Nucleobase carbonyl groups are poor Mg²⁺ inner-sphere binders but excellent monovalent ion binders—a critical PDB survey

Abstract

Precise knowledge of Mg²⁺ inner-sphere binding site properties is vital for understanding the structure and function of nucleic acid systems. Unfortunately, the PDB, which represents the main source of Mg²⁺ binding sites, contains a substantial number of assignment issues that blur our understanding of the functions of these ions. Here, following a previous study devoted to Mg²⁺ binding to nucleobase nitrogens, we surveyed nucleic acid X-ray structures from the PDB with resolutions ≤2.9 Å to classify the Mg²⁺ inner-sphere binding patterns to nucleotide carbonyl, ribose hydroxyl, cyclic ether, and phosphodiester oxygen atoms. From this classification, we derived a set of “prior-knowledge” nucleobase Mg²⁺ binding sites. We report that crystallographic examples of trustworthy nucleobase Mg²⁺ binding sites are fewer than expected since many of those are associated with misidentified Na⁺ or K⁺. We also emphasize that binding of Na⁺ and K⁺ to nucleic acids is much more frequent than anticipated. Overall, we provide evidence derived from X-ray structures that nucleobases are poor inner-sphere binders for Mg²⁺ but good binders for monovalent ions. Based on strict stereochemical criteria, we propose an extended set of guidelines designed to help in the assignment and validation of ions directly contacting nucleobase and ribose atoms. These guidelines should help in the interpretation of X-ray and cryo-EM solvent density maps. When borderline Mg²⁺ stereochemistry is observed, alternative placement of Na⁺, K⁺, or Ca²⁺ must be considered. We also critically examine the use of lanthanides (Yb³⁺, Tb³⁺) as Mg²⁺ substitutes in crystallography experiments.

INTRODUCTION

As established during recent decades, Mg²⁺ is of high relevance to the molecular ecosystem regulating nucleic acids folding, architecture, and function (Cate et al. 1997; Draper 2004, 2013; Klein et al. 2004; Woodson 2005; Freisinger and Sigel 2007; Auffinger et al. 2011; Bowman et al. 2012; Erat et al. 2012; Sigel and Sigel 2013; Marcia and Pyle 2014; Nierhaus 2014; Zhou et al. 2017). Still, for an exact understanding of the roles played by Mg²⁺, a precise structural knowledge of its binding modes is required. This knowledge is typically derived from solution and crystallographic experiments. However, solution studies are unable to locate Mg²⁺ with atomic precision and poorly distinguish between Mg²⁺ inner- and outer-sphere binding that are both important to nucleic acid structure and function (Draper 2004). Thus, crystallographic structures deposited to the PDB (Berman et al. 2016) remain the main source of information regarding Mg²⁺ binding modes. Nonetheless, correctly assigning an ion to an experimental electron density pattern is notoriously difficult and, unfortunately, the PDB embeds a significant number of well- and not so well-documented assignment errors (Williams 2005; Wlodawer et al. 2008, 2013, 2018; Kleywegt 2009; Cooper et al. 2011; Joosten et al. 2012; Pozharski et al. 2013; Dauter et al. 2014; Echols et al. 2014; Jain et al. 2015; Weichenberger et al. 2015; Minor et al. 2016; Raczynska et al. 2016; Rupp 2016; van Beusekom et al. 2016; Richardson et al. 2018). Incorrectly identified ions considerably bias database analysis results. As such, some efforts to correct these issues have been previously described (Zheng et al. 2015). The authors of this PDB survey (September 2014) established that only ≈15% of the ≈100,000 Mg²⁺ binding sites identified in RNA crystallographic structures should be considered as trustworthy. Later, it was recognized that a significant portion of the remaining Mg²⁺ binding sites does not satisfy the strict stereochemical criteria associated with Mg²⁺ (see below). Hence, it has been suggested that Mg²⁺ binding sites should be reexamined in the light of revised validation checklists (Leonarski et al. 2017).

Here, we pursue efforts to assess the reliability of Mg²⁺ assignments in crystal structures of RNA, DNA, and nucleobase-containing metabolites with resolutions ≤2.9 Å by examining the binding of Mg²⁺ to nucleobase and ribose oxygen atoms (Leonarski et al. 2017). This study aims to establish a “prior knowledge” data set of Mg²⁺ binding modes to be used to validate existing ion attributions in crystallographic but also NMR and cryo-EM structures, and to “limit” future solvent density pattern misinterpretations. This has been done by enforcing stereochemical criteria derived from the “almost” invariable Mg²⁺ octahedral coordination geometry (Chen et al. 2015). Indeed, even if some binding sites could at first glance seem well suited for Mg²⁺, the absence of trustworthy structural references in PDB structures of appropriate resolution should arouse reasonable doubts regarding their legitimacy and make one wonder if these binding sites would not better accommodate monovalent ions (Na⁺, K⁺) or transition metals (Leonarski et al. 2017) as emphasized by some nonambiguous examples presented in this study.

Throughout, we use the Mg²⁺ binding site nomenclature described in Zheng et al. (2015). O2/O4/O6, N1/N3/N7, O2′/O3′/O4′/O5′, OP1/2 atoms are respectively labeled O_b, N_b, O_r, and O_ph atoms; their combination (O_b.N_b, 2O_b or cis-2O_ph.O_b, …) leads to the naming of binding sites. The direct binding of Mg²⁺ to oxygen atoms of phosphate groups that represent the primary nucleic acid binding locations (Klein et al. 2004; Sigel and Sigel 2010; Zheng et al. 2015) will be covered elsewhere.

RESULTS AND DISCUSSION

PDB overview of direct Mg²⁺ to carbonyl oxygen atom (O_b) contacts

Here, we investigate the potential of Mg²⁺ to establish direct (inner-sphere) contacts to nucleobase O2/O4/O6 (O_b) oxygen atoms of carbonyl groups. In a set of ≈5250 nucleic acid structures with resolution ≤2.9 Å, we identified ≈64,500 Mg²⁺ ions with a 1.0 occupancy and B-factors in the 1–79 Ų range. Out of those, 9325 (≈14.5%) and 664 (≈1%) establish at least one contact with a d(Mg²⁺…O_b) ≤ 3.5 Å and a d(Mg²⁺…O_b) ≤ 2.3 Å coordination distance, respectively. The largest part of these Mg²⁺ is found in RNA and only a few were assigned to DNA (Table 1).

TABLE 1.

Number of nonredundant Mg²⁺/Na⁺…O_b contacts in PDB structures (resolution ≤2.9 Å)

In RNA, most of the 658 Mg²⁺ with d(Mg²⁺…O_b) ≤ 2.3 Å bind to (G)O6 and (U)O4 atoms and only a small proportion to (C)O2 and (U/T)O2 atoms (70%, 25%, 4%, and 1% contacts, respectively). With this distance criterion, only ≈2% of these Mg²⁺ establish more than one contact to O_b atoms. Note that we categorized ≈40% of them as redundant (see Materials and Methods) leaving a relatively small sample of nonredundant Mg²⁺ binding sites involving O_b atoms to be analyzed (≈400 in total). In the following, we focus on nonredundant binding sites.

