Evolution of cation binding in the active sites of P-loop nucleoside triphosphatases

The activity of cellular nucleoside triphosphatases (NTPases) must be tightly controlled to prevent spontaneous ATP hydrolysis leading to cell death. While most P-loop NTPases require activation by arginine or lysine fingers, some of the apparently ancestral ones are, instead, activated by potassium ions, but not by sodium ions. We combined comparative structure analysis of P-loop NTPases of various classes with molecular dynamics (MD) simulations of Mg-ATP complexes in water and in the presence of potassium, sodium, or ammonium ions. In all analyzed structures, the conserved P-loop motif keeps the triphosphate chains of enzyme-bound NTPs in an extended, catalytically prone conformation, similar to that attained by ATP in water in the presence of potassium or ammonium ions bound between alpha- and gamma-phosphate groups. The smaller sodium ions could not reach both alpha- and gamma-phosphates of a protein-bound extended phosphate chain and therefore are unable to activate most potassium-dependent P-loop NTPases.

3 Introduction crystal structures of P-loop NTPases with bound NTPs and their analogs (11,12,(60)(61)(62), see also 126 Fig. 1. The initial structure of the Mg-ATP complex was optimized in vacuum using the PM3 127 Hamiltonian. After that, 1,200 water molecules and 6 monovalent cations (K + , Na + or NH 4 + ) were 2.2 Å for Na + , 2.6 Å for K + , and 2.7 Å for NH 4 +  binding to the phosphate groups, at least two distinct binding sites for M + ions could be identified 145 ( Fig. 2A). One of them was formed by the oxygen atoms of β-and γ-phosphates, and the other site 146 involved the oxygens of α-and γ-phosphates. We refer to these binding sites as the BG and AG 147 sites, respectively. Additionally, M + ions were often found close to the distal end of the phosphate 148 chain, where they contacted one or more oxygen atoms of the γ-phosphate (the G site(s), Fig. 2A). 149 To characterize M + binding in the AG and BG sites, we measured the distances from each M + ion to 150 the nearest oxygen atoms of the two respective phosphate residues (R AG and R BG distances in Fig.   151 2B). Site occupancy was estimated, as shown in Fig. 2C-E, from the number of M + ions located in 152 the proximity of the binding site at each moment of the simulation. In the BG site, binding of any 153 M + ion produced a prominent maximum in the R BG distribution. The R BG values peaked at the same 8 distance as the maxima of the distribution of distances to separate oxygens (Fig. S1), which 155 indicates that the cations in the BG site simultaneously formed coordination bonds with two oxygen 156 atoms. Similarly, in the AG site, the NH 4 + and Na + ions produced peaks in the R AG distribution plots 157 with the maxima at 2.7 Å and 2.3 Å, respectively. For K + ions, the corresponding peak with a R AG 158 value of 2.6 Å was wide. Still, the distributions of the distances between cations and individual 159 oxygen atoms of the triphosphate chain show that α-and γ-phosphates had the most contacts with 160 K + ions, see graphs for O 2A and O 1G in Fig. S1 (hereafter, the atom names follow the CHARMM 161 naming scheme (66) and the recent IUPAC Recommendations (67), as shown in Fig. 1D and Fig. 162 S1). 163 While occupying the same binding sites, M + ions bound with different affinity that decreased in the 164 order of Na + > NH 4 + > K + (Table S2). Higher affinity of ATP to Na + ions, as compared to K + and 165 NH 4 + ions, was previously observed in several experimental studies, albeit in the absence of Mg 2+ 166 (Table S2). For each M + ion, MD simulation data indicated much lower occupancy of the AG site 167 than of the BG site; the average occupancy of the BG site was estimated to be 0.95 for Na + , 0.72 for 168 NH 4 + , and 0.5 for K + , compared to the average occupancy of the AG site of 0.15 for Na + , 0.2 for 169 NH 4 + , and 0.05 for K + (Fig. 2C-E). 170 The reasons for the weak K + -binding in the AG site could be, in principle, clarified by structural and 171 thermodynamic analysis of the conformations of the Mg-ATP complex with two K + ions bound. 172 Such an analysis, however, was hindered by the scarcity of the respective MD simulation frames. 