Allosteric Inhibition of Human Ribonucleotide Reductase by dATP Entails the Stabilization of a Hexamer.

Ribonucleotide reductases (RNRs) are responsible for all de novo biosynthesis of DNA precursors in nature by catalyzing the conversion of ribonucleotides to deoxyribonucleotides. Because of its essential role in cell division, human RNR is a target for a number of anticancer drugs in clinical use. Like other class Ia RNRs, human RNR requires both a radical-generation subunit (β) and nucleotide-binding subunit (α) for activity. Because of their complex dependence on allosteric effectors, however, the active and inactive quaternary forms of many class Ia RNRs have remained in question. Here, we present an X-ray crystal structure of the human α subunit in the presence of inhibiting levels of dATP, depicting a ring-shaped hexamer (α6) where the active sites line the inner hole. Surprisingly, our small-angle X-ray scattering (SAXS) results indicate that human α forms a similar hexamer in the presence of ATP, an activating effector. In both cases, α6 is assembled from dimers (α2) without a previously proposed tetramer intermediate (α4). However, we show with SAXS and electron microscopy that at millimolar ATP, the ATP-induced α6 can further interconvert with higher-order filaments. Differences in the dATP- and ATP-induced α6 were further examined by SAXS in the presence of the β subunit and by activity assays as a function of ATP or dATP. Together, these results suggest that dATP-induced α6 is more stable than the ATP-induced α6 and that stabilization of this ring-shaped configuration provides a mechanism to prevent access of the β subunit to the active site of α.

Found in all organisms, RNRs are complex allosteric enzymes that supply precursors for DNA replication and repair at controlled levels needed for maintaining the fidelity of these processes.[1−3] Because of their essential role in DNA biosynthesis, RNRs are excellent targets for anticancer, antiviral, and antiparasitic drugs. In particular, human RNR is the target for a number of inhibitors that are currently in clinical use for chemotherapy.[4,5] However, despite their medical importance, the lack of structural information on the active and inactive forms of many of these enzymes presents a major challenge in the design of new therapeutics.

RNRs are classified by the metallocofactor essential for initiating the intricate, radical-mediated mechanism of nucleotide reduction.[6,7] Eukaryotes and many aerobic bacteria utilize class Ia RNRs, which require two subunits, α and β, for activity (Figure A). α contains three nucleotide-binding sites, while β harbors a diferric-tyrosyl radical cofactor (Y176• in human RNR). For each turnover, the two subunits associate and engage in a long-range radical transfer process over 35 Å, involving multiple proton-coupled electron transfer (PCET) steps and redox active amino acids. In this process, the Y• cofactor in β oxidizes a conserved cysteine in the active site of α to generate a thiyl radical that initiates nucleotide reduction via a conserved mechanism.[8] Class Ia RNRs can reduce any of the four ribonucleoside 5′-diphosphates (ADP, CDP, UDP, GDP) that bind to the active site of α, whereas substrate specificity is controlled by the binding of deoxyribonucleoside 5′-triphosphates (dATP, TTP, dGTP) and adenosine triphosphate (ATP) to an allosteric site in α known as the specificity site (Figure A). Both ATP and dATP can promote CDP reduction when bound to the specificity site. However, their binding at a second allosteric site in the so-called N-terminal cone domain of α modulates the overall rate of reduction (Figure A, green domains). Binding of ATP at this activity site stimulates nucleotide reduction, whereas dATP inactivates the enzyme. Together, these allosteric sites act to maintain the proper balance of intracellular deoxyribonucleotide pools.

Figure 1

Structures of human RNR. (A) Each monomer of human α2 contains three nucleotide-binding sites (PDB: 3HNC, 3HND, 3HNE). The activity-regulating site is housed in the N-terminal cone domain (colored in green). Both the C-terminus and N-terminus are disordered in structures of β2 (dotted lines, PDB: 2UW2). (B) Crystal structure of human dATP-induced α6 determined in this study. (C) Scattering from 4 μM α in the presence of 1 mM CDP yields an ab initio shape reconstruction (yellow) that is consistent with the crystal structure of human α2. (D) Scattering from 4 μM α in the presence of 50 μM dATP and 1 mM CDP yields a shape reconstruction (yellow) that is consistent with the crystal structure of human α6.

