Crystal structure of a human mitochondrial deoxyribonucleotidase

articles

© 2002 Nature Publishing Group http://www.nature.com/naturestructuralbiology

Crystal structure of a human mitochondrial deoxyribonucleotidase Agnes Rinaldo-Matthis1, Chiara Rampazzo2, Peter Reichard2,3, Vera Bianchi2 and Pär Nordlund1 Published online: 23 September 2002, doi:10.1038/nsb846

5′ nucleotidases are ubiquitous enzymes that dephosphorylate nucleoside monophosphates and participate in the regulation of nucleotide pools. The mitochondrial 5′-(3′) deoxyribonucleotidase (dNT-2) specifically dephosphorylates dUMP and dTMP, thereby protecting mitochondrial DNA replication from excess dTTP. We have solved the structure of dNT-2, the first of a mammalian 5′ nucleotidase. The structure reveals a relationship to the HAD family, members of which use an aspartyl nucleophile as their common catalytic strategy, with a phosphoserine phosphatase as the most similar neighbor. A structure-based sequence alignment of dNT-2 with other 5′ nucleotidases also suggests a common origin for these enzymes. Here we study the structures of dNT-2 in complex with bound phosphate and beryllium trifluoride plus thymidine as model for a phosphoenzyme– product complex. Based on these structures, determinants for substrate specificity recognition and the catalytic action of dNT-2 are outlined.

In mammalian cells, ribo- and deoxyribonucleotides are synthesized de novo from low molecular weight precursors1. They can also be provided by salvage of nucleosides or nucleobases arising from enzymatic degradation of nucleic acids or nucleotides present in the diet and transported by the blood to cells. When entering the cells, nucleosides are trapped via phosphorylation by nucleoside and nucleotide kinases, yielding dNTPs and NTPs for DNA and RNA synthesis. However, these phophorylation reactions, carried out by cellular kinases, are opposed by the dephosphorylating reactions catalyzed by nucleotidases. In mitochondria, DNA replication depends on the import of precursors from the cytosol, either as deoxyribonucleotides or as deoxyribonucleosides. The mitochondria contain two specific kinases (deoxyguanosine kinase and thymidine kinase), which are responsible for the phosphorylation of all four deoxyribonucleosides2,3. The importance of the mitochondrial kinases was recently highlighted by the discovery that their genetic malfunction leads to human diseases characterized by a depletion of mitochondrial DNA as a result of a deficiency of precursors4,5. Also, an excess of precursors results in severe inborn errors of metabolism with aberrant mitochondrial DNA replication6,7. Thus, a strict regulation of synthesis and catabolism of deoxyribonucleotides is a prerequisite for mitochondrial DNA replication. We recently cloned and characterized a mitochondrial deoxyribonucleotidase (dNT-2) with a high specificity for the dephosphorylation of dUMP and dTMP8. The enzyme was discovered by its 52% sequence identity with a cytosolic deoxyribonucleotidase (dNT-1)9. Together with the cytosolic thymidine kinase, dNT-1 forms a substrate cycle that regulates the size of pyrimidine deoxyribonucleotide pools for nuclear DNA replication10. dNT-2 and the mitochondrial thymidine kinase probably form a similar substrate cycle inside mitochondria, protecting DNA replication from an excess of dTTP. Insight into the properties of nucleoside kinases and nucleotidases is of direct medical importance to develop nucleoside analogs for the treatment of cancer and viral diseases.

Phosphorylation by kinases activates these analogs, and nucleotidases counteract the activation. In this respect, dNT-2 might be of particular interest for minimizing the mitochondrial toxicity of nucleoside analogs, which has become a major problem in HIV therapy11. Ideally, the concentration of phosphorylated analogs should be minimized in mitochondria, which may be accomplished by choosing analogs whose 5′ phosphates are rapidly and specifically degraded by dNT-2. As an example, azidothymidine-5′-phosphate, the most commonly used anti-HIV nucleoside analog, is a poor substrate for dNT-2 (unpub. data). Insight into the structural determinants for catalysis and the specificity of dNT-2 should be helpful in finding nucleoside analogs with diminished mitochondrial toxicity. In addition to dNT-2 and dNT-1, five other human 5′ nucleotidases have been clearly defined via cloning of their cDNAs. All seven enzymes differ in tissue specificity, subcellular location, primary structure and substrate specificity. One is a ubiquitous ectonucleotidase (eNT) anchored to the surface of the plasma membrane and is involved in the production of adenosine as an extracellular messenger12. Five of these nucleotidases, including dNT-1, occur in the cytosol and one (dNT-2) occurs in mitochondria. dNT-1 and dNT-2 are both ubiquitous enzymes that show a preference for the dephosphorylation of dUMP and dTMP, although this specificity is less pronounced for dNT-1 (ref. 9). Among the other cytosolic enzymes, the ubiquitous ‘high KM-nucleotidase’ (cN-II) shows a preference for dephosphorylation of GMP and IMP. cN-II has been studied most extensively but its physiological function remains unclear13. Three other cytosolic enzymes, two with a preference for dephosphorylation of AMP (cN-IA and cN-IB14,15) and the other with a preference for CMP and UMP (PN-1), show a limited cellular localization: cN-IA to muscle, heart and, to a lesser extent, brain, cN-IB to testicle and PN-1 to erythrocytes. cN-IA is believed to be involved in energy metabolism; and PN-1, in the degradation of RNA during erythrocyte maturation. The genetic loss of PN-1 results in a severe hemolytic anemia16.

