified map showed improved connectivity and clear side chain electron density, which allowed the sequence alignment and connectivity to be deciphered. However, several segments of the protein were missing in the experimental electron density; the initial model was about 75% complete. Refinement of the initial model against the 1.7 Å native data set was carried out in CNS using the protocols for twinned data. The refinement procedure involved successive rounds of simulated annealing refinement, temperature factor refinement, and model rebuilding. The S63A structure minus water molecules and three residues before and after the site of pyruvate formation was used as the molecular replacement model for the wild-type proenzyme structure. Molecular replacement was carried out in CNS (27) using all data between 3.4 and 10 Å. Model building and refinement were carried out as for the S63A structure. The final refinement statistics are shown in Table III.
EcAdoMetDC Modeling-Templates for a homology model of the E. coli enzyme were chosen from the known AdoMetDC structures. The S68A human AdoMetDC structure (PDB code 1MSV) was separated into two templates corresponding to the N-and C-terminal domains. The highest sequence similarity with EcAdoMetDC was 19% for the TmAdoMetDC sequence (13% identity). The ␤-sheet domains of the TmAdoMetDC and the N-and C-terminal halves of the S68A human AdoMetDC were structurally superimposed using LSQ (28), and a ClustalW alignment (29). The sequence alignment of the three templates with EcAdoMetDC were manually adjusted based on the structural superposition. Ten homology models were built from the three template structures using Modeler version 6 (30, 31) with 2 cycles of slow MD annealing (MD_LEVEL ϭ refine_4, LIBRARY_SCHEDULE ϭ

RESULTS
Overall Structure-The structure of the TmAdoMetDC S63A non-processing mutant was solved by MAD phasing using Se-Met-containing crystals. The wild-type proenzyme structure was solved by molecular replacement. Both the proenzyme and the S63A mutant crystallize in the trigonal space group R3 with unit cell dimensions a ϭ 104.9 Å and c ϭ 69.5 Å. The asymmetric unit contains one complete dimer with overall dimensions of ϳ42 Å ϫ 33 Å ϫ 25 Å. Crystallographic data and refinement statistics are shown in Tables II and III. A representative section of the electron density is shown in Fig. 1.
The architecture of each protomer (designated A and B) is a two-layer ␣␤-sandwich with an anti-parallel ␤-sheet flanked by two ␣-helices and one short 3 10 -helix (Fig. 2). The ␤-sheet consists of six stands, with a strand order of ␤2␤3␤4␤5␤1␤6. The two ␣-helices (␣2, ␣3) are amphipathic and pack tightly against the outer face of the ␤-sheet. The N-terminal methionine and the C terminus, spanning residues 119 -130, are disordered in all monomers. Pro 51 adopts a cis conformation in all subunits; no disulfide bonds are observed.
The Dimer Interface of TmAdoMetDC Proenzyme-The dimer interface of TmAdoMetDC forms the putative active sites and buries ϳ3100 Å 2 of surface area. The monomers are related by noncrystallographic 2-fold symmetry and interact via a face-to-face ␤-sheet interaction resulting in an overall topology of a four layer ␣␤␤␣ sandwich fold (Fig. 3, A and B). The region between the ␤-sheets contains several pockets of bound water and a grouping of charged residues including His 7 , Glu 11 , His 64 , His 68 , Asp 79 , His 110 , and Arg 112 , several of which are conserved in the class 1B AdoMetDC enzymes (Fig. 4). The two active site serine residues (Ser 63 ) are separated by ϳ16 Å and are connected by a hydrogen-bonding network involving His 64 , His 68 , His 64 Ј, His 68 Ј, and a water molecule (Fig. 4; the primed residues refer to the B protomer). The charged cluster stabilizes the dimer interface through hydrogen bonding. The N⑀ of Arg 112 forms a hydrogen bond with Glu 11 Ј in the adjacent sheet. The Arg 112 NH1 nitrogen atom interacts directly with His 7 and Asp 79 Ј. Asp 79 Ј is hydrogen bonded to His 7 Ј and through a water molecule to Asp 79 , His 64 Ј, and His 64 . His 64 Ј is directly hydrogen bonded to His 68 . His 110 and the 2-fold related His 110 Ј, also found near the charged cluster, form a hydrogen bond further stabilizing the dimer interface.