Although less frequently assigned than Mg²⁺, monovalent ions such as Na⁺ and K⁺ were described to be part of the RNA and DNA solvation shell (Table 1). Among others, these ions recurrently contact (G)O6 atoms within quadruplex structures (Largy et al. 2016) and, as inferred from molecular dynamics simulations, they also frequently contact (G)O6 atoms belonging to the major groove of CpG steps within helical motifs (Auffinger and Westhof 2000; Pan et al. 2014; Auffinger et al. 2016; Šponer et al. 2018).

d(Mg²⁺…O_b) distance histograms highlight recurrent Mg²⁺ misidentifications

From a stereochemical point of view, if Mg²⁺ were strongly interacting with C=O_b groups, the sharp peak around 2.1 Å seen in the d(Mg²⁺…O_w) histogram derived from the CSD (Fig. 1A; CSD or Cambridge Structural Database: a repository for crystallographic structures of small molecules [Groom and Allen 2014; Groom et al. 2016]) would also appear in the d(Mg²⁺…O_b) PDB-derived histogram (Fig. 2). Instead, the latter displays a broad peak centered around 2.9 Å that overlaps with the 2.3–3.8 Å oxygen atom exclusion zone, a zone where, in principle, the second coordination shell oxygen atoms should not penetrate (see Materials and Methods and Fig. 1A). Therefore, this histogram exposes important ion misidentification issues (Leonarski et al. 2017).

FIGURE 1.

Mg²⁺/Na⁺ first hydration shells obey strict stereochemical rules. (A) Distance histograms for d(Mg²⁺…Ow) (top) and d(Na⁺…Ow) (bottom) derived from the Cambridge Structural Database (CSD; version 5.38; R-factors ≤5%) (Groom and Allen 2014). No disordered, error containing, polymeric or powder structures were considered. The water exclusion zones and the second coordination shells are marked by gray and blue rectangles, respectively. (B) Ultra-high-accuracy X-ray structures of Mg(H₂O)₆²⁺ (top) and Na(H₂O)₆⁺ (bottom) illustrating similarities between Mg²⁺ and Na⁺ octahedral first hydration shells (Gerasimchuk and Dalley 2004; Hennings et al. 2013). (C) In scale schematic representation of the radius of the Mg(H₂O)₆²⁺ (green) and Na(H₂O)₆⁺ (magenta) first cordination shells. The dashed circle marks the ≈2.8 Å d(H₂O…O_w) average distance and the radius of the less well-defined K(H₂O)_n⁺ first hydration shell.

FIGURE 2.

Mg²⁺/Na⁺ distance histograms to nucleobase O2/O4/O6 (O_b) carbonyl oxygens. (A) Top and bottom: d(Mg²⁺…O_b) and d(Na⁺…O_b) histograms derived from PDB structures with resolution ≤2.9 Å. Only ions with 1.0 occupancies and B-factors in the 1–79 Ų range were considered. The different ion binding and oxygen atom exclusion zones are colored in accordance to Figures 1A, 2B. Note that the provided boundaries are indicative. (B) Scheme showing ion binding and exclusion zones to nucleobase O_b atoms. The (G)O6 atom is taken as an example. The exclusion zone (gray) illustrates the fact that ions are rarely observed close to the C = O_b axis (see C = O_b…Mg²⁺ angle values in Figs. 3, 4; Supplemental Fig. S1). Thus, ions placed in this conical exclusion zone should be considered with caution.

To analyze more precisely these histograms (Fig. 2), we note that Mg²⁺ with coordination distances in the 2.3–2.6 Å range may correspond to misidentified Na⁺ ions that are characterized by ≈2.4 Å coordination distances and a well-defined octahedral coordination shell (Fig. 1). In the same distance range, Mg²⁺ could also correspond to misidentified Ca²⁺ that are sometimes part of crystallization buffers. The latter ion usually adopts an irregular coordination shell comprising 7/8 atoms (Marcus 1988) or, more rarely, an octahedral coordination shell similar to that of Na⁺/Mg²⁺ (Kennedy et al. 2004; Kolev et al. 2018).

Mg²⁺ modeled with coordination distances in the 2.6–3.2 Å range may correspond to misidentified K⁺/NH₄⁺/H₂O since all have coordination distances around 2.8 Å (Auffinger et al. 2016). In order to differentiate them, it is important to note that water and NH₄⁺ have a coordination number of four while that of K⁺ ranges from six to eight (Page and Di Cera 2006; Harding et al. 2010). Mg²⁺ in the 3.2–3.8 Å distance range could correspond to misassigned anions or to crystallographic artifacts (Auffinger et al. 2004a; D'Ascenzo and Auffinger 2016). Finally, the broad peak around 4.2 Å (Figs. 1A, 2A) is attributable to ions—not necessarily Mg²⁺—establishing water-mediated contacts named outer-sphere or second shell contacts. The peaks in the 3.2–4.2 Å may also be attributable to other solvent molecules present in crystallization buffers (Weichenberger et al. 2015).

To further refine our stereochemical criteria, we identified an angular exclusion “cone” with C = O_b…Mⁿ⁺ angles in the 160–180° range (Fig. 2), suggesting that binding of Mg²⁺/Na⁺ with C = O_b…Mⁿ⁺ angles >160° should be interpreted with caution. Indeed, in the CSD (Supplemental Fig. S1) and PDB examples described below, the C = O_b…Mg²⁺/Na⁺ angle values in the 100°–160° range support the validity of this coordination criterion.

Incidentally, we note that identification issues are not restricted to ions in direct contact with nucleic acids but were also observed for hexahydrated ions that bind through water-mediated contacts, even in structures with resolutions ≤2.0 Å (Supplemental Fig. S2). Supplemental Figure S2C shows Na(H₂O)₆⁺, with d(Na⁺…Ow) in the 2.08–2.15 Å range, probably assigned in place of Mg(H₂O)₆²⁺ (Wang et al. 2016). On the opposite, Supplemental Figure S2D shows Mg(H₂O)₆²⁺ with d(Mg²⁺…O_w) ≈ 2.4 Å that is more likely to be Na(H₂O)₆⁺. These data underscore that assignment errors are not limited to divalent ions but can affect all ionic species (Supplemental Fig. S2E,F) and highlight two assignment errors related to high-resolution CSD structures pointing to the unfortunate fact that no database is error free and, thus, completely trustworthy (Spek 2009; Minor et al. 2016).

Direct binding of Mg(H₂O)₅²⁺ to carbonyl oxygen atoms (O_b) is rare

With d(Mg²⁺…O_b) ≤ 2.3 Å, we identified a limited number of nonredundant binding sites (96 occurrences) where Mg²⁺ binds to an O_b atom. Among those, 72, 13, 9, and 2 Mg²⁺ are at direct contact distance with (G)O6, (U)O4, (C)O2, and (U)O2 atoms, respectively. Only 19 of them are pentahydrated with d(Mg²⁺…Ow) ≤ 2.3 Å. This number does not change when a 2.4 Å distance criterion is applied. Eight of these Mg(H₂O)₅²⁺ appear in structures with resolution ≤2.0 Å (Fig. 3) and with Mg²⁺ close to the nucleobase plane. These Mg(H₂O)₅²⁺ present well-defined solvent densities forming a complete octahedral coordination shell and are, therefore, candidates for “prior-knowledge” Mg²⁺ binding sites.

FIGURE 3.

