Research Article

A Glycyl Radical Site in the Crystal Structure of a Class III Ribonucleotide Reductase

See allHide authors and affiliations

Science  05 Mar 1999:
Vol. 283, Issue 5407, pp. 1499-1504
DOI: 10.1126/science.283.5407.1499


Ribonucleotide reductases catalyze the reduction of ribonucleotides to deoxyribonucleotides. Three classes have been identified, all using free-radical chemistry but based on different cofactors. Classes I and II have been shown to be evolutionarily related, whereas the origin of anaerobic class III has remained elusive. The structure of a class III enzyme suggests a common origin for the three classes but shows differences in the active site that can be understood on the basis of the radical-initiation system and source of reductive electrons, as well as a unique protein glycyl radical site. A possible evolutionary relationship between early deoxyribonucleotide metabolism and primary anaerobic metabolism is suggested.

Ribonucleotide reductases (RNRs) are essential enzymes for all life, constituting the only de novo catalytic path for the production of the deoxyribonucleotides required for DNA synthesis. The chemically very difficult replacement of the ribose 2′-OH in ribonucleotides by hydrogen is made possible through the use of free-radical chemistry. The presently characterized RNRs can be divided into three classes on the basis of their primary structure and, more importantly, their radical generation mechanism (1). Class I RNRs are composed of two dimeric subunits, R1 and R2. They generate a stable tyrosyl free radical on the R2 subunit through activation of molecular O2 by a dinuclear iron center. This radical is most likely transferred over a long hydrogen-bonded radical transfer pathway to the active site on the R1 subunit (2, 3). Class II enzymes are monomers or homodimers and use adenosylcobalamin to generate transient 5′-deoxyadenosine and cysteine radicals (4).

Class III RNRs are expressed by certain strictly or facultatively anaerobic bacteria under anaerobic conditions (5). They have quaternary structure α2β2(6, 7). The active site and allosteric regulatory sites lie in the large α2 subunit (gene product NrdD, molecular size in phage T4 136 kD, 2 × 605 amino acids), whereas the small β2 subunit, also known as the activase (NrdG, molecular size in T4 36 kD, 2 × 156 amino acids), carries out homolytic cleavage of S-adenosylmethionine (AdoMet) in association with an iron-sulfur cluster to generate the stable glycyl radical near the COOH-terminus of NrdD, at residue 580 (7–9). Exposure of the active complex to oxygen results in inactivation by cleavage adjacent to the glycyl radical and loss of the COOH-terminal 25 residues (10). In class I and II reductases the original radical is transferred to a thiyl radical in the active site (4, 11), and this is also believed to be the case for class III. The reducing equivalents for reduction of the ribonucleotide in class I and II are delivered from thioredoxin or glutaredoxin and are shuttled to the active site by way of two Cys residues in the COOH-terminus of the large subunit (3,12). In class III reductases, formate serves as the overall reductant (13), being oxidized to CO2. Class III RNRs have no significant overall sequence similarity to class I or II enzymes in the putative active-site domain, but sequence homology in a nucleotide-binding regulatory domain present in some RNRs, combined with similarities in allosteric behavior and radical chemistry (1, 5), have been used to argue for a common evolutionary origin for the three classes.

The concept of an RNA world preceding the DNA world of today is based on the versatility of the RNA molecule for both storage of genetic information and as a catalyst. Supporting the RNA world hypothesis is the existence of metabolic paths for de novo synthesis of RNA precursors, whereas DNA precursors are only produced through reduction of ribonucleotides by RNR. Making the reasonable assumption that an autotrophic production of deoxyribonucleotides was present at an early stage of the DNA world, an understanding of the evolutionary relationships between the various RNR families could give essential information on the nature of the transition between the RNA and DNA worlds. Several lines of evidence point to an evolutionary kinship between class I and class II RNRs: conservation of five essential Cys residues (3, 14, 15), participation of a thiyl radical in the reaction (4, 11), and a significant level of overall sequence homology between class I and recently determined class II sequences from Archaea and deeply rooted Eubacteria (16). However, no strong evidence has been available for an evolutionary relationship between class III and the others, and hence for a unified three-dimensional structure for RNRs. The fact that class III enzymes use the primitive metabolites formate and AdoMet, as well as the presence of an iron-sulfur cluster, suggests that class III may be the most closely related of all modern-day RNRs to the so-called “ur-reductase,” the postulated common ancestor of all three classes (1). In this scenario, the advent of oxygen on Earth, being lethal to the class III RNR–dependent organisms, would lead to the entry of the more “modern” class I and II systems, using oxygen-resistant radical cofactors and dithiol reduction chemistry.