Comparison of the TmAdoMetDC Proenzyme and S63A Mutant Structures-The overall structures of the TmAdoMetDC proenzyme and the non-processing S63A mutant are very similar with an r.m.s.d. of 0.35 Å for 236 superimposed C␣ atoms. In the wild-type proenzyme structure, residues Glu 62 and Ser 63 at the cleavage site are connected by continuous electron density (Fig. 1), with their C␣ atoms separated by 3.8 Å. The electron density and stereochemical constraints indicate that these atoms are connected through an amide linkage and that the wild-type enzyme has not undergone partial processing.
Residues near the Glu 62 -Ser 63 Cleavage Site-Residues 60 -63 form a type II ␤-turn that connects the ends of two adjacent antiparallel ␤-strands, ␤3 and ␤4. The B-factors for the main chain atoms in this region do not deviate significantly from the average values reported in Table III. However, some mobility for the side chain of Ser 61 was observed and the residue was modeled with two conformations in final model. The S63A mutation of TmAdoMetDC traps the enzyme in the proenzyme form by removing the hydroxyl moiety needed for the N 3 O acyl rearrangement. Both the wild-type and S63A AdoMetDC structures reveal many of the interactions that promote the N 3 O acyl rearrangement leading to the formation of the pyruvoyl group. These interactions involve the sulfhydryl of Cys 83 , which is 3.7 Å from the carbonyl oxygen atom of Glu 62 , and the hydroxyl oxygen atom of Ser 55 Ј, which is 4.2 Å from the carbonyl oxygen atom of Glu 62 (Fig. 5). In the wild-type AdoMetDC proenzyme structure the hydroxyl oxygen atom of Ser 63 is 3.5 Å from the carbonyl carbon atom of Glu 62 . The Ser 63 hydroxyl oxygen atom is also in position to donate a hydrogen bond to the carbonyl group of Ser 61 .
Homology Modeling of E. coli AdoMetDC-The results of the homology modeling indicate that the core structure of the E. coli enzyme will likely be a 6 stranded anti-parallel ␤-sheet with at least two flanking ␣-helices. The creation of the active site requires that two EcAdoMetDC monomers associate via a face-to-face ␤-sheet interaction in a manner analogous to the TmAdoMetDC dimer. There are a number of insertions in the E. coli enzyme with respect to TmAdoMetDC for which there is no significant degree of sequence homology in any of the other classes of AdoMetDC. These insertions comprise residues 1-12, 29 -38, 85-111, 124 -131, and 189 -264 (Fig. 6). The 27 amino acid insertion, residues 85-111, is proline-rich and contains a high concentration of negatively charged residues and immediately precedes the serine residue, Ser 112 , involved in the nonhydrolytic serinolysis and pyruvoyl moiety formation. The high concentration of negatively charged residues and proxim-  ity to the cleavage site indicates that it may play a role in binding Mg 2ϩ , which stimulates the decarboxylation reaction in the class 1A AdoMetDCs. Weak secondary structural predictions combined with poor sequence homology to any known structure for residues 189 -264 prevent useful homology modeling in this region.

Effects of Mutations of Conserved Residues on Processing-
The processing of the TmAdoMetDC was studied by in vitro translation at 30°C followed by heating at 65°C. There was virtually no processing (less than 1%) of the proenzyme during the 30 min synthesis period at 30°C but substantial conversion to the ␣␤ form at 65°C. As expected, the mutation S63A, which removes the source of the pyruvate, completely prevented processing. Mutant S55A was cleaved more rapidly than the wild type as was mutant C83A (although slower than S55A). The mutant H68A was cleaved much more slowly than the wild type but the addition of hydroxylamine which is known to cleave ester bonds led to the cleavage of this mutant (Fig. 7) and the conversion was then equivalent to that of the wild type, which was not affected by hydroxylamine.