Pentahydrated Mg²⁺ contacting O2/O4/O6 (O_b) carbonyl oxygen atoms with d(Mg²⁺…O) ≤ 2.3 Å showing extensive H-bond networks. The green marks indicate Mg²⁺ with appropriate density patterns and stereochemistry. Only 11 Mg(H₂O)₅²⁺ with d(Mg²⁺…O) ≤ 2.3 Å were characterized in structures with resolution ≤2.0 Å—see PDBid: 4XWF, 4Z4D, 4Z4E, 4W5O, 1DFU, 2VJV, 2R8S. (A) Mg²⁺ binding to a Hoogsteen edge (G)O6 atom. A water molecule establishes an H-bond (dashed cyan line) with the (G)N7 atom. (B) Mg²⁺ binding to a Hoogsteen edge (U)O4 atom. (C) Mg²⁺ binding to a Watson–Crick edge (C)O2 atom. A symmetry-related first shell water molecule establishes an H-bond (dashed cyan line) with the (C)N3 atom. (D) View of a dense H-bond network stabilizing a group I intron core motif (Ye et al. 2008); see also Figure 5B in Juneau et al. (2001). (E) Extended view of the H-bond network at a DNA/protein interface around the Mg²⁺ shown in A (see also Supplemental Fig. S3). The protein residues are gray. In D and E, the first and second shell H-bonds are, respectively, blue and green, and the second shell waters are cyan.

In the 2.0–2.9 Å resolution range, 11 Mg(H₂O)₅²⁺ nonredundant binding sites were identified. All are in ribosomes (Supplemental Table S1) with Mg²⁺ displaying a distorted coordination shell. Four Mg(H₂O)₅²⁺ (E. coli; 2.8–2.9 Å resolution) present proper coordination distances to O_b atoms. However, at the same location in a 2.1 Å resolution E. coli structure (Noeske et al. 2015), a d(Mg²⁺…O_b) ≈ 2.5 Å distance underlines the difficulties of assigning ions with confidence in ribosomes; see Figure 5 in Leonarski et al. (2017).

FIGURE 5.

RNA structures with a high Na⁺ content. Na⁺, K⁺, and Mg²⁺ are shown in magenta, red, and green, respectively. O_b bound Na⁺ can be found in the following structures with resolution ≤2.0 Å: PDBid: 1J8G, 1J6S, 3ZP8, 2R21, 5B2O/P, 3DIL, 4RKV, 4RNE, 4RJ1. (A) A 1.55 Å resolution hammerhead structure with 16 Na⁺ (Anderson et al. 2013). (B) A 1.9 Å resolution lysine riboswitch structure with 29 Na⁺, 3 K⁺, and 1 Mg²⁺ (Serganov et al. 2008). (C) A 2.4 Å resolution structure of the Haloarcula marismortui 50S with 85 Na⁺, 3 K⁺, and 138 Mg²⁺ (Klein et al. 2004; Gabdulkhakov et al. 2013). (D) The rRNA and protein chains shown in C were hidden to highlight the ionic content of this structure.

A few pentahydrated Mg²⁺ ions are involved in large H-bond networks

The few contacts described above in structures with resolution ≤2.0 Å (Fig. 3A–C) are part of large H-bond networks involving ion-coordinated water molecules. In a group I intron P4-P6 domain (Ye et al. 2008), a Mg(H₂O)₅²⁺ is almost completely isolated from the surrounding solvent (Fig. 3D). The five first shell waters form 10 H-bonds (seven to phosphate groups, two to N7 atoms, and one to water). Intriguingly, this Mg²⁺ contributes to the stabilization of a local fold by holding five phosphate groups through second shell contacts. This particular configuration suggests that other energetic factors overrule local charge neutralization effects, as also inferred for sulfate-binding proteins where the anion is recognized through the formation of regular H-bonds to neutral amino acid groups (Pflugrath and Quiocho 1988; Hirsch et al. 2007; D'Ascenzo and Auffinger 2016). Thus, it appears that the ability of an ion to establish water-mediated second shell contacts can overrule electrostatic considerations (Auffinger et al. 2003, 2004b; Bowman et al. 2012).

A second Mg²⁺ binding pattern involving a single direct (G)O6 contact appears at a DNA protein interface (Fig. 3E). The H-bond network involves only one water-mediated contact to a phosphate group. In other structures, comparable Mg(H₂O)₅²⁺ binding patterns to ApUpG steps were noted (Supplemental Fig. S3). Hence, Mg(H₂O)₅²⁺ to O_b binding, although rare, is possible when the solvated ion occupies a tight binding pocket formed by RNA and/or protein residues establishing multiple H-bonds with the Mg²⁺ hydration shell.

Is tetrahydrated Mg²⁺ coordination to two carbonyl oxygens (2O_b) relevant?

2O_b binding sites are not frequent in PDB structures. We identified only eight nonredundant Mg²⁺ with two or more O_b contacts and d(Mg²⁺…O_b) ≤ 2.3 Å. The only 2O_b site observed in a nonribosomal structure, i.e., a group I intron (Ye et al. 2008), displays both d(Mg²⁺…O_b) below 2.3 Å while all distances to waters are in the 2.3–2.5 Å range, suggesting the presence of Na⁺ (Supplemental Fig. S4). No other Mg²⁺ binding to dinucleotide steps with appropriate coordination distances was observed. Thus, if existing, this binding mode is certainly highly uncommon. Elsewhere, we reached similar conclusions regarding the O_b.N_b binding mode that implies monovalent cations rather than Mg²⁺; see Figure 6B in Leonarski et al. (2017). A rare trans-2O_b binding motif (Supplemental Fig. S5) was identified in the CSD where the crystallization conditions often drastically differ from those used for biomolecular systems (Marino et al. 2016).

FIGURE 6.

Re-refinement of a Mg²⁺ binding site in a high-resolution luteoviral pseudoknot fragment (Pallan et al. 2005). (A) Note that the original distances are not in agreement with the presence of a Mg²⁺, and the coordinated water molecule positions do not coincide with electron density peaks. A subsequent unrestrained refinement with Na⁺ led to more appropriate distances for this ion. (B) Positive and negative peaks in the original (PDB deposited) F_o – F_c maps around the pentahydrated ion hint to refinement issues. Such peaks should not appear in high-resolution structure F_o – F_c maps.

This finding looks surprising since it is well appreciated that Mg²⁺ ions are often bridging oxygen atoms (O_ph) of adjacent phosphate groups and are known to form bidendate RNA-Mg²⁺ clamps (Petrov et al. 2011). This discrepancy can be rationalized if we consider that anionic groups (O_ph) are better Mg²⁺ binders than carbonyl groups (O_b atoms; see below) and that the binding of Mg(H₂O)₄²⁺ at a 2O_b site formed by consecutive nucleotides would prevent, in most instances, the formation of optimal ion-coordinated water-mediated contacts such as those shown in Figure 3D,E.

Mg²⁺ binding to O_b and phosphate/carbonyl anionic oxygens

In proteins and nucleic acids, anionic oxygen atoms are commonly considered as primary Mg²⁺ binders (Zheng et al. 2008, 2015, 2017; Bowman et al. 2012). Patterns involving O_b and O_ph atoms with d(Mg²⁺…O) ≤ 2.3 Å are slightly more frequent than those involving O_b/N_b atoms since we identified 48 (O_ph.O_b), 36 (2O_ph.O_b), and 4 (3O_ph.O_b) nonredundant binding sites. For O_ph.O_b, a simple pattern is observed when Mg²⁺ coordinates in trans (trans-O_ph.O_b) to an OP and a (G)O6 atom (Fig. 4A). In ribosomes, few examples of cis-O_ph.O_b and cis-2O_ph.O_b patterns were noted (Fig. 4B,C). Potential fac-3O_ph.O_b binding sites are infrequent (Fig. 4D). Fac- refers to the isoform where the three O_ph atoms are on the same side of the coordination octahedron (Zheng et al. 2015).

FIGURE 4.