In recent years, the role of the oligomerization state in the allosteric regulation of class Ia RNRs has been a central question.[9−17] In all class Ia RNRs examined to date, β is a stable dimer (β2). In the case of the most studied class Ia RNR from Escherichia coli, α is also a dimer (α2) when effectors are bound to the specificity site. Recent studies on E. coli RNR have demonstrated that, as originally proposed,[18] α2 and β2 assemble into a compact but transient α2β2 complex, which is active for turnover.[9,14] In contrast, the binding of dATP at inhibitory concentrations to the N-terminal cone domains leads to the formation of an unusual ring-shaped α4β4 complex in E. coli RNR,[9,19] in which each β2 is held at “arm’s length” from the active sites of α2, preventing radical transfer and hence inhibiting the enzyme. However, despite structural homology across all class Ia RNRs, eukaryotic RNRs do not form the same quaternary structures as the E. coli enzyme. Early analytical ultracentrifugation (AUC) studies showed that both ATP and dATP induce higher order oligomers of mammalian RNRs containing only the α subunit,[20,21] in contrast to E. coli RNR, which require both subunits to form oligomers.[22,23] The role of oligomerization in mammalian α was later revisited by Cooperman and co-workers with dynamic light scattering (DLS) and AUC. In these studies, they observed large oligomers, which they attributed to α4 formation in the presence of dATP and α6 in the presence of high ATP (10 mM).[11,13] Although α tetramerization in eukaryotic RNRs has not been reported since, α hexamerization has been reported in the presence of both activating and inactivating effectors by multiple mass-estimation techniques, including size-exclusion chromatography (SEC),[10,12,15] gas-phase electrophoretic mobility analysis (GEMMA),[16] and multiangle light scattering (MALS).[12] These observations have led to the proposal that α6 can be active or inactive, depending on the effectors that are bound.[12]

Structural studies targeted at elucidating the quaternary structure of eukaryotic RNRs have been challenging and thus far limited to inhibitory conditions. Notably, in a recent study by Fairman et al., cocrystallization of Saccharomyces cerevisiae α with dATP led to a crystal lattice that could be formed by two different hexameric arrangements of α (Protein Data Bank or PDB entry 3PAW), one in which three copies of α2 form a ring with the active sites facing inward (Figure S1A) and another in which three copies of α2 form an inverted structure with the active sites facing outward (Figure S1B).[12] In the same study, mutagenesis on human α and electron microscopy (EM) data on Saccharomyces cerevisiae α supported the ring-shaped structure with the inward-facing active sites (Figure S1A) as the dATP-inhibited form of α6. On the basis of mutagenesis and tryptophan fluorescence quenching data, it was further proposed that the ATP-induced α6 configuration would have a different packing arrangement than that formed with dATP. However, no structural characterizations in the presence of ATP were reported. Also, this study relied on the S. cerevisiae enzyme for their structural studies of the α6 state.

In this present study, we investigate the quaternary structure of human RNR using three complementary structural techniques. We present the first X-ray crystal structure of a dATP-induced α6 from human RNR. The crystal lattice giving rise to this structure supports only one hexameric arrangement, in which the active sites face inward. By measuring small-angle X-ray scattering (SAXS) at increasing concentrations of dATP and ATP, we further provide evidence that this hexameric arrangement forms in solution from the dimeric state in the case of both effectors. At high ATP concentrations (>1 mM), however, the ATP-induced α6 readily interconverts with a newly identified filament state, which we describe by both SAXS and electron microscopy. Addition of β2 also promotes large-scale structural changes to α6 in the presence of ATP, but not dATP. Although the physiological relevance of the filament state is not known, our results show that the human α subunit can readily accommodate multiple structural states in the presence of ATP and β2. These results support a unified mechanism of allosteric inhibition for the class Ia RNRs, in which dATP is the only effector that favors stabilization of a single conformation, one in which β2 likely has limited access to the active sites of α6.

Experimental Procedures

Protein Expression and Purification

His6-tagged human α and β were isolated using a modified procedure from previous studies,[10,15] while β was reconstituted following published protocols with minor modifications.[15] To remove the N-terminal His6-tag from as-purified α, 2 U thrombin (Novagen) was mixed with 1 mg α, followed by incubation at 4 °C for 6–8 h. Removal of the His6 tag was previously shown to have a negligible effect on the specific activity of α,[10] and hence, the tagged and untagged α constructs are interchangeable with the one exception being the fraction of filaments at high ATP concentration (as discussed in the Results). In the case of human β, which is prone to proteolysis, the His6 tag was kept intact to aid in the purification of full-length β from fragments. For additional details, see Supporting Information.

