In Vivo Reshaping the Catalytic Site of Nucleoside 2′-Deoxyribosyltransferase for Dideoxy- and Didehydronucleosides via a Single Amino Acid Substitution*

Nucleoside 2′-deoxyribosyltransferases catalyze the transfer of 2-deoxyribose between bases and have been widely used as biocatalysts to synthesize a variety of nucleoside analogs. The genes encoding nucleoside 2′-deoxyribosyltransferase (ndt) from Lactobacillus leichmannii and Lactobacillus fermentum underwent random mutagenesis to select variants specialized for the synthesis of 2′,3′-dideoxynucleosides. An Escherichia coli strain, auxotrophic for uracil and unable to use 2′,3′-dideoxyuridine, cytosine, and 2′,3′-dideoxycytidine as a source of uracil was constructed. Randomly mutated lactobacilli ndt libraries from two species, L. leichmannii and L. fermentum, were screened for the production of uracil with 2′,3′-dideoxyuridine as a source of uracil. Several mutants suitable for the synthesis of 2′,3′-dideoxynucleosides were isolated. The nucleotide sequence of the corresponding genes revealed a single mutation (G → A transition) leading to the substitution of a small aliphatic amino acid by a nucleophilic one, A15T (L. fermentum) or G9S (L. leichmannii), respectively. We concluded that the “adaptation” of the nucleoside 2′-deoxyribosyltransferase activity to 2,3-dideoxyribosyl transfer requires an additional hydroxyl group on a key amino acid side chain of the protein to overcome the absence of such a group in the corresponding substrate. The evolved proteins also display significantly improved nucleoside 2′,3′-didehydro-2′,3′-dideoxyribosyltransferase activity.

Two different enzymes have been described in Lactobacillus helveticus and Lactobacillus leichmannii; they are NDT I, specific for purines, and NDT II, which transfers dR between purine-purine, pyrimidine-pyrimidine, and purine-pyrimidine (5)(6)(7). The transferase reaction proceeds via a ping-pong bi-bi mechanism in which the first product is released leaving a covalent deoxyribosyl-enzyme intermediate before the second substrate combines with it (5, 8, 9, ). The stereospecificity of the reaction (only the ␤ anomer is formed) and the tolerance of NDT II for structural variations in the acceptor base have been exploited to synthesize various nucleoside analogs (10 -14).
Enzymatic synthesis catalyzed by NDTs (10,14) has several advantages over chemical methods. Indeed, chemical glycosylation increases the difficulty of obtaining the correct regioselectivity generating few secondary products (15). Furthermore, the instability of some intermediates under the conditions used for synthetic manipulations may be problematic, as demonstrated by the acid lability of purine deoxyribonucleosides.
Whereas NDTs accept different nucleobases from azole derivatives (16,17) to size-expanded purines (18,19), they are highly specific for 2-deoxyribose. When ribose is the sugar moiety, the nucleophilic oxygen atom of the catalytic Glu hydrogen bonds to the O 2 Ј atom of the ribonucleoside and is, thus, inactive (20). The 3Ј-OH of the sugar interacting with the catalytic Glu seems critical for proper orientation of the sugar moiety and optimal catalysis (21). In its absence the transfer reaction is much slower (22). Deoxyribosyltransferase activity is also low with nucleoside analogs modified at the 3Ј position of the sugar (23). Because several nucleoside analogs acting as antiviral or anticancer agents have modifications on their sugar moiety and despite the number of reported synthetic methods, the development of alternative approaches is still relevant.
Here, we address the question of sugar tolerance of NDT at the 3Ј position by selecting enzymes with improved 2,3dideoxyribosyl (ddR) transfer activity between various bases. An Escherichia coli strain auxotrophic for uracil (U) and unable to grow in the presence of dideoxyuridine (ddU) as a source of nucleobase was first genetically engineered. NDT mutants from two distantly related lactobacilli, Lactobacillus fermentum (Lf-NDT) and L. leichmannii (Ll-NDT) were then selected for the use of ddU as uracil source. A single amino acid substitution, A15T in Lf-NDT or G9S in Ll-NDT, allowed the conversion of the nucleoside 2Ј-deoxyribosyltransferase activity to a nucleoside 2Ј,3Ј-dideoxy-or 2Ј,3Ј-didehydro,2Ј,3Ј-dideoxyribosyltransferase activity.

