Halogen Bonds Involved in Binding of Halogenated Ligands by Protein Kinases

Analysis of 664 known structures of protein kinase complexes with halogenated ligands revealed 424 short contacts between a halogen atom and a potential protein X-bond acceptor, the topology and geometry of which were analyzed according to the type of a halogen atom (X = Cl, Br, I) and a putative protein X-bond acceptor. Among 236 identified halogen bonds, the most represented ones are directed to backbone carbonyls of the hinge region and may replace the pattern of ATP-like hydrogen bonds. Some halogen-π interactions with either aromatic residues or peptide bonds, that accompany the interaction with the hinge region, may possibly enhance ligand selectivity. Interestingly, many of these halogen-π interactions are bifurcated. Geometrical preferences identify iodine as the strongest X-bond donor, less so bromine, while virtually no such preferences were observed for chlorine; and a backbone carbonyl as the strongest X-bond acceptor. The presence of a halogen atom in a ligand additionally affects the properties of proximal hydrogen bonds, which according to geometrical parameters get strengthened, when a nitrogen of a halogenated ligand acts as the hydrogen bond donor.


INTRODUCTION
Post-translational modifications (PTMs), both reversible and irreversible, may affect intracellular localization of proteins, regulate their interactions with protein or non-protein partners, modulate their catalytic activity, or select some of them for degradation.In general, PTMs increase proteome diversity by at least an order of magnitude, when compared to the transcriptome, and even more so relative to the genome.They also enable rapid response or adaptation to extracellular factors, contributing to signal transduction and regulation of numerous cellular pathways.The most frequent modifications include glycosylation, lipidation, methylation, N-acetylation, S-nitrosylation and sumoylation.A particular role is played by reversible protein phosphorylation.The residues most susceptible to phosphorylation are serine, threonine and tyrosine, less frequently histidine (Klumpp & Krieglstein, 2002;Besant et al., 2003;Steeg et al., 2003;Besant & Attwood, 2005;Ciesla et al., 2011), and rarely aspartate (Wagner & Vu, 2000;Lapek et al., 2015), cysteine (Pannifer et al., 1998;Feng et al., 2008), lysine (Mat-thews, 1995;Khorasanizadeh, 2004;Besant et al., 2009) or arginine (Fuhrmann et al., 2009;Elsholz et al., 2012).Protein kinases, which catalyze phosphorylation of proteins, display a large spectrum of substrate specificities.Most use ATP as a phosphate donor, albeit some may accept GTP (Ventimig & Wool, 1974).
Protein kinases are attractive molecular targets for drug design (Cohen, 2002), since they are playing key roles in the regulation of many cellular processes, including the cell cycle, growth and apoptosis.To date, most promising protein kinase inhibitors are small ATP-competitive molecules (Zhang et al., 2009;Fabbro, 2015), which bind at the highly conserved ATP-binding site.
Nonetheless, most of the currently used protein kinase inhibitors locate, at least partially, in the ATP-binding pocket.These ligands must mimic the overall properties of the ATP molecule, i.e. they are locally flat, preferably aromatic, and capable of hydrogen bond formation and efficient electrostatic interactions with residues that form the protein kinase ATP-binding site.According to the Lock-and-Key analogy originally postulated in 1894 by Fischer, and further extended to the Induced-Fit Theory (for review see Koshland, 1994), which is a biochemical equivalent of the Pauli exclusion principle: a low-mass ligand should fit to a binding site attainable for ligands in the solvent phase.Van der Waals (vdW) interactions, both attractive and repulsive, are short-range contacts that control binding events (Barratt et al., 2005), favor-ing the ligands that fit to the protein binding site.Electrostatic interactions between a protein and a ligand are dominated by short contacts between charged groups (known as salt bridges, formally zero order term in multipole expansion of electrostatic interactions).Their contribution to the Gibbs free energy of ligand binding approaches 40 kJ/mol (Hendsch & Tidor, 1994).However, the subsequent moments in multipole expansion related to static (charge-dipole, dipole-dipole, etc) or induced-charge distributions (i.e.Debye and London forces), stacking interactions (electron correlation in proximal π-electron systems), hydrogen and halogen bonding, may also contribute significantly.The significance of these interactions is well described, with the exception of halogen bonding, the contribution of which is still under debate (Eckenhoff & Johansson, 1997;Liu et al., 2005;Voth et al., 2007;Memic & Spaller, 2008;Kraut et al., 2009;Zou et al., 2009;Hauchecorne et al., 2010;Sarwar et al., 2010;Carter & Ho, 2011;Hardegger et al., 2011;Aakeroey et al., 2013;Poznanski et al., 2014), with estimates of the