RNR Activity Assay

The activity of His6-tagged α or β was determined by measuring the reduction of [5-3H] CDP in the presence of 5-fold excess of the other subunit. To correlate activity with the SAXS studies, 4 μM α (or β) was added to an assay mixture containing 20 μM β (or α), 2 mM [5-3H] CDP, 3 mM ATP, 25 μM human thioredoxin 1 (hTrx1), 0.2 μM human thioredoxin reductase 1 (hTrxR1), 2 mM NADPH in a modified assay buffer (50 mM HEPES, 5% glycerol, 15 mM MgCl2, 150 mM KCl, pH 7.6). Before mixing, all the components were preincubated at 37 °C for 1 min. After initiation by [5-3H] CDP addition, the 180-μL reaction mixture was incubated at 37 °C for 2 min, and a 40-μL aliquot was quenched by boiling for 2 min. Four aliquots were taken at 0.5, 1.0, 1.5, and 2.0 min, respectively. After centrifugation, the supernatant (36 μL) was transferred to a new tube, and the dephosphorylation reaction was performed by incubation at 37 °C for 2 h with 14 U calf alkaline phosphatase (Roche) and 400 nmol of carrier deoxycytidine (dC) in dephosphorylation buffer (75 mM Tris-HCl, 0.15 mM EDTA, pH 8.5). The amount of dC was quantified as described previously.[24] In addition, for assaying α at other concentrations (0.5, 1, 10 μM), the [α]:[β] ratio was kept at 1:5 for all experiments to facilitate comparison of α activities, while [CDP] was varied in order to ensure substrate saturation (Table S1).

Crystallography

Protein solutions containing 12 mg/mL His6-tagged α in 50 mM HEPES pH 7.6, 15 mM MgCl2, 100 mM KCl, 10 mM DTT, 5% v/v glycerol, 500 μM dATP, and 3 mM CDP were prepared on ice and incubated at 37 °C for 2 min. Crystallization experiments were then performed by hanging drop vapor diffusion at 18 °C. Protein solution (1 μL) was mixed with precipitant solution (1 μL) containing 100 mM HEPES pH 7.0, 32% v/v Jeffamine M600, and 10 mM DTT and suspended over a 400 μL reservoir of precipitant solution. Crystals appeared within a week and were flash cooled in liquid N2 without additional cryoprotection. Diffraction images were collected at the Cornell High Energy Synchrotron Source (CHESS) F1 Station on an ADSC Quantum 270 CCD detector, indexed and integrated with XDS, and scaled with XSCALE (Table S2).[25] The structure was solved to 9 Å resolution by molecular replacement in Phaser[26] using a previously solved structure of α2 (3HNC chains A and B) with all waters and ligands removed as the search model. A solution with one α2 in the asymmetric unit was found. Refinement at 9 Å resolution was performed in PHENIX[27] using four atom groups for both rigid body and group B-factor refinement: for each α chain, the N-terminal domain (residues 1–90) and the rest of the polypeptide (residues 91–742) (Table S2). Refinement beyond rigid body was not carried out due to the low resolution. Coordinates and structure factors have been deposited in the PDB as entry 5D1Y.

Small-Angle X-ray Scattering

SAXS was performed at the CHESS G1 station using a 10.5 keV 250 μm × 250 μm X-ray beam. Data were collected on a Pilatus 100 K detector with a sample-to-detector distance of 1.5 m. The transmitted intensity was measured on a PIN diode beamstop, as previously described.[25] SAXS measurements on His6-tagged α were performed in the standard in vitro assay buffer (50 mM HEPES, pH 7.6, 15 mM MgCl2, 5% glycerol). All other SAXS measurements were performed with untagged α or His6-tagged β in a modified assay buffer (50 mM HEPES, pH 7.6, 15 mM MgCl2, 5% glycerol, 13 mM KCl). Each α sample was prepared by first incubating solutions of α, CDP, and ATP (or dATP) separately at 37 °C for 1 min, then adding CDP and ATP (or dATP), in that order, to the α solution. The mixed solution was then equilibrated by incubating at 37 °C for 1 min, followed immediately by data collection. Samples containing both subunits were prepared in a similar way, except that DTT was added to the α stock solution to bring the final DTT concentration in the reaction mixture to 5 mM. CDP and ATP (or dATP) were added to the α stock solution in the same manner as described above. β (preincubated for 30 s at 37 °C) was then added, and data collection was initiated within 30 s of mixing. For every sample, a carefully matched background solution was prepared containing all ingredients in the protein buffer. Samples were loaded into an in vacuo oscillating flow cell,[26] and multiple 1 and 5 s exposures were taken per sample as previously described.[27] Data collection was completed within 4 min of loading each sample. The scattering images were integrated about the beam center and normalized by the transmitted intensities following previously described image correction procedures.[27] Superimposable exposures that did not display apparent radiation-induced changes were averaged. Background scattering collected on matched buffer was subtracted from the protein solution scattering to produce the one-dimensional protein scattering profile, I(q), as a function of q, where q = 4π/λ sin θ; 2θ is the scattering angle, and λ is the X-ray wavelength. The quality of the background subtraction was evaluated as described previously.[27]

Analysis of structural parameters, singular value decomposition analysis, and multistate fitting of experimental scattering curves were performed using custom code written in MATLAB, as described previously.[9,28] The zero-angle scattering intensity I(0) and overall radius of gyration Rg were obtained from a Guinier approximation (eq )[29] to the low q region of the scattering profiles. Rg and I(0) are determined from linear fits to Guinier plots, ln(I(q)) versus q2, in the region qmax*Rg < 1.3.In the case of elongated proteins (e.g., filaments), a similar approximation can be made to I(q)*q (eq ).[29] Here, a cross-sectional radius of gyration, Rc, can be determined from a linear fit to the cross-sectional Guinier plot, ln(I(q)*q) versus q2.