Mg²⁺ contacting both carbonyl (O_b) and anionic oxygen atoms (O_ph, O_coo) with d(Mg²⁺…O) ≤ 2.3 Å. The green marks indicate Mg²⁺ with appropriate densities and stereochemistry. (A) Trans-O_ph.O_b: Mg²⁺ binding to a guanine Hoogsteen edge. A water molecule establishes an H-bond (dashed cyan line) with the (G)N7 atom. (B) Cis-O_ph.O_b: Mg²⁺ binding to a guanine Hoogsteen edge. A water molecule establishes an H-bond (dashed cyan line) with the (G)N7 atom. (C) A ribosomal cis-2O_ph.O_b binding site. (D) A ribosomal fac-3O_ph.O_b binding site. Note the coordination distances ≈2.2 Å that suggest the use of restraints (Supplemental Table S2). (E) A cis-2O_coo.O_b binding site at a protein–RNA interface. The contact with a first shell water molecule and a (U)N3-H group, that was considered as difficult to establish (see Fig. 3C), is here replaced by a (U)N3-H…O=C(Asp) H-bond. (C–E) For nomenclature, see Zheng et al. (2015); O_coo corresponds to oxygen atoms of carboxylate groups.

In nucleic acid/protein complexes, carbonyl groups belonging to carboxyl(ate) Asp, Glu (O_coo), or carboxamide Asn and Gln side chains (O_cno) as well as to the peptidyl backbone C=O group (O_bb) may also contact Mg²⁺. In a DNA/protein complex, Mg²⁺ establishes three contacts to water molecules, two to O_coo atoms, and one to an (U)O2 atom (Fig. 4E). At present, an exhaustive classification of binding sites involving O_b and anionic oxygen atoms is out of reach given the high level of uncertainty associated with these binding modes, the ion identity and the rarity of reliable Mg²⁺ binding site in structures with appropriate resolution.

No evidence for Mg²⁺ binding to ribose and backbone oxygen (O_r) atoms

Besides the binding of Mg²⁺ to O_b atoms, we also explored Mg²⁺ binding to O_r atoms defined as oxygen atoms belonging to the nucleic acid ribose group (O2′, O4′) and backbone (O3′, O5′). In the ≤2.9 Å resolution range, we found no Mg²⁺-to-O_r binding site with appropriate coordination distances. In support of this observation, it has been reported that direct interactions between divalent metals and hydroxyl groups are very weak and that the strength of the hydroxyl–metal interaction increases with the decreasing charge of the coordinating atom, suggesting that monovalent ions would better interact with hydroxyl groups than Mg²⁺ (Al-Sogair et al. 2011). Many studies targeted at assessing the involvement of metals in catalytic mechanisms have been conducted. For instance, the participation of a metal ion in the ribosomal catalytic mechanism that would facilitate nucleophilic attack by binding to a 2′-OH group has been investigated through ion substitution experiments (Schmeing et al. 2005). The authors concluded that their crystallographic data are most consistent with a model where a water molecule and not a mono or divalent ion interacts with an active site ribose O2′ atom. These data question the involvement of divalent metal ions interacting with hydroxyl groups in at least some ribozymes (Lilley 2011; Ward et al. 2014).

Critical evaluation of the use of Yb³⁺ and other lanthanide ions as Mg²⁺ substitutes

In contrast to what is reported above, a peculiar and rare example of Mg²⁺ binding to the O3′ group of a terminal guanine is seen in a 1.7 Å resolution protein/RNA complex; see Supplemental Figure S6A (Gan et al. 2008). In this specific context, the O3′ hydroxyl group may take an anionic form and turn into a more appropriate ligand for Mg²⁺. Similar examples are exceptional in the CSD/PDB. However, in a recent group II intron lariat structure (PDBid: 5J01, resolution: 3.4 Å); based on Yb³⁺ anomalous signals, a Mg²⁺ was modeled at a bonding distance of O2′/O3′ ribose atoms of a terminal uridine; see Supplemental Figure S6B (Costa et al. 2016). Similar Mn²⁺ and lanthanide-based strategies were used to explore the RNA ionic landscape (Adams et al. 2004; Stahley and Strobel 2005; Toor et al. 2008; Kazantsev et al. 2009; Wang 2010; Marcia and Pyle 2012, 2014; Bénas et al. 2014; Robart et al. 2014). In these crystallographic studies, it was assumed that Yb³⁺ is a good Mg²⁺ mimic. However, the coordination distance of Yb³⁺ to water (≈2.3 Å) is closer to that of Na⁺ (≈2.41 Å) than to that of Mg²⁺ (2.06 Å). Moreover, the Yb³⁺ coordination number derived from high-resolution CSD structures is dominantly eight, sometimes seven or nine (Cossy et al. 1989; Thuéry 2009) and exceptionally six as observed in a handful of specific chemical contexts; see Supplemental Figure S6C,D (Lundberg et al. 2010). Thus, the observation that Yb³⁺ may bind to group I/II introns and other ribozyme hydroxyl groups does not warrant that Mg²⁺ is present at these sites.

However, it can also be hypothesized that this binding mode occurs only in structural contexts unique to ribozymes. In the group II intron mentioned above (Supplemental Fig. S6B), Mg²⁺ is bound to three phosphates and its charge may be sufficiently delocalized to allow binding of two (see Supplemental Fig. S6B), eventually deprotonated, 2′-OH groups belonging to a terminal ribose. Yet, we suggest that, in the absence of high-resolution data, caution should be exerted in interpreting crystallographic ion binding motifs such as those encountered at ribozyme catalytic sites since binding principles of Mg²⁺, lanthanides, and other ions are still incompletely understood.

Mg²⁺ does not bind to carbonyl oxygen (O_b) atoms of metabolites containing nucleobases

By using the Relibase⁺ program to search the PDB (Hendlich et al. 2003), we checked if Mg²⁺ to O_b binding is associated with nucleobase-containing metabolites. In the ≤2.9 Å resolution range, ≈11,000 metabolites (with G/C/U/T nucleobases) were identified with only one binding site with d(Mg²⁺…O_b) ≤ 2.3 Å. This unique site occurs in a structure of a human signaling protein involving GDP (guanine-diphosphate; Supplemental Fig. S7). Unfortunately, this structure and the four related PDB releases (2005/6) from the same group are not currently associated with a publication record, an issue addressed by Wlodawer et al. (2018). Moreover, this structure contains 12 Ca²⁺ and five Mg²⁺. Most of the latter display a high electron density peak and a tetrahedral coordination with ≈2.1 Å coordination distances that better match a transition metal such as Zn²⁺. Therefore, this Mg²⁺ binding site has to be interpreted with caution given the serious solvent attribution issues reported for this structure. The fact that no obvious Mg²⁺ binding to nucleobase metabolite O_b or N_b atoms, as shown previously (Leonarski et al. 2017), could be characterized, supports the claim that the nucleobase Mg²⁺ binding potential is poor, as also inferred from solution studies on nucleoside and nucleotides (Sigel and Kapinos 2000; Sigel and Sigel 2010).

Direct Na⁺ binding to carbonyl oxygens (O_b) is possible

Since we established that reliable instances of Mg²⁺ binding to O_b atoms are few, we propose potential hexacoordinated substitutes for these ions. As already mentioned, the d(Mg²⁺…O_b) histograms (Fig. 2) suggest that Na⁺ or K⁺ binding to O_b atoms is more likely than Mg²⁺ binding. First, we provide a few examples of Na⁺ to O_b binding. Yet, since misattributions are also an issue for monovalent ions (Supplemental Fig. S2), we focus on structures with resolution ≤2.0 Å.