DISCUSSION
Comparison with Class 2 AdoMetDCs-The TmAdoMetDC proenzyme monomer was submitted to the DALI server (32) to identify other proteins of similar structure. The most significant similarity found was to the C-terminal domain of the human AdoMetDC structure (18). A comparison of the dimeric TmAdoMetDC and the human (␣␤) protomer reveals remarkably similar topologies (Fig. 3, A and B). Fig. 3C shows a sequence alignment according to a structural superposition of TmAdoMetDC and the N-and C-terminal halves of the human AdoMetDC. Although TmAdoMetDC shows less than 8% sequence identity to either half of the human AdoMetDC, the secondary structures show strong correlation. The TmAdoMetDC dimer can be superimposed on the human AdoMetDC S68A proenzyme monomer (PDB code 1MSV) (33) with an r.m.s.d. of 2.0 Å for 156 aligned C␣ carbon atoms. The main differences are a result of several insertions in the human AdoMetDC accounting for 74 additional residues relative to the TmAdoMetDC dimer. The insertions occur primarily relative to the C terminus of each TmAdoMetDC monomer. The insertions form two additional ␤-strands in both N-and C-terminal halves of the human AdoMetDC (␤7, ␤8, ␤15, and ␤16), an additional ␣-helix in the N-terminal half (␣3), as well as a loop region that connects the Nand C-terminal halves of the human structure. The strands ␤7 and ␤15 are involved in the (␣␤) 2 dimer interface of the human AdoMetDC.
Key differences between the T. maritima and human structures are associated with the removal of the second active site of the human AdoMetDC. Overall the T. maritima structure is in better agreement with the C-terminal half of the human enzyme than with the N-terminal half. The ␤2-␤3 and ␤4-␤5 loops of the TmAdoMetDC are considerably longer than in the N-terminal domain of the human AdoMetDC. The corresponding loops in the C-terminal half of the human AdoMetDC (␤10-␤11 and ␤12-␤13) overlap well and are involved in forming the substrate binding pocket in the human enzyme (20). Also found in this region of the human structure are the catalytically important residues Ser 229 and His 243 (19,34). In TmAdoMetDC, the structurally equivalent residues are Ser 55 and His 68 , which are highly conserved in all class 1 AdoMetDCs.

Comparison with other Pyruvoyl-dependent Enzymes-The
TmAdoMetDC structure was compared with the three other published pyruvoyl-dependent enzyme structures, those of Lactobacillus 30a HisDC (13,14), E. coli AspDC (15,16) and M. jannaschii ArgDC (17). None of these proteins show similarity in sequence, structure, or oligomeric state with TmAdoMetDC and only ArgDC and HisDC are homologous to each other. Although the monomers show little structural similarity, a few common features can be noted. AdoMetDC, HisDC, AspDC, and ArgDC are ␣/␤ proteins that catalyze an amino acid decarboxylation reaction (reviewed in Ref. 17). All four enzymes undergo an autocatalytic, intramolecular self-cleavage reaction that generates a pyruvoyl group in a loop between two ␤-strands. In AdoMetDC and AspDC the pyruvoyl group is generated between strands from the same ␤-sheet, whereas in HisDC and ArgDC the strands are from different ␤-sheets. The absence of any clear structural similarity between AdoMetDC, HisDC, and AspDC supports the hypothesis of convergent evolution in the creation of their similar catalytic functions.
Implications for Processing-The mechanism of pyruvoyl group formation is well established (13,15) (Fig. 8). The selfprocessing reaction of pyruvoyl-dependent enzymes is initiated through an internal serinolysis in which the serine hydroxyl group participates in a nucleophilic attack at the carbonyl carbon atom of the preceding residue. The resulting oxyoxazolidine intermediate then undergoes a rearrangement to the ester (6,35). The proposed proenzyme mechanism has been validated by the observation of a trapped ester intermediate in the crystal structure of AspDC (15,16) and in the human H243A AdoMetDC structure (19). The overall mechanism for the TmAdoMetDC self-processing reaction is expected to be essentially the same as in the other pyruvoyl-dependent en-zymes. In the cleavage reaction of TmAdoMetDC, an ester intermediate is proposed to form when the side chain oxygen of Ser 63 performs a nucleophilic attack on the main chain carbonyl carbon atom of Glu 62 . A ␤-elimination occurs across the C␣ and C␤ of Ser 63 , resulting in a dehydroalanine at residue 63 and a new C terminus at residue 62. Residue 63 is then converted to the pyruvoyl residue by a two-step reaction with water (Fig. 8).