For conformationally homogeneous samples, the pair distance distribution function, P(r), was calculated from the experimental I(q) with the indirect Fourier transform method[29] implemented in the program GNOM.[30] Low-resolution models of protein structures were generated from the GNOM outputs (with a high-resolution limit of qRg ≈ 8) using the ab initio reconstruction programs DAMMIF[31] and DAMMIN.[32] DAMAVER[33] was used to align ab initio models, reject outliers, and average the remaining models to produce the most probable model. Structural parameters and shape reconstruction statistics are summarized in Table S3. Theoretical scattering curves were calculated from atomic coordinates using FoXS.[34] The disordered C-terminus of α (residues 743–792) and several other missing residues (292–293) were modeled into the structure of the dATP-induced α6 with Modeller,[35] and conformational sampling was performed with constant-temperature molecular dynamics (300 K) in AllosMod-FoXS.[36] In the model that best fit the SAXS data, the C-terminal tails were allowed to relax and take up space within the α6 ring.

Electron Microscopy

Specimens for electron microscopy were prepared by incubating 6 μM α with 1, 3, or 10 mM ATP in assay buffer at 37 °C for 2 min followed by 20-fold dilution in assay buffer containing a matched concentration of ATP. The diluted mixture (5 μL) were applied to a thin carbon film supported by a 300 mesh Cu/Rh grid (Ted Pella) that had been glow discharged immediately before use.[37] The grid was washed three times with a stain solution containing 2% uranyl acetate (Ted Pella) and 0.2% trehalose. A second carbon layer was applied, excess stain blotted with filter paper, and the specimen air-dried following standard protocols.[38] Images were acquired at 50000× magnification on a Tecnai F20 microscope (FEI) operated at 120 kV equipped with a 4096 × 4096 pixel CCD camera (Gatan).

Results

Crystal Structure of the dATP-Induced Hexamer

The subunit arrangement of the dATP-induced α6 was determined by X-ray crystallography. Although crystallization of human α in the presence of dATP has been documented, the structure could not be solved due to poor diffraction.[12] We find that in agreement with previous observations,[12] cocrystallization of human α with 500 μM dATP yields crystals that diffract to very low resolution (∼8 Å), much of which is masked by the beamstop in a typical macromolecular crystallography beamline. However, by utilizing a small beamstop to collect diffraction data at low resolution, a 9-Å structure could be solved by molecular replacement using a previously reported human α2 structure as the search model (PDB: 3HNC)[12] (Table S2). At this low resolution, structural details such as side chain and ligand positions cannot be inferred. However, the subunit arrangement can be determined unambiguously (Figure S2). The final structure reveals a ring-shaped α6 (Figure B) formed by three α2 subunits making contacts at the N-terminal cone domains containing the activity-regulating site that binds dATP (Figure B, green). Importantly, the crystal lattice giving rise to this new structure supports only one hexameric arrangement, one that is similar to the inward-pointing architecture in the 6.6-Å resolution structure of the dATP-induced α6 from S. cerevisiae RNR[12] (Figure S1A). The human dATP-induced α6 structure also shares features with the dATP-inhibited α4β4 complex from E. coli,[9,19] which is a ring of a similar size with the active sites facing inward. As in the human α6, the N-terminal cone domains are also involved in making the subunit contacts in this E. coli structure.

dATP and ATP Produce Similar Hexamers

With the structure of the dATP-induced human α6 determined, SAXS was employed to compare the structural effects of ATP and dATP on human α in solution. SAXS is a solution-based structural technique that is uniquely suited for investigating quaternary structure under varying solution conditions, particularly when mixtures of species may be present.[9] Using an in-vacuum oscillating flow cell, scattering was measured from α at very dilute concentrations close to the physiological range (0.5–1 μM monomer).[39] All samples contained 1 mM CDP, the natural substrate for both ATP and dATP when bound to the specificity site.

