Structures of eukaryotic ribonucleotide reductase I provide insights into dNTP regulation

Abstract

Ribonucleotide reductase catalyzes a crucial step in de novo DNA synthesis and is allosterically controlled by relative levels of dNTPs to maintain a balanced pool of deoxynucleoside triphosphates in the cell. In eukaryotes, the enzyme comprises a heterooligomer of α₂ and β₂ subunits. The α subunit, Rnr1, contains catalytic and regulatory sites. Here, we report the only x-ray structures of the eukaryotic α subunit of ribonucleotide reductase from Saccharomyces cerevisiae. The structures of the apo-, AMPPNP only-, AMPPNP–CDP-, AMPPNP–UDP-, dGTP–ADP- and TTP–GDP-bound complexes give insight into substrate and effector binding and specificity cross-talk. These are Class I structures with the only fully ordered catalytic sites, including loop 2, a stretch of polypeptide that spans specificity and catalytic sites, conferring specificity. Binding of specificity effector rearranges loop 2; in our structures, this rearrangement moves P294, a residue unique to eukaryotes, out of the catalytic site, accommodating substrate binding. Substrate binding further rearranges loop 2. Cross-talk, by which effector binding regulates substrate preference, occurs largely through R293 and Q288 of loop 2, which are analogous to residues in Thermotoga maritima that mediate cross-talk. However loop-2 conformations and residue–substrate interactions differ substantially between yeast and T. maritima. In most effector–substrate complexes, water molecules help mediate substrate–loop 2 interactions. Finally, the substrate ribose binds with its 3′ hydroxyl closer than its 2′ hydroxyl to C218 of the catalytic redox pair. We also see a conserved water molecule at the catalytic site in all our structures, near the ribose 2′ hydroxyl.

Eukaryotic ribonucleotide reductase (RNR) is an enzyme composed of α₂ and β₂ subunits that catalyzes a crucial step of de novo DNA synthesis by converting nucleoside diphosphates to deoxynucleoside diphosphates (1, 2). Tight control of dNTP pools is vital for cell viability; because of the crucial role RNR plays in balancing the relative levels of dNTPs, it is highly regulated transcriptionally (3), allosterically (4–6), by compartmentalization of the various subunits within the cell (7, 8), and, in Saccharomyces cerevisiae, by its protein inhibitor Sml1 (9–12). The molecular basis for these processes is not fully understood.

Rnr1, the α subunit of RNR, contains the catalytic site, the substrate-specificity site, and the activity site (Fig. 1A), whereas the β subunit houses a tyrosyl radical required for RNR activity (13, 14). An elegant mechanism of specificity cross-talk determines substrate preference based on the nucleotide effector bound at the specificity site (4, 15, 16). Brown and Reichard (2) have proposed that the effectors ATP and dATP bind at the activity site and activate or inhibit, respectively. They also bind at the specificity site and select for pyrimidine substrates, whereas thymin triphosphate (TTP) and dGTP select for GDP and ADP, respectively. More recently, evidence for a different role for the activity site has been proposed (17).

Fig. 1.

Structure of Rnr1. (A) The dimer of apo Rnr1. Rnr1 monomers are yellow and green; dGTP (violet) and ADP (blue) from the cognate complex are shown in the specificity and catalytic sites. The three-helix insert and C-terminal insert are red and blue, respectively. (B) 1.0 σ 2F_o–F_c electron density for effector (blue density), loop 2 (red density), and substrate (green density) for the dGTP–ADP complex.

Although several structures have been reported for prokaryotic Rnr1s (4, 15, 16, 18–21), until now, no eukaryotic structure has been available. In general, it has proven easier to solve Rnr1s with effector (4, 15, 16, 21) than with substrate (4, 16). A class III Rnr1 (15) has also been observed with dCTP bound to the catalytic site, but its binding mode is probably either inhibitory or nonphysiological. Presently, structures of Rnr1 bound to a specificity effector and its corresponding substrate (a cognate effector–substrate pair) are available only for Class II RNR (from Thermotoga maritima) (4) or as a single class Ia structure from Escherichia coli that has a poorly defined substrate bound at the catalytic site (16).