The active-site fold is conserved. We chose to work with the G580A mutant of NrdD in order to avoid the O2-dependent cleavage associated with the native enzyme and thus allow purification and crystallization of the intact complex under aerobic conditions (17). The structure was solved by a combination of multiple isomorphous replacement and multiple wavelength anomalous dispersion methods, with three derivatives (Table 1). Refinement was continued with a higher resolution data set obtained from a crystal soaked in deoxyadenosine 5′-triphosphate (dATP) and cytidine 5′-triphosphate (CTP). The current refined model has an Rfactor of 23.0% and Rfree of 30.5%, with good geometry (Table 2). The relatively high R factors are due to the presence of several disordered regions.

Table 1

Data collection and processing statistics. A first native data set was collected at the Swiss-Norwegian beamline of the European Synchrotron Radiation Facility to 3.0 Å. A higher resolution data set extending to 2.75 Å was later obtained at beamline X12B of the NSLS from a fresher crystal soaked in 4 mM dATP and 4 mM CTP for 2 days. Data were integrated and merged with the HKL package (46), and further data analysis was performed with programs from the CCP4 package (47). For details of the structure solution and refinement, see (48).

View this table:
Table 2

Statistics for refinement of the NrdD model. The two values given for number of reflections represent the working and free sets, respectively. An acceptable percentage of the residues in structure native 1 (76.6%) and of those in structure native 2 (79.7%) lie in the core regions of the Ramachandran plot as defined in (49).

View this table:

The refined model of T4 NrdD consists of residues 27 to 542 and 571 to 586, as well as the allosteric effector dATP. There is density for a further 25 residues, most probably lying between residues 542 and 571, but which cannot yet be identified because of disorder. The fold is depicted in Fig. 1A. Despite the lack of significant sequence homology between class I and III RNRs, the fold is based around the same 10-stranded β/α barrel observed in the structure of the class I large subunit, R1 (15) (Fig. 1), consisting of two parallel five-stranded β sheets arranged in an antiparallel manner, linked by an extended loop known as the “finger loop” protruding through the center of the barrel, which is flanked on the outside by helices. 163 Cα atoms from the barrel cores (31% of NrdD) can be superimposed with a root-mean-square deviation of 1.5 Å (18). In view of the mechanistic and sequence similarities between classes I and II outlined above, we can therefore anticipate that all three classes, despite their substantially different cofactors and quaternary structures, will have the same active-site fold and likely be of common evolutionary origin (19).

Figure 1

A comparison of the folds of the active-site domains of class I RNR (R1) and class III RNR (NrdD) (15). β strands are colored green, α helices are red (except where noted), and loops are white. Only one monomer of each dimer is shown. (Left) NrdD. The allosteric effector molecule dATP is shown in a ball-and-stick representation. The “finger loop” is drawn as a yellow coil. The lengthened helices αA and αB, and the inserted helix α4, at the dimer interface, are colored light blue. For clarity, the COOH-terminal region after the end of the last β strand is not shown. (Right) R1. Only the β-barrel domain is depicted (residues 225 to 734). The substrate GDP is depicted in the center of the barrel and the allosteric ef- fector dTTP at the edge (20), as ball-and-stick models. Helices αA and αB, which form a four-helix bundle at the dimer interface, are shown in light blue.

The major differences in topology between NrdD and R1 lie at the dimer interface, in particular between β strands B and C, which represents part of the substrate specificity allosteric site in both enzymes. Structural changes induced by deoxynucleoside triphosphate (dNTP) binding at this site have been suggested to mediate substrate specificity on the neighboring monomer through interaction of loops across the dimer interface (20). In R1, the dimer-stabilizing interactions come mainly from a short four-helix bundle consisting of helices A and B from each subunit. In NrdD, two long helices, α4 and αB, replace the simple crossover helix αB of the repeated β-α motif in the R1 barrel (Fig. 1). The inserted helix α4 is 22 residues long and precedes αB, which is lengthened from 12 residues in R1 to 26 in NrdD (Fig. 1). These long antiparallel helices pack against each other at the two-fold axis (Fig. 2A), the main contacts being made by the parallel pair α4 and α4′ (where the prime indicates an element from the second monomer). The dimer interface is greatly remodeled through shifts in these helices with respect to the strands of the barrel (Fig. 2C), such that the monomers are oriented very differently to each other in NrdD. In particular, the dimer axis is rotated by almost 90°. Also, 1725 Å2 of surface area are buried in the NrdD dimer as opposed to only 744 Å2 in R1 (21).