173 Therefore, we have conducted additional MD simulations with positional restraints applied to the 174 cations. We have conducted 10-ns simulations of an ATP molecule with Mg 2+ in the βγ coordination 175 and K + in the BG site, and of the same system but with the addition of the second K + ion in the AG 176 site. Positional restrains were applied to K + and Mg 2+ ions and to one of the atoms of the adenine 177 base. Binding of the second K + ion in the AG site was found to stabilize all three phosphate groups 9 in a near-eclipsed conformation, with the phosphorus-oxygen bonds of the α-phosphate group 179 almost coplanar to the respective bonds of β-and γ-phosphates (Fig. S3, Table S4). In this 180 conformation, the distance between the oxygen atoms of α-and γ-phosphates was short enough to 181 accommodate the second K + ion. As shown in Fig. S3, binding of the second K + ion in the AG site 182 promotes the transition of the phosphate chain into the almost fully eclipsed conformation by 183 approximately 60 meV or 5.7 kJ/mol. 184 We were mostly interested in the βγ conformations of the Mg-ATP complex that are typical for P- 185 loop NTPases. To sample enough βγ conformations, we have conducted an additional series of 25 186 independent 20-ns long MD simulations, with and without M + ions (Table 1). These data were used 187 to define the shape of the phosphate chain of the βγ-coordinated Mg-ATP complex.
188 189 190 The MD simulation data were used to compare the geometry of the ATP phosphate chain in the 191 presence and in the absence of different M + ions. 192 Cleavage of the bond between β-and γ-phosphates is believed to proceed via a planar transition 193 complex, whereby the P B -O 3B -P G angle widens (41, 44, 51-53, 68-70). Another important feature of 194 the Mg-ATP complex is the curvature of the phosphate chain, which can be characterized by the P A -195 P G distance (Fig. 1D). 196 In Fig. 3, values of the P B -O 3B -P G angle and P A -P G distance are plotted as a function of the 197 simulation time. conformation could be seen, differing in the particular oxygen atoms of the phosphate chain 209 involved in the tridentate coordination of the Mg 2+ ion (Fig. S4). 210 In each of the sampled conformations, the Mg-ATP complex was characterized by distinct P A -P G 211 distances and P B -O 3B -P G angles, which depended on the nature of the added monovalent cation (Fig. 212 3, Table 1). While all M + ions seemed to contract the phosphate chain, it was more extended in the 213 presence of K + than in the presence of NH 4 + or Na + . Furthermore, Na + and NH 4 + ions could induce 214 an even more compressed, curled conformation of the Mg-ATP complex with even shorter distances 215 between P A and P G atoms. Such curled conformations of the phosphate chain were not observed 216 either in the presence of K + ions or in the absence of M + ions (Fig. 3, Table 1). 217 In short MD simulations that started from the same βγ conformation (simulations 5-8 in Table S3), 218 we did not observe significant differences in the lifetime of the βγ conformation between systems 219 with different cations (Table S5). For the βγ conformation of the Mg-ATP complex, the largest P A -220 P G distances, up to 5.5 Å, were observed in simulations without M + ions (Fig. 3, 4). Presence of M + 221 11 ions in the simulation system led to a significant decrease of the P A -P G distances (Fig. 3, 4, Table 1). 222 Among the studied cations, K + ions allowed for the longest P A -P G distances. The P B -O 3B -P G angles 223 in the βγ-coordinated Mg-ATP complexes did not differ significantly between simulations with 224 different cations or without cations added (Fig. 3, 4, Table 1). 225 12 228 Binding in the catalytic site of a P-loop NTPase imposes constraints on the Mg-NTP complex, so 229 that only particular conformations of the phosphate chain are allowed. These conformations appear 230 to be catalytically prone, since NTP binding to an inactive P-loop domain ( To characterize the conformations of the phosphate chain in the active sites of P-loop proteins, we 247 used the same parameters as for the MD simulation data, namely the P A -P G distance (or the 248 corresponding distances in substrate analogs) and the value of the P B -O 3B -P G angle (or the 13 corresponding angles in substrate analogs). Using these two parameters as coordinates, we mapped 250 the conformations attained by NTP-like molecules in the crystal structures (separately shown and 251 described in Fig. S5) on the heat maps for all four systems, calculated from MD simulations (Fig. 4). 