We report the only eukaryotic Rnr1 structure from S. cerevisiae, solved by multiwavelength anomalous dispersion (MAD) phasing (22) to 2.2-Å resolution. The six structures we present include the apo and complexes with AMPPNP (a nonhydrolysable analogue of ATP), AMPPNP–CDP, AMPPNP–UDP, TTP–GDP, and dGTP–ADP. Our structures reveal the class I Rnr1 with fully ordered substrates bound at the catalytic site and show details of loop rearrangements that reposition crucial residues to facilitate specificity cross-talk between the specificity and catalytic sites.

Results and Discussion

Tertiary Structure of Rnr1.

The yeast Rnr1 structure contains the characteristic 10-stranded α/β-barrel (23) formed by two antiparallel β sheets comprising parallel strands A–E and F–J (Fig. 1A; and see Figs. 4–6, which are published as supporting information on the PNAS web site). Yeast Rnr1 is similar to prokaryotic Rnr1s; it aligns with E. coli and Salmonella typhimurium Rnr1 structures with rms deviations of 1.7 and 1.8 Å, respectively, and with the monomeric class II Lactobacillus leichmannii Rnr1 with an rms deviation of 1.8 Å; see Fig. 4 for a structure-based alignment. Compared specifically with the E. coli structure, yeast Rnr1 has two insertions, a three-helix insert (THI) common to all eukaryotic sequences examined and some Class II and Class III Rnr1 sequences and a C-terminal insert (CI) of 118 residues, 28 of which are shared with other eukaryotes (Figs. 1A and 4–6). The electron density for the CI in the apo form ends 48 residues beyond what is visible in the E. coli structure, leaving 92 residues of the C terminus undetermined. In this structure, the CI crosses the THI.

The Catalytic Site.

The previous E. coli structure of Rnr1, with a partially visible GDP at the catalytic site (16), showed the 2′ and 3′ OH of the substrate ribose near a catalytic cysteine, asparagine, and glutamate (on the finger loop) as well as a catalytic redox pair of cysteines (on βA and βF, see Table 1, which is published as supporting information on the PNAS web site). Structures of S. typhimurium Class Ib Rnr1 with effector only (21), L. leichmannii Class II monomeric apo Rnr1 (20), and T. maritima Class II Rnr1 with cognate effector–substrate pairs (4) showed similar arrangements at the catalytic site. The role of the catalytic cysteines has been described in ref. 24.

Our four cognate-pair effector–substrate structures were produced by soaking crystals of Rnr1 with appropriate nucleotides and contain fully ordered, reduced Class I Rnr1 catalytic sites (Fig. 1B shows a 2F_o–F_c map of the dGTP–ADP complex). As previously observed in T. maritima, the ribose ring adopts a 3′-endo conformation in all substrates in our cognate pairs (in contrast, the ribose in the E. coli structure was modeled as unpuckered). The 2′ and 3′ OH of the ribose are near the catalytic N426 and E430, the free-radical recipient C428, and C218, the near member of the catalytic redox pair (C218 and C443; see Table 1). This redox pair is reduced in our structures (Fig. 2A–E). C443 is 6 Å from C218 and 9.4 Å away from the ribose 2′ OH and, consequently, not shown in Fig. 2. Interestingly, we observe a strongly bound water molecule 2.5–2.6 Å from the 2′ OH of the ribose in all cognate effector–substrate complex structures (Fig. 2) and at the same location in the apo structure. This water molecule is coordinated by four hydrogen bonds: to the 2′ OH of the ribose, to the amide nitrogen of L427, to the side chain of N426, and to the carbonyl of G247. Compared with the E. coli (16) Class I structure with a partially defined substrate and the T. maritima (4) Class II structure, our TTP–GDP and AMPPNP–purine substrate complexes show C218 closer to the ribose’s 3′ OH than to its 2′ OH. In the dGTP–ADP complex, C218 is only slightly nearer to the 3′ OH. In contrast, in E. coli, the near redox cysteine C225 is 4.9 Å from the 3′ OH and 3.3 Å from the 2′ OH, whereas in T. maritima, the equivalent cysteine C134 is 4.6 Å from the 3′ OH and 4.1 Å from the 2′ OH. It may be that, in the yeast structures, the ribose’s hydrogen bond to the catalytic-site water molecule (Fig. 2) twists its 2′ OH away from and its 3′ OH toward C218, relative to the E. coli and T. maritima structures, raising the possibility that the prokaryotic and eukaryotic catalytic mechanisms could be subtly different.