Figure 2

Altered dimer interfaces of NrdD caused by insertion and elongation of helices and loops with respect to R1. (A) Dimer interactions in NrdD. The view is perpendicular to the two-fold axis, which is shown as a blue vertical rod. The dimer axis in R1 is shown for comparison, as a red rod approximately perpendicular to the page. Monomer 1 is drawn in beige, monomer 2 in aquamarine. Helices αA, α4, and αB are drawn in rust and dark blue for monomers 1 and 2, respectively. The COOH-terminal loop and partially disordered area are drawn in shades of purple. The COOH-terminal loop of the left-hand monomer is mostly hidden, on the rear of this view. The allosteric effector dATP is represented as a CPK model, in the same colors as the respective monomers. (B) Close-up stereo view of the helix interactions at the dimer interface. The helices are anchored at the top by a hydrophobic cluster consisting of residues Trp154, Ile156, and Tyr162 of each monomer. In contrast, the middle portions of the helices contain many basic residues that protrude into a putative solvent-filled channel. Other secondary-structure elements involved in dimer-forming interactions closer to the body of the barrels are loops 90 to 100, 185 to 192, 221 to 228, and the NH2-terminal helix α1, on the bottom of the dimer as seen in (A). (C) A large shift in the orientation of the dimer-forming helices relative to the three strands of the β barrel to which they are connected. This shift, together with the lengthening and insertion of helices, is primarily responsible for the almost 90° rotation of the dimer axis with respect to R1. The molecules are oriented to best superpose strands A to C of the barrel (seen from left to right). Residues from R1 are drawn in purple, from NrdD in aquamarine.

The class III active site. Figure 3A shows the active site of NrdD with modeled guanosine 5′-diphosphate (GDP), obtained by superimposing the coordinates of R1 with bound GDP (20) on NrdD. Residues in the vicinity of GDP that are conserved in all known NrdD sequences are depicted, as well as potentially relevant, nonconserved residues. The substrate of class III reductases are NTPs, but we have conservatively chosen not to model the third phosphate here.

Figure 3

(A) The active site of NrdD. The view is looking down into the β barrel, with Cys290 at the tip of the finger loop. In the center of the picture is GDP, as modeled by superimposing the Cα coordinates of R1 in complex with GDP (20) on NrdD. Side chains of all strictly conserved residues near the putative NTP-binding site and others that may be of importance are shown. Strictly conserved residues are drawn with light gray C atoms. The widely, but not totally, conserved Tyr441 and Ser292 are drawn with darker gray C atoms. Conserved Gly residues are indicated by coloring the coil regions red. The upper stretch contains Gly124 and Gly125, the lower stretch Gly289. (B) Sketch of the active site of R1, for comparison. Functionally important residues as identified by mutagenesis, and theoretical and structural studies, are drawn. Gly253, important in providing space for the substrate base moiety, is drawn as a short stretch of red coil.

A reaction mechanism for RNRs has been proposed that could be general at least for classes I and II (22) and that is given weight by functional and structural (20) studies [for a recent discussion see (23)]. In class I RNRs, three conserved cysteine residues in the active site are essential for activity (24). Cys439 (Escherichia coli numbering) lies at the end of a loop that penetrates through the center of the barrel, and is suggested to be responsible for abstraction of a H atom from C3′ of the substrate ribose to initiate reduction (25) (Fig. 3B). Cys225 and Cys462 lie in close proximity, at equivalent positions on strands βA and βF in opposite halves of the barrel (15), and deliver two reducing equivalents to the substrate, being thereby oxidized to a disulfide (Fig. 3B). Only two of these essential Cys residues are conserved in class III, namely Cys290(equivalent to Cys439 in R1) at the tip of the loop and Cys79 (equivalent to Cys225) on strand βA. The location of Cys290 in the active site is consistent with results obtained from [γ-32P]8-N3ATP labeling (26). In place of the third Cys (Cys462 in R1) is found Asn311, conserved in all NrdD sequences. Because disulfide bond formation is a critical step in the proposed reaction mechanism (20, 22), it is clear that it cannot be general for all classes. However, the sequential reactivation process in class I and II RNRs, in which reduction of the active-site disulfide by a Cys pair near the COOH-terminus is followed by reduction of the COOH-terminal disulfide by thioredoxin or glutaredoxin (12, 15), is not necessary in NrdD, whose overall reductant is formate (13). On the basis of the current structure, the most plausible suggestion is that formate binds directly in the active site, where it would be well placed to pass reducing equivalents through Cys79 (27).

