fHow ATP and dATP act as molecular switches to regulate enzymatic activity in the prototypic bacterial class Ia ribonucleotide reductase

Class Ia ribonucleotide reductases (RNRs) are subject to allosteric regulation to maintain the appropriate deoxyribonucleotide levels for accurate DNA biosynthesis and repair. RNR activity requires a precise alignment of its α2 and β2 subunits such that a catalytically-essential radical species is transferred from β2 to α2. In E. coli, when too many deoxyribonucleotides are produced, dATP binding to RNR generates an inactive α4β4 state in which β2 and α2 are separated, preventing radical transfer. ATP binding breaks the α−β interface, freeing β2 and restoring activity. Here we investigate the molecular basis for allosteric activity regulation in the prototypic E. coli class Ia RNR. Through the determination of six crystal structures we are able to establish how dATP binding creates a binding pocket for β on α that traps β2 in the inactive α4β4 state. These structural snapshots also reveal the numerous ATP-induced conformational rearrangements that are responsible for freeing β2. We further discover, and validate through binding and mutagenesis studies, a previously unknown nucleotide binding site on the α subunit that is crucial for the ability of ATP to dismantle the inactive α4β4 state. These findings have implications for the design of allosteric inhibitors for bacterial RNRs.

the conformational equilibrium away from 44 toward 22 (Ando et al., 2011). Importantly, RNR variants in which residue substitutions prevent formation of the 44-ring are no longer allosterically regulated by dATP (Chen et al., 2018), indicating that dATP-induced 44-ring formation is causative of the allosteric inhibition. Previous structural studies have indicated that ATP and dATP bind to the same site within the cone domains of 2: a structure of E. coli class Ia 70 RNR in the 44 state co-crystallized with dATP (Zimanyi et al., 2016(Zimanyi et al., , 2012 shows the nucleotide in the same site that was occupied by AMP-PNP (an ATP mimic) in a structure of 2 (Eriksson et al., 1997), raising the question of how the binding of nucleotides that differ by one hydroxyl group can lead to such dramatic oligomeric state changes.

80
The N-terminal regulatory "cone" domain containing the allosteric activity site is colored in green. (B) β2 from the structure of the active α2β2 complex (PDB: 6W4X) (C) A compact α2β2 structure is required for catalytic activity (PDB: 6W4X). This complex is in equilibrium with the free subunits and an inactive α4β4 complex (PDB: 5CNS) which is formed when dATP binds to the allosteric activity site. The presence of ATP promotes formation of the active complex. 85 al., 1997). The C root-mean-squared deviation (RMSD) for each structure comparison is shown in Table S4. 120 Figure 2. Overall structures of 2-dATP and 2-ATP agree with that of the previously reported structure of the E. coli class Ia RNR 2 subunit. The regulatory cone domain is colored in a darker shade of the overall structure. Overlay of α2 with AMP-PNP bound at the allosteric activity site (PDB ID: 3R1R) 125 colored in blue/dark blue, α2-dATP colored in teal/dark teal (this work) and α2-(ATP)2 colored in light green/dark green (this work) with nucleotides shown as spheres.
dATP binds similarly to activity site regardless of the presence of 2. The cone domain is at the N-termini of the 2 subunits and is comprised of a -hairpin and a four-helix bundle. We find 130 that dATP binds to the cone domains of the 2 structure in an analogous manner as was observed previously in the structure of the dATP-inhibited 44 state (Zimanyi et al., 2016). The 2 structure presented here is of higher resolution and contains an intact dATP molecule (the dATP molecule was hydrolyzed to dADP during 44 crystallization), and thus affords a more complete picture of dATP binding to the E. coli class Ia RNR enzyme (Fig. 3A,B, Fig. S4A,B). Specificity for the 135 adenine base of dATP is generated by hydrogen bonds between the side chain carboxylate of E15 (acceptor) and N6 (donor) and the backbone amide NH of N18 (donor) and N1 (acceptor).
The base is additionally held in place by packing interactions with residues V7 and I17 (-hairpin), I22 (helix 1), and F49 and I58 (helix 3) (Fig. 3D). The deoxyribose moiety sits between helix 1 and 3 of the four-helix bundle with a single hydrogen bond made between O3´ and helix 3 residue 140 H59. -hairpin residues K9 and R10 and helix 4 residue K91 provide charge neutralization and electrostatic interactions with the phosphates of dATP, which also coordinate a Mg 2+ ion. The only difference in the coordination environment between the 2 and 44 structures is the side chain of T55, which adopts different rotamer conformations that form different hydrogen bonds (Fig.   3A,B). This variation may be due to subtle changes in coordination caused by the loss of the gamma phosphate of dATP over the time course of the 44 crystallization.  The first ATP molecule binds to site 1 by making very similar interactions as dATP (Fig. 3). The 175 hydrogen bonds to the adenine base are identical (Fig. 3B,C). The presence of the 2´-hydroxyl group results in a slight adjustment upward of the ribose ring of ATP compared to the ribose in dATP due to the packing of the ribose against I22 of helix 3 (Fig. 3D, E). This subtle shift of the ribose results in the loss of a hydrogen bond between the ribose 3´-hydroxyl group and H59, whose side chain flips 90° and now hydrogen bonds to a water molecule (Fig. 3F). The 180 phosphates of ATP also sit slightly higher in the activity site, with favorable hydrogenbonding/electrostatic interactions being made by K9, R10, and K91 (Fig. 3C). As a result of these subtle movements, the -hairpin is slightly shifted (Fig. 3F).
The second nucleotide-binding site (site 2) is directly adjacent to site 1 within the cone domain, 185 sandwiched between helix 1 and helix 4 (Fig. 4A). The phosphate groups of the two ATP molecules create an octahedral coordination environment around a central Mg 2+ ion. The additional negative charge from the second triphosphate moiety is stabilized by electrostatic/hydrogen-bonding interactions from K9 and K21 (Fig. 4B). R24 is also within 4 Å of the phosphate groups, although it does not make direct contacts in the crystal structure. The site 190 2 ATP ribose makes hydrogen-bonding interactions to the backbone carbonyl of F87 through O2´ and O3´, suggesting specificity at this site for ribonucleotides (Fig. 4C). Unlike in site 1, there are no specific contacts to the adenine base; instead, it is held in a hydrophobic pocket between helix 1 and 4.