Fig. 2.

The catalytic site. (A) A schematic of ribose in the catalytic site of the dGTP–ADP complex. Carbon is yellow, oxygen red, nitrogen blue, and sulfur green. (B–E) Substrate binding (loop 2 is shown on the right). Secondary structure: dGTP–ADP, magenta; TTP–GDP, blue; AMPPNP–CDP, orange; and AMPPNP–UDP, yellow. Interacting atoms are colored as above, except phosphate, magenta; substrate carbons, cyan; protein Cα carbons, as in secondary structure. (B) Stereoview of dGTP–ADP. (C) Stereoview of TTP–GDP. (D) Stereoview of AMPPNP–CDP. (E) Stereoview of AMPPNP–UDP.

A network of hydrogen bonds to phosphates and ribose hydroxyls grips the common portion of the substrate. The 2′ OH of the ribose hydrogen-bonds the carbonyl of S217. Four hydrogen bonds pin the 3′ OH: one to Nδ2 of N426, one to Sγ of C218, and two to the side chain of E430. The phosphates of the different substrates bind similarly: the α phosphate to main-chain amides from P607 to A609, the β phosphate to main-chain amides and side-chain hydroxyls of S610, T611 from the N terminus of α24, and S202 from the N terminus of α12. Equivalent helices were seen to bind substrate β phosphates in E. coli and T. maritima structures; the helix dipole is thought to stabilize phosphate binding, suggesting similar stabilization by yeast Rnr1. The side chain of L445 makes van der Waals contacts with the ribose and α phosphate, whereas the side chains of M606 and A201 make van der Waals contact with the phosphates, cradling them against the main chain (Fig. 2).

Specificity Cross-Talk.

Specificity cross-talk takes place through rearrangement of loop 1 and loop 2 (Fig. 3). Loop 1 interacts solely with effectors bound at the specificity site, whereas loop 2 spans the specificity and catalytic sites. Loop 2 contacts the bases of the effector and the substrate, adopting unique conformations in each effector–substrate complex, providing specificity (Fig. 3). The sequence of loop 2 is identical for yeast, mouse, and human Rnr1s, indicating a common mechanism for eukaryotic specificity cross-talk (Fig. 4). In the apo structure, loop 2 is approximately planar and stands straight up between the catalytic site and the bottom of loop 1, such that P294 would hinder substrate binding because of steric clashes (Fig. 3A) that are severe for the larger purines (<1 Å). In contrast, for the pyrimidines, the closest distance to loop 2 is 2.3 Å. Comparing the structures of apo, AMPPNP only, and AMPPNP–CDP or AMPPNP–UDP reveals that binding of effector twists and pulls loop 2 out of the catalytic site, and subsequent substrate binding draws a portion of loop 2 back partway to interact with the substrate base (Fig. 3A).

Fig. 3.

Specificity-site interactions. (A) Loop-2 rearrangements. Substrate (Left) and loop 2 and effector (Right) are shown for AMPPNP–UDP (yellow). Loop 2 is shown for apo (black), AMPPNP only (gray), and AMPPNP–CDP (orange). P294 is shown for apo and AMPPNP only. Q288 and R293 are shown for AMPPNP–CDP and AMPPNP–UDP. Cα spheres are shown for Q288, R293, and P294 in all structures. (B–E) Specificity effector binding. Colors of secondary structure cartoons are as in Fig. 2. Loop-1 carbons are yellow; Loop-2 carbons, light blue; other interacting atoms are colored as in Fig. 2, except effector carbons, which are green. (B) Stereoview of dGTP–ADP. (C) Stereoview of TTP–GDP. (D) Stereoview of AMPPNP–CDP. (E) Stereoview of AMPPNP–UDP.