1Department of Biochemistry and Biophysics, Stockholm University, S-106 91 Stockholm, Sweden. 2Department of Biology, University of Padova, I-35131 Padova, Italy. 3Department of Biochemistry, Medical Nobel Institute, MBB, Karolinska Institute, S-171 77 Stockholm, Sweden.

Correspondence should be adressed to P.N. email: [email protected] nature structural biology • volume 9 number 10 • october 2002

779

articles


a

Fig. 1 Structure and sequence alignment of the dNT-2. a, The dNT-2 dimer, with residues from the α6, β3 and the loop between α7 and β4 forming the dimer interface. In the active site, the bound thymidine product is found. The gray ball represents the Mg2+ ion. b, Structurebased sequence alignment of the dNT-2 (TrEMBL entry Q9NPB1), dNT-1 (Q9NP82), PN-1 (Q9P0P5), cN-IA (Q9BXI3), cN-IB (Q91YE9) and cN-2 (SWISS-PROT entry P49902) sequences. All sequences are human proteins except cN-Ib, which is from mouse. Dots represent gaps in the sequences, and blanks indicate that the sequence has been excluded in this area. The above numbering follows the amino acid sequence in the dNT-2 protein. Blue fields represent residues identical in all six sequences, and yellow fields mark residues identical in only some of the sequences. Circles show amino acids involved in catalysis. The sequence alignment was done using ALSCRIPT43.

b

Aside from the relationships between dNT-1 and dNT-2 and between cN-IA and cN-IB, the various mammalian 5′ nucleotidases show no statistically significant sequence similarities. No direct structural information is available for any of the mammalian 5′ nucleotidases, but the structure of a bacterial enzyme, homologous to eNT, has been determined17. This structure displays an α/β fold with a dimetal center in the active site and is related to the purple acid phosphatase and the serine/threonine protein phosphatase families17,18. Relatively little is known about the catalytic mechanism of the soluble nucleotidases. All enzymes except eNT depend on Mg2+. cN-II and PN-1 were shown to also have phosphotransferase activity13,19,20. Recently Allegrini et al.21 demonstrated the formation of a covalent phosphoenzyme intermediate for cN-II and localized the phosphate group to a peptide containing an Asp-X-Asp-X-Thr/Val motif. They proposed that this short motif, which can also be found in other 5′ nucleotidases, including dNT-2, was related to a similar motif in a phosphoserine phosphatase (PSP)22,23. PSP is a member of the haloacid dehalogenase (HAD) super family of enzymes, where most members use an aspartyl nucleophile as their common catalytic strategy. Here we report the structure of human dNT-2, the first structure of a mammalian soluble monophosphate nucleotidase, in complex with phosphate, with beryllium trifluoride (BeF3) and with BeF3–thymidine nucleoside. The structures provide a detailed view of the active site and allow a discussion of the features of the catalytic mechanism and substrate specificity, as well as the evolutionary relationships to other soluble nucleotidases. 780