In this resolution range, 25 RNA and 105 RNA/protein structures containing Na⁺ were identified. Among those, 13 contain Na⁺ with d(Na⁺…O_b) in the 2.3–2.6 Å range (Fig. 1). One hexacoordinated Na(H₂O)₄⁺ at a 2O_b site (Supplemental Fig. S8A) was identified in a 1.55 Å resolution hammerhead ribozyme structure (Anderson et al. 2013) that contains a total of 16 Na⁺ and no divalent ion (Fig. 5A). This best resolution PDB hammerhead structure is at odds with the remaining 22 hammerhead structures that contain no or only divalent/trivalent ions. It could be argued that the crystallization buffer (1.7 M Na⁺ malonate and 10 mM MgCl₂) plays a significant part in displacing Mg²⁺ in favor of Na⁺. However, an RNA tridecamer (resolution: 1.3 Å; Supplemental Fig. S8B) and a lysine riboswitch (resolution: 1.9 Å; Fig. 5B) were crystallized in buffers containing Mn²⁺ and 100 mM NaCl or Mg²⁺, K⁺ and 100 mM Na⁺ citrate, respectively (Timsit and Bombard 2007; Serganov et al. 2008). Out of a total of 29 Na⁺, the latter structure contains 17 hexacoordinated Na⁺ with 10 displaying d(Na⁺…O) in the 2.3–2.6 Å range. Hence, these structures and that of the hammerhead ribozyme demonstrate that Na⁺ can bind to nucleic acid systems with a well-defined octahedral coordination. Though, the reasons as to why these structures that were crystallized in the presence of Mg²⁺ display so many Na⁺ are not understood. The usual explanation stating that Mg²⁺ easily displace monovalent cations does not hold here. Interestingly, the fact that Na⁺ can take over the role of Mg²⁺, even at strong binding sites, was recognized very early (Jack et al. 1977; Quigley et al. 1978). In contrast, Z-DNA structures with resolutions <1.3 Å derived from crystals with a 200 mM MgCl₂ or CaCl₂ content do not show evidence of bound mono or divalent ions (Luo et al. 2017; Harp et al. 2018). Supplemental Figure S8C,D displays a further misattribution case where a Mg²⁺ bridges two O_b atoms belonging to stacked nucleobases with 2.6 Å coordination distances that better characterize Na⁺ but may also correspond to K⁺ given a high σ peak value at the binding location.

A particularity of a few Na⁺-containing structures relates to the occurrence of metallic clusters recruiting two or more Na⁺ with d(Na⁺…Na⁺) in the 3.1–3.7 Å range (Supplemental Figs. S8E, S9). These Na⁺ clusters (Timsit and Bombard 2007; Serganov et al. 2008) represent a neglected category of two-metal binding motifs (Glusker et al. 2001). An example of a Na⁺ cluster mistaken for a Mg²⁺ cluster in a protein is given in Figure S6D of Wlodawer et al. (2018), stressing the probable widespread occurrence of dimetallic Na⁺ clusters in biomolecular systems, a fact to remember during the electron density interpretation process.

The monovalent ion count in ribosomes is underestimated

Overall, the examples we reported show that monovalent ions such as Na⁺ are identifiable in nucleic acid systems even in presence of Mg²⁺. They back up the results of a seminal study by Klein and Steitz regarding the ionic distribution of a 2.4 Å resolution H. marismortui 50S ribosome structure (Klein et al. 2004). The authors stated that, besides Mg²⁺, ribosomes are surrounded by monovalent ions that can be clearly distinguished from divalent ions, based on their coordination patterns and the anomalous signals of Rb⁺/Cs⁺ derivatives (Fig. 5C,D). To establish a primary ion-binding classification, these authors claimed that both Na⁺ and K⁺ lack preferred coordination geometries. If this is reasonable for K⁺, which shows irregular binding patterns with coordination numbers ranging from six to eight, it is less reasonable for Na⁺. Hence, Na⁺ can easily be mistaken for Mg²⁺.

A quick survey of the data presented by Klein and Steitz shows that several Mg²⁺ were assigned to octahedral coordination patterns with d(Mg²⁺…O) in the 2.3–2.8 Å range (see Supplemental Fig. S10A,B), while Na⁺ were assigned to irregular coordination patterns better matching K⁺ (see Supplemental Fig. S10C,D). In a subsequent refinement (PDBid: 4V9F; resolution: 2.4 Å; Gabdulkhakov et al. 2013) of the original H. marismortui 50S structure, the ion coordination patterns and distances remained unchanged.

Mg²⁺ assignments can obliterate monovalent ions—a case for re-refining ion binding sites

As inferred from above, Na⁺ is a better match than Mg²⁺ to hexacoordinated solvent electron densities when d(Mg²⁺…O) is in the 2.3–2.6 Å range. However, no clear-cut ion identification rule can be provided even with d(Mⁿ⁺…O) ≤ 2.3 Å. This is illustrated by the 1.34 Å resolution luteoviral pseudoknot structure that contains two modeled Mg²⁺, one coordinating to a G(O6) atom (Fig. 6) and the other being hexahydrated (Pallan et al. 2005). While for the first ion, water coordination distances are in the 2.1–2.3 Å range, d(Mg²⁺…O6) is close to 2.4 Å. Based on deposited electron density maps, we observed that the positions of ion coordinated waters do not overlap with electron density peaks resulting in positive and negative blobs in the F_o – F_c maps. Therefore, we suspected that the d(Mg²⁺…Ow) in the 2.1–2.3 Å range were inappropriate and performed a basic “unrestrained” refinement of the structure with phenix.refine using default settings (Afonine et al. 2012). While d(Mg²⁺…O6) remained unchanged at 2.4 Å, d(Mg²⁺…Ow) drifted toward 2.36–2.54 Å suggesting to swap the originally assigned Mg²⁺ for Na⁺. For the hexahydrated ion, d(Mg²⁺…Ow) also drifted toward the 2.39–2.52 Å range, a distance consistent with the presence of Na⁺, this ion being present in the crystallization buffers (50 mM Na⁺ cacodylate). Interestingly, the latest version of PDB_REDO, a web service devoted to improving the fit of old and new models to crystallographic data, confirms the coordination distances we obtained but does not yet propose alternative ions that would better fit the data issued from subsequent refinements (Joosten et al. 2009, 2012, 2014; van Beusekom et al. 2018).

K⁺, a ubiquitous but difficult to assign ion that binds to O_b atoms

K⁺ is unfrequently assigned in crystallographic structures, although it is generally considered as the dominant intracellular monovalent cation (Nierhaus 2014; Auffinger et al. 2016). K⁺ is known to bind to specific RNA pockets, as inferred from several nucleic acid X-ray structures (Basu et al. 1998; Batey and Doudna 2002; Conn et al. 2002; Klein et al. 2004; Auffinger et al. 2016). However, its detection remains difficult since K⁺ lacks a well-defined and regular coordination pattern and because its ≈2.8 Å coordination distance overlaps with those of water and NH₄⁺ molecules (Zheng et al. 2017). Therefore, besides direct observation of K⁺ anomalous signals (Tereshko et al. 2001; Egli et al. 2002; Ennifar et al. 2003; Stahley et al. 2007), substitution strategies involving Tl⁺, Rb⁺, and Cs⁺ have sometimes been used (Klein et al. 2004; Marcia and Pyle 2014). Yet, these strategies did not prevent ion misidentifications such as those shown in Supplemental Figure S10C,D.