195
Creation of second ATP binding site involves a coordinated movement of three side chains. The transition from the apo-or dATP-to the two ATP-bound form of the cone domain requires a dramatic and coordinated shift of three key residues-H59, F87 and W28-all three of which adopt different rotamer conformations upon ATP binding (Fig. 5, Fig. S5). As mentioned above, the side chain of H59 hydrogen bonds to O3´ of dATP but is flipped out of the activity site 200 when ATP is bound. This new position of the H59 side chain is in close proximity to the side chain of F87, and F87 must adopt a flipped-down position. This flipped-down position of F87 brings its side chain into the pocket occupied by the W28 side chain, causing W28 to flip 90° sideways. The net result of these side chain movements is formation of a new nucleotide-binding site in the cone domain. F87 movement creates the binding pocket for the ribose of the second ATP molecule 205 (Fig. 5, Fig. S5) and W28 movement creates a cavity for the base of the second ATP. The adenine base is nicely sandwiched between side chains of W28 and F97 as a result of W28's movement.

210
F97 does not change, as dATP only binds in site 1. (B & C) Space filling models of the interaction of the dATP/ATP molecules with helix 4 and the W28, H59, F87, and F97 residues.
Equilibrium binding assays are consistent with two higher affinity ATP binding sites at the activity site and one lower affinity ATP binding site at the specificity site. The presence of 215 two ATP molecules at the activity site was unexpected. Previous ultrafiltration binding studies by Ormö and Sjöberg (Ormö and Sjöberg, 1990) indicated that two ATP molecules bind per , which was interpreted as signifying the binding of one ATP molecule to the specificity site and one to the activity site. The 2-ATP structure shows three ATP molecules per  in total: one ATP in the specificity site and, as described above, two ATP molecules in the activity site. To pursue the 220 possibility that a lower-affinity ATP binding site might have gone undetected if the ATP concentration was not high enough in the previous ultrafiltration binding studies, we have re-run these assays using a higher concentration of 3 H-ATP and using non-linear regression to analyze the data. In this assay, the concentration of free and bound 3 H-ATP is determined after separation of the protein by centrifugation in a spin filter, allowing for determination of equilibrium binding 225 parameters. The resulting binding curves from multiple experimental runs were fit with a one-state binding model that assumes all binding sites are equivalent. This simplifying assumption is necessary due to the impracticality of fully sampling the binding curve. Using this approximation, the Kd for ATP binding at 25 °C was estimated to be 158 ± 37 μM with the maximum number of binding sites being 6.8 ± 0.6 per 2 (Fig. 6). This number is consistent with this structure, which 230 shows 3 ATP molecules per  (6 ATP molecules per 2).
To verify the presence of multiple binding sites for ATP at the activity site, the ultrafiltration experiment was repeated in the presence of 100 or 500 μM dGTP, which should bind at full occupancy to the specificity site given that the Kd for dGTP at the allosteric specificity site is 0.77 235 μM (Ormö and Sjöberg, 1990). In the presence of dGTP, the Kd for ATP was measured to be 120 ± 52 μM with a maximum of 3.8 ± 0.5 binding sites per 2 (Fig. 6). These data are consistent with the binding of two ATP molecules in each cone domain. No difference in binding was observed at the two different dGTP concentrations, suggesting that the specificity site is indeed saturated with dGTP under these conditions. Importantly, these data also indicate that the binding sites in 240 the cone domain are the higher affinity sites for ATP, further suggesting that the initial binding study was most likely reporting on two ATP molecules binding to the activity site. We can now also explain the previous observation that ATP binding is cooperative (Brown and Reichard, 1969b;Ormö and Sjöberg, 1990), which was hard to explain if ATP was binding to sites that are more than 40 Å apart, but easy to explain if ATP molecules are coordinated by a single Mg 2+ ion 245 and if the binding of the first ATP creates the binding site for the second ATP, as it appears to do.
Thus, in terms of the question of how the difference of one hydroxyl group (ATP vs dATP) can destabilize the 44 ring and shift the equilibrium back to an active 22 state, these structures and these binding data suggest that it is the one extra hydroxyl group of ATP, and one whole extra molecule of ATP, which are responsible for 44 destabilization.  Ormö and Sjöberg (Ormö and Sjöberg, 1990) was used with modifications to determine the equilibrium binding parameters for ATP. Each point is an independent measurement. The sample consisted of 7-20 μM α2, 3 H-

255
ATP binding to 2 alone is in gray circles. 3 H-ATP binding to 2 in the presence of 100 or 500 μM dGTP is in black squares. These dGTP concentrations saturate the specificity site allowing for the determination of activity-site-only binding parameters. The amount of bound nucleotide was found by subtracting the amount of free nucleotide from total nucleotide and resulting binding curves from multiple experimental runs were fit with a one-state binding model that assumes all binding sites are equivalent. Using this approximation,

260
the Kd for ATP binding at 25 °C was estimated to be 158 ± 37 μM with the maximum number of binding sites being 6.8 ± 0.6. In the presence of dGTP, the Kd for ATP at the allosteric activity site was measured to be 120 ± 52 μM with a maximum of 3.8 ± 0.5 binding sites. These data are consistent with the binding of two ATP molecules at each activity site.

