To assess Nrk specificity in vitro, recombinant human Nrk1 and Nrk2 were expressed in E. coli and purified by immobilized metal chelate affinity chromatography. As shown in Table 1, the enzymes discriminate between substrates almost entirely in the K_M term. Nrk1 has a k_cat of approximately 0.5 s⁻¹ irrespective of substrate, and Nrk2 possesses a k_cat of approximately 1 s⁻¹ irrespective of substrate. Nrk1 strongly favors NR as a substrate, displaying a 340-fold preference for NR over cytidine in the K_M term and a ∼500-fold preference over either cytidine or uridine in the k_cat/K_M term. Tiazofurin, the prodrug form of the toxic NAD⁺ analog TAD, is a relatively good Nrk1 substrate with a k_cat/K_M of 1300 s⁻¹M⁻¹, which represents 19% of the second-order rate for NR conversion to NMN (6800 s⁻¹M⁻¹). Moreover, Nrk1 shows little preference for ATP (6800 s⁻¹M⁻¹) over GTP (5000 s⁻¹M⁻¹) as phosphodonor in formation of NMN.

Whereas these data would classify Nrk1 as an NR and tiazofurin:ATP or GTP kinase, the data establish Nrk2 as an ATP-specific NR, tiazofurin, and uridine kinase. As shown in Table 1, with GTP as the phosphodonor, Nrk2 has only 1.5% of the NR phosphorylating activity with respect to ATP. Tiazofurin (4500 s⁻¹M⁻¹) is phosphorylated as well as NR (3900 s⁻¹M⁻¹), and uridine (850 s⁻¹M⁻¹) is within 5-fold of NR. The fact that Nrk1 and Nrk2 are distinct from uridine/cytidine kinases [19] is underscored by the poor cytidine monophosphate-forming activity of each enzyme.

To understand the basis for substrate specificity of human Nrk1, we prepared a selenomethionyl form of human Nrk1 and grew single crystals of a complex of the enzyme with Mg²⁺·ADP. A crystal, which had the symmetry of C222₁, was subjected to 0.9793-Å synchrotron X-radiation, and produced nearly complete diffraction data to 1.9-Å resolution (Table 2). Single-wavelength anomalous scattering [23] allowed location of Se sites [24] and phasing [25] to produce an interpretable experimental electron density map of the Nrk1 monomer prior to model building. The 1.95-Å refined protein model includes residues 1–82 and 92–189 of the 199–amino acid polypeptide, plus ADP, Mg²⁺, and 72 water molecules with B factors between 9 and 37 Ų.

As shown in Figure 1A, Nrk1 consists of a five-stranded β sheet flanked on one side by α helices, E and A, and on the other side by helix B. Additionally, the monomeric enzyme contains a lid domain consisting of helix C and D connected by a 12– amino acid loop. The five-stranded sheet is entirely parallel and is formed from strands 2, 3, 1, 4, and 5 in the primary sequence. Earlier [4], we detected sequence similarity with uridine/cytidine kinase and pantothenate kinase. Indeed, the DALI structural similarity server [26] revealed Nrk1 to be a structural homolog of a variety of Rossmann fold-containing metabolite kinases including human uridine/cytidine kinase 2 Uck2 [27], E. coli pantothenate kinase panK [20], Bacillus stearothermophilus adenylate kinase [28] and E. coli gluconate kinase [29]. A structural superposition of Nrk1 and Uck2 is provided in Figure 1B.

The ADP-binding site, including P-loop [30] sequence Gly-Val-Thr-Asn-Ser-Gly-Lys-Thr (residues 10–17), is shown in close-up in Figure 2A. The guanidino group of Arg132, the ε amino group of Lys16, and the hydroxyl of Thr18 coordinate the β and α phosphates of ADP, whereas the hydroxyl of Thr17, a β phosphate oxygen, and four well-ordered water molecules coordinate the magnesium ion. The adenine ring of ADP lies between Arg128 and Glu174. Accounting for the ATP/GTP nonspecificity of Nrk1, the 2 carbon of adenine is solvent exposed such that the 2 amino group of guanine would not appear to preclude binding in the same manner. However, in the ATP-specific Nrk2 sequence, Glu174 is replaced with Arg. Indeed, in a large study of amino acid propensity in adenine and guanine-binding sites, Arg was found to be localized at adenine sites and to be largely excluded from guanine sites [31].