For instance, it has been found that the major groove cleft of the cis-WC G•U pair (Fig. 7C) is often occupied by water (Auffinger and Westhof 1998; Mueller et al. 1999), but also by K⁺ (Klein et al. 2004; Fan et al. 2005). However, Mg²⁺ establishing direct contacts with G(O6) and U(O4) atoms, was sometimes unduly modeled at this location. Figure 7D shows d(Mg²⁺…O) in the 2.6–3.3 Å range that point to the presence of K⁺. In a re-refinement of Cas9–RNA/DNA structures (Anders et al. 2014; Olieric et al. 2016), it was observed that a fraction of the eight ions originally modeled as Mg²⁺ display anomalous peaks above the 10 σ level (Supplemental Fig. S11). It was not possible to unambiguously identify the corresponding ion from the anomalous peak height or the local stereochemistry. However, X-ray fluorescence (XRF) analysis confirmed the presence of K⁺ in crystals that were obtained by using a buffer containing 250 mM KCl, 300 mM KSCN, and 5 mM Mg²⁺. Based on this evidence, six Mg²⁺ were reassigned as K⁺. However, three uncorrected structures with Mg²⁺ assignments (PDBid: 4UN3/4/5) remain in the PDB, jeopardizing subsequent reinterpretations of the ionic environment of this system (Minor et al. 2016; Rupp et al. 2016; van Beusekom et al. 2016).

FIGURE 7.

K⁺ binding to O_b atoms in the major groove of cis-WC G•U pairs. (A) K⁺ binding to a guanine Hoogsteen edge; coordination number: 8. (B) K⁺ binding to the major groove of a G•U pair; coordination number: 8. (C) K⁺ binding to the major groove of a G•U pair. To highlight the 5.0 σ level K⁺ density, the radius of the K⁺ sphere (magenta) has been reduced. (D) Mg²⁺ modeled in the major groove of a G•U pair that could correspond to a monovalent cation. The electron density pattern is too imprecise for an unambiguous assignment. (B,C,D) (Green) G•U carbons; (cyan) G•U H-bonds; (yellow) other carbon atoms.

Ions in the ribosomal peptidyl and decoding sites: Mg²⁺ or K⁺?

In ribosomes, Mg²⁺ located close to important functional elements (Hsiao and Williams 2009; Bowman et al. 2012; Petrov et al. 2012) are of particular interest. In the first H. marismortui 50S X-ray structures, a K⁺ bridging two guanine Hoogsteen edges (Fig. 8A) was modeled close to the peptidyl transferase center or PTC (Nissen et al. 2000; Klein et al. 2004). It was proposed but not confirmed that this ion plays a role in the catalytic mechanism by stabilizing tautomeric nucleotide forms. The coordination distances for this ion with (G)O6 and (G)N7 are in the 2.8–3.3 Å range and unambiguously point to the presence of K⁺. This K⁺ has been consistently assigned to the same location in 44 H. marismortui PDB structures (Supplemental Table S1). Overall, this unique K⁺ binding site demonstrates the monovalent ion ability to fit within well-defined structural notches (Auffinger et al. 2016).

FIGURE 8.

rRNA K⁺ binding sites. (A) K⁺ close to the peptidyl transferase center has been recurrently assigned in H. marismortui 50S structures (Supplemental Table S1). The guanine K⁺ coordination is comparable to that shown in Fig. 7A. (B) A recurrently assigned Mg²⁺ with K⁺ characteristics in the ribosomal decoding center of several Thermus thermophilus structures.

At odds, an ion close to the decoding center with similar coordination distances to oxygens (Fig. 8B) has been systematically assigned to Mg²⁺ (Murphy and Ramakrishnan 2004; Weixlbaumer et al. 2007; Rozov et al. 2015, 2016a). This ion assignment is made in structures with resolutions in the 2.95–3.30 Å range from which it is really problematic to infer light ion binding. However, its coordination distance to the (C518)O2 and (G530)O6 atoms in the 2.7–3.3 Å range point to the presence of K⁺. In a recent review, it was stated that these ions were modeled as Mg²⁺ but that their precise identity could not be established (Rozov et al. 2016b). To us, the likelihood for this ion to be K⁺ is high, although it has been labeled Mg²⁺ in crystal structures or “M” for metal in related publications. This binding site is representative of ion misattributions in ribosomes (Klein et al. 2004; Noeske et al. 2015).

Although at first glance, such issues may not seem important for the interpretation of crystal structures, one could envisage that they may change the outcome of molecular dynamics simulations since the Mg²⁺ stabilization effect is much greater than that of K⁺ (Hayatshahi et al. 2017). Henceforth, it is highly probable that a decoding site modeled with one or the other ion would behave in significantly different ways and change our perception of the energetics and dynamics of the ribosomal decoding center (Lind et al. 2017). Thus, it is recommended to avoid using structures that contain ions with poor stereochemistry for initiating modeling studies (Hashem and Auffinger 2009).

Remarks regarding the MgRNA database and proposals for a revised set of “prior-knowledge” Mg²⁺ binding sites

The MgRNA database was designed to build an exhaustive and comprehensive classification of Mg²⁺ binding sites (Zheng et al. 2015). In its present state, MgRNA lists 41 inner-sphere coordination patterns among which 16 are associated with O_b atoms. It has already been documented that MgRNA significantly overestimates the binding of Mg²⁺ to nucleobase N_b atoms, a category that regroups the N1/N3/N7 atoms (Leonarski et al. 2017). The same issues originating from too lenient selection criteria are observed for Mg²⁺ to O_b binding. These issues are related to (i) the inclusion in the analyzed sample of structures with resolution >3.0 Å and sometimes >4.5 Å, from which it is impossible to infer the position of light ions or water molecules; (ii) the fact that Mg²⁺-to-O_b coordination distances significantly exceeding 2.3 Å were not discarded; (iii) that artificially restrained ions with d(Mg²⁺…Ow) = 2.18 Å (see Materials and Methods section and Supplemental Table S2) were not excluded; (iv) that several binding modes were described based on only a handful of ions displaying often inappropriate stereochemistry; (v) that no attempts to consider redundancy were made.

To illustrate the above statements, we note that in the O_b data set (468 occurrences), over 50% of the binding sites are redundant. In the remaining structures with resolution <2.9 Å, 15 nonredundant octahedral coordination sites (≈5%) were identified that display often suboptimal coordination stereochemistry. Besides, MgRNA categorizes some infrequent binding modes and defines nine binding types comprising between one and 10 occurrences. These should not be used to define “prior-knowledge” categories, especially when associated with low resolution and poor stereochemistry. In the same line, the 13 MgRNA binding modes involving the O_r atoms (O2′, O3′, O4′, O5′) should not be considered as “prior-knowledge” Mg²⁺ binding motifs (Rupp 2016). More detailed information on issues related to the current MgRNA version is given in Leonarski et al. (2017). These findings are summarized in Table 2 for the O_b atoms and Supplemental Tables S3, S4 for O_r and N_b atoms, respectively.

TABLE 2.

MgRNA (Zheng et al. 2015) Mg²⁺ binding sites involving O_b atoms and correspondence with present study

About statistics

The few instances of Mg²⁺ binding sites with appropriate stereochemistry we characterized do not really allow the collection of meaningful statistics. However, we believe that the trends noted by the MgRNA authors are somewhat preserved although the number of binding sites is considerably reduced because many of them, including binding of Mg²⁺ to hydroxyl groups, need urgently to be discarded. It remains possible that, from forthcoming nucleic acid structures, we may be able to derive slightly different binding principles leading to an extension of those described here.