265
Identity of nucleotide in site 1 can be uncoupled from nucleotide binding in site 2 through a W28A substitution. With the confirmation from the binding assays that two ATP molecules bind to the activity site, we next sought to test the importance of ATP binding at site 2 to the overall activity regulation of E. coli class Ia RNR. We generated individual alanine variants of residues W28, F87, and F97, which comprise the second ATP binding site (Fig. 4). As described above,

270
F87 and W28 side chains move between rotamer conformations that alternatively block the second ATP from binding (residue positions in yellow in Fig. 5) and support the second ATP binding (residue positions in pink in Fig. 5), as signaled by the position of H59, which moves in response to the presence of ATP versus dATP in site 1. We reasoned that substitutions of either F87 or W28 with Ala would create room for ATP binding in site 2 regardless of the position of H59 275 and thus be independent of the presence of ATP in site 1. In other words, F87A and W28A substitutions would be expected to uncouple the binding of ATP to site 2 from the identity of the nucleotide effector in site 1. To test this idea, we obtained structural data for W28A-2 with ATP, with dATP/ATP, and with dATP/GTP (Table S1, S3). First, we wanted to determine if ATP could enter site 2 with dATP in site 1 in a W28 construct, so we grew crystals in the presence of a 280 mixture of 3 mM to 5 mM ATP and 1 mM dATP. The resulting 3.40-Å-resolution structure revealed clear density for two nucleotides in the activity site, but the identity of the site 1 nucleotide was ambiguous given the low resolution of the structure (Fig. S6A,B). Due to the lack of clarity about the identity of the nucleotide in site 1, this structure could not be used as evidence that a W28A substitution leads to site 1, 2 uncoupling. Unfortunately, attempts to obtain a high-resolution 285 structure with ATP and dATP have so far been unsuccessful.
Thus, we tried a different approach, co-crystallizing W28A 2 with 1 mM dATP and then soaking with 3 mM GTP. As described above, site 1 is specific for a nucleotide with an adenine base (see  Table S1, S3). With the W28A substitution, the side chain of F87 is positioned down, creating the site 2 binding pocket, and allowing for nucleotide binding, despite the fact that H59 has not moved (Fig. 7). The density in site 2 is 295 consistent with GTP binding in a very similar conformation as found for ATP in the WT 2-(ATP)2 structure (Fig. S6D,H). We also confirmed that W28A substitution does not alter ATP binding in site 2 by determining a crystal structure of W28A 2 with ATP as the only effector at 2.6-Å resolution (Table S1, S3). We find that the W28A substitution results in a more open pocket for ATP binding at site 2, but no structural perturbations in the bound ATP molecule are apparent 300 ( Fig. S6E,F,G,H). All site 1 residues and site 2 residues F87 and F97 are identical to WT 2-(ATP)2 ( Fig. 7, S6E,F,G,H). Therefore, the W28A substitution appears successful in uncoupling nucleotide binding in site 2 from the identity of the nucleotide in site 1 without substantially altering ATP binding to site 2.

305
Although not the focus of this report, the W28A 2-(ATP)2/CDP structure at 2.6-Å resolution structure also provides a higher-resolution picture of the active site bound to the CDP/ATP substrate/effector pair (Fig. S7). The hydrogen bonding pattern that affords substrate specificity is the same as that described previously (Zimanyi et al., 2016).

315
Substitutions W28A, F87A, and F97A at the second ATP binding site disrupt activity regulation. With the knowledge that the W28A substitution uncouples nucleotide identity at site 1 from that of site 2, we proceeded to investigate the relevance of site 2 to allosteric regulation of activity. If site 2 is not important in either dATP-induced activity down-regulation or ATP-induced 320 activity up-regulation, then the W28A RNR variant should behave like WT protein. As controls, we also investigated an F87A RNR variant, which should also uncouple site 1 and site 2, and an F97A RNR variant that should not lead to site 1, 2 uncoupling, but would be expected to decrease the affinity of ATP binding in site 2 due to loss of the stacking interaction of the adenine base with the F97 side chain (Fig. 4). Our results described below indicate that site 2 is, in fact, critical for 325 the ability of RNR to be regulated by the ratio of dATP to ATP, but is not important for activity under ATP-only conditions, and is not important for the ability of dATP to down-regulate RNR when no ATP is present (Fig. 8). In particular, our data show that under activating conditions (3 mM ATP as specificity and activity effector), these three enzyme variants had activities comparable to wild-type 2 with CDP as substrate (Fig. 8A). In the presence of inhibitor dATP,

330
F97A was inactivated to a degree similar to wild-type 2 (5-10% of maximal), and the activity of W28A and F87A was reduced somewhat less (20-25% of maximal) (Fig. 8A). Overall activity regulation in these three mutants is thus only moderately perturbed when ATP and dATP are used as effectors in isolation.

340
RNR variants that are similarly active with 3.0 mM ATP (black bars) as WT, deactivate with 0.175 mM dATP like WT (white bars), although not as fully as WT, and unlike WT, re-activate when 3.0 mM ATP is added to 0.175 mM dATP (grey bars). (B) Specific activity of WT and the W28A variant RNR assessed for a wider range of ATP and dATP conditions. Whereas WT RNR (dark grey bars) is largely inactive when the dATP concentration is 0.175 mM or higher, regardless of the amount of activator ATP, the W28A variant (light gray bars) is very sensitive to the addition of ATP and shows activity even when ATP concentrations are ten-fold less than dATP concentrations. (C) Specific activity of WT and the W28A variant RNR in the presence of inhibitor dATP (0.5 mM) with activator ATP replaced by 1.0 mM GTP. The substitution of W28 for alanine appears to alter the specificity of site 2 for ATP such that GTP can now upregulate enzyme activity. dATP and ATP concentrations are shown below the graph. Error bars shown are standard 350 deviations.
To evaluate the activating effect of ATP in combination with dATP, we tested a condition containing 3 mM ATP and 0.175 mM dATP (Fig. 8A), which are physiologically relevant concentrations for E. coli. With 3 mM ATP and 0.175 mM dATP, wild-type 2 and F97A activity 355 are stimulated only slightly relative to 0.175 mM dATP alone, and remain low relative to 3 mM ATP alone. In contrast, W28A and F87A variants in 3 mM ATP and 0.175 mM dATP both recover near maximal activity under these conditions. This finding is consistent with the ability of ATP to bind to site 2 in W28A and F87A variants even when dATP is bound in site 1, and is consistent with the importance of ATP binding in site 2 to the up-regulation of RNR activity. Notably, our 360 structural data suggest that site 2 is restricted to ribonucleotides and these data support this conclusion. If dATP was binding to site 2 in either the W28A or F87A RNR variant, then these proteins would not show these WT-like levels of dATP inactivation. Therefore, site 2 does not appear to be important to the ability of RNR to be turned off by dATP, but is important for the ability of RNR to be turned on by ATP.