Effector Interactions.

In yeast Rnr1, as in the previous Class I structures from S. typhimurium and E. coli, specificity effector binds at both ends of a four-helix bundle comprising α helices A and B from one monomer and equivalent helices A′ and B′ from the other monomer (see Figs. 5–7 and Table 2, which are published as supporting information on the PNAS web site). In our complex structures, the effector fits into a hydrophobic pocket whose floor (below and in front of the nucleotides in Fig. 3B–E and behind the nucleotide in Fig. 7) consists of I228 and I231 from αA and Y285 and V286 from αB′. I231, under the ribose, is conserved in other Class I structures, as is I228, under the base, V286 corresponding to alanine in S. typhimurium and cysteine in E. coli. Y285 is conserved in S. typhimurium, where it is believed to hinder ATP binding, because it would sterically interfere with the 2′ OH (21). In the yeast structure, this side chain hydrogen-bonds a well coordinated water molecule that stabilizes its position 0.5 Å further from the effector 2′ carbon, allowing it to hydrogen-bond the 2′ OH of AMPPNP; this interaction is unique to yeast. In E. coli, this residue is serine, but its position is largely occupied by F281 from a different helix, αB. This phenylalanine is orthogonal to the plane of the tyrosine side chain it replaces (see Fig. 7).

As in prokaryotic Rnr1 structures, loop 1 of yeast Rnr1 folds over the effector, fitting I262 above the base. This residue is conserved in E. coli and S. typhimurium. The ribose’s 3′ OH hydrogen-bonds the side chain of D226, an interaction conserved in the other Rnr1 structures. In the yeast structures, K243 from the C terminus of αA′ makes a strong salt bridge to the effector’s α phosphate; this interaction is a hydrogen bond in S. typhimurium and T. maritima. In E. coli, this residue is serine, but the interaction is conserved via K246, one turn upstream on αA, which makes a salt bridge to the α phosphate.

Effector binding initiates changes in loop-2 structure (Fig. 3). The greatest chemical differences between the effector bases are proximal to the bottom of loop 2, where their unique patterns of hydrogen bonds with loop-2 residues may be a key to selectivity (Fig. 3C–F). In the yeast dGTP–ADP complex (Fig. 3B), the guanine base makes five hydrogen bonds to the bottom of loop 2, involving main-chain atoms of Y285 as well as G289 and G290, residues that are considerably rearranged in this structure, and the side chain of D287 (Fig. 3C). dGTP is the only effector that hydrogen-bonds the side chain of D287. In the yeast TTP–GDP complex (Fig. 3C), the base makes four hydrogen bonds to backbone atoms from D287, Q288, and G289 of loop 2 and N270 of loop 1. A fifth hydrogen bond, and the only one from a side chain to TTP, is from N270. This residue is threonine in E. coli and makes only van der Waals interactions with effector; in S. typhimurium, this residue is serine and makes a backbone hydrogen bond but only van der Waals contact via its side chain; in T. maritima this residue is a serine and makes a backbone hydrogen bond, whereas its side chain makes a second-sphere hydrogen bond to N3 of the base. In E. coli, the TTP base is close to loop-2 residues but does not form any hydrogen bonds with them. In S. typhimurium, the base makes one hydrogen bond to N246 of loop 2; in T. maritima, the base makes two hydrogen bonds to backbone carbonyls V200 and K202 from loop 2.

In the yeast AMPPNP only complex (data not shown), the base makes two hydrogen bonds: to D287 and Q288 of loop 2. However, in the AMPPNP–CDP complex (Fig. 3D), Q288 moves completely out of the specificity site to the catalytic site. The adenine ring then makes two hydrogen bonds with D287, which takes Q288’s place. Loop 2’s conformation in the AMPPNP–UDP complex is almost identical to AMPPNP–CDP (Fig. 3A), and AMPPNP–UDP’s effector forms the same hydrogen bonds (Fig. 3E). In all of the effector–substrate complexes, D287 hydrogen-bonds the bases of the respective effectors, usually with its backbone atoms. We observe 1 Mg²⁺ bound to the effector in all complexes.