Mg²⁺ assignment and validation checklist

At this point, we hope that it has become clear that the characterization of each Mg²⁺ to nucleobase binding occurrence needs experimental validation rather than circumstantial evidence derived from low resolution data and limited occurrence. Thus, we feel that it is essential to update existing validation checklists. In Table 3, we adapt our previous Mg²⁺-to-N7 validation checklist to O_b atoms (Leonarski et al. 2017).

TABLE 3.

“Ion-to-O_b” assignment and validation checklist in complement to the “ion-to-N7” checklist (Leonarski et al. 2017)

As before, we stress that the chosen cutoff distances are merely indicative and may be modulated regarding the specific structural context. Thus, d(Mg²⁺…O) ≈2.4 Å distances are borderline and must be considered with caution. Marginal ion-to-oxygen distances may also find their origin in hidden crystallographic disorder (multiple conformations, partial/mixed occupancies…), the presence of unexpected solvent molecules contaminating purification or crystallization buffers (Borek et al. 2003; Giegé 2013; Dauter et al. 2014; Weichenberger et al. 2015; Niedzialkowska et al. 2016; Moon et al. 2017), or even inappropriate refinement protocols that remain to be documented.

The use of restraints during crystallographic refinement often introduces major interpretational biases (Leonarski et al. 2017). In particular, we strongly discourage the use of the widespread 2.15/2.18 Å d(Mg²⁺…O) default distance restraint values in the CCP4 ener_lib.cif library (Supplemental Table S2). When distance restraints are used during refinement, it becomes problematic to assign with confidence Mg²⁺ or Na⁺ to a given electron density pattern (Zheng et al. 2014, 2017; Leonarski et al. 2017). When needed, one should use CSD derived distance restraints (Zheng et al. 2014, 2017; Leonarski et al. 2017) since distances from the PDB were shown not to be reliable (Fig. 2). Unfortunately, structures refined by using inappropriate restraints remain in the PDB and represent a serious hazard for the less experienced users (Cooper et al. 2011; Dauter et al. 2014; Minor et al. 2016; Leonarski et al. 2017).

It is also important to mention that the use of restraints for structures in the >2.5 Å resolution range is often justified by the fact that diffraction data are unable to determine metal-to-N/O coordination distances with the required precision (Harding 2001). However, this should be avoided, especially when the risk of Na⁺/Mg²⁺ misidentification is high. In such instances, crystallographers should consider that Na⁺/K⁺/Ca²⁺ are rightful alternatives to the placement of Mg²⁺ and refrain from drawing firm conclusions regarding ion identity. When ion identity remains ambiguous and when, nevertheless, the electron density pattern points to the presence of a metal, the “M” marker should be used (Rozov et al. 2016b).

In the future, significant help in assigning solvent molecules should stem from better use of anomalous signals (Leonarski et al. 2017). Thanks to continual improvements in anomalous difference measurements through specialized beamlines, constant accuracy improvement of X-ray detectors, and more efficient software (Storoni et al. 2004; Thorn and Sheldrick 2011; Weinert et al. 2015; Olieric et al. 2016; Wagner et al. 2016; Leonarski et al. 2018), it might become possible to make use of weak Na⁺/Mg²⁺ signals when high resolution is available. With greater likelihood, the detection of the anomalous signals of heavier ions such as K⁺, Ca²⁺, Cl⁻, and SO₄²⁻ will be facilitated (Ennifar et al. 2003; Auffinger et al. 2004a; Mueller-Dieckmann et al. 2007; Thorn and Sheldrick 2011; Echols et al. 2014; D'Ascenzo and Auffinger 2016). In the meantime, we advocate for the deposition of diffraction images for all relevant X-ray measurements, including heavy atom soaks, to allow reprocessing of the data and to check for weak anomalous signals. We suggest also to define a marker that would help to differentiate ions that were placed based on native data from those that were modeled by using anomalous signals from different ions (Grabowski et al. 2016). Indeed, as discussed earlier (Leonarski et al. 2017), ion substitution experiments are not always sufficiently reliable to confirm Mg²⁺ binding sites since the binding preference of lanthanides or soft ions like Mn²⁺ does not systematically match those of the harder Mg²⁺. This has already been suggested in an early study of tRNA ion binding where the authors noted in a sobering manner: “A comparison of the magnesium, cobalt and manganese binding sites gives reason to doubt the idea that these last two mimic magnesium in their binding properties” (Jack et al. 1977).

In order to get clues about the ionic composition of the crystals, the systematic use of X–ray fluorescence (XRF) analysis should be encouraged (Olieric et al. 2016) as well as the exploration of the local environment of a metal by extended X-ray fine absorption structure (EXAFS) techniques (Hensley et al. 2011; Hummer and Rompel 2013). Interestingly, inductively coupled plasma emission spectroscopy (ICP-ES) has been used to eliminate the presence of possible divalent metal contaminants (Mn, Ni, Zn, Pb, and Cd) in tRNA crystals (Jovine et al. 2000).

SUMMARY AND CONCLUDING REMARKS

The combined data presented in this and an earlier investigation (Leonarski et al. 2017) interrogate recurrent assumptions made in interpreting solvent electron density maps within nucleic acid structures. These assumptions or “disruptive nudges”—to divert a concept made popular by Richard Thaler (Thaler 2000; De Bondt et al. 2018)—have led in our opinion to the deposition in the PDB of a significant number of nucleic structures with exaggerated Mg²⁺ contents at the expense of the assignment of monovalent cations (Na⁺, K⁺) and other small solvent molecules. Here, we stress that the possibility that other ions (Na⁺, K⁺, Ca²⁺, …) could fit solvent electron density patterns should be systematically envisaged, especially for sites displaying borderline stereochemistry and that ion substitution experiments should be interpreted with caution given the rising number of documented instances emphasizing deceiving effects associated with ion replacement strategies (Jack et al. 1977; Leonarski et al. 2017).

We suggest that, before inferring ion binding from low-resolution crystallographic structures, a set of trustworthy binding sites derived from high-resolution structures or, in other words, a set of “prior-knowledge” binding sites must be defined. This has been tentatively proposed in Table 2 for O_b atoms and Supplemental Tables S3, S4 for O_r and N_b atoms. In the current state of the art, “prior-knowledge” binding sites are difficult to collect since Mg²⁺/Na⁺/K⁺ misattributions are observed even in structures with resolutions <2.0 Å. In the current PDB data set, through replication of errors, the recurrence of a given binding site may unfortunately not warrant its reliability and, therefore, it is important to keep redundancy issues in mind. More specifically, present data imply that the assertion that “the coordination of Mg²⁺ by nucleobases should be considered as a significant factor in the stabilization of RNA structure” (Zheng et al. 2015) must be approached with caution and should not be used to support claims regarding the implication of nucleobase carbonyl groups in catalytic mechanisms (Liu et al. 2017).

This study raises the following interrogation: Why, given the high number of accessible nucleic acid carbonyl groups, do we observe such a small number of Mg²⁺ binding sites involving nucleobases? We propose that carbonyl groups (as well as hydroxyl groups) are poor Mg²⁺ binders but excellent monovalent binders. For instance, both quadruplexes and K⁺ ion channels use the same principles based on monovalent binding to carbonyl groups to fulfill their function (Auffinger et al. 2016).

A more speculative rationale for the poor binding occurrence of Mg²⁺ to O_b atoms can be proposed. If these quite frequent oxygen atoms would be linked to efficient Mg²⁺ binding sites, such binding occurrences would be particularly abubdant imposing a high Mg²⁺ consumption by ribosomes in the cell. Thus, a limited occurrence of Mg²⁺ binding to nucleobase atoms may be required in order not to impede crucial folding and assembly steps and allowing structural fluidity at critical regions of these molecular machines.