Two further important side chains in class I RNRs are absent in NrdD (Fig. 3). Glu441 (E. coli R1 numbering) is suggested to participate in base-catalyzed elimination of a water molecule across the C2′-C3′ bond to remove the 2′-OH group. This is supported by structural studies (20), site-directed mutagenesis (28), chemical studies with substrate analogs (29), and quantum mechanical calculations (30). Asn437 has been suggested to participate in a hydrogen-bonded chain important in the latter stages of reduction (20). Asn437 and Glu441 are replaced in the NrdD sequence by Met288 and Ser292, neither of which have their side chains oriented to interact with the modeled substrate. Arg291 is also directed away from the substrate. No suitably oriented acidic side chain can be found in the active site to take on the role of Glu441 (31). It is possible that formate plays the dual role of proton donor or shuttle in the early stages of the reaction (replacing Glu441), and hydrogen donor at a later step (replacing Cys462) (32).

Other strictly conserved residues in the active site of T4 NrdD are shown in Fig. 3A. His64 and His66 almost certainly participate in phosphate binding, the former probably to the γ-phosphate. The base is likely to stack on top of Phe194, which in turn stacks on Phe231. The lack of a side chain at Gly125 and Gly126 is important to make room for the base (20).

The glycyl radical is proximal to the active site. In the initial 3 Å electron density maps, it was not possible to build a model for the COOH-terminal 65 residues. However, data collected from fresher crystals soaked with dATP and CTP revealed clear density for residues 571 to 586 (Fig. 4A). These form a long loop directed into the barrel, of which the tip, containing the mutated glycyl radical site Ala580, meets the tip of the “finger loop” in the active site (Fig. 4B). This places Ala580surprisingly close to Cys290, the putative radical cysteine. The distance between Cα of Ala580 and Sγ of Cys290 is only 5.2 Å. This suggests that a radical could be generated on Cys290 in the native enzyme through short-range transfer of a H atom to Gly580, in contrast to class I, where the stable tyrosyl radical is generated at a buried site in the separate R2 protein and transferred by way of a long radical transfer pathway involving at least six amino acid side chains. Taking into account the possible steric effects of the Gly580→Ala mutation, direct abstraction is a likely possibility, but neighboring residues may also participate in radical transfer (33).

Figure 4

(A) The glycyl radical site in the COOH-terminal region of NrdD. Electron density for the COOH-terminal loop, which extends into the top of the barrel toward the active site. Also drawn is Cys449. The Sγ atoms of Cys449 and Cys579 lie at a distance of 4.0 Å from each other. The figure shows the beginning of a break in the electron density between them that may be due to a covalent modification. The map has coefficients |F o| − |F c| and is calculated after removal of the displayed atoms from the phasing model, application of small random shifts to the atoms, and conjugate gradient energy minimization. The contour level is 3σ. (B) Conformation of the COOH-terminal loop and its interactions with the body of NrdD. The loop is colored beige and the rest of the protein light blue. The finger loop extends up from the bottom of the figure and is colored brown. Possibly significant hydrogen bonds are indicated; the bond between Arg291 and Cys579 is rather long, at 3.6 Å, but could be significant within the experimental coordinate error at this resolution. (C) Sequence alignment of the sequence preceding the finger loop and of the COOH-terminal region in NrdD and PFL. Only two representative NrdD sequences and one PFL sequence are shown. nrddt4, bacteriophage T4 NrdD (SwissProt P07071); nrddec,E. coli NrdD (P28903); pflec, E. coli PFL (P09373); tbssa, benzylsuccinate synthase from Thauera aromatica (GenBank 3184131); tutd: TutD protein from T. aromatica (independently identified, also a benzylsuccinate synthase; GenBank 3127057); tdce, bifunctional PFL/2-ketobutyrate formate lyase from E. coli (P42632). Key to consensus sequence (cons.): bold letters represent residues conserved in more than 80% of known sequences, lowercase letters represent conservation in more than half the sequences; φ, hydrophobic; (+) positively charged. (D) Anchoring of the COOH-terminal loop by interactions that suggest an evolutionary relationship to pyruvate formate lyase and other glycyl radical–containing enzymes. The color scheme is the same as that in (B). The residues Arg577, Met265, and Asp268 are conserved in all NrdD sequences and in all but one of the PFL sequences.

The conformation around Ala580 is partly determined by two hydrogen bonds from the main chain nitrogens of residues 580 and 581 to the side chain of Glu446. The latter is completely conserved in all known NrdD sequences. It lies at the NH2-terminus of helix αH, which in R1 hydrogen bonds to the α-phosphate of the GDP substrate (20). Glu446 may participate in fine-tuning the position of the glycyl radical relative to Cys290 in response to substrate binding, which would thus act as a trigger for Cys radical formation.