Growth of Bacterial Strains and DNA Manipulation
Bacteria were routinely grown in MS minimal medium (24) or LB (25). When necessary, ribo-and deoxyribonucleosides were added at a final concentration of 0.3 mM. Antibiotics were added at the following concentrations: gentamycin and tetracycline (10 g/ml), ampicillin (100 g/ml), chloramphenicol and kanamycin (30 g/ml). All DNA manipulations were performed according to Sambrook et al. (25). Preparation of L. fermentum CIP 102980T DNA, construction of the genomic DNA bank, and screening for NDT activity were as previously described (7).

General Procedure for Gene Deletion
A DNA fragment containing 1-kilobase upstream and downstream of the gene to be deleted was first amplified from the E. coli MG1655 genome by PCR with a high fidelity polymerase and cloned into pCR-XL-TOPO (Invitrogen). The gene to be deleted was then replaced by a DNA fragment containing an antibiotic resistance gene using restriction enzymes and T4 DNA ligase. Finally, a PCR-generated fragment containing the antibiotic resistance gene flanked by the DNA region surrounding the gene to be deleted was used to recombine into the E. coli chromosome as described by Chaveroche et al. (26). Oligonucleotides ycEL, dinR, yceR, and dinL (Table 1) and a 1.1-kilo-base EcoRI fragment from pUC1318Ap/gm (gift of D. Mazel) conferring resistance to gentamycin were used to construct the MG1655 ⌬pyrC::gm strain. Oligonucleotides codBL, cynTR, codBR, and cynTL (Table 1) and a 1.2-kilobase DNA fragment from pUC4K (GE Healthcare) conferring resistance to kanamycin were used to construct strain MG1655 ⌬codA::km. Deletions were confirmed by PCR and by testing the auxotrophy for uracil for MG1655 ⌬pyrC::gm or by testing the sensitivity to 5-fluorouracil (10 g/ml) and the resistance to 5-fluorocytosine (10 g/ml) for the MG1655 ⌬codA::km.

Construction of ndt Mutant Libraries and Selection of the Nucleoside 2,3-Dideoxyribosyltransferase
L. fermentum-Oligonucleotides T7prom and T7term (Table 1) were used to amplify the ndt gene from plasmid pLF6 (pBam3 carrying the L. fermentum ndt gene, GenBank TM accession number AY064168) using the GeneMorph PCR mutagenesis kit (Stratagene). The cycle parameters were: 1 cycle of 5 min at 95°C, 30 cycles with 3 steps: 30 s at 95°C, 30 s at 51.5°C, 3 min at 72°C; 1 cycle of 10 min at 72°C. Two concentrations of DNA template were used, 10 ng and 10 pg.
The purified PCR products were digested for 2 h at 37°C with BamHI and NdeI restriction enzymes and purified after separation on an agarose gel with the QIAquick gel extraction kit (QIAgen). The purified digested PCR products were ligated to pSU19N (a derivative of the general purpose cloning vector pSU19 (28) that contains an NdeI site, CATATG, ATG being the start codon of the ␤-galactosidase ␣ peptide) digested with the same restriction enzymes and purified by the same procedures. The ligation products were dialyzed against water on Millipore filters before transformation of strain PAK9 by elec- In Vivo Selection of a N-2,3-Dideoxyribosyltransferase troporation. Electroporated cells were incubated in LB medium supplemented with uracil for 1 h at 37°C. The cells were washed twice with MS medium before plating on solid MS agar medium supplemented with chloramphenicol, ddU, and C. Plates were incubated 4 days at 37°C. Colonies were reisolated on the same medium before preparation of the plasmid DNA. L. leichmannii-The GeneMorph libraries were constructed as described above for L. fermentum except that plasmid pSH1 (29) was used as DNA template. Three other libraries were constructed (i) by adding various concentrations of manganese in the PCR reaction, (ii) by adding various concentrations of manganese and by altering the nucleotides ratio in the PCR reaction (30), and (iii) by introducing the purine analog 1-(2-deoxy-␤-Dribofuranosyl)-imidazole-4-carboxamide 5Ј-triphosphate in the PCR reaction (31). Purification of the PCR products, digestion, subcloning in the PSU19N plasmid, transformation, and selection were as described above.