To determine the nature of nucleoside phosphorylation by Nrk1, we co-crystallized the Nrk1 enzyme with NR, NR·Mg²⁺·AppNHp, NMN, and tiazofurin, and performed molecular replacement and refinement to obtain high-resolution models. Despite the typical hinge motions identified in Rossmann fold-containing metabolite kinases upon substrate-binding [32], Nrk1 was neither opened nor closed by any ligands examined. Pairwise comparisons of the α carbon coordinates indicated that all structures are within a root mean square difference of less than 0.4 Å. As shown in Figure 2B, which is derived from a 1.32-Å crystal structure, NR is bound with the 2′ and 3′ hydroxyl groups recognized by bidentate interactions from Asp56 and Arg129 and with the carboxylate of Asp36 accepting an apparent hydrogen bond from the NR 5′ hydroxyl. Such a hydrogen bond could serve to activate the 5′ oxygen toward bond formation with the γ phosphorous atom of a bound ATP substrate.

As shown in Figure 2C, in the 1.92-Å refined crystal structure of Nrk1 bound to the hydrolysis-resistant ATP analog AppNHp with Mg²⁺ and NR, the γ phosphate—recognized by side chains of Thr12, Lys16, Tyr134, and Arg132—is positioned for potential in-line transfer to the 5' oxygen of NR. In this structure, the carboxylate of Asp36 is a direct Mg²⁺ ligand. In the NMN product complex (Figure 2D), all four side chains “formerly” associated with the γ phosphate of AppNHp are associated with the α phosphate of NMN, suggesting that these residues may be optimally aligned to stabilize a putative pentacoordinate phosphorane transition state that is resolved either by collapse to ATP + NR or by the formation of ADP + NMN products.

In the NR (Figure 2B) and NMN (Figure 2D) substrate and product complexes, Asp36 is oriented toward the 5′' oxygen, suggesting a role in activating the acceptor oxygen and promoting bond formation. In the absence of NR or NMN and in the presence of the ADP product (Figure 2A), Asp36 stabilizes a Mg²⁺-associated water molecule. Curiously, Asp36 has yet a third unique conformation in the inactive bi-substrate analog complex (Figure 2C). In addition, Glu98 appears to have a key role in organizing a stable water ligand of Mg²⁺. To test the hypothesis that Asp36 and Glu98 (residues 35 and 100 in Nrk2) might be essential for function, we constructed human nrk1-D36A, nrk1-E98A, nrk2-D35A, and nrk2-E100A alleles for evaluation in yeast. These mutants, alongside wild-type NRK1 and NRK2 controls, were introduced into yeast strain BY278 in which NRK alleles were expressed from the GAL1 promoter on a LEU2 plasmid, the endogenous NRK1 gene was deleted, and a QNS1 gene was provided on a URA3 plasmid. In this system, a functional NRK gene allows a yeast cell grown in the presence of 10 μM NR to lose the QNS1 gene with associated URA3 marker, as scored by resistance to 5-fluoro-orotic acid [4]. As shown in Figure 3, the conserved Asp and Glu residues are required for function of Nrk enzymes in vivo. To exclude the possibility that the conserved Glu residues are required for folding and are potentially dispensable after protein biosynthesis, we expressed and purified nrk1-E98A and nrk2-E100A mutant proteins in E. coli. As shown in Figure S1, the conserved Glu is not required for soluble expression, accumulation, purification, or concentration. Although nrk1-E98A and nrk2-E100A proteins behaved precisely as did wild-type enzymes in purification, their activity in ATP-dependent phosphorylation of NR was below the level of detection of our assay. Thus, Asp36 (Asp35 in Nrk2) and Glu98 (Glu100 in Nrk2) are essential residues for function in vivo. The demonstrated post-biosynthetic role for the conserved Glu and the conserved active-site positions of Glu and Asp strongly argue for roles in catalysis.

Data in Table 1 show that Nrk1 has strong specificity for nucleosides containing a carboxamide group two bond lengths away from N1, such as NR and tiazofurin, and features that may discriminate against the 2- and/or 4-substitutions found in cytidine and uridine. Indeed, in crystal structures of Nrk1 bound to NR and NMN, it is clear that the 4-amino group of cytidine or the 4-oxy group of uridine could not be accommodated without rearrangement, because these constituents would clash with the carbonyl oxygen of Gln135, which is in a cis peptide linkage with Pro136. This unique backbone conformation is unlikely to be conserved by the uridine-accepting Nrk2 enzyme, which has a Thr-Val sequence in this position, likely to be in the typical trans conformation. The van der Waals clash between a modeled 4-oxy group of uridine and the Gln135 carbonyl oxygen is shown in Figure 4.