The NrdG subunit and AdoMet most likely bind on the outside of the loop, on the other side of Glu446 (Fig. 4B). If the conformation of native NrdD is the same as that seen in the G580A mutant, both of the H atom positions on Gly580 would be relatively inaccessible for direct H atom abstraction by a 5′-deoxyadenosyl radical derived from AdoMet, because of steric hindrance caused by the hydrogen bonding of Glu446 to the main chain nitrogens of residues 580 and 581. This suggests that some conformational change occurs before H atom abstraction. The buried nature of Gly580 is in agreement with the very slow solvent exchange observed in D2O (7,34). More structural and spectroscopic studies are needed on the NrdG subunit and its interactions with AdoMet.

The small-subunit NrdG remains elusive. The (4Fe-4S) cluster that interacts with AdoMet to generate Gly580 is formed only in dimeric NrdG (9). We observe no NrdG in crystals of the complex (35). Assuming coaxial aligment of dimeric NrdD and NrdG, the most plausible position for NrdG is on the lower face of NrdD as seen in Fig. 2A. The distance between symmetry-equivalent Gly580 residues in the dimer is ∼50 Å, which makes it difficult to imagine how a single iron-sulfur cluster positioned on the dimer axis could generate a 5′-deoxyadenosyl radical that would then abstract a H atom from Gly580, because the distance to the (4Fe-4S) cluster would be about 20 Å. It seems likely that two CxxC motifs, disordered in the present structure, preceding the Gly radical site (residues 543 to 546 and 561 to 564, respectively) are part of a structural motif involved in radical generation. Mutagenesis of any of these cysteines to Ser results in lack of glycyl radical–dependent cleavage in the presence of oxygen (36).

Hints of an evolutionary relationship to other glycyl radical enzymes. Despite very limited sequence homology, NrdD intriguingly shares many of its radical-generation properties with pyruvate formate lyase (PFL), an enzyme involved in anaerobic fermentation (37). The latter is also a large dimeric protein with a glycyl radical near the COOH-terminus (38). This radical is generated by direct abstraction of the Pro-S hydrogen of the glycine by a 5′-deoxyadenosyl radical produced through cleavage of AdoMet by a small iron-sulfur protein (39) with high homology to NrdG (3). However, the activase is monomeric, rather than dimeric as in class III RNRs (40).The residues immediately surrounding the glycyl radical site in PFL are highly homologous to NrdD (41), and the product of the PFL reaction, formate, is the overall reductant of NrdD. These similarities raise the question whether they could be evolutionarily related. In addition, two further short regions of sequence homology are observed (Fig. 4C) (3).

The NrdD structure suggests a role for these conserved residues. Asp268, at the beginning of strand βE, makes a salt bridge to Arg577 which also appears to be essential for loop stabilization (42). The side chain of Met265 provides a buttressing interaction between the last strand of the β barrel, βJ, and the extended portions of the COOH-terminal loop (Fig. 4D). It participates in a hydrophobic cluster involving Val540, Ile575, and Leu582, which are hydrophobic in all NrdD and PFL sequences. This suggests a distant but significant structural homology between NrdD and PFL. Electron paramagnetic resonance spectroscopy of PFL inactivated by substrate analogs and dioxygen (43) has suggested that in PFL the Gly and Cys radicals are in rapid exchange and lie spatially close to each other, which is consistent with the structure of the Gly-Cys radical site in NrdD. The methyl group of Ala580 in NrdD replaces the Pro-S hydrogen of Gly580 in the native enzyme and faces toward the surface. If we assume a similar structure for PFL, this may also explain why this proton is stereospecifically removed in PFL (44). PFL also provides an alternative model for NrdD-NrdG interaction. A NrdG dimer could interact “end-on” with one of the NrdD monomers in such a way that the (4Fe-4S) cluster approaches the glycyl radical site, similar to the interaction, although transient, of monomeric PFL activase with the PFL dimer (40).

Recent results have shown that NrdD and PFL belong to a growing family of glycyl radical enzymes, which now contains members involved in toluene metabolism and threonine degradation (45). Arg577 is also conserved in these systems (Fig. 4D). It may thus be that the structure of NrdD represents a common framework for the whole family. Asn311 is also conserved in PFL, which binds pyruvate, but not in the enzymes where a carboxylate moiety is not involved, raising the possibility that it is part of a common carboxylate binding site. An evolutionary link to PFL, a likely participant in early pre-aerobic glycolysis, raises the possibility of a progression from a potent radical-based catalytic machinery in primary metabolism to the first system performing the difficult reduction of the 2′ carbon-oxygen bond of ribonucleotides.

  • * To whom correspondence should be addressed. E-mail: derek{at}, par{at}


View Abstract

Navigate This Article