Titration of dATP into 4 μM α in the presence of 1 mM CDP leads to a dramatic change in the shape of the scattering curves and a concomitant increase in the zero-angle scattering intensity (Figure A, blue to red), indicative of a large structural change involving an increase in mass. Over the course of the dATP titration, much of the structural change occurs at low μM dATP. Above 8 μM dATP, the scattering curves are nearly superimposable. The large set of scattering curves allows for the determination of the number of interconverting species in a model-independent manner via singular value decomposition (SVD). Consistent with the presence of iso-scattering points (Figure A), SVD produces two significant singular vectors (Figure S3, magenta), suggesting that the structural change caused by dATP is an interconversion of two oligomerization states. The initial state (at 0 μM dATP) agrees well with known crystal structures of the dimer (Figures C and S5A), whereas the final state (above 8 μM dATP) agrees well with the crystal structure of the human hexamer (Figures D and 5B). Interestingly, although mammalian α has been reported to dissociate in the absence of effectors,[13,15,16] we find that substrate binding is sufficient to induce dimerization, possibly due to reduced conformational disorder. Linear combinations of the scattering at 0 μM dATP and at 50 μM dATP fit well to the set of curves shown in Figure S6A, thus indicating that the structural change caused by dATP can be described as a dimer–hexamer transition with a midpoint near the equimolar point (1 dATP per α) (Figure B). Only hexamers are observed above 8 μM, which corresponds to the full occupancy point (2 dATP’s per α). Importantly, intermediates, including α tetramers, are not observed in these data.

Figure 2

Both dATP and low ATP induce hexamerization of human α. (A) SAXS profiles measured as 0–50 μM dATP is titrated into 4 μM α in the presence of 1 mM CDP (blue to red curves) display iso-intensity points, suggesting a two-state transition. (B) Fitting linear combinations of α2 and α6 scattering to the titration data provides relative fractions of the two species and shows a midpoint near equimolar [dATP]:[α]. (C) Activity in the presence of dATP is biphasic with a loss in activity beyond equimolar [dATP]:[α]. (D) SAXS profiles measured as 0–1 mM ATP is titrated into 4 μM α in the presence of 1 mM CDP (blue to red curves) are consistent with a two-state dimer–hexamer transition. (E) Fitting linear combinations of α2 and α6 scattering to the titration data provides relative fractions of the two species as a function of ATP. (F) A steady increase in activity is observed up to 1 mM ATP and continues to slowly increase at higher ATP.

Figure 5

Model for the interconversion of oligomerization states in E. coli class Ia RNR compared with that of human RNR. In E. coli RNR, α2 does not form higher order oligomers without β2. Addition of dATP and β2 leads to a ring-shaped α4β4, which prevents β2 from interacting with the active site of α2 (“β2 held at arm’s length” model), whereas addition of ATP favors a compact α2β2 complex that is active. In human RNR, α forms higher order oligomers without β2. Both ATP and dATP cause α to hexamerize, but these hexamers have variable stability. The increased stability of dATP-induced α6 prevents β2 from gaining access to the active site of α2 (“β2 exclusion” model), whereas the decreased stability of ATP-induced α6 allows β2 access to the active site of α2, creating active RNR species.

Under identical solution conditions, ATP was titrated into 4 μM α in the presence of 1 mM CDP. Addition of ATP up to 1 mM leads to scattering curves (Figure D, blue to red) that are highly similar to the set obtained from the dATP titration experiment (Figure A). SVD of these ATP titration data also yields two significant singular vectors, indicative of a two-state transition that is largely complete at 1 mM ATP (Figure S4, magenta). Notably, the calculated scattering of an inverted hexamer model in which the active sites point outward (Figure S1B) yields a poor fit to the scattering data at 1 mM ATP (Figure S5B). In contrast, the scattering at 1 mM ATP is nearly superimposable (up to a q value of 0.15 Å–1) with that measured in the presence of 50 μM dATP (Figure S5B), suggesting that ATP forms a hexamer that is indistinguishable from the dATP-induced α6 at low resolution. Furthermore, linear combinations of the same two scattering curves used to fit the dATP titration data also fit well to the ATP titration data (Figure S6B), indicating that sub-millimolar concentrations of ATP cause a dimer–hexamer transition similar to that observed with micromolar concentrations of dATP.