Interactions with Substrate Bases.

Loop 2 contacts substrate bases distal to the ribose ring, in the region of greatest chemical variation. R293 and Q288 of loop 2 seem to provide most of the selectivity by their interactions with the base (Fig. 2B–E, and Table 1). In our dGTP–ADP structure, the side chains of Q288 and R293 make crucial interactions with the adenine ring (Fig. 2B). Q288’s Nε2 atom hydrogen-bonds the N1 atom of the adenine edge-on. R293’s Nη2 atom also hydrogen-bonds the adenine N1 and helps position the side chain of Q288 by hydrogen-bonding its Oε1 atom. These interactions are unique to the ADP substrate. The N6 atom on the adenine participates in a second-sphere hydrogen bond, via a water molecule, with the carbonyl oxygen of P294 (Fig. 2B), which moves 5.7 Å from its position in the apo structure (Fig. 3A).

In our TTP–GDP complex (Fig. 2C and Table 1), residues R293 and A296 from loop 2 make van der Waals interactions with the base through their main chains and Cβ atoms. Additionally, N1 and N2 of the GDP base hydrogen-bond a water that also hydrogen-bonds the carbonyl oxygen atoms of R293 and G295 of loop 2. The N2 of GDP hydrogen-bonds the carbonyl oxygen of G246, an interaction that ADP, which lacks a hydrogen-bond donor at that position, cannot make. We propose that this interaction with G246 is crucial for discriminating between GDP and ADP.

The pyrimidines make van der Waals contacts with Q288 (Fig. 2D and E and Table 1). O2 and N3 of UDP (Fig. 2E) make an additional water-mediated hydrogen bond to Oδ1 of Q288. In the purine-bound structures, the side chains of Q288 and R293 hydrogen-bond each other, and Q288 also hydrogen-bonds the carbonyl of S242, whereas, in the pyrimidine-bound structures, the side chain of Q288 hydrogen-bonds G246 instead. The larger purines make extra contact with A296 that the pyrimidines do not. In the dGTP–ADP, TTP–GDP, and AMPPNP–UDP structures, water molecules make second-sphere hydrogen bonds between loop 2 and the substrate base.

Residues equivalent to D287, Q288, and R293 are reported to be involved in substrate recognition in the T. maritima Class II Rnr1 structures (K202, Q203, and R207) (4). However, the interactions between these residues and the four substrates are not well conserved between yeast and T. maritima. R207 (R293 in yeast) interacts with the substrate phosphates in the GDP- and CDP-bound complexes, and is disordered in the UDP- and ADP-bound complexes. In yeast, R293 interacts only with the base of the substrate in our ADP- and GDP-bound structures and is within 5.3 Å of CDP and 5.9 Å of UDP. In T. maritima, Q203 (Q288 in yeast) directly contacts only the UDP and CDP substrates, whereas, in our yeast cognate-pair complexes, Q288 contacts all substrates except GDP. Moreover, in the T. maritima dGTP–ADP structure, K202 (D287 in yeast) interacts with the substrate, whereas, in the yeast dGTP–ADP structure, the side chain of D287 hydrogen-bonds the effector. These differences may result from sequence variation in loop 2 between species. Only 50% of the loop 2 residues are conserved (4), with major changes, including insertion of N291 and substitution of an arginine by P294 in eukaryotes (Fig. 4).

The differing bases of the substrates have several conserved interactions. All bases hydrogen-bond the amide nitrogen of G247. The side chains of L427 and C428 pave the bottom of the base portion of the catalytic site (below the substrate in Fig. 2B–E). The side chain of A296 from loop 2 makes a van der Waals contact with the purine bases but not the pyrimidines, which do not extend sufficiently close to loop 2. G246, G247, and the main chain of S217 compose the back of the base-binding site (behind the substrate in Fig. 2 B–E).

Implications.