The overall structure of dNT-2 The structure of dNT-2 was solved using the MAD method on a crystal soaked with mercury and containing thimerosal24. The structure was initially modeled and refined at 1.8 Å resolution using data of a crystal from the native phosphate-containing mother liquor, yielding a structure with a phosphate ion and a Mg2+ ion bound at the active site. The final model of this structure contains 193 out of 196 amino acids (amino acids 34–227). Amino acids 1–32 constitute the mitochondrial localization leader sequence and were not part of the expressed construct (Fig. 1a). This structure has been refined to an R-factor of 15.7% and Rfree of 18.8%. Two structures from crystals soaked with BeF3 serve as a model for a potential phosphoenzyme intermediate. The structure of one of the BeF3-soaked crystals was determined at 1.9 Å, with an R-factor of 16.9% and Rfree of 21.5%. One of the BeF3 complexes was also soaked with thymidine, representing a nucleoside product. The BeF3–thymidine-containing structure was determined at 2.8 Å resolution, with an R-factor of 14.2% and Rfree of 20.6%. All structures had 90% of the residues in the most favorable regions, and the remaining residues are in the additionally allowed regions (Table 1), according to the Ramachandran plot. The crystal structure of dNT-2 reveals a possible dimeric structure (Fig. 1a), consistent with gel filtration data. An extensive interaction surface between the two monomers in the crystal lattice is formed with a crystallographic two-fold axis. This interaction buries an accessible surface of 1,300 Å2 and yields a dimeric molecule with the dimensions ∼60 × 50 × 40 Å3. Each monomer is composed of a large and small domain. The large domain, comprising amino acids 33–42 and 111–227, nature structural biology • volume 9 number 10 • october 2002

articles


a

b

Fig. 2 Comparison of the dNT-2 structure with PSP (PDB entry 1F5S). a, The overall structure of dNT-2 (right) is closely related to the structure of PSP (left). Both structures contain a phosphate and an Mg2+ ion bound in the active sites. b, Catalytic machineries of dNT-2 and PSP. The active sites of dNT-2 and PSP are structurally similar. This provides support that the formation of the aspartyl-phosphate adduct also occurs in the dNT-2 reaction cycle, with Asp 41 serving as the nucleophile. Figure was made using MolScript44.

forms a α/β Rossmann fold, except for 2 loops, residues 175–187 and 194–203, which are helices in the canonical Rossmann fold. The small domain forms a truncated four-helix bundle inserted between strand β1 and helix α5 of the large domain. Two loops, residues 42–49 and residues 97–110, connect the two domains. The dimer interaction is formed primarily by α6, β3 and the loop between α7 and b4 from the core domain. The interaction surface is relatively hydrophobic, involving residues Tyr 141, Tyr 144, Tyr 138 and Phe 136. Strand β3 forms a trans domain antiparallel β-sheet. In addition, several residues from the loop between α7 and β4 form part of the dimer interface. A Dali search25 revealed that dNT-2 has a similar structure to several HAD family proteins — such as 2-haloacid dehalogenase (HAD)26 with an r.m.s. deviation of 3.1 Å (71 amino acids), phosphonoacetaldehyde hydrolase (P-ald)27 (r.m.s. deviation = 3.3 Å for 89 residues), PSP (r.m.s. deviation = 3.1 Å, for 95 residues) and CheY28 (r.m.s. deviation = 1.9 Å for 38 residues) — despite no significant sequence similarity of these proteins to dNT-2. The structural superposition is found mainly in the Rossmann domains, but the similarity to dNT-2 in topology of the small domain is also found for PSP (Fig. 2a) and to some extent for P-ald. Structure of the active site of dNT-2 The active site of dNT-2 is found in a cleft between the two domains and is accessible to solvent. In the active site of the native structure, we can detect distinct nonprotein density, which we have assigned as a bound phosphate ion and a Mg2+ ion, with the phosphate coordinating directly to the Mg2+(Fig. 3a,b). The main components of the active site are constituted by loop 1 (residues 41–50), the C-terminal part of strand β4 (residues 175–176), the C-terminal part of strands β2 (residues 130–132) and β3 (residues 163–165), the N-terminal part of α4 (residues 96) and the loop between helices α2 and α3 (residues 71–78). The Mg2+ ion is coordinated to the carboxylate side chains of Asp 41 and Asp 176 and the main chain carbonyl of Asp 43. In nature structural biology • volume 9 number 10 • october 2002

addition, Mg2+ is coordinated by three exogenous ligands, the phosphate ion and two densities modeled as water molecules, completing an octahedral coordination. The exogenous ligand trans to Asp 41, modeled as a water molecule, extends partially into a fragmented density in the active site pocket, indicating that a low molecular weight adduct may be binding in the active site at partial occupancy or in a disordered mode. At the present resolution, however, the assignment of this density remains uncertain. An extensive hydrogen-bonding network coordinates the phosphate ion in the active site (Fig. 2b). Positive charges are contributed by NH of Lys 165 and the Mg2+ ion. Thr 130 makes a hydrogen bond to the phosphate, and the phosphate ion is found at a short distance (