The carboxamide-containing preferred nucleoside substrates of Nrk1 are tiazofurin and NR. As shown in Figure 5A, the two nucleosides are bound almost superimposably by Nrk1. The pyridine and thiazole moieties of NR and tiazofurin, respectively, are stacked between the phenol rings of Tyr55 and Tyr134. An additional aromatic interaction with Phe39 allows the polar carbonyl oxygen and amino groups of NR and tiazofurin to be exposed to solvent (Figure 5B). Because carbonyl oxygen and amino groups are isosteric at 1.9-Å resolution, we looked for an electrostatic interaction that might uniquely orient the carboxamide function of NR and tiazofurin, and found no interacting residue within hydrogen-bonding distance. We therefore considered the possibility that Nrk enzymes might phosphorylate the isosteric but nonisoelectronic NR analog, NaR. Consequently, NaR was synthesized and examined as an in vitro substrate of Nrk1 and Nrk2. As shown in Table 1, NaR is phosphorylated by Nrk1 and Nrk2 with highly similar kinetics with respect to those for NR phosphorylation. In assays of each human enzyme, NaR is favored over NR by slight K_M advantages offset by slight k_cat disadvantages. In k_cat/K_M terms, Nrk1 has 60% of the activity and Nrk2 has 138% of the activity with NaR versus NR. Thus, human Nrk1 and human Nrk2 are dual-specificity, NR and NaR kinases.

The steric complementarity of Nrk1 with NR and the lack of electrostatic exclusion of NaR by human Nrk1 and Nrk2 suggested that NaR might be a synthetic NAD⁺ precursor vitamin. Should NaR be utilized by yeast cells, it would be conceivable that NaR is a previously unrecognized metabolite, such that the utility of NaR salvage might have played a role in the evolution of Nrk specificity. Additionally, our discovery of Nrk1-dependent [4] and Nrk1-independent [15] NR utilization suggested that, should NaR support the vitamin requirement of de novo pathway–deficient yeast cells, there could be two different metabolic pathways for NaR utilization. As shown in Figure 6A and 6B, the bna1 mutant in de novo biosynthetic enzyme 3-hydroxyanthranilic acid dioxygenase is a Na auxotroph [33] that can also be supported by 10 μM NR, thus providing an assay for vitamin activity of NaR. As reported earlier [4], NR can bypass the requirement of glutamine-dependent NAD⁺ synthetase, Qns1 [21]. Also shown in Figure 6B, NR keeps a bna1 mutant alive in two different ways because cells have two NR salvage pathways [15]. The first NR salvage pathway goes through Nrk1, which allows NR-dependent viability in a bna1 mutant deleted for Npt1, which is the Na phosphoribosyltransferase. The second NR salvage pathway depends on the NR-splitting activities of Urh1 and Pnp1, followed by nicotinamidase and Npt1 activities. This pathway allows a bna1 nrk1 double mutant to retain viability. The two NR salvage pathways are schematized in the lower right section of Figure 7.

As shown in Figure 6C, 10 μM NaR can also be used by bna1 mutants, establishing that NaR is a transportable NAD⁺ precursor vitamin. Genetic control over NAD⁺ biosynthesis in the yeast system allowed us to establish that NaR is not simply used as Na, not contaminated by or converted to NR, and is used via a unique set of enzymes including Nrk1, Urh1 and Pnp1, and Qns1. If NaR were merely a source of Na, the bna1 npt1 mutant would fail to grow on NaR. However, Figure 6C clearly shows that NaR supports the growth of bna1 npt1 mutant yeast cells. Moreover, if NaR were either contaminated by NR or converted to NR by any cellular process, then NaR would support the growth of the qns1 mutant. As shown in Figure 6C, NaR fails to support the growth of the qns1 mutant. Whereas NaR shares with NR the ability to be utilized by bna1 npt1 and bna1 nrk1 mutants, the Qns1 requirement for NaR indicates that NaR metabolites must flow through nicotinic acid mononucleotide (NaMN) and nicotinic acid adenine dinucleotide (NaAD) as schematized in Figure 7.

Vitamin activity of NaR in bna1 npt1 and bna1 nrk1 mutant strains can be explained by two pathways for NaR utilization. NaR is phosphorylated by Nrk1 with highly similar kinetics to those of NR (Table 1). Thus, phosphorylation of NaR by Nrk1 produces NaMN in a pathway that is independent of Npt1. By analogy with the recently described Nrk1-independent NR utilization pathway, we hypothesized that NaR is a substrate of the nucleoside hydrolase and nucleoside phosphorylase activities of Urh1 and Pnp1, which are responsible for virtually all Nrk1-independent NR salvage [15]. However, whereas yeast Nrk1-independent NR salvage requires nicotinamidase, the corresponding pathway for NaR would simply produce Na from NaR, which would be salvaged by the Preiss-Handler pathway [34], consisting of Na phosphoribosyltransferase Npt1 [7], Nma1,2, and Qns1.