Instability of the ATP-Induced hexamer

Although α hexamerization is favored up to 1 mM ATP (Figure D,E), additional conformational changes are observed at higher ATP concentrations. Increasing the ATP concentration to 10 mM in the presence of 4 μM α and 1 mM CDP leads to an upturn in the scattering at low q and a loss of features representing the hexamer, such as the peak at q = 0.07 Å–1 arising from cross-ring interference (Figure A). These changes are consistent with the growth of aggregates that are highly anisotropic in shape. Interestingly, this effect is more pronounced for α constructs with an N-terminal hexahistidine (His6) tag. Although the tag has negligible effects on the dimer–hexamer transitions in the presence of dATP or submillimolar ATP (Figures S3, S4), the presence of the tag leads to a greater fraction of anisotropic aggregates at >1 mM ATP. In fact, the scattering from 4 μM His6-tagged α displays a total loss of features at 10 mM ATP and can be approximated as a highly elongated species with a cross-sectional radius of gyration (Rc) of 28.1 Å (Figure S5C). In agreement with the SAXS data, electron micrographs of 0.3 μM His6-tagged α incubated with 1, 3, and 10 mM ATP illustrate the progressive conversion of hexamers to prominent filaments (Figure B). The formation of filaments would explain previous observations of aggregation-like behavior in eukaryotic α occurring in the presence of ATP.[16] Whereas a loss of scattering features is observed at millimolar concentrations of ATP (Figure A), these features become more defined when dATP is added to 4 μM α in the presence of 3 mM ATP (Figure C), suggesting that the filament formation is reversible. Together, these results demonstrate that the ATP-induced α6 can readily interconvert with other conformational states, whereas dATP favors the hexameric form.

Figure 3

High ATP induces a third α oligomerization state. (A) SAXS profiles measured from 4 μM α in the presence of 1 mM CDP and high ATP (1–10 mM) display changes in scattering (blue to red curves) that can be attributed to the conversion of hexamers into a highly elongated species. (B) Electron micrographs of 0.3 μM His6-tagged α in 1, 3, and 10 mM ATP show the conversion from primarily hexamers to filaments. The scale bar corresponds to 50 nm. (C) The SAXS profile measured from 4 μM α in 3 mM ATP, 1 mM CDP (blue) becomes more defined when 50 μM dATP is added to the mix (red) and becomes superimposable with that of 4 μM α in 50 μM dATP, 1 mM CDP (black dotted). Although subtle, the change in scattering is consistent with the reversal of filaments into hexamers.

Activity in the Presence of dATP and ATP

In an effort to correlate oligomerization with the specific activity of α, dCDP formation was measured as a function of dATP and ATP in the presence of 4 μM α, a 5-fold excess of β, 0.5 mM [5-3H]-CDP, human thioredoxin, and human thioredoxin reductase. To prevent precipitation of the subunits at these high concentrations, the assay buffer was modified with the addition of 150 mM KCl. In the absence of dATP or ATP, we find that turnover occurs at a very low rate of 0.080 ± 0.014 s–1 (initial points in Figure C,F). Titration of dATP leads to a 3.2-fold increase up to 4 μM dATP (equimolar with α), followed by a steady decrease to 0.045 ± 0.007 s–1 at 16 μM dATP (Figure C). On the basis of previously reported dissociation constants of 0.07 and 1.5 μM for dATP binding to the specificity and activity sites, respectively, in murine RNR,[40] our observation of biphasic behavior in activity can be attributed to dATP first binding the specificity site, which stimulates CDP reduction, followed by dATP binding to the activity-regulating site causing inhibition. In contrast, titration of ATP leads to a steady increase in turnover up to 3 mM ATP (Figure F), and then maintenance of this high level of activity out to 10 mM ATP (Figure S7), the maximum concentration of ATP we tested. This trend in activity data is as expected from a positive effector and is consistent with recent work on murine RNR.[16]

Before comparing these activity data collected in the presence of both α and β (both subunits are required for activity) with oligomeric state data on α alone, the influence of β on the oligomeric state of α must be established. If, in the presence of dATP, β has no effect on the oligomeric state of α, then Figure data can be interpreted as indicating that α6 formation and inhibition are correlated (Figure B,C). As we describe below, we find this correlation to be valid in the case of dATP. In the presence of ATP, however, we show below that the presence of β does affect the oligomeric state of α, and thus we cannot definitively say based on Figure E,F data that ATP-induced α6 are active.

β Has a Large Effect on the Structure of α Oligomers in the Presence of ATP, but Not in the Presence of dATP

To address the affect of β on the structures of α oligomers, we prepared multiple-turnover reaction mixtures of the two subunits under both activating (3 mM ATP, 1 mM CDP, 5 mM DTT) and inhibiting conditions (50 μM dATP, 1 mM CDP, 5 mM DTT). Scattering was then measured immediately after a short incubation of the reaction mixtures. Here, DTT was used to reduce the redox active cysteines in α that are oxidized during each turnover. Under the activating conditions (3 mM ATP, 1 mM CDP, 5 mM DTT), we expect ∼25% of the CDP to be converted to dCDP during the course of the SAXS measurement. As no changes in scattering were observed during the course of data collection, this low level of turnover does not appear to affect the structure or background scattering to an appreciable extent.