252 In the top row (Fig. 4A) In MD simulations in the presence of K + and NH 4 + ions, the distribution of the conformations of 14 transition state analogs (Fig. 4B). Only the conformations of the transition state analogs with 274 severely widened (>135°) P B -O 3B -P G angle were not matched by the MD-derived conformations. 275 Altogether, Fig. 4 shows that the conformational space of phosphate chain conformations, as seen in 276 P-loop NTPases, overlapped much better with conformations seen in the MD simulations of Mg-277 ATP with K + and NH 4 + ions than with conformations obtained with Na + ions. site between the β-and γ-phosphates (the BG site) is always occupied by the amino group of the 290 conserved P-loop lysine residue, whereas the binding site between the α-and γ-phosphates (the AG 291 site) could be occupied, in the crystal structures, by either a K + or Na + ion (Fig. 5B), or an amino 292 group of an activating lysine residue, or the guanidinium group of arginine (Fig. 5C) analogs and K + , Na + , or NH 4 + ions bound in the active site (Table 3). For each such structure, we 312 checked the shape of the phosphate chain and the coordination sphere of the cation in the AG site. 313 In all these structures, the distances between P A and P G atoms (or between the corresponding 314 mimicking atoms) were in the range of 4.9-5.3 Å for the non-hydrolyzable analogs and 5.3-5.6 Å for 315 the transition state analogs (Table 3). These values are similar to the P A -P G distances observed in 316 MD simulations of the Mg-ATP complex in the presence of K + ions (Fig. 3, 4 and Table 1). 317 The majority of K + -activated NTPases, as well as the unique family of the Na + -adapted dynamin-318 related GTPases, belong to the TRAFAC class of P-loop NTPases (2), where the binding of the M + 16 ion is assisted by the so-called K-loop (20). This loop goes over the nucleotide binding site and 320 provides two backbone carbonyl groups as additional ligands to the M + coordination sphere (purple 321 cartoon and sticks in Fig. 1B,C). To our surprise, very few structures of K + -dependent GTPases of 322 the TRAFAC class contained K + ions in their AG sites (cf Table S1 and Table 3). Furthermore, in 323 most cases, the K + loops were either unresolved or distorted (Fig. S6). Separate crystal structures 324 with and without activating K + ion were available only for the tRNA modification GTPase MnmE 325 see Table 3 and  (Table 3, Fig. 1C, Fig. 6). These structures contain fully resolved K-loops, which allowed 17 us to compare the structures of K + -dependent and Na + -adapted P-loop NTPases with the results of 344 our MD simulations. In MD simulations, presence of Na + ions led to contracted phosphate chain 345 conformations (Fig. 3, 6A), whereas crystal structures of dynamins showed extended conformations 346 of the phosphate chain even with a Na + ion bound (Fig. 6B). In dynamin-like proteins, as in other P-347 loop NTPases, the phosphate chain is in the catalytically prone extended conformation owing to its 348 stabilization by the residues of the P-loop, so that the Na + ion interacts with the γ-phosphate but 349 cannot reach the oxygen atom of the α-phosphate (Table 3, Fig. 6B). The ability of dynamins to 350 keep the Na + ion in the AG position appears to be due to the changes in the K-loop and its The stabilization of an NTP molecule at the P-loop in an extended conformation dramatically 387 increases the rate of hydrolysis even in the absence of an activating moiety. In Ras-like GTPases, 388 binding of GTP to the P-loop accelerates the rate of hydrolysis by five orders of magnitude (74, 75). 389 Delbaere and coauthors noted that, in a bound NTP molecule, the β-and γ-phosphates are in an 19 eclipsed state owing to the interaction with the Mg 2+ ion and conserved Lys residue of the P-loop. In 391 this state, β-and γ-phosphates repel each other, which could explain the higher hydrolysis rate (85, 392 86). Hence, P-loop-bound conformations of the phosphate chains ( Fig. 5) are catalytically prone. 