This work provides a molecular basis for understanding how eukaryotic Rnr1 helps maintain balanced dNTP pools. The four effector–substrate-bound structures show the only fully ordered, reduced catalytic site of Class I Rnr1. The observations of altered position of substrate with respect to C218 and a conserved catalytic-site water molecule offer a eukaryotic template for theoretical analysis of the catalytic mechanism. Our structures of apo-, AMPPNP only-, AMPPNP–CDP-, and AMPPNP–UDP-bound Rnr1 demonstrate that effector binding occurs in concert with P294 rearrangement opening the catalytic site and indicate that loop-2 conformation is determined jointly by binding of effector and substrate.

Materials and Methods

Protein Purification and Crystallization.

The yeast Rnr1 expression plasmid (9) was kindly provided by Rodney Rothstein (Columbia University, New York). Yeast Rnr1 was overexpressed in E. coli BL21(DE3) pLysS strains as described in ref. 25. The cells were lysed by using the freeze–thaw method, and the protein was purified by using peptide-affinity chromatography as described in ref. 26.

Yeast Rnr1 was crystallized in space groups P2₁2₁2 and F4₁32 by using the hanging-drop method at 298 K. The well solution for the P2₁2₁2 form was 0.1 M sodium acetate, pH 6.5, 20–25% PEG 3350, and 0.2 M ammonium sulfate. The well solution for the cubic form was 0.1 M sodium acetate, pH 4.6, and 2.0 M sodium formate. In both cases, 1 μl of the well solution was mixed with 1 μl of protein solution at a concentration of 20 mg/ml. The nucleotide complexes were obtained by soaking the P2₁2₁2 crystals for 3 h in mother liquor containing 20 mM nucleotides plus 10 mM DTT and 10 mM MgCl₂. The dATP–CDP soaked crystals did not diffract to high resolution; the use of the ATP analogue AMPNP in our soaking experiments gave improved resolution.

Data Collection.

Data for the apo form were collected at the Industrial Macromolecular Crystallography Association Collaborative Access Team beamlines, whereas the MAD, AMPPNP only, and dGTP–ADP complex data were collected at the BioCARS beamlines at the Advanced Photon Source. Data for the remaining nucleotide complexes were collected at our in-house x-ray facility by using an R-AXIS IV++ imaging plate mounted on a Rigaku rotating anode with X-stream cooling. Three hundred sixty degrees of data were collected for each of the MAD data sets to obtain maximum redundancy. Although the R_symm of the last shell of data are >50% for the peak and inflection wavelengths, we phased at 3.5 Å to aid with map interpretation. All data sets were collected at 100 K. The data were integrated and scaled by using the program hkl2000 (27) (see Table 3, which is published as supporting information on the PNAS web site).

Structure Determination and Refinement.

The structure of yeast Rnr1 was determined by the MAD method (22) using a HgBr₂ derivatized crystal of the P2₁2₁2 form (Table 3). Initial bromine heavy-atom positions were derived by using the program solve (28). These sites were refined, and phases were calculated with the program sharp (29). The phases were further improved by solvent flattening and extended to 2.6 Å by using the program solomon (30). The apo structure was determined by using the molecular replacement (MR) method (31) implemented in the program phaser (30) using the MAD structure of the unliganded P2₁2₁2 form (the native structure) as the search model.

The complex crystals are all isomorphous to the native P2₁2₁2 form, and the structures were directly determined by the difference Fourier technique (see Table 4, which is published as supporting information on the PNAS web site). The graphics program o (32) was used for model building interspersed with refinement by using both cns (33) and refmac (30). The final models were all evaluated with the program procheck (34), and 99.1% of all residues were in the allowed region of the Ramachandran plot, with >85.3% in the most favored region. Simulated annealing omit maps were calculated by using the program cns (33) to assess the correctness of the structures.

Figures were prepared with the programs pymol (35), ligplot (36) molscript (37) raster3d (38), and topdraw (www.doe-mbi.ucla.edu/People/Software/topdraw.html). The sequence alignment was performed by using clustalw 1.8 from the Baylor College of Medicine (BCM) Search Launcher (http://searchlauncher.bcm.tmc.edu/multi-align/multi-align.html). Structure-based alignments were conducted by using lsqman (32).

Abbreviations

MAD

multiwavelength anomalous dispersion

RNR

ribonucleotide reductase

TTP

thymin triphosphate