We find that addition of β2 to α under inhibiting conditions (50 μM dATP, 1 mM CDP, 5 mM DTT) leads to a scattering curve (Figure A, red) that is greater in overall intensity but otherwise highly similar to that of the dATP-induced α6 ring alone (Figure A, black). Notably the peak at q ≈ 0.07 Å–1, which arises from the ring-shaped hexameric arrangement, remains unperturbed and thus suggests that the dATP-induced α6 ring maintains its shape even with the addition of β2. Interestingly, the scattering under these conditions (Figure A, red) cannot be explained as a simple sum of the scattering from the α6 ring (Figure A, black) and free β2 (Figure A, green), indicating that the two subunits associate. In particular, a 1.6-fold increase is observed in the forward scattering intensity or I(0), which is larger than the expected value of 1.1 in the case where β2 cannot associate with the dATP-induced α6 ring but is smaller than the expected value of 2.3 in the case where three copies of β2 fully associate with the α6 ring to form a homogeneous solution of α6β6. These results indicate that some fraction of β2 is associated with the dATP-induced α6 ring, which is consistent with previous GEMMA results that suggest a dATP-induced α6β2 species in murine RNR[16] as well as an EM reconstruction of a dATP-induced species from S. cerevisiae RNR in which a single copy of β2 appears bound to the α6 ring.[12] Although in this S. cerevisiae RNR structure β2 appeared to be inside the ring, our data on the human protein in solution is not consistent with an interior binding site for β2. In particular, we observe an increase in radius of gyration (Rg) from 72.9 ± 1.7 to 79.3 ± 3.0 Å with the addition of β2 (Figure S8), which cannot be explained by the formation of a α6β2 complex, in which β2 is bound within the ring as the predominant species. As Rg represents the root-mean-square distance of all electrons in a particle, binding of β2 to the interior of the α6 ring would lead to an overall decrease in apparent Rg, and this value would be further reduced by the presence of any unbound β2. Thus, although we cannot definitively identify a single structure for the inhibited complex as the scattering from our dATP sample likely contains both free and associated β2, we can say that an α6β2 state with β2 outside the ring is a likely component (Figure B).

Figure 4

Human β can alter the structure of α oligomers substantially in the presence of activating effector, ATP, but not in the presence of dATP. (A) The dATP-induced α6 (black) is represented by the SAXS profile of 4 μM α in 50 μM dATP, 1 mM CDP. Addition of equimolar β to 4 μM α under inhibitory conditions (50 μM dATP, 1 mM CDP, 5 mM DTT) results in scattering (red) that shares similar features as that of the dATP-induced α6 (black). In contrast, addition of equimolar β to 4 μM α under activating conditions (3 mM ATP, 1 mM CDP, 5 mM DTT) leads to scattering (blue dotted) that deviates greatly from that of the dATP-induced α6 (black). For reference, the scattering from 16 μM β in 1 mM CDP is shown in green, normalized by a factor of 4 to facilitate comparison with other curves. (B) Model for the differential ability of human β to affect the oligomeric state of human α. In human RNR, both ATP and dATP can promote the formation of ring-shaped α6 in the absence of β2. β2 cannot disrupt the dATP-induced α6 and hence has poor access to the active sites lining the interior of the ring-shaped α6. By contrast, addition of β2 can disrupt the ATP-induced α6 ring. Although the active conformation is not yet known, it is plausible that the increased dynamics under active conditions allows for greater access of β2 to the active sites.

In contrast to the small changes observed when β is added to dATP-inhibited hexamers (black to red curves in Figure A), the addition of 4 μM β to 4 μM α in the presence of ATP leads to a dramatically different scattering curve (Figure A, blue dotted). As in the case of the ATP-induced filaments, here, we see an upturn at low q and a loss of scattering features, indicating that there is aggregation-like behavior involving highly anisotropic species. Although β2 may bind to α filaments, any scattering from smaller species (such as α2β2) would be masked by the presence of aggregates, which scatter more strongly. Thus, we cannot unambiguously identify the active form. However, comparing the effects of the two activity-regulating effectors leads to the important observation that the dATP-induced α6 maintains its shape in the presence of the β subunit, whereas the ATP-induced α6 does not (Figure B). This difference may explain why both effectors can form similar hexameric rings and produce different outcomes in activity.