393 Here, we showed that monovalent cations occupy specific well-defined sites (AG and BG sites, Fig.   394 2A) in the vicinity of the triphosphate chain even in the absence of enzymes. Fig. S3 shows that 395 binding of the second K + ion in the AG site can bring the phosphate chain into the fully eclipsed 396 conformation, which has been previously suggested to be particularly catalytically productive (43). 397 These data could explain why larger ions, such as K + and Rb + , were shown to be more efficient than 398 the smaller Na + and Li + ions in accelerating transphosphorylation even in the absence of enzymes 399 (50), see Table S2. contract the phosphate chain that is fixed by the P-loop, which explains why Na + ions are not 414 competent in most P-loop NTPases. 415 The affinity of the AG site to K + ion is intrinsically low (Table S2, (20). 419 In the case of dynamins, the ability to bind either a Na + ion or a K + ion in the AG site was earlier 420 traced to several mutations (20). Specifically, in dynamins, (i) the conserved Asn in the P-loop is The observed absence of K + ions from most structures of K + -dependent P-loop NTPases (Fig. S6) 430 could be due to several reasons, including their absence from the crystallization medium. For 431 example, in one of the structures of the K + -dependent GTPase Era, which was crystallized in the 432 absence of K + ions (PDB: 3R9W, (87)), the potential K + binding site contains a water molecule (id 433 624) that is 2.9-3.4 Å away from six potential K + ion ligands. Owing to the presence of a full-434 fledged K + -binding site, we included this structure in Table 3 (see also Fig. S9). Even when K + ions 435 were present in the crystallization medium, the electron density difference between the K + ion (18 21 electrons) and the water molecule (10 electrons) is often insufficient to easily distinguish their 437 relative contributions to the diffraction pattern (37). Thus, at 60% occupancy, the K + ion cannot be 438 distinguished from a water molecule (88). However, in most crystal structures of K + -dependent 439 GTPases (Table S1), not only the M + ion is absent, but the entire K-loop is either unresolved or crucial for the cation binding, since this loop provides two backbone oxygen atoms as ligands for the 22 cation. We believe that the same mechanism could be involved in the activation of other K + -461 dependent NTPases (Table 2), whereby the proper conformation of the K-loop and functionally 462 relevant K + binding could be promoted by interaction with the activating protein or RNA. 463 In RecA-like recombinases (Fig. S8), the K + ion in the AG site is coordinated by a conserved Asp 464 residue, which is responsible for the K + -dependent activation (89). This residue (Asp302 in PDB: 465 2F1H) is provided by the adjacent monomer within the RadA homooligomer that assembles upon 466 interaction of RecA proteins with double-stranded DNA. Thus, in RecA-like recombinases, the K + -467 binding sites differ from those in K + (or Na + )-dependent TRAFAC NTPases, but, similarly to 468 TRAFAC NTPases, appear to attain functionality upon the interaction with the activating partner 469 that provides ligands for the K + ion. 470 In conclusion, in P-loop NTPases, the activating amino groups of Arg/Lys residues or K + ions 471 occupy the AG sites similarly to the K + and NH 4 + ions seen in MD simulations of Mg-ATP in water. 472 In addition, the very formation of the M + -binding site next to the P-loop appears to require to speculate that the P-loop could have been shaped in K + -and/or NH 4 + -rich, but Na + -poor 493 environments, which would favor the extended conformations of unbound (free) NTPs. Indeed, the 494 smallest ion is this study -Na + -is known to exhibit the strongest binding to the phosphate chain, 495 which has been reproduced in our MD simulations (Table S2,   When an NTP molecule is bound to a P-loop NTPase, the catalytically-prone extended conformation 503 of its phosphate chain is fully determined by the interactions with the residues of the P-loop itself. 