Overexpression of Wild-type and Engineered Lf-NDT and Ll-NDT and Purification of Recombinant Proteins
L. fermentum-Oligonucleotides PAK 5 and PAK 6 ( Table 1) were used to amplify the L. fermentum ndt gene using plasmid pLF6 as DNA template in a standard reaction.
Oligonucleotides ϩ (A15Sϩ, A15Cϩ, or A15Vϩ) and T7 term, oligonucleotides Ϫ (A15SϪ, A15CϪ, or A15VϪ) and T7 prom (Table 1) were used in two separate PCR reactions. An aliquot of each PCR was used as a template for a third PCR reaction using T7 prom and T7 term as primers.
L. leichmannii-Oligonucleotides PAK 7 and PAK 8 ( Table  1) were used to amplify the L. leichmannii ndt gene using plasmid pSH1 as DNA template in a standard reaction. Each PCR product was digested with NdeI and BamHI (underlined in the oligonucleotide sequences), purified using the Qiaquick PCR purification kit (Qiagen), and inserted by ligation into plasmid pET24a digested with the same restriction enzymes.
Oligonucleotides PAK22 and either PAK14 (Gly 3 Arg, Gly 3 Cys) or PAK18 (Gly 3 Thr) or PAK19 (Gly 3 Trp) ( Table 1) were used in a PCR reaction with plasmid pSH1 as DNA template. The PCR products were directly used to transform strain ␤2033 (7). After transformation of strain ␤2033, plasmid DNA was extracted and sequenced. Plasmids with the correct sequence were used to transform strain Bli5. Cultures of Bli5 carrying the different pETndt plasmids were grown as described (7). Nucleoside 2-deoxyribosyltransferase was purified as described by Porter et al. (32). The protein concentration was measured by the Bradford assay with bovine serum albumin as the standard.

Analytical Procedures
NDT Assays-All reactions were carried out at 37°C in a 1-cm path length quartz cuvette and a 1-ml final volume of medium containing 50 mM MES buffer (pH 6.0), 0.5 mM 2Ј-deoxyribonucleosides or 2Ј,3Ј-dideoxyribonucleosides, and 1 mM base. The reaction was started by adding appropriate quantities of NDT and followed by the absorbance change at 290 nm. The specific activities were calculated using the following ⌬⑀ values at 290 nm: (d)dC/A, 1.35; (d)dT/A, 1.4; (d)dU/C, 1.5; (d)dG/T, 4.9. When dI or ddI was the sugar donor, a 5 mM concentration of compound was used. The resulting hypoxanthine was further oxidized to uric acid by xanthine oxidase (0.2 unit). The specific activity was calculated using an ⑀ value for uric acid of 12 10 3 mol⅐cm Ϫ1 at 290 nm. One unit corresponds to the formation of 1 mol of product/min at 37°C in 50 mM MES buffer (pH 6.0).
Mass Spectrometry-Ion spray mass spectra were recorded on an API 365 mass spectrometer (Perkin-Elmer-Sciex Thornhill, Canada). Samples dissolved in water/methanol/formic acid (50/50/5) were introduced at 5 l/min with a syringe pump (Harvard Apparatus, South Natick, MA). The mass spectra were acquired from m/z 1100 to 1700 with a scan step of 0.