Nucleophilic attack by the alcohol on the amide would be assisted by deprotonation of Ser 63 and protonation of the carbonyl oxygen atom of Glu 62 . In the TmAdoMetDC proenzyme structure, there is a hydrogen bond between the Ser 63 side chain and the carbonyl oxygen atom of Ser 61 . The lack of a stronger base in this structure, as well as in the models of the human proenzyme AdoMetDC (33), strongly suggests that the serine hydroxyl is not activated by base catalysis in the first step of the self-processing reaction.
Mechanical strain in the loop preceding the cleavage site has been implicated in the self-processing reactions of HisDC and AspDC (14,16). In contrast to the conformational changes observed for the wild-type and S25A AspDC proenzyme structures (16), the structures of human AdoMetDC processed enzyme, ester intermediate and proenzyme structures (18,19,33) showed little evidence for conformational change on the region of pyruvoyl formation. The TmAdoMetDC proenzyme and S63A mutant structures resemble the human AdoMetDC S68A proenzyme structure. In these structures, the cleavage site is preceded by a standard type II ␤-turn with no evidence of strain before or after the turn. These observations suggest that strain may play a lesser role in the processing of AdoMetDCs than in AspDC and HisDC.
Structural and mutagenesis studies of the human AdoMetDC have identified several residues important in its processing reaction. These include Cys 82 , His 243 , and Ser 229 . The S229A mutant does not process, indicating that the hydroxyl group of residue 229 is required for processing (34). The H243A mutant processes very slowly, but the proenzyme from this mutant is readily split by hydroxylamine, indicating the proenzyme in the ester form (34). Models of the human proenzyme and the oxyoxazolidine intermediate structures suggest that Cys 82 may initially assist the nucleophilic attack and that after the oxyoxazolidine ring forms the Cys 82 hydrogen bond is replaced with a Ser 229 hydrogen bond (33). Because the ester is trapped in the H243A mutant, the side chain of His 243 was identified as the most likely candidate involved in the H␣ proton abstraction of the ␤ elimination step (19,34). Key resi-  (21). Hydrogen bonds are shown as maroon dotted lines and water molecules as red spheres. This figure was prepared with Molscript (41,43) and Raster3D (42). dues involved in the enzymatic processing were identified based on proximity to the active site and a structural superposition of the T. maritima and human AdoMetDCs. In the T. maritima structure the highly conserved residues Cys 83 , Ser 55 Ј, and His 68 Ј are in positions to fulfill analogous roles in the processing of the bacterial enzyme. The H68A TmAdoMetDC mutant showed similar behavior to the human H243A mutant (Fig. 7B), providing further evidence for the involvement of His 68 in the H␣ proton abstraction of the ␤ elimination. Based on a sequence alignment with the E. coli enzyme, Cys 83 corresponds to the putative active site cysteinyl residue, Cys 140 , identified by mass spectrometry in a substratedependent inactivation (9). Cys 83 also lies within the only conserved region between the eukaryotic and prokaryotic enzymes (TCGX (4 -6) KA). The role of Ser 55 in processing was not supported by site-directed mutagenesis, with the S55A mutant appearing to cleave faster than the wild-type enzyme. However, it is not known at this stage if the cleaved mutant has undergone proper processing to generate the pyruvoyl moiety.