504 An extended phosphate chain could bind (activating) K + /NH 4 + ions or amino groups of Lys/Arg in 505 its AG site, but is too stretched to bind Na + ions. As argued by Lupas and colleagues, one of the 506 possible mechanisms for the emergence of diverse classes of P-loop NTPases could be a 507 24 combination of the same "original" NTP-binding P-loop domain with different partners that could 508 promote the insertion of an activating moiety into the active site (4). This suggests that K + ions 509 and/or amino groups were available as activating cofactors during the emergence of P-loop 510 NTPases. Hence, the P-loop motif itself may have been shaped by the high levels of K + and/or NH 4 + 511 ions in the habitats of the first cells. Since the emergence of the P-loop motif happened at the very 512 beginning of life, when the ion-tight membranes were unlikely to be present, the match between the 513 shape of the P-loop and large cations of K + and NH 4 + is consistent with our earlier suggestions on 514 the emergence of life in terrestrial environments rich in K + and nitrogenous compounds (38,39). 515 The activating Arg/Lys residues are usually provided upon interactions of the P-loop with another 516 domain of the same protein, or an adjacent monomer in a dimer or an oligomer, or a specific 517 activating protein, or DNA/RNA (  of K + ions to the "naked" AG site of the ATP molecule (Fig. 2C, S1 -S4). This poor K + binding 526 manifests itself also in the need to use very high (>>100 mM) levels of potassium salts to activate 527 the K + -dependent P-loop NTPases in the absence of their physiological activating proteins or RNA 528 (33,37). As our comparative structure analysis showed, the functional K-loop in such NTPases is 529 distorted in the inactive (apo-) state (Fig. S6), but attains its functional shape and eventually binds 530 the cation upon the interaction with the activating partner ( Fig. S7-S9). The interaction with the 25 activator, however, must be highly specific to prevent the activation of hydrolysis in response to an 532 occasional binding to a non-physiological partner. It indeed seems to be specific; Table 3 lists   533 structures of the eukaryotic translation initiation factor eIF5B in which a kind of a K-loop formed 534 not via their functional interaction with the ribosome, but through non-physiological crystal-packing 535 contacts (37). Although these quasi-K-loops bind different monovalent cations, the corresponding 536 structures contain GTP molecules, indicating the absence of hydrolytic activity. In addition, the 537 respective P A -P G distances are shorter than those in the structures of P-loop NTPases in their active 538 conformations ( In spite of the long evolution of P-loop-NTPases, only in a single known case, in eukaryotic 542 dynamins, the enzyme can be activated both by K + and Na + ions (20,48,49). The adaptation to Na + 543 ions required at least 3 mutational changes in the highly conserved parts of the protein, see (20) and 544 Fig. 6. The low probability of this combination of changes may explain why just this one case of 545 Na + -adaptation is known. In contrast, Arg/Lys residues are widespread as activators of P-loop 546 NTPases, see Table 2 and Shalaeva et al., submitted). In a few cases (e.g. in TRAFAC class 547 NTPases) it was possible to trace how Arg residues replaced K + ions in the course of evolution in 548 different lineages (38,39). The recruitment of an Arg/Lys residue as an activating moiety is 549 relatively simple and makes the catalysis independent of the oscillations of K + and Na + levels in the 550 cell. βγ coordination of the Mg 2+ ion, which is typical for the P-loop NTPases, but also on their 556 interaction with tridentate αβγ-coordinated Mg-ATP complexes (Fig. 3, 4A, S4, Table 1). 557 The tridentate αβγ-coordination is found, for instance, in K + -dependent chaperonin GroEL and 558 related proteins. Unlike P-loop NTPases, the GroEL from E. coli and the related chaperonin Mm-559 cpn from Methanococcus maripaludis were inhibited by Na + ions even when Na + was added over K + 560 (97). In the crystal structures of GroEL, K + ion was identified in the position that corresponded to 561 the AG site of our MD simulations, cf the right structure in Fig. 3B   bound to at least one Lys (indicating that the nucleotide is indeed bound to the P-loop and the P-loop 29 Lys residue is not mutated). Custom MatLab scripts were also used to measure the shape of the 622 phosphate chain in each NTP-like substrate or the transition state-mimicking molecule. 