Discussion

Since the first observations of nucleotide-dependent oligomerization in pioneering studies of class Ia RNRs,[20−23,41] studies by Cooperman and co-workers have spurred renewed interest in elucidating its role in allosteric activity regulation.[9−16,19,42] These studies have shown that eukaryotic RNRs form a different set of quaternary structures than the E. coli class Ia RNR and, in particular, demonstrate the need to better resolve the oligomerization states of the α subunit. In this present study, we used solution-based SAXS to monitor the conformational changes of human α as a function of ATP and dATP. Because scattering is additive at the dilute protein concentrations used in SAXS, the large data sets generated by the titration experiments allow for model-independent SVD analysis. From this analysis, we find that dATP induces a dimer–hexamer transition with no tetramer intermediate. In addition, a similar dimer–hexamer transition is seen at sub-millimolar ATP. Furthermore, both ATP and dATP induce a ring-shaped hexamer that is consistent with our low-resolution X-ray crystal structure (Figure B). Although structurally similar, however, the dATP-induced α6 and the ATP-induced α6 differ in their relative stabilities. Whereas the latter is easily disrupted by the presence of the β subunit or by high concentrations of ATP, the former maintains its ring structure even when β subunit is added.

The inability to detect tetramers between the dimeric and hexameric forms of α in our SAXS data would at first seem to contradict previous observations. Seminal studies performed on RNR subunits purified from Ehrlich tumor cells and calf thymus were the first to show that both dATP and ATP can produce higher order oligomers of α.[20,21,41] In these studies, less generalized forms of the Svedberg equation were used to estimate mass from the sedimentation coefficients (s20,w), leading to its underestimation. Specifically, use of the Atassi-Gandhi approximation to the Svedberg equation[43] led to a tetrameric mass assignment to a 15.2 S species observed in calf thymus RNR treated with dATP.[21] In the case of murine RNR, use of the Svedberg equation for a smooth sphere[44] again led to the assignment of a tetrameric mass for the 16 S species observed in the presence of dATP and a hexameric mass for a broad 23 S peak observed in the presence of ATP.[11,13] However, both equations only apply to highly globular proteins. With the structural data now available from this work and others,[10,12] we know that α oligomers cannot be approximated as such. Using modern software for hydrodynamic calculations,[45] we find that the ring-shaped hexamer observed in our crystal structure has a theoretical s20,w value of ∼16 S. This result strongly suggests that the previously observed 15–16 S species likely arose from ring-shaped α hexamers, whereas higher s20,w values would be consistent with our observations of α filament formation. Use of structural information in hydrodynamic calculations thus brings consensus to previously inconsistent observations.

With the absence of tetramers established in eukaryotic RNRs, we find that a major structural difference in human RNR observed with ATP or dATP is in the relative stabilities of α6 rings. Despite differences in quaternary structure, our observations with human RNR are reminiscent of our previous observations with the E. coli class Ia RNR (Figure ). In E. coli RNR, dATP is uniquely able to stabilize a single species: the ring-shaped α4β4 complex (Figure ).[9] Furthermore, due to its transient nature, the active α2β2 complex in E. coli RNR readily interconverts with other species, including α4β4, under activating conditions.[9,14] Similarly, our results suggest that the α6 ring forms the basis of the allosterically inhibited form of human RNR and that this species is able to interconvert with other species in the presence of ATP (Figure ). Indeed, contrary to previously reported electron micrographs of the S. cerevisiae RNR (EM DataBank entry 1807), in which β2 appeared to bind the interior of the α6 ring,[12] our SAXS results suggest that such a configuration is not likely to be predominant in dATP-inhibited human RNR. We instead propose that the interior α6 ring is largely inaccessible to β2, particularly if we account for the presence of functionally essential tails in both subunits that are disordered and thus not visible in existing crystal structures. Modeling the disordered C-terminal tails of α within the α6 ring not only reduces the accessibility of the active sites, but also further improves the agreement between the theoretical and experimental scattering (Figure S9 and Supporting Information). Our results thus suggest that in both E. coli and human RNR, allosteric inhibition entails keeping β2 away from catalytic positions on α, either by forming a ring structure with β to hold it “at arm’s length” or by forming a ring without β, such that β is excluded (Figure ). Although the oligomeric states involved are not conserved between these two RNR species, the net result is the same: β is prevented from providing a radical species necessary to initiate catalysis in the active site of α.

Our results serve to elucidate the structural basis of allosteric inhibition in human RNR, and at the same time, raise a number of fascinating questions. Specifically, we observe mixtures of species in the presence of both subunits and ATP, some or all of which could be physiologically relevant and/or catalytically active. Although the physiological relevance of α filaments remains to be seen, with the recent visualization of filamentous states of CTP synthetase and IMP dehydrogenase in cells,[46,47] a case can be made for exploring whether filaments observed in vitro can also exist in vivo. Understanding the relevance of RNR oligomeric states as well as the kinetics of structural interconversions will ultimately be essential for understanding RNR function. With the recent progress made for eukaryotic RNRs from this work and others,[10−12,15,16] we are entering a new phase of discovery into this complex class of enzymes.