To test the hypothesis that NaR utilization depends on Nrk1, Urh1, and Pnp1, i.e., the same enzymes that initiate NR salvage [15], we grew wild-type, npt1 mutant, nrk1 mutant, urh1 pnp1 mutant, and nrk1 urh1 pnp1 mutant cells in vitamin-free media and in vitamin-free media supplemented with 10 μM NaR. As shown in Figure 6D, NaR elevates NAD⁺ levels in all strains except the nrk1 urh1 pnp1 mutant. In the wild-type strain, NaR elevated intracellular NAD⁺ from 0.70 ± 0.04 mM to 1.18 ± 0.10 mM, an increase of 480 μM. This can be compared with the 1.21 mM increase in intracellular NAD⁺ that is produced by growing wild-type cells in 10 μM NR [15]. Elimination of Nrk1 or both Urh1 and Pnp1 produced an identical 40% decline in the ability of NaR to elevate NAD⁺. In the nrk1 mutant, NAD⁺ was elevated from 0.70 ± 0.03 mM to 0.99 ± 0.05 mM, whereas NAD⁺ was elevated from 0.70 ± 0.01 mM to 0.99 ± 0.02 mM in the urh1 pnp1 double mutant. Two strains, namely the npt1 mutant (0.59 ± 0.001 mM), which is deficient in the NAD⁺ salvage necessitated by Sirtuin activity [7], and the nrk1 urh1 pnp1 mutant (0.60 ± 0.02 mM), which is deficient in NR utilization and salvage [15], have baseline NAD⁺ concentrations in vitamin-free media that are 110 μM and 100 μM lower than those of the other strains, respectively. However, whereas the npt1 mutant was increased to 0.73 ± 0.03 mM with addition of NaR, the triple mutant in NR utilization was unable to obtain an increase in NAD⁺ concentration in response to NaR (0.59 ± 0.02 mM).

Thus, NaR is a synthetic vitamin precursor of NAD⁺ that is phosphorylated by Nrk enzymes in vitro (Table 1) and utilized in vivo (Figure 6). In vivo utilization of NaR is not limited to the Nrk pathway producing NaMN, because NaR can fulfill the vitamin requirement of a bna1 nrk1 mutant, and NaR can elevate NAD⁺ in cells without Nrk1. Just as NR salvage goes through Nrk and Urh1/Pnp1 pathways [15], the nucleoside-splitting activities of Urh1 and Pnp1 and Nrk1 must be eliminated to block NaR utilization. Yeast NAD⁺ biosynthetic pathways updated to include NaR utilization are schematized in Figure 7.

His-tagged human Nrk1 and Nrk2 proteins were expressed and purified from E. coli strain BL21(DE3) as described [4]. Kinetic analyses were performed in 20 mM HEPES, pH 7.5, 100 mM NaCl, 5 mM MgCl₂ with 1 mM ATP or GTP as phosphodonor and with varying concentrations of nucleoside substrates. Reactions were initiated by Nrk1 or Nrk2 enzyme sufficient to convert 1% to 10% of the input nucleoside to nucleoside monophosphate in 30 min incubations at 37 °C. Products were quantified by anion exchange high-performance liquid chromatography (HPLC) as described [4] and kinetic parameters were determined from Lineweaver-Burke plots.

Nrk1 (30 mg/ml) was crystallized by 1:1 sitting drop vapor diffusion (18 °C) against the reservoir solutions listed in Table 2. Crystals were cryo-protected in 1:1 paratone and mineral oil. Diffraction data (Table 2) were reduced to intensities with the HKL2000 suite [36], and the first Nrk1 structure was solved de novo as described in the text. ARP/wARP [37] was used for model building, and PHASER [38] was used for molecular replacement of subsequent Nrk1 structures. Geometric restrains for NR, NMN, and tiazofurin were generated on the PRODRG server [39]. Restrained refinement using REFMAC [40], geometric validation using MOLPROBITY [41], and manual rebuilding using COOT [42] were performed iteratively until convergence (Table 2). Coordinate alignments were performed by secondary structure matching [43] within COOT. Molecular graphics were produced with PyMOL [39]. Structure factors and coordinates have been deposited in the Protein Data Bank.