365
We further tested the W28A variant under different ratios and concentrations of dATP and ATP ( Fig. 8B) and found that the W28A variant is fully active in the presence of 1.0 mM ATP under extremely high concentration of dATP (0.5 mM or 1 mM). With a Kd of ~ 6 M for dATP binding (Brown and Reichard, 1969b;Ormö and Sjöberg, 1990) and ~120 M for ATP binding to site 1, 370 concentrations of 1.0 mM dATP and 1.0 mM ATP should be completely inhibitory, but W28A RNR is fully active. Activity does start to go down when the concentration of ATP is lowered to 0.1 mM and dATP is kept at 1.0 mM. Again, these data indicate the importance of site 2 to successful activity regulation by dATP and ATP.

375
Given the structural observation that GTP binds to site 2 in W28A-2, we investigated whether GTP can reverse dATP-inactivation, and it can (Fig. 8C). Under conditions where WT RNR is inactive (0.5 mM dATP and 1.0 mM GTP), W28A-2 is active, although not as active as W28A-2 with 0.5 mM dATP and 1.0 mM ATP. These results indicate that GTP binding to site 2 can reverse dATP-induced inhibition, but that GTP does not bind as well as ATP to this site. It is very 380 interesting that the presence of a ribonucleotide in site 2, regardless of identity, can restore RNR activity. Of course, under physiological conditions, the concentration of ATP is much higher than the concentrations of other ribonucleoside triphosphates (Bochner and Ames, 1982;Buckstein et al., 2008), such that binding of other NTPs in site 2 is unlikely to be relevant in vivo. This result does, however, serve to confirm the importance of site 2 for up-regulation of RNR activity. Again, 385 these data suggest that allosteric regulation by dATP and ATP is not just due to the difference of one hydroxyl group. The hydroxyl group is extremely important, but it is not the whole story. Both the canonical activity site (site 1) and the newly discovered site 2 are critical for allosteric regulation.

ATP binding leads to an increase in the helix 2 helicity, which in turn destabilizes the 44
structure. We next considered the molecular basis for 44 destabilization by the binding of two molecules of ATP by comparing 44 and 2-(ATP)2 structures. There are three changes in the 405 cone domain of note when comparing these structures. First, in the presence of the two molecules of ATP, the 9 KRDG 13 motif of the -hairpin of the cone domain is shifted upward ~2.3 Å (~6° to ~7º) at the tip relative to the pivot point at V7 between these two structures (Fig. 9A). As mentioned above, as compared to dATP bound in site 1, the ribose of the site 1 ATP and the ATP phosphates sit higher in the activity site, and the site 1 ATP phosphates now share a coordinating the position of K9 and R10 of the 9 KRDG 13 motif. R10 and K9 both shift up and K9 no longer contacts the adenine base of the nucleotide in site 1 (Fig. 3A,B), instead contacting a -phosphate of the site 2 ATP (Fig. 4B,C). As a result, the 9 KRDG 13 motif of the -hairpin of the cone domain shifts upward. This upward shift results in the second change of note: an approximately equal and 415 opposite downward shift (2.5 Å) of residues Y50-I53 near the base of the -hairpin (Fig. 9B). This downshift of residues Y50-I53, in turn, releases a strain on the residues at the end of helix 2, which can now fold up into an additional helical turn, which is the third change (Fig. 9B).
Consequently, helix 2 has an extra turn in the 2-(ATP)2 structure than it does in the dATPinactivated 44 structure (Fig. 9C).