RESULTS
Genetic Selection for a Nucleoside 2Ј,3Ј-Dideoxyribosyltransferase Activity-Random mutant libraries of ndt genes from L. leichmannii and L. fermentum were constructed. The L. leichmannii ndt gene was used because the x-ray structure of Ll-NDT is known and the function of several amino acids involved in substrate binding and catalysis are defined by sitedirected mutagenesis experiments (21,33). L. fermentum ndt gene was also mutagenized, because Lf-NDT is quite distant from Ll-NDT (33% identity, data not shown), whereas the amino acids important for the transfer of the 2Ј-deoxyribose are conserved. Different mutagenesis protocols (see "Experimental Procedures") were used to generate libraries having an almost unconstrained spectrum of mutations with a variable frequency (between 1 to 10 nucleotide changes per sequence).
A functional screen allowing the selection of variants was established in E. coli for the reaction ddU ϩ C % ddC ϩ U. This reaction was chosen because we hypothesize that the determinants of the base specificity are different from those for sugar recognition and neither ddU nor ddC interfered with the growth of E. coli. Consequently, variants with improved dideoxyribose transfer between U and C should also have an enhanced transferase activity between any bases.
The engineered strain should be auxotrophic for uracil (U) and unable to grow on minimal medium supplemented with cytosine (C), ddC, or ddU as a source of uracil. This strain (PAK9) was obtained by deleting the pyrC gene, which codes for dihydroorotase along with the genes coding for cytosine deaminase (codA) and cytidine deaminase (cdd) (Fig. 1). PAK9 will grow if one of the following conditions is fulfilled; (i) ddU is cleaved to ddR and U by uridine or thymidine phosphorylase (udp or deoA) or (ii) an NDT exchanges ddR between two nucleobases, U and X, according to the reaction ddU ϩ X % ddX ϩ U.
The random mutant libraries were screened. Ten clones from L. fermentum out of 1.2 ϫ 10 6 grew on MS supplemented with ddU and C. Fig. 2 illustrates the growth differences of the PAK9 strain expressing the L. fermentum ndt (Fig. 2B, c) or the mutated ndt gene (Fig. 2B, a). Because lactobacilli NDT In Vivo Selection of a N-2,3-Dideoxyribosyltransferase JULY 18, 2008 • VOLUME 283 • NUMBER 29 enzymes can marginally transfer ddR between bases (22), a residual growth of strain PAK9 with ddU as the uracil source was observed (Fig. 2B, c). However, variants of NDT with improved dideoxyribosyl transfer activity have been isolated as having a growth advantage over the wild type. Nucleotide sequences of these clones showed the same G 3 A mutation at position 43 of their coding region which results in the substitution of Ala-15 by a Thr in the corresponding protein.
L. leichmannii ndt variants were isolated using the same procedures. All variants carried the same mutation changing the ninth codon, GGT (Gly) to an AGT (Ser).
Activity of the wild-type and engineered NDTs-The enzyme activities were determined for both wildtype L. leichmannii and L. fermentum NDTs with different purines and pyrimidines as donor and acceptor substrates ( Table 2). The highest specific activity was obtained with the couple dC/A, as previously observed with L. helveticus and L. leichmannii NDT (7,34,35). Pyrimidines appear to be better donors of deoxyribose, and purines are often better acceptors. Deoxyinosine is a poor substrate as for the L. helveticus NDT (7). The transfer of ddR by either Ll-NDT or Lf-NDT is between 