Implications for Activation of Processing-Wild-type TmAdoMetDC exists in the crystal structure as the proenzyme form and cleaves only slowly with heating. It is not currently known if heating the enzyme results in formation of the pyruvoyl moiety or just results in enzyme cleavage. Heating of TmAdoMetDC in the presence of E. coli cell lysate (results not shown) results in an increase in the rate of cleavage. The optimal growth temperature for T. maritima is 80°C (36), making heat-induced activation a biologically reasonable possibility for TmAdoMetDC. However, the class 1B B. subtilis enzyme is also found in a primarily unprocessed state and heat-induced activation is not biologically relevant in this case (10). Studies of TmAdoMetDC processing resulted in several conflicting observations. The mutation of Ser 55 or Cys 83 to alanine, which by analogy to human AdoMetDC are important for processing, results in faster cleavage and the S55A mutant cleaves even without heating. The observations could be explained by the requirement of a yet to be identified processing factor. Efforts to identify a processing factor are ongoing.
Implications for the Mechanism of Decarboxylation-In the catalytic decarboxylation reaction of AdoMetDC, the pyruvoyl moiety functions as an electron sink in a manner similar to that of the cofactor PLP. The decarboxylation reaction begins with the formation of a Schiff base between that pyruvoyl cofactor and the ␣-amino group of the substrate. The resulting electron sink facilitates the removal of the ␣-carboxylate. After loss of CO 2 , the Schiff base is hydrolyzed to yield the decarboxylated product. In the absence of a processed TmAdoMetDC structure, a structural superposition of the TmAdoMetDC proenzyme and the human AdoMetDC complexed with the methyl ester of AdoMet (MeAdoMet; PDB code 1I7B) provides some insight into the residues involved in substrate binding and catalysis in the T. maritima enzyme. The pyruvoyl group in the human AdoMetDC is located in a cleft containing a cluster of highly conserved amino acids and is highly shielded from the solvent. The structure of the human MeAdoMet complex revealed key active site residues involved in substrate binding and catalysis (20). Three residues, Cys 82 , Ser 229 , and His 243 , that are known to be important for either the processing or decarboxylation reaction are located near the 5Ј-linker that joins the ribosyl 5Ј position to the pyruvoyl group through the CϭN Schiff base linkage. The absolutely conserved glutamate residue, Glu 247 , forms two hydrogen bonds to the ribosyl group and the phenyl groups of Phe 7 and Phe 223 are involved in stabilizing the purine ring of the adenine moiety (20). The terminal carboxylate of the ␤-chain forms a hydrogen bond to the amino group of the adenine moiety further stabilizing substrate binding. In the complex, Cys 82 is located 3.6 Å from the ␣-carbon of the substrate analog, making it the closest residue capable of protonating the decarboxylated Schiff base intermediate and regenerating the pyruvate during normal catalysis.
The catalytic pyruvoyl group in TmAdoMetDC is generated from Ser 63 , which is found near the dimer interface in the loop FIG. 6. Sequence alignment of TmAdoMetDC with EcAdoMetDC prepared using ClustalW (29) and EPSript (44). Asterisks indicate the serine that forms the pyruvoyl moiety. connecting strands ␤3 and ␤4. The pyruvoyl group would thus be located on the edge of a large pocket between the sheets of the (␣␤) 2 dimer interface. Glu 72 Ј, an absolutely conserved residue, is the structural equivalent of Glu 247 in the human AdoMetDC and is expected to stabilize binding of the substrate by hydrogen bonding to the ribose hydroxyl atoms. Phe 49 Ј in the TmAdoMetDC (highly conserved as Phe or Tyr) is analogous to Phe 223 in human AdoMetDC. Trp 70 Ј is located near the active site pocket and might be homologous to human AdoMetDC Phe 7 following a conformation change. Cys 83 , the structural equivalent of the human Cys 82 , is located near the active site and in addition to its role in processing likely plays a role as a proton donor for catalysis.