Trimethylsilyl trifluoromethanesulfonate (1.039 g, 4.4 mmol; Sigma-Aldrich; http://www.sigmaaldrich.com) was slowly added to ethyl nicotinate (0.9 ml, 6.6 mmol; Sigma-Aldrich) and 1,2,3,5-tetra-O-acetyl-β-D-ribofuranose (1.4 g, 4.4 mmol; Sigma-Aldrich) in 50 ml anhydrous methylene chloride at room temperature, stirred under argon. The mixture was heated to reflux for 8 h. TLC (CH₂Cl₂: MeOH: TEA=5: 0.3: 0.05) stained with 10% H₂SO₄ in MeOH showed the disappearance of the ribofuranose and appearance of the presumed product, 2′, 3′, 5′-triacetyl ethyl NaR in a single spot at lower mobility relative to the front. After evaporation of methylene chloride, product (25 mg, 0.05 mmol) was added into 0.9 ml of 312 mM NaOEt in EtOH on ice to form O-ethyl β-NaR. After mixing well, the reaction was stored at −20 °C overnight. The reaction was quenched with addition of acetic acid to neutralize the pH. After organic solvent was removed in vacuum, the residue was dissolved in water and extracted with cyclohexane to remove organic impurities. The aqueous phase was then concentrated 10-fold, made to 150 mM in phosphate buffer, and provided with 10 μl of pig liver esterase (13 units; Sigma-Aldrich) to release NaR in a 25 °C overnight incubation. NaR was purified by C-18 HPLC. NaR was assayed by MALDI MS, in positive ion detection mode, and was observed as the protonated molecular ion (predicted mass-to-charge ratio m/z= 256.08, observed m/z= 256.1). Other assignable fragmented ions detected included protonated Na. The entire m/z spectrum (% peak height) was 256.1 (1.81%), 228.0 (48.5%), 207.1 (25.4%), 146.1 (10.3%), and 124.0 (13.9%). We used a molar extinction coefficient of 6411 cm⁻¹ (260 nm) for NaR and 4305 cm⁻¹ (259 nm) for NR.

Yeast strain BY278, which contains qns1 deletion covered by plasmid pB175 (QNS1 and URA3) and which contains nrk1 deletion, has been described [4]. pB450 and pB459, which are LEU2 plasmids for expression of human NRK1 and NRK2 cDNAs under GAL1 promoter control [4], were used as templates for site-directed mutagenesis to produce nrk1-D36A (pHC12), nrk1-E98A (pHC10), nrk2-D35A (pHC13), and nrk2-E100A (pHC11). BY278 was transformed with each plasmid and the empty p425GAL1 control. After passage on galactose media, transformants were streaked on synthetic complete, galactose media with 5-fluoroorotic acid and 10 μM NR [4] to score the function of NRK alleles. Isogenic strains for the NaR utilization study were BY165-1d (qns1) [4], KB046 (bna1 in the deletion consortium background [44]), KB056 (nrk1 deleted from KB046), and JS949 (bna1 npt1, a gift of Jeffrey S. Smith, University of Virginia, United States). The four strains were grown in synthetic media with 3 μM Na plus 10 μM NR, washed in saline, and then cultured to exhaustion in vitamin-free media [15]. To assay utilization of NR or NaR, strains grown to exhaustion in vitamin-free media were streaked on vitamin-free synthetic media supplemented with 10 μM NR or NaR and photographed after 3 d at 28 °C. NAD⁺ measurements were performed as described [15] with isogenic yeast strains grown in vitamin-free media and vitamin-free media supplemented with 10 μM NaR to an optical density (OD)_{600 nm} of 1. Strains were wild-type BY4742, KB009 (nrk1 in the deletion consortium background [44]), KB008 (npt1 in the deletion consortium background [44]), PAB047 (urh1 pnp1 [15]) and PAB038 (urh1 pnp1 nrk1 [15]).

The Swiss-Prot (http://www.ebi.ac.uk/swissprot) accession numbers for proteins in this paper are: human MTAP (Q13126); human Nrk1 (Q9NWW6); human Nrk2 (Q9NPI5); human NP (P00491); human Uck2 (Q9BZX2); S. cerevisiae Bna1 (P47096); S. cerevisiae Meu1 (Q07938); S. cerevisiae Npt1 (P39683); S. cerevisiae Nrk1 (P53915); S. cerevisiae Pnc1 (P53184); S. cerevisiae Pnp1 (Q05788); S. cerevisiae Qns1 (P38795); S. cerevisiae Urh1 (Q04179); E. coli gntK (P46859); E. coli panK (P0A615); and B. stearothermophilus adk (P27142). The Protein Data Bank (PDB) (http://www.rcsb.org/pdb) accession numbers for human Nrk1 are 2QSY, 2QT1, 2QT0, 2QSZ, and 2P0E; for human Uck2 is 1UJ2.