420
Thinking about these conformational changes in the opposite direction, the order of events would be: dATP replaces ATP in site 1 and shifts the -hairpin down; residues Y50-I53 shift up in response; and helix 2 unwinds due to the strain. Thus, the -hairpin appears to act as a lever responding to the presence of dATP or 2 ATP molecules, alternatively pulling on and unwinding 425 helix 2, or relaxing and refolding helix 2 (Fig. 9C). This change in helicity also alters the positions of residues H46 and I47 dramatically. dATP binding unwinds helix 2, which creates a binding pocket for residues of the subunit, stabilizing an 44 structure. To understand the significance of helix 2 unwinding in 430 the present of dATP, we evaluated the − interface in the previously determined 44 structure (Zimanyi et al., 2016(Zimanyi et al., , 2012. We find that the unwinding of helix 2 creates a hydrophobic binding pocket for residue I297 of the  subunit to bind, in which I297 packs against I47 (Fig. 9D). When helix 2 is fully wound, as it is in the apo structure (Fig. 9E, PDB: 1R1R (Eriksson et al., 1997)) and as it is in the α2-(ATP)2 structure (Fig. 9F), there is no binding pocket for I297 of  between 435 helix 1 and 2. The turn of the helix 2 physically blocks the -binding site. Additionally, polar residue H46 is pointing toward the binding site for I297, making for an unfavorable interaction, and I47 is unavailable to make a favorable one as it is pointing is the opposite direction ( Fig. 9D-F). Thus, the unwinding of helix 2 of the cone domain of  appears key for the formation of a binding site for the  subunit, creating not only the room for  to bind, but also swapping out an unfavorable 440 interaction (H46) for a favorable one (I47).
Considering the conformational changes in reverse: the binding of two molecules of ATP would destabilize the 44 inactive state by shifting the -hairpin lever up, releasing the strain on helix 2.
As helix 2 refolds, the hydrophobic pocket for residue I297 of  is lost, and the favorable interaction with I47 is replaced with an unfavorable one (H46) to ensure 's departure. The overall interface is small, ~525 Å 2 , making it relatively easy for small changes to break  and  apart.
Although the above mechanism of helix 2 unwinding and re-winding in the presence of dATP and ATP, respectively, beautifully explains how the binding surface for the  subunit can alternately 450 be exposed and tucked away, and all ATP-bound structures show helix 2 fully wound (Fig. S8), we were puzzled as to why our structure of 2-dATP, described above, did not show an unwound helix 2. The structure of dATP-bound 44 shows the unwound helix 2, but we would expect helix 2 unwinding to proceed  binding, since helix unwinding appears to create the  binding site. The lack of helix unwinding in the 2-dATP structure cannot be attributed to differential contacts made 455 by dATP, since as noted above, the same contacts are observed in the 2 structure as in the 44 structure (Fig. 3A, B). Thus, we investigated whether lattice contacts in the crystal might be restricting movement of the −hairpin in the 2-dATP structure, such that helicity of helix 2 would be unchanged, and in fact, residues at the base of the −hairpin (Y50-I53) are involved in lattice contacts (Fig. S8).
We therefore sought a different crystal form for 2 in which the -hairpin region of the cone domain is less restricted by lattice contacts. Finding new crystal forms for 2 has historically been problematic due to the fact that the 2 subunits are not very soluble in the absence of the subunits. In fact, the first crystal form of 2 was obtained through co-crystallization of the 2 465 subunit with a short peptide (residues Y356-L375) that contained the sequence of the  subunit C-terminus (Uhlin and Eklund, 1994). With this in mind, we were able to obtain a new crystal form of 2 by fusing the 35 C-terminal residues of  (342-376) to the C terminus of an 8-residue truncated  (2-C35) (Fig. S9). The overall structure of this crystal form with dATP bound at 2.10-Å resolution (Table S1, S2) is essentially identical to that observed above for free 2; however, 470 the cone domain is slightly less restricted by lattice contacts near the -hairpin (Fig. S8E). With the restraints of the crystal lattice relaxed, helix 2 adopts the unwound conformation as in the 44 complex (Fig. 9G). The hydrophobic pocket vacated by H46 is left unfilled in this structure, and residues 19-27 (helix 1) and 35-43 (helix 2) that interact with 2 in the 44 complex are exposed to solvent. This structure thus demonstrates that specific contacts with 2 are not required for helix

Discussion
Allosteric activity regulation in enzymes often involves the movement of side chains in an active site in response to the binding of an allosteric effector in order to either increase or decrease 480 enzyme activity. With this in mind, the incredible oligomeric state change associated with E. coli class Ia RNR activity regulation is even more impressive (Fig. 1C). E. coli class Ia RNR's mechanism for allosteric regulation of activity can be described as a game of 2 keep away, preventing radical transfer and thus activity by keeping 2 at arms-length, trapped in a ring structure. But, how is it that the dATP binding stabilizes the trapped ring state whereas, ATP, a 485 molecule that differs only by a single hydroxyl group, allows the ring to fall apart, freeing 2. The work presented here suggests that the difference is one hydroxyl group and a whole second molecule of ATP and indicates that protein conformational changes that are involved in controlling ring stability are considerable. In fact, one could describe the molecular mechanism involved as the protein-equivalent of a Rube Goldberg machine.

490
A key component of this Rube Goldberg machine is H59. Previously, Sjöberg and co-workers showed that a H59A variant was not able to discriminate effectively between dATP and ATP and implicated H59 in triggering allosteric regulation (Rofougaran et al., 2008). Studies on mammalian RNR, and the equivalent residue (D57), have additionally showed the importance of this residue 495 in allosteric regulation (Caras and Martin, 1988;Reichard et al., 2000). Now, through this work, we can explain how H59 communicates the presence of dATP or ATP and triggers the appropriate response in the E. coli class Ia RNR enzyme. We can now also describe that response, which we find involves three sets of conformational changes: the H59-triggered forming/unforming of allosteric effector site 2; the upward/downward tilting of the -hairpin and accompanying 500 relaxing/tugging of helix 2; and the winding/unwinding of helix 2 and accompanying sealing/revealing of the  subunit binding pocket on . domain: H59 hydrogen bonds to the ribose (R) hydroxyl group of dATP (yellow); site 2 is blocked by Phe87 and Trp28; β-hairpin (black) is anchored down via contacts between dATP and R10 and K9; and helix 2 (green) is unwound, creating binding surface for . (B) ATP-bound cone domain: H59 does not engage in hydrogen bonding to the site 1 ribose (R) of ATP (pink) and H59 is tilted; F87 and W28 have moved and site 2 is open with ATP (grey) bound; the β-hairpin (black) is pushed up due to presence of 2ATPs; the strain on helix 2 (green) is decreased and helix 2 is rewound. (C) The binding surface for β (orange) is alternatively created by dATP (yellow) pulling the β-hairpin (black) down, and hidden by 2 molecules of ATP pushing the β-hairpin (black) up, which unwinds and rewinds helix 2 (green) respectively.
In terms of stabilizing the 44 inactive ring-like state, H59 contributes by communicating the 515 presence of dATP to Phe87 and Trp28 such that they adopt positions that block allosteric effector site 2 (Fig. 10a, Fig. S5). With allosteric effector site 2 closed for nucleotide binding, dATP in site 1 anchors the tip of the -hairpin down by contacts made to R10 and K9. With the tip down, a strained helix 2 is unwound, and the binding surface for the  subunit is stabilized (Fig. 10a,c).