Comparison with the E. coli AdoMetDC Homology Model-The EcAdoMetDC proenzyme is longer than TmAdoMetDC by ϳ130 residues. The additional residues form several insertions but the majority of these residues are contained in a C-terminal extension of about 70 residues. The E. coli enzyme has a tetrameric structure and the residues found in the C-terminal extension could form additional ␤-strands and helices that may be involved in intersubunit interactions analogous to the human AdoMetDC dimer. Based on a structural superposition Cys 140 , Ser 77 Ј, His 117 Ј in EcAdoMetDC align with active site residues Cys 82 , Ser 229 , and His 243 , respectively, in human AdoMetDC and Cys 83 , Ser 55 Ј, and His 68 Ј, respectively, in TmAdoMetDC.
Implications for Protein Evolution-The human AdoMetDC (␣␤) 2 dimer contains two active sites, each comprised of residues from both chains (␣ and ␤) of one protomer, with the two active sites being separated by 54 Å. Structural homology between the N-and C-terminal halves was observed, each being comprised of an eight-stranded anti-parallel ␤-sheet with flanking helices. The two halves of the protomer are related by an approximate 2-fold symmetry axis and are connected by a single covalent link. This suggested that the class 2 AdoMet-DCs evolved from an oligomer of identical chains and that the gene was duplicated and fused to form a single polypeptide chain. Gene duplication has become a well accepted mechanism for protein evolution, with numerous similar examples, such as arginine kinase (37) and PLP-independent amino acid racemases (38). The TmAdoMetDC structure provides further evidence for gene duplication and fusion in the evolution of class 2 AdoMetDCs. In the (␣␤) 2 dimer of TmAdoMetDC, the two active sites are comprised of residues from the ␣and ␤-chains of two protomers. Superposition of the human and TmAdoMet-DCs reveals an insertion of a glycine residue between Asp 236 and Thr 238 in the human enzyme (see Fig. 3C), which otherwise would align with the Glu 62 -Ser 63 cleavage site, thus altering the conformation of the ␤11-␤12 loop and resulting in loss of the processing site in the C-terminal half of the structure. The loops in this region are also shortened, further obliterating the second active site.
The structures of the human and potato AdoMetDCs, along with sequence alignments of the class 2 enzymes, suggest that AdoMetDCs from higher organisms have an unusual site containing numerous buried charges. This site is located between the two central ␤-sheets. In the human enzyme, this site binds putrescine, which stimulates both processing and decarboxylation reactions. Two mechanisms for putrescine stimulation have been proposed (19). The binding of putrescine may serve to properly orient the ␤-sheets by balancing the large number of negative charges in the buried charge site. The conformational changes could be propagated to the active site, which is located at the opposite end of the ␤-sheets. Alternatively, the observed hydrogen-bonding network between the charged site and the active site in the human structure could be altered by putrescine binding and could affect the relative orientation of critical residues in the active site. The stimulation of processing and activity of the potato enzyme is not affected by putrescine, but has basal levels of activity similar to the putrescineactivated human enzyme (39). The crystal structure of the potato enzyme revealed the addition of three positively charged residues not present in the human charged site, Arg 18 , Arg 114 , and His 294 (21). The arginine residues presumably play a similar role to putrescine by balancing charge and maintaining the hydrogen-bonding network that leads to the active site (Fig. 4). The structure of the T. maritima enzyme and sequence alignments of the bacterial AdoMetDCs (class 1A and 1B) reveal an analogous site containing a high concentration of charged residues. This site is located between ␤-sheets at the dimerization interface and forms a hydrogen-bonding network that may allow communication between the two active sites through a charge relay.
Finally, a recent bioinformatics study of the enzymes involved in the plant polyamine biosynthetic pathway revealed bacterial origins for all of the enzymes involved with one exception, AdoMetDC (40). The plant S-adenosylmethionine decarboxylase was identified as a potentially eukaryote-specific enzyme form, showing a high degree of sequence homology to other class 2 AdoMetDCs. As noted previously, the class 1 AdoMetDCs found in bacteria have almost no detectable sequence homology to the eukaryotic form of the enzyme. The TmAdoMetDC structure provides evidence of a bacterial ancestor for the eukaryotic AdoMetDCs, thus demonstrating that all enzymes involved in the plant polyamine biosynthetic pathway have bacterial origins.