As such, we advocate for a greater awareness of the fact that monovalent ions can easily be mistaken for Mg²⁺. We strongly believe that strict compliance to well-established stereochemical rules (Fig. 1) may lead to less misidentifications. Indeed, such issues were shown to considerably blur our understanding of nucleic acid ion binding principles Therefore, we must correct our perception of the existing ionic equilibrium around nucleic acids.

Artificial intelligence or machine learning technologies could certainly help to disentangle these difficult ion assignment issues provided that their algorithms are nurtured by sound data (Kowiel et al. 2018). At least, such techniques may help to recognize that current structural databases are far from an error free state and suggest reprocessing some of the underlying experimental data. With current statistics, we might reach counterproductive conclusions regarding the roles of ions in nucleic acids (Zheng et al. 2015). This might significantly impact domains related to the development of molecular dynamics force fields that have to rely on a rigorous interpretation of experimental data for calibration purposes (Panteva et al. 2015; Lemkul and MacKerell 2016; Casalino et al. 2017; Li and Merz 2017) and domains related to the automatic detection and classification of ion binding sites (Brylinski and Skolnick 2011; Lemkul et al. 2016; Casalino et al. 2017; Cunha and Bussi 2017; Sun et al. 2017). For these strategies to be successful, a “prior-knowledge” database of validated Mg²⁺ to nucleic acid binding modes derived from high-resolution structures is urgently needed.

MATERIALS AND METHODS

All ≈5250 nucleic acid crystal structures deposited to the Protein Data Bank (PDB; February 2017) with resolution ≤2.9 Å were searched for Mg²⁺ binding to purine and pyrimidine O2/O4/O6 carbonyl oxygen atoms, hereafter named O_b atoms (Zheng et al. 2015). It is well established that Mg²⁺ has an octahedral coordination sphere with a stringent d(Mg²⁺…Ow) ≈ 2.06 ± 0.03 Å coordination distance and a second hydration shell around 4.2 Å that is marked by a shallow peak in the CSD distance histogram (see Fig. 1; Markham et al. 2002; Harding et al. 2010). The clearly identifiable gap in the 2.3–3.8 Å range, between the first and second coordination shell peaks, defines an oxygen atom “exclusion zone.”

To account for crystallographic inaccuracies, we used a rather tolerant d(Mg²⁺…O_b/Ow) ≤ 2.3 Å criterion for our PDB searches. This distance criterion is more stringent than the d(Mg²⁺…N_b) ≤ 2.4 Å criterion used in an earlier study (Leonarski et al. 2017). The latter cutoff choice was based on the fact that d(Mg²⁺…N_b) is often assumed to be ≈0.1 Å longer than d(Mg²⁺…O) (Harding et al. 2010; Leonarski et al. 2016). The present d(Mg²⁺…O_b/Ow) ≤ 2.3 Å cutoff has the added benefit to allow for a better differentiation of Mg²⁺ versus Na⁺ oxygen binding given that d(Mg²⁺/Na⁺…O) ≈2.06/2.41 Å (Fig. 1A). For Na⁺, as for Mg²⁺, a shallow second coordination peak ≈4.3 Å is observed that is associated with a less marked oxygen “exclusion zone.” Remarkably, Na⁺ displays in numerous instances a well-defined octahedral coordination shell (Fig. 1B), a fact that is not always fully appreciated (Klein et al. 2004). A pentahydrated coordination for Mg²⁺ is excessively rare and may not be observed in biological contexts (Chattopadhyay et al. 2009) with the exception of chlorophyll, where Mg²⁺ coordination requires the assistance of chelatase enzymes (Chen et al. 2015).

To increase the reliability of our structural sample, we downsized our resolution cutoff from 3.0 to 2.9 Å. This 0.1 Å shift resulted in the exclusion of ≈400 (roughly 10% of the total) structures including ≈40 redundant ribosomes (Supplemental Table S1). This is advisable since resolutions ≥3.0 Å are by far not ideal for accurate light ion placement (Z ≤ 12; i.e., Na⁺ or Mg²⁺). Other authors selected even more cautious cutoff criteria by stating that, at resolutions >2.5 Å, unambiguous placement of light ions is not reasonable (Harding et al. 2010; Harding and Hsin 2014).

Ions with B-factors ≥79 Ų were excluded from the statistics since such high B-factors do not warrant unambiguous ion characterization (see Supplemental Material). We further excluded ions with B-factors ≤1.0 Ų and occupancies ≠ 1.0 unless otherwise specified since the assignment of such ions is not reliable—see a 3.0 Å resolution rRNA structure (PDBid: 1FJG) that displays Mg²⁺ occupancies in the 0.22–1.47 range (Carter et al. 2000).

As for Mg²⁺ to N7 binding (Leonarski et al. 2017), we applied a 1.0 Å out-of-nucleobase plane cutoff since Mg²⁺ tend to bind to the lone pairs of carboxyl oxygen atoms and should therefore lie in the nucleobase plane. On the other hand, monovalent cation binding is not restricted to the nucleobase plane as exemplified by K⁺ binding to quadruplex structures (Largy et al. 2016). Finally, for all ions for which crystallographic assignment is unclear, 2F_o − F_c and F_o − F_c electron density maps were inspected (Gutmanas et al. 2014; Velankar et al. 2016). For a quick assessment of ion binding stereochemistry, we also used the CheckMyMetal website that permits to rapidly visualize the coordination of all ions present in a PDB structure and suggests meaningful ion replacements but does not allow to observe electron densities and does not read files in the mmCIF format that are associated with the large ribosomal structures (Zheng et al. 2014, 2017).

Nonredundant Mg²⁺ binding sites were tagged as follows. Two nucleotides from different structures at a comparable Mg²⁺ binding site and sharing the same residue numbers, chain codes, trinucleotide sequences, ribose puckers, backbone dihedral angle sequences (with the g+, g−, t categorization) and syn/anti conformations, were considered similar and the one with the best resolution was considered nonredundant. In case of matching resolutions, the nucleotide with the lowest B-factor was selected. Similarly, if in a same structure, two nucleotides involved in a comparable Mg²⁺ binding site and located in different biological units shared the same residue numbers and trinucleotide sequences (with different chain codes) as well as ribose puckers, backbone dihedral angle sequences, and syn/anti conformations, they were considered as similar and the one corresponding to the first biological unit was marked as nonredundant. To further limit redundancy in ribosomal structures, we restricted our analysis to a single biological assembly (see Supplemental Material).

Three nonredundant sets were calculated with 2.3/2.6/3.5 Å cutoffs for d(Mg²⁺/Na⁺…O_b) (Table 1). Note that our redundancy criteria were designed for analyzing local structural features and must be distinguished from more global structure-based “nonredundant” criteria (Leontis and Zirbel 2012). Indeed, structures embedding RNA systems with identical sequences are frequent in the PDB. However, these structures often differ by the solvent composition of the buffers in which they were crystallized (Supplemental Table S1). Homemade programs were used to collect data relative to ion binding sites. More precisely, in-house PyMOL (The PyMOL Molecular Graphics System, Schrödinger, LLC) and Perl scripts were used to download and analyze nucleic acids from the PDB, as well as to extract and categorize information relating to ion binding sites. PyMOL was used to apply symmetry operators and to visualize data. Phenix.refine (Afonine et al. 2012) was used for X-ray refinement, as indicated in the appropriate section.

SUPPLEMENTAL MATERIAL

Supplemental material is available for this article.