520
The molecular mechanism by which ATP frees the  subunit from the 44 ring is surprisingly elaborate. With RNR turned off, the ratio of ATP to dATP will increase, and ATP, despite its lower affinity for allosteric effector site 1 (Kd of ~120 M), will displace dATP (Kd of ~ 6 M) (Brown and Reichard, 1969b;Ormö and Sjöberg, 1990). Due to the extra hydroxyl group on the ribose of ATP, the ribose sits higher in allosteric effector site 1 to prevent a steric clash between the 2' hydroxyl 525 group and I22 (see Fig. 3F). This repositioning of the ribose, which has also been reported for human RNR (Fairman et al., 2011), breaks the hydrogen bond between H59 and the 3' hydroxyl group. The side chain of H59 tilts to one side.
The tilting of the H59 side chain starts a Rube Goldberg-like mechanism in which H59 movement 530 results in the movement of the side chain of F87, which in turn results in the movement of the side chain of W28. With the side chains of F87 and W28 repositioned, the second ATP can bind to previously unavailable site 2 (Fig. 10b). With six ATP phosphates now positioned around one Mg 2+ ion, the -hairpin tip is pushed up, and with the -hairpin lever tipped up, the strain on helix 2 residues is decreased and helix 2 re-winds, sealing away I47 and the binding pocket for the  535 subunit (Fig. 10c). Ensuring the departure of , helix 2 re-winding swaps hydrophobic residue I47 with a polar and potentially charged H46. Thus, the Rube Goldberg mechanism concludes with an exchangeable protein surface that alternatively attracts (I47-outward facing) and repels (H46outward facing) the  subunit. Consistent with the ability of an Ile-to-His exchange at a small interface (~525 Å 2 ) to disrupt that interface, is the previous observation (Chen et al., 2018) that 540 site-directed single substitutions of residues at this interface (e.g. L43Q, S39F, E42K) abolish ring formation.
It is quite impressive from a protein design prospective that a 95-residue domain has two "hideaway" binding sites, one for a nucleotide effector (site 2) and the other for a protein subunit, 545 and that these sites are in communication. The exposure of one binding site necessitates that the other is sealed. When ATP is bound and site 2 is exposed, the -subunit binding site is tucked away. Conversely, when dATP is bound and site 2 is tucked away, the -subunit binding site is exposed. Importantly, we were able to demonstrate that site 2 is relevant to allosteric activity regulation through a W28A substitution. Although dATP inhibits W28A RNR in the absence of any 550 ribonucleoside triphosphate, addition of ATP or GTP restores activity for W28A RNR under conditions that are inhibitory for the WT protein. In other words, when site 2 is open, the identity of the nucleotide in site 1 does not matter. Thus, like the H59A RNR variant (Rofougaran et al., 2008), W28A RNR does not discriminate between dATP and ATP. Although unexpected, the involvement of two ATP binding sites in the cone domain of E. coli class Ia RNR has a certain 555 structural and chemical logic to it; when ATP levels rise in response to RNR inactivity, that the binding of two ATP molecules per  subunit, rather than one ATP molecule, are required to shift the conformation equilibrium from 44 to 22.
It is too early to say whether human class Ia RNR will utilize one or two ATP molecules in its 560 allosteric regulatory mechanism. W28 is not conserved in human RNR and there is no evidence for the unwinding of helix 2 in the formation of the − interface of the dATP-inhibited 6 ring . D57 (equivalent of H59) appears to be responsible for signaling the presence of dATP versus ATP Caras and Martin, 1988;Reichard et al., 2000), but the molecular response that follows is unknown. Given that the interface involved in 565 formation of inactive rings is different in human RNR (−) than in E. coli (−), we do not expect the molecular mechanism to be the same. This difference in the nature of the inhibited states is exciting and has potential application for selective RNR inhibitor design. FDA-approved drug hydroxyurea and prodrugs gemcitabine and clofarabine either inhibit human RNR and not E. coli or both human and E. coli class Ia RNR. There are no FDA approved inhibitors that are specific 570 for bacterial RNRs.
Molecules that stabilize the inactive ring structures of RNR would be expected to be successful RNR inhibitors. Notably, clofarabine triphosphate, which is used in the treatment of pediatric acute leukemia (Pession et al., 2010), inhibits human RNR with the generation of "persistent hexamers" 575 (Aye et al., 2012) and does not inhibit the 44-forming E. coli class Ia RNR. Recently, we showed that the class Ia RNR from N. gonorrhea (NgRNR) forms 44 inactive rings that are analogous to the rings formed by the E. coli enzyme, and that compounds that inhibit NgRNR are not crossreactive with human RNR (Chen et al., 2018). Although there is no structure of NgRNR bound to these compounds, we do know that N. gonorrhea develops resistance to them when mutations occur in NgRNR at the − interface of the 44 ring, suggesting that these compounds target an inactive ring structure (Chen et al., 2018).
The work presented here should aid in the development of compounds that target, stabilize and thereby increase the lifetime of inactive 44 ring structures of bacterial RNRs. In particular, our 585 studies suggest that small molecules that prevent site 2 from opening, by blocking W28 from moving, for example, would stabilize the inactive 44 ring structure. The inactive ring structure should also be stabilized through maintenance of the -hairpin in the downward position or maintenance of the unwound conformation of helix 2. In contrast, compounds that stabilize an open site 2 would be expected to impede down-regulation by dATP and lead to persistently active 590 bacterial RNRs. We hope that the molecular information presented here will facilitate the development of new antibiotic compounds. Antibiotic resistance is an eminent threat (CDC, 2018; Willyard, 2017) and RNR is a largely unexplored target to address this threat.
For the structures of wild-type 2 bound to dATP and ATP, high-purity 100 mM solutions of ATP and dATP were purchased from USB Corporation or Invitrogen. A high-purity 100 mM solution of dGTP was purchased from USB Corporation for ultrafiltration assays.