623 To characterize M + -binding sites of P-loop proteins, we have searched the available literature data 624 for cation-dependent activities of the respective proteins, with the results summarized in Table S1.     Mean values and standard deviations of P A -P G distance (in Å) and the P B -O 3B -P G angle (in degrees) 978 were measured over the respective parts of the simulations. Simulation periods corresponding to βγ 979 and αβγ conformations were identified by tracking distances between Mg 2+ and non-bridging 980 oxygen atoms of the phosphate chain (Fig. S3); simulation periods corresponding to the "curled" 981 conformation were identified from P A -P G distance tracks and visual inspection of the phosphate 982 chain shape (Fig. 3). Data for the αβγ coordination of the Mg 2+ -ATP complex and conformations 983 with curled phosphate chain were calculated from simulations 1-4 in Table S3; characterization of 984 the βγ-coordination was based on simulations 5-8 in Table S3, see Table S5 for further details.    shown on the right. The analysis was performed as in Fig. 2B. The color scheme is as in Fig. 1 Table S3). B. Data from 1056 4x20 ns simulations of Mg-ATP in βγ conformations (no. 5-8 in Table S3, Table S5).  Table S3.

Shape of the phosphate chain in the structures of P-loop NTPases
Dihedral angle is an angle between two planes that is defined by four atoms. Values of dihedral angles between phosphate groups were defined as follows: Ψ α-β = ∠O 2A -P A -P B -O 2B ; Ψ β-γ = ∠O 1B -P B -P G -O 1G ; and Ψ α-γ = ∠O 1A -P A -P G -O 3G , see also Fig. S3. During the analysis of MD simulation data, the average and standard deviation values for dihedral angles were obtained by fitting the angle distribution histograms with normal functions, using the MatLab function "fit". All distributions were fitted with one-term Gaussian models, except for the Ψ β-γ angle in case of the Mg-ATP with two K + bound; this distribution was fitted with a two-term Gaussian, and parameters are shown for the highest peak. Distribution histograms and fitted curves are shown in Figure S3.
* The rotation of α-phosphate is unrestricted and the corresponding dihedral angles can take any values between -180° and 180°. 20 20 20 For each system, 25 independent 20-ns MD simulation runs were conducted, each starting with the Mg-ATP complex in the βγ conformation. Stability of the βγ conformation was tracked by measuring the distance from the Mg 2+ ion to the nearest oxygen atom of α-phosphate, and the time periods during which the βγ conformation was retained were compared between different systems. The one-way ANOVA analysis did not reveal any significant dependence of the stability of the βγ-coordination on the monovalent cation present. For each monovalent cation, the βγ-coordination was retained during the whole 20 ns in at least four cases (shown by bold numbers). These simulations were used to characterize the shape of the phosphate chain of ATP with βγ-coordination of the Mg 2+ ion (Table 1 and Figure 5B). were the same 2.7 Å for K + and NH 4 + ions, while for Na + this distance was 2.2 Å. For the NH 4 + ion, the distance was measured from each oxygen atom to the nitrogen atom of NH 4 + . There are two ester bond oxygens in the phosphate chain, but only the oxygen (O 3B ) that connects βand γphosphates was seen involved in the cation binding, it interacted more often with K + and Na + than with NH 4 + . Monovalent cations were found near oxygen atoms of γ-phosphate more often than near oxygens of β-and α-phosphates. B. Free energy of the cation binding as a function of the distance from the phosphate chain, as estimated from the probability data in Fig. S2A. In addition to the two binding sites at the distances of approx. 4 Å and 6 Å, the free energy plot revealed a less pronounced third binding site at a distance of approx. 8-9 Å from the phosphorus atoms. The most prominent is the first peak, corresponding to cation binding around the phosphate chain, within the 4 Å distance of at least one of the phosphorus atoms, so further focus was specifically on cation binding around the phosphate chain. Gaussian models, except for the Ψ β-γ angle in case of Mg-ATP with two cations bound, this distribution was fitted with a two-term Gaussian, parameters for the highest peak are shown.

