Enhancing the Latent Nucleotide Triphosphate Flexibility of the Glucose-1-phosphate Thymidylyltransferase RmlA*

Nucleotidyltransferases are central to nearly all glycosylation-dependent processes and have been used extensively for the chemoenzymatic synthesis of sugar nucleotides. The determination of the NTP specificity of the model thymidylyltransferase RmlA revealed RmlA to utilize all eight naturally occurring NTPs with varying levels of catalytic efficiency, even in the presence of nonnative sugar-1-phosphates. Guided by structural models, active site engineering of RmlA led to alterations of the inherent pyrimidine/purine bias by up to three orders of magnitude. This study sets the stage for engineering single universal nucleotidyltransferases and also provides new catalysts for the synthesis of novel nucleotide diphosphosugars.

Carbohydrates are vital in nature, not only for energy metabolism, but also as structural scaffolds, recognition motifs, solubility aids, and functional modulators (1,2). Yet, despite the vast structural and functional diversity of natural glycoconjugates, they are constructed via a common biosynthetic theme. Specifically, sugars are attached to most proteins, lipids, carbohydrates, and small molecules by glycosyltransferases, which with few exceptions use sugar nucleotides as the monosaccharide donors (3)(4)(5)(6)(7). These sugar nucleotides are constructed from sugar-1-phosphates and NTPs by sugar-1-phosphate nucleotidyltransferases, also referred to as sugar nucleotide pyrophosphorylases (EC 2.7.7.-), providing the precursors (usually ADP-, CDP-, GDP-, UDP-and dTDP-glucoses, as well as GDP-mannose, GDP-fucose, and UDP-N-acetylglucosamine) central to nearly all glycosylation-dependent processes (4,6).
Nucleotidyltransferases are prevalent in nature (there are currently ϳ14,000 known and putative nucleotidyltransferase sequences in GenBank TM (8)), are often allosterically controlled, and generally proceed via an ordered bi-bi mechanism. For example, the forward reaction catalyzed by Salmonella glucose-1-phosphate thymidylyltransferase (RmlA) (9) proceeds via a direct S N 2 attack upon the NTP ␣-phosphate by an ␣-Dsugar anomeric phosphate to provide the desired sugar nucleotide and pyrophosphate (Fig. 1). Nucleotidyltransferases from both prokaryotes and eukaryotes have reported flexibility toward variant sugar phosphates in vitro (10 -18), and the uniquely broad sugar-1-phosphate tolerance of RmlA has been exploited for the synthesis of diverse UDP-and dTDP-based sugar nucleotide libraries and enhanced via structure-based engineering (9 -13). To date, more than 30 different sugar-1phosphates have been reported as substrates for RmlA variants (9 -13, 19).
The corresponding pyrimidine-based sugar nucleotide libraries have served as the foundation for a process known as natural product glycorandomization (Fig. 1B), an enzymatic strategy to exchange natural product sugars with diverse sugar arrays (18). To date, this strategy has been applied toward the diversification of glycopeptide, coumarin, and macrolide antibiotics (19 -24), the anthelmintic avermectin (25) and enediyne anticancer agents (22). Yet, although there exist limited reports wherein sugar-1-phosphate guanylyl-or adenylyltransferases displayed moderate sugar-1-phosphate flexibility (15), the lack of a general strategy to generate diverse purine-based sugar nucleotide libraries excludes the corresponding diversification of many natural products glycosylated by purine sugar nucleotide-dependent glycosyltransferases (26 -32).
Recently, thermophilic uridylyl-and thymidylyltransferases were revealed to utilize alternative nucleotides, including both ribo and deoxyribo variants of one or more purine nucleotides (33,34). Although no single enzyme has been reported to utilize all eight naturally occurring NTPs, these pioneering studies, along with a few prior reports of nucleotidyltransferase NTP flexibility (35)(36)(37)(38)(39)(40)(41), suggest the potential to further expand the uniquely promiscuous RmlA toward the goal of engineering a single "universal" nucleotidyltransferase. As a first step, the indepth characterization of the NTP specificity of wild-type (wt) 3 RmlA described herein revealed wt RmlA to utilize all eight naturally occurring nucleoside triphosphates as substrates with varying levels of catalytic efficiency, even in the presence of non-native sugar-1-phosphates. Based upon a composite nucleotidyltransferase purine-binding structural model, RmlA was subsequently engineered to provide mutants that displayed altered pyrimidine/purine biases by up to three orders of magnitude, as measured by apparent k cat /K m values. This study sets the stage for the production of diverse purine-based sugar nucleotide libraries and will not only enhance the prospects of natural product glycorandomization but may also facilitate the production of new reagents for glycobiology.