610
Construct and protein preparation. Untagged 2 and 2 were prepared as previously described (Salowe et al., 1987;Salowe and Stubbe, 1986). The concentrations of 2 and 2 were determined using ε280 of 189 and 131 mM -1 cm -1 , respectively; unless noted otherwise, all molar concentrations correspond to the subunit dimer.
For His6-W28A-2, an estimated ε280 of 182 mM -1 cm -1 was used to determine the final protein concentration.

620
The 2-C35 fusion construct was made by sequential megaprimer mutagenesis (Xu et al., 2003) in which double-headed PCR primers were used to amplify a 144 bp insert encoding the C-terminal tail of  (the C-terminal 35 residues) from a plasmid encoding wild-type . These megaprimers were subsequently used to amplify the wild-type His6--encoding plasmid, while excising the final eight residues of the  C-terminal tail. The final construct contains an N-terminal His6-2 (Minnihan et al., 2011). As the resulting construct has no additional tryptophan or tyrosine residues, ε280 of 189 mM -1 cm -1 was used to determine the final protein concentration. Crystallization of W28A-2. Crystals of W28A-2 were identified in sparse matrix trays as described for wild-type 2 above. For screening and optimization of the ATP/CDP co-crystal complex, His6-W28A-2 at 60 µM in assay buffer (50 mM HEPES pH 7.6, 15 mM MgCl2, 1 mM EDTA) was pre-incubated with 10 mM ATP and 1 mM CDP for 20 min at ~25 ºC before mixing X-ray data collection. Diffraction data for wild-type 2-dATP and 2-ATP were collected at the Advanced Photon Source (APS) on beamline 24ID-C on a Quantum 315 CCD detector at 680 100 K. The W28A-2-(ATP)2/CDP, W28A-2-(dATP/ATP), W28A-2-(dATP/GTP), and wild-type 2-C35-dATP datasets were collected at APS beamline 24ID-C on a Pilatus 6M detector (Dectris) at 100 K. Diffraction data were indexed, integrated, and scaled using HKL2000 (Otwinowski and Minor, 1997), with statistics shown in Table S1.

Crystallization of wild-type
Structure solution and refinement. The method of structure solution is described 685 individually for each structure below. Model refinement statistics for wild-type 2 structures are found in Table S2 and model refinement statistics for W28A 2 structures are found in Table S3.
For all structures, multiple rounds of refinement were performed with phenix.refine (Adams et al., 2010) in the SBGRID software package (Morin et al., 2013). For all structures, refinement consisted of rigid body, positional, and individual B factor refinement. Translation-libration-screw 690 (TLS) B factor refinement was used for all structures except W28A-2-(ATP)2/CDP. Manual rebuilding and geometry correction was performed in Coot (Emsley et al., 2010). Simulated annealing composite omit maps calculated in Phenix were used to validate modeling of ligands in all structures. For structures with resolution of 3 Å or better, waters were placed automatically in Phenix with manual editing and placement in Coot (Emsley et al., 2010). Ligand restraint files 695 were obtained from the Grade Web Server (Global Phasing Ltd.). Coordination distances for Mg 2+ ions were explicitly defined at 2.1 Å with loose restraints. All structural figures were made in PyMOL v. 1.7 and v. 2.3.7 (Schrödinger, LLC). Refinement statistics for each final model are given in Table S2 (wild-type cone domain) and Table S3 (W28A mutants).
The 2-dATP co-crystal structure was solved by molecular replacement in Phaser (McCoy 700 et al., 2007) at 2.55-Å resolution using a single  monomer (chain A) from a previously solved structure of 2 (PDB ID 3R1R) (Eriksson et al., 1997) as the search model. The 2-ATP co-crystal structure was solved by molecular replacement in Phaser at 2.62-Å resolution using the refined 2-dATP co-crystal structure with all ligands and waters removed. Since the crystal form of these two structures is essentially identical, the cross-validation sets for the 2-ATP structure was 705 preserved from the 2-dATP structure. For both structures, the resolution was extended to the full range after partial model building and refinement at lower resolution. CNS 1.3 (Brünger et al., 1998) was used for early stages of refinement. ATP or dATP and Mg 2+ ions were placed into omit density prior to addition of water molecules. In both 2-dATP and 2-ATP models, two  chains are present in the asymmetric unit in a physiological dimer. Loose non-crystallographic symmetry 710 (NCS) restraints on both coordinate positions and B factors were used throughout refinement and then removed during the final rounds. Residues 5-736 (of 761) are present in each chain of both structures. 1-2 additional residues are visible in some chains at the N-terminus, but are poorly structured. The C-terminal 25 residues are thought to form a flexible tail essential for the rereduction of the active site disulfide upon turnover.