2 and 4 orders of magnitude lower than the transfer of 2-deoxyribose depending on the donors and acceptors ( Table 2). With 2Ј-deoxyribose donors, the specific activity of the purified Ll-NDT G9S and Lf-NDT A15T mutants was between 2 and 30% that of the wildtype activity, depending on the deoxyribose donor and acceptor pairs. In contrast, the transfer of ddR between two bases is enhanced by a factor of 10 -250 (Table 2). Remarkably, the transfer of ddR is enhanced by a factor of at least 100 with three different pairs (indicated in bold in Table 2). These results confirm that growth differences on minimal medium with ddU as the uracil source between strains carrying wild-type or mutated NDTs are due to improved activity of the mutated enzyme (Fig.  2). It should also be mentioned that the NDT mutants display an improved activity for the transfer of 2,3-didehydro-2,3dideoxyribose as illustrated by the d4T ϩ A exchange reaction, the specific activity of Lf-NDT A15T (9.28 units/mg of protein) being enhanced by a factor of 250 compared with wild-type Lf-NDT (0.036 unit/mg of protein) (highlighted in bold in Table 2).
Kinetic Characterization of the Wild-type and Engineered NDTs-Considering the measured specific activity (Table 2), the kinetic constants of L. leichmannii and L. fermentum NDTs were then determined with the best substrates, adenine as glycosyl acceptor and dC or ddC, respectively, as glycosyl donor. The K m for adenine was between 40 and 80 M when measured using constant and saturating concentrations of dC or ddC irrespective of the enzyme tested (data not shown). Similarly, the K m for dC or ddC when measured with constant and saturating concentrations of adenine varied by a factor of two (Table 3). However, k cat and consequently the k cat /K m were greatly affected. Indeed, when dC was used as a donor, the catalytic  efficiency (expressed by the k cat /K m ratio) of the Lf-NDT A15T and the Ll-NDT G9S was 1 order of magnitude lower than the corresponding wild-type enzymes. Conversely, when ddC was used as the sugar donor, a 2 orders of magnitude increase of catalytic efficiency was observed in the mutant Lf-NDT A15T in comparison to the wild-type Lf-NDT. We also analyzed the 2,3-didehydro-2,3-dideoxyribose exchange reaction between U and C. The K m for d4U was similar for both the wild-type and Lf-NDT A15T mutant, whereas the k cat was 75-fold higher with the A15T mutant (38.9 s Ϫ1 ) as compared with the wild-type enzyme (0.51 s Ϫ1 ).
Effect of Other Amino Acid Substitutions on NDT Activity-To determine whether in vivo selected mutations expressed the highest transfer activity within 2Ј,3Ј-dideoxynucleosides and nucleobases, the Gly-9 of Ll-NDT and the Ala-15 of Lf-NDT were substituted by site-directed mutagenesis by other polar or hydrophobic amino acids. The activity of the new variants is reported in Table 4. Ser appears as a good alternative for Thr in the case of Lf-NDT as the rate of transfer of ddR from ddC to A of the two mutants (i.e. A15T and A15S) is practically identical. Moreover, the specific activity of Lf-NDT A15S with dC and A as substrates is closer to that of the wild-type enzyme than that of the Lf-NDT A15T mutant. In the case of Ll-NDT, Ser remains a better substituent for Gly-9 than Thr, probably due to their size difference, namely Ser being closer to Gly than Thr. As expected, neither Cys, nor the other non-polar amino acids were capable of improving 2Ј,3Ј-dideoxyribose transfer reaction within any explored substrate pairs.