Protein Expression and Purification-Salmonella enterica typhimurium LT2 wt RmlA and all engineered RmlA mutants were expressed as N-His 6 fusion proteins from pET28a-based expression plasmids (Novagen, Madison, WI) in Escherichia coli BL21(DE3) in a manner similar to previous methods (17)(18)(19)(20)(21). Specifically, an overnight starter culture of LB medium containing 50 g/ml kanamycin was directly inoculated from a glycerol stock of the desired expression strain. After growth overnight (37°C, 250 revolutions/min), this culture was diluted 1:100 with 2ϫ YT medium (42) containing 50 g/ml kanamycin typically to a total volume of 1 liter. The large scale culture was subsequently grown (37°C, 250 revolutions/min) to midlog phase (A 600 ϳ 0.6), at which point isopropyl-␤-D-thiogalactopyranoside was added to a 1 mM final concentration. Growth was continued for an additional 2-4 h, and the cells were collected by centrifugation (15 min, 5000 ϫ g) and resuspended in 100 ml of 50 mM sodium phosphate, pH 8.0, containing 300 mM NaCl and 20 mM imidazole on ice. The cells were lysed via incubation with 1 mg/ml lysozyme (ϳ50,000 units/mg; Sigma) for 30 min on ice followed by sonication (VirSonic 475, Virtis, Gardiner, NY; 100 W, 4 ϫ 30 s pulses, ϳ1 min between pulses) on ice. Protein was purified with nickel-nitrilotriacetic acidagarose resin or spin columns (Qiagen, Hilden, Germany) using manufacturer's protocols. As RmlA is not stable in the elution buffer (50 mM sodium phosphate, pH 8.0, containing 300 mM NaCl and 250 mM imidazole), the buffer was exchanged with 20 mM Tris-HCl, pH 7.5, containing 1 mM EDTA, 200 mM NaCl, and 10% (v/v) glycerol via PD-10 gel filtration columns (GE Healthcare, Uppsala, Sweden). Purified RmlA was subsequently concentrated to Ͼ10 mg/ml, flash frozen in liquid nitrogen, and stored at Ϫ80°C. Protein concentrations were determined by Bradford assay (Bio-Rad) using bovine serum albumin as a standard and a molecular mass of 34.6 kDa for RmlA (43). Spectrophotometric determination of protein concentration (calculated ⑀ 280 ϭ 33,350 cm Ϫ1 M Ϫ1 ) (44) were consistent with the Bradford assay.
RmlA Mutagenesis-RmlA mutants were generated with the QuikChange mutagenesis kit (Stratagene, La Jolla, CA) using the wt RmlA-pET28a parent expression plasmid as the template and the primers indicated in Table 1. Individual progeny plasmids were confirmed by DNA sequencing to carry the desired mutations. Driven by these mutant plasmids, the corresponding RmlA variants were subsequently expressed and purified as described above.
Enzymatic Reactions-The enzyme assay was accomplished via slight modification of previously reported methods (10 -13). In a typical reaction, sugar-1-phosphate (10 mM final) and enzyme (0.01-50 M) were mixed with NTP (0.01-40 mM) in the presence of MgCl 2 ([NTP] ϩ 5 mM) and 10 units/ml inorganic pyrophosphatase (Sigma) in 100 mM Tris HCl, pH 8.0 (20 l final volume). Typical reactions were analyzed after 10 min Natural or unnatural free sugars are enzymatically converted to nucleotide sugar donors via a two-step process catalyzed by a set of promiscuous anomeric kinases and sugar-1-phosphate nucleotidyltransferases, respectively. The inherent substrate flexibility of most natural product glycosyltransferases subsequently allow the ability to use unique sets of sugar nucleotides to generate libraries of differentially glycosylated natural products. The sphere represents any given natural product, and a key to this process is the ability of all participating enzymes to accommodate substrate diversity.
Enhancing the Nucleotide Flexibility of RmlA at 37°C, and reaction times were extended for some slow reacting sugar/NTP combinations. Reactions were stopped by the addition of 80 l of prewarmed HPLC buffer and heat inactivation (95°C, 5 min). The amount of enzyme used and the incubation time were adjusted so that the reactions never proceeded to Ͼ10% turnover for kinetics or 30% for activity assays. Equivalent reactions with heat-denatured enzyme displayed no product formation as determined by HPLC. Product Characterization-Reactions were analyzed via analytical HPLC (Supelcosil LC18-T, Supelco, Bellefonte, PA; 5 m, 250 ϫ 3 mm, 40 mM phosphoric acid, adjusted to pH 6.5 with triethylamine, with a gradient of 0 -10% MeOH over 20 min, 0.8 ml/min, A 254 ). Due to earlier elution, a 5-min 0% MeOH isocratic hold was performed prior to starting the gradient for CTP-containing reactions. For nucleotide sugar products with commercially available standards, HPLC peak identity was confirmed by co-elution. The retention times for sugar-nucleotide product peaks under the above HPLC program are listed in supplemental Table S1. The identities of representative new reaction products were confirmed by mass spectrometry (supplemental Table S2).
Kinetic Measurements-Pseudo-first order kinetics were obtained by fixing glucose-1-phosphate at a saturating concentration of 10 mM (9) and titrating NTPs. At least eight different concentrations in the range of 1 ⁄ 4 ϫ K m to 4 ϫ K m (0.09 -20 mM for dTTP, 0.2-40 mM for others) were assayed in triplicate. Reaction rates were confirmed to be linear over twice the incubation time. Activities were corrected for time and enzyme concentrations, and the kinetic curves were fit to the Michaelis-Menten equation using SigmaPlot with the Enzyme Kinetics module (SPSS, Chicago, IL).
Structural Models of Active Site Mutants-Atomic coordinates of enzyme structures were obtained from the RSCB Protein Data Bank (for RmlA with dTTP, 1IIM (9); RmlA with UDP-Glc, 1IIN (9); and MobA with GTP, 1FRW (45)). Point mutations were simulated by the mutation function of Swiss-PdbViewer software (46). Models of non-crystallographic ligands were created by manual overlay of nucleotide bases so that they were in the same plane as that of the crystallographic base. Mutant side chain positions were selected from the rotamer set provided, and no additional energy minimization was performed on the structures. Figures were generated with VMD (47) and rendered with POV-Ray software.