715
The W28A-2-(ATP)2/CDP structure was solved by molecular replacement in the Phenix implementation of Phaser (McCoy et al., 2007) with a 3.60-Å resolution cutoff. Due to the large changes in the overall conformation upon binding of the substrate CDP, the best search model was a single 2 dimer from the wild-type 44-dATP/CDP structure (PDB ID 5CNS) (Zimanyi et al., 2016). The initial Rfree for this molecular replacement solution was 0.41. The resolution was 720 extended to 2.60 Å after rigid body refinement and manual rebuilding of the model. Simulated annealing and real-space refinement were used early on until the model converged. Eight 2 dimers are present in the asymmetric unit, organized as two separate dimer-of-dimer units. A stable 4 oligomeric state has never been observed in solution for E. coli class Ia RNR, and analysis of the overall structure by the PISA server suggests that the only stable assembly is the 725 2 dimer and not 4. The overall structure of the  subunit's (/)10 barrel closely resembles that observed in the substrate/effector-bound 44 complex, despite the complete absence of 2 from this crystal form. The cone domain was partially deleted and rebuilt manually during refinement.
ATP and Mg 2+ ions were placed into omit density. NCS restraints were used throughout refinement. Composite omit maps were used to verify the final structure, especially to ensure model bias did not influence the rebuilding of the cone domain. The N-terminal His6 tag and thrombin cleavage site and residues 737-760 at the C-terminus are disordered in all chains.
Residues 645-652, which form a flexible -hairpin, are poorly ordered in four of the eight chains and have been omitted where there is no clear density.
The W28A-2-dATP/ATP and W28A-2-dATP/GTP structures were solved by molecular 735 replacement in Phaser (McCoy et al., 2007) with the wild-type 2-(ATP)2 structure as the model with no resolution cutoff. This crystal form contains lattice contacts that are similar to those observed in the wild-type 2-ATP and 2-dATP structures, but four molecules are present in the asymmetric unit instead of two, and the length of the b-axis is increased by 20 Å (a ~17% increase). The C-terminal tail of  (24 residues) is disordered in all of the chains of each structure 740 along with 3-4 residues at the N-terminus, and the N-terminal His6 tag and thrombin cleavage site.
There is no clear reason for the change in space group that is apparent from the crystal packing, but it is possible that the presence of the N-terminal His6 tag disrupts some crystal contacts in this crystal form, although the N-terminus is not ordered in either W28A or wild-type free 2 structures.
To prevent overfitting, strict NCS for both coordinate positions and B factors was maintained 745 throughout refinement. All ligands were removed prior to molecular replacement and were rebuilt based on omit maps. Due to the low resolution and similarity of the starting model, few structural changes were observed during refinement. dATP was placed in the specificity site based on omit maps. ATP/dATP and GTP/dATP were placed according to the positions of the ATP/ATP pair in the wild-type 2-(ATP)2 structure. To assess whether the structure was phase-biased by the 750 choice of 2-(ATP)2 as the molecular replacement model, 2-dATP was also tested. The molecular replacement solution was substantially worse in this case, but could be corrected with rigid body refinement of the hairpin and rotamer flips of W28 and F87. Composite omit maps were used to verify ligand placement, the conformation of F87, and the W28A mutation. The N-terminal His6 tag and thrombin cleavage site and residues 737-760 at the C-terminus are disordered in all 755 chains. All other residues are present in all four chains of both structures.
The 2-C35-dATP/CDP structure was solved to 2.10-Å resolution by molecular replacement in Phaser (McCoy et al., 2007) using an 2 structure that contained two bound peptides that mimicked the sequence of the β tail (PDB ID 1R1R) (Eriksson et al., 1997) with no resolution cutoff. The first 20 residues of 2 were not included in the search model. The initial Rfree 760 for this starting model was 0.29. The crystal form contains two molecules in the asymmetric unit.
Only one of the two cone domains is fully structured (beginning after the N-terminal His6 tag and thrombin cleavage site); the other cone domain has no density for residues 1-19 and 47-58 due to a close crystal contact. Residues 738-753 of the  tail and 342-364 ( numbering) of the attached  tail are disordered, but the remainder of the  tail, 365-376 ( numbering), renumbered 765 as 1365-1376, is bound as previously observed for both the  peptide in the 2 structure and for the  tail in the 44 complex (Eriksson et al., 1997;Zimanyi et al., 2016). No NCS restraints were used as there were substantial differences in the cone domains in the two protomers. dATP was modeled into the final structure based on omit maps. At both specificity and activity sites, a Mg 2+ ion coordinates three phosphate oxygens of dATP and three water molecules. Composite omit 770 maps generated in Phenix (Adams et al., 2010) were used to verify the cone domain conformations and effector ligands.
Determination of equilibrium ATP binding parameters by ultrafiltration. The ultrafiltration method described by Ormö and Sjöberg (Ormö and Sjöberg, 1990)  wash, the sample consisting of 7-20 μM 2 (in assay buffer with 5 mM DTT) and 50-1000 μM 3 H-ATP with a specific activity of either 690 cpm nmol -1 or 3090 cpm nmol -1 in a total volume of 150 μL was added to the filter. The solution was equilibrated in a 25 °C water bath for 5 min. and a 25 μL aliquot was taken for determination of total nucleotide concentration. The sample was then centrifuged at 12,000 × g for 1 min. 25 μL of the filtrate was then taken to determine free 785 nucleotide concentration by scintillation counting. To isolate the binding events at the activity site, 100 or 500 μM dGTP was included in the sample before equilibration. The amount of bound nucleotide was found by subtracting the amount of free nucleotide from the total nucleotide. Data were plotted as a saturation binding curve and analyzed using non-linear regression and a onesite specific binding model in the program Prism (GraphPad).

795
Standard active and inactive conditions contained either 3 mM ATP or 175 µM dATP as allosteric effectors, respectively. Concentrations of nucleotides used in titration experiments are given in the Fig. 8 legend. Linear fitting of the initial rates was performed in the Cary WinUV Kinetics program (Varian/Agilent) and data were plotted in MATLAB (Mathworks).

940
(A) Substrate specificity is regulated allosterically via the binding of deoxynucleotides. Under low concentrations of dATP, CDP and UDP reduction is favored. TTP in turn leads to GDP reduction and dGTP leads to ADP reduction. When the concentration of dATP increases, it can bind to a second allosteric site, the activity site, which inhibits enzymatic activity. ATP can compete for binding at this activity site to restore activity. (B) Enzymatic turnover in class Ia RNR involves a series of redox-active cysteine pairs. The 945 reducing equivalents for nucleotide reduction are initially provided by a pair of cysteines (C225/462) in the active site that is oxidized to form a disulfide concomitant with product formation. This disulfide is reduced by a second pair of redox-active cysteines (C754/759) found at the C-terminus of the 2 subunit. Ultimately, the disulfide between C754 and C759 is reduced via the thioredoxin/thioredoxin reductase pair together with NADPH, thus allowing for additional rounds of turnover.