D.
Coupling between cation binding in the AG site and rotation of γ-phosphate relative to αand β-phosphates. Data from 10-ns MD simulations with restraints on the positions of K + ions (see the text).The top graph shows free energy calculated from normalized probabilities of ATP conformations and plotted as function of the dihedral angle between γ-and β-phosphates. The bottom plot displays free energy of coupling the binding of the second K + ion with the γphosphate rotation, calculated as the difference between the free energy plots shown on the top graph. The lowest energy value was set to zero. These plots show that the presence of second K + ion in the AG site induces a near-eclipsed state of the phosphate chain, by bringing both Ψ α-β and Ψ α-γ angles close to 0°, at the expense of Ψ β-γ , which increases slightly (see Table S4). Binding of the second K + ion in the AG site stabilizes this almost eclipsed state by ~60 meV.  conformations that correspond to different steps of hydrolysis, to the conformations that correspond to the separated reaction products. In the latter case, the distance between P A and 'P G ' exceeds 5.5Å, indicating complete separation of the γ-phosphate-mimicking group from ADP/GDP; whereas the "P B -O 3B -P G " angle decreases due to the displacement of the former ester oxygen atom that becomes the free terminal oxygen of β-phosphate. For comparison with the MD data, we only considered the major cluster of true transition state-like conformations. Figure S6. Active sites of P-loop NTPases with established K + -dependent activity (see Table   S1 for the full list and references). Each of the proteins shown has both Asn residues that were shown to be associated with binding of monovalent cations in related proteins (34). Switch I, including the K-loop, and its flanking regions are shown in magenta, switch II motif DxxG is shown in orange. NTP-like molecules are shown as sticks, Mg 2+ ions are shown as green spheres, water molecules in the area of supposed cation binding are shown as red spheres.

Statistical Analysis
To analyze conformations of Mg-ATP complexes in the presence of different cations, we selected MD simulation fragments of similar length with the same type of interaction between the Mg 2+ ion and the triphosphate chain. In each case ~160 ns of MD simulation were taken to characterize a particular Mg-ATP conformation; if needed, results of several independent simulations were merged to collect enough data, see Fig. S10A, B and S11A, B for examples.
For the MD simulation data, we calculated autocorrelation functions ( Fig. S10C and S11C).
Given the correlation times obtained, independent frames were extracted to calculate characteristic values for the separate conformations of ATP. For the systems without additional monovalent cations, every N-th frame was taken for the calculation, with N defined by the correlation time. For the systems with monovalent cations, only frames in which at least 1 monovalent cation was bound to the phosphate chain were taken, with at least N frames between measurements. A monovalent cation was considered to be bound when it was within a binding distance from at least one oxygen atom of the phosphate chain, with binding distances defines a s follows: 2.4 Å for Na + and 3.2Å for K + and NH 4 + .    The null hypothesis was that the Pα-Pγ distances in ATP are the same in the αβγ-coordinated and "curled" βγ-coordinated Mg-ATP complexes, respectively.
* The null hypothesis is NOT rejected, no significant difference between samples  The null hypothesis was that Pα-Pγ distances are similar for the αβγ-coordinated and βγcoordinated, "curled" Mg-ATP complexes.