DISCUSSION
Crude extracts from L. helveticus or L. leichmannii have been used to synthesize a large number of deoxyribonucleosides with modified acceptor bases or sugar residues. Some of these analogs have been shown to have antiviral, antimicrobial, or antitumor activity (11, 23, 36 -38). However, other enzymes such as deaminases present in the crude extracts can interfere with the synthesis (39). The activity of NDT is also generally low with nucleoside analogs as compared with natural 2Ј-deoxyribonucleosides (22,37). Thus, an evolved NDT with better activity on analogs could be very advantageous. To initially demonstrate the evolvability of the NDT enzyme, 2Ј,3Ј-dideoxynucleosides were chosen because they are poor substrates for NDT (22,40), and some, ddI and ddC, are of therapeutic interest. Although the three-dimensional structure of L. leichmannii NDT is known, mutation(s) required for an efficient transfer of ddR between bases is not easily predictable (21). Thus, a genetic selection was established in E. coli. The difficulty in the establishment of a screen with unnatural substrates resides in the substrate/product toxicity or in the marginal activity of some metabolic enzymes. The selection was based on the res-

units/mg of protein) of nucleoside 2-deoxyribosyltransferase from L. leichmannii and L. fermentum with adenine (A), cytosine (C), and thymine (T) as acceptors and various 2-deoxy-and 2,3-dideoxynucleosides as donors
Transfer of ddR is enhanced by a factor of at least 100 with three different pairs (indicated in bold). WT, wild type.   toration of uracil auxotrophy of an E. coli strain deleted of the pyrC, codA, and cdd genes with dideoxyuridine as a source of uracil. This selection was applied to several libraries of mutated ndt genes from two lactobacilli strains including a randomized library of the ␤1 strand (from amino acid 6 to 10) of L. leichmannii NDT. Mutants having a growth advantage (Fig. 2B) carry the same kind of amino acid change resulting in the addition of a hydroxyl group to the glycine at position 9 for L. leichmannii or alanine at position 15 for L. fermentum. Remarkably, L. fermentum position 15 matches with position 9 of L. leichmannii (Fig. 3). More than 132 sequences from 107 species, only one from Methanosarcina mazei (Q8PW97), contains a glycine/serine substitution (Pfam 22.0 July 2007). However, the assignment of nucleoside 2-deoxyribosyltransferase function for this deduced protein is questionable because essential catalytic residues (such as Tyr-7, Asp-92, and Asn-123 in leichmannii NDT) are absent. It is then very unlikely that natural N-deoxyribosyltransferases cleave the glycosidic bond of dideoxynucleosides efficiently. We cannot exclude that other single or multiple mutants may exist since the different mutagenesis techniques and the selection may have introduced a bias. A single amino acid change is enough to gain activity with dideoxynucleosides as substrate. The specificity for the base is unchanged, and only the sugar recognition is modified. In the native NDT, the deoxyribose binding site is lined by three acidic residues; that is, the catalytic Glu-98, which interacts with the 3ЈOH, and Asp-92 and Asn-123, which make contact with the 5ЈOH of the sugar. None of these hydroxyl groups is required for the catalysis (21) but may be involved in H-bonds helping to position the sugar in the proper orientation for optimal catalysis (21). Gly-9, even if it does not make contact with deoxyribose, is localized in the active site of NDT. The addition of a hydroxyl group to Gly-9 (or Ala-15 in L. fermentum NDT) should not alter dramatically the geometry of the catalytic site of NDT. Indeed, the affinity for 2Ј-deoxynucleosides is almost identical for native NDTs and mutants (Table 3). On the contrary, the k cat is diminished in mutants. This could be explained by the formation of two H-bonds between the OH group of Ser-9 and the 3Ј-OH groups of the 2Ј-deoxyribose and also with the carboxylate group of the catalytic Glu-98 (Fig. 4), which consequently slow down the catalysis. Another additional explanation is a weak steric clash due to the presence of the OH group of Ser-9.

Substrates
As mentioned earlier, the 3Ј-OH of the sugar that interacts with the catalytic Glu-98 is important for the proper orientation of the sugar for optimal catalysis. When absent, the reaction is much slower. The transfer of 2Ј,3Ј-dideoxyribose is improved significantly if the missing 3Ј-OH group is brought by the side chain of Ser-9 (or of Thr-15 in L. fermentum NDT) and to a lesser extent by Thr-9 or Ser-15, respectively.
Differences in catalytic activities or in affinities due to the presence of a single hydroxyl group have been reported for mutants of human nucleoside diphosphate kinase (41), for transition state analogs of adenosine deaminase (42), and for mutants of the E. coli DNA polymerase I (43   Structural impact of the substitution glycine to serine was modeled using MODELLER 7 (49) and PDB1F8Y as a template with the bound ligand included during the simulation as a rigid block. Model quality and protein-ligand interaction was evaluated using the program ViTO (50). The protein model was generated using coordinates from the crystal structure of L. leichmannii NDT with 5-methyl-2Ј-deoxypseudouridine: 1F8Y. Potential H-bonds between the serine, the glutamic acid, and the 3ЈOH group of the deoxyribose are drawn with dashed lines.
absence of specificity is due to the compensation of the missing 3Ј-OH group of the nucleotide by the phenolic OH of Tyr (43). This resembles the principle of substrate-assisted catalysis (44,45). Although the initial amino acid was not essential for catalysis, the activity with deoxynucleoside was lower than the wildtype enzyme. The OH group of serine or threonine does not replace the OH group of 2Ј-deoxynucleosides in terms of H-bonds. However, it compensates this loss by stabilizing the intermediate in a different way, allowing a gain of activity. The Lf-NDT A15T enzyme was used to prepare at the mmol scale and in good yield (up to 70%) 2Ј,3Ј-didehydro-2Ј,3Јdideoxyadenosine and 2Ј,3Ј-didehydro-2Ј,3Ј-dideoxyinosine from d4U (data not shown). This enzymatic synthesis represents an efficient alternative method to organic syntheses (46 -48) that often require several steps from ribonucleosides. Furthermore, radiolabeled derivatives could be synthesized for metabolic studies in both animal models and human subjects.
Our selection strategy should be applicable to other nucleosides analogs such as those substituted at 3Ј or 5Ј positions of the sugar moiety. The newly selected variants will allow the diversification and expansion of the number of analogs of biological interest.