RESULTS AND DISCUSSION
Protein Expression and Purification-The wild-type and mutant enzymes were expressed in soluble form at high levels (10 -100 mg/liter of culture) and were purified to Ͼ95% purity, as estimated by Coomassie-stained SDS-PAGE (supplemental Fig. S1). No significant differences in yield were noticed between the expression of wt RmlA and any of the RmlA mutants described herein.

NTP Specificity of Wild-type RmlA-Previous
RmlA studies indicate a preference for the pyrimidine nucleoside triphosphates dTTP and, to a lesser extent, UTP using standard assay conditions (9 -13, 49). Recent work on thermophilic uridylyland thymidylyltransferases (Thermus caldophilus UDP-sugar pyrophosphorylase, or Usp, and Sulfolobus tokodaii ST0452, respectively) revealed the conversion of alternative nucleotides, including both ribo and deoxyribo variants of one or more purine nucleotides (33,34). Specifically, Usp was found to turn over glucose-1-phosphate with UTP, ATP, GTP, CTP, and dTTP (33). Alternatively, whereas ST0452 could not utilize ATP, CTP, or GTP, the enzyme showed good activity with UTP and the four deoxynucleoside triphosphates dATP, dTTP, dGTP, and dCTP (34). Given the high degree of amino acid similarity between Usp or ST0452 to RmlA (44 and 36%, respectively), these pioneering studies prompted a re-evaluation of the RmlA purine nucleotide specificity.
Pseudo-first order Michaelis-Menten kinetic analysis revealed RmlA to turn over all eight "natural" NTPs in the presence of glucose-1-phosphate (Table 2), although with appreciably reduced activity (ϳ15-560-fold reduction in the apparent k cat value) in comparison to dTTP. The apparent K m for variant NTPs, including UTP, was also notably higher (ϳ13-50-fold) than that for dTTP. This cumulative analysis translates to a drastic reduction in the apparent RmlA specificity constants (k cat /K m ) for variant NTPs ranging from ϳ12-fold (UTP) to Ͼ15,000-fold (ATP) and is noteworthy for a number of reasons. First, although RmlA has been used for the efficient synthesis of a variety of "unnatural" UDP-sugars (9 -13), this study reveals the dTTP bias of RmlA to be dictated by K m . Second, this study highlights the apparent K m values of all alternative natural NTPs to increase ϳ10-fold, whereas the large differences in the apparent k cat values bias RmlA toward dTTP and UTP. Third, a comparison between the determined kinetic parameters for deoxyribose-containing nucleotides (dTTP, dATP, dCTP, and dGTP) and ribose-containing nucleotides (UTP, ATP, CTP, and GTP, respectively) suggests the ribose 2Ј-hydroxyl to contribute ϳ10-fold to the overall RmlA apparent nucleotide specificity (k cat /K m ). This is consistent with the previously reported influences of 2Јhydroxylation upon nucleotidyltransferase activity (50 -52). Finally, it should be noted that, because of the high apparent K m values for alternative nucleotides, preliminary assays were unsuccessful until the level of Mg 2ϩ was adjusted according to Morrison (53), lending support to the theory that the true nucleotidyltransferase substrate is a Mg 2ϩ ⅐NTP complex (54).

Enhancing the Nucleotide Flexibility of RmlA
Sugar-1-phosphate Specificity with Variant NTPs-In light of the newly discovered ability of RmlA to employ variant NTPs, the specific activity of RmlA in the presence of a variety of unnatural sugar phosphate/NTP combinations was assessed. Turnover was observed with nearly all sugar phosphate/NTP combinations examined, although with reductions in overall efficiencies up to 10 7 -fold (Fig. 2). Consistent with previous studies (10,13), wt RmlA was found to be least tolerant of sugar C3/C4 substitutions in the presence of alternative NTPs, and alteration of both the NTP and sugar-1-phosphate typically exceeded an additive effect, previously noted as "adversely cooperative" (10). Yet, despite the adverse cooperativity and large reductions in catalytic efficiency, the clean production of this new set of unnatural sugar nucleotides could be accomplished by simply increasing enzyme concentration and incubation time (supplemental Fig. S2).
Engineering RmlA NTP Specificity-Examination of RmlA structural homologs emphasized that guanylyltransferases, such as E. coli MobA (which catalyzes the phosphate-phosphate coupling reaction highlighted in Fig. 3A (55)) employ an aspartate to bind the GTP base-pairing face (45). In RmlA, the structural equivalent of this moderately conserved MobA Asp-71 (

Enhancing the Nucleotide Flexibility of RmlA
to engineering a universal nucleotidyltransferase. Thus, Q83D (Fig. 3E) and the isosteric Q83N mutations were pursued. In addition, mutants that incorporated smaller nonpolar (Q83A) or polar (Fig. 3F, Q83S) as well as larger isoelectronic substitutions (Q83E) at this position were also studied.
The activities of this mutant series, in the presence of alternative NTPs, was compared (Table 3), and the salient kinetic parameters for uniquely active mutants were subsequently determined ( Table 2). Consistent with our model, changing glutamine to aspartic acid (Q83D) increased the guanine/thymidine bias (as measured by the ratio of specificity constants) by approximately three orders of magnitude, wherein this altered specificity derives in large part from a 6.5-fold improvement in apparent K m for GTP and a corresponding 21-fold increase in the dTTP K m ( Table 2). In a similar manner, substituting glutamine for a smaller amino acid (Q83S) favored the adenine/thymidine bias by approximately three orders of magnitude, wherein dATP improvements were predominately apparent K m -derived, whereas ATP improvements were primarily apparent k cat -dictated (Table 2). In this latter case, the effect of mutations on apparent k cat value was not entirely expected, as reduction in steric overlap and the alteration of hydrogen bonds were anticipated to mainly influence substrate and product binding. Surprisingly, no significant increase in (d)CTP turnover was observed in any of the mutants tested (Table 3). Given thymidine and cytidine are isosteric and the alternative cytidine base-pairing face could easily be accommodated by a simple rotation of the wt RmlA glutamine side chain, this suggests the glutamine to be locked into position or perhaps implicates that other residues (e.g. the backbone nitrogen of glycine 88) may contribute to thymidine/cytosine discrimination.

CONCLUSIONS
The determination of the RmlA NTP specificity revealed this catalyst to utilize all eight naturally occurring NTPs with varying levels of catalytic efficiency, even in the presence of nonnative sugar-1-phosphates. The uniquely broad synthetic utility of RmlA was further "generalized" by structure-based engineering. The ability to modulate the in vitro specificity of RmlA is consistent with the theory that enzymes have evolved to be perceived as specialists in the context of a discrete in vivo environment (56) but are perhaps not far-removed from more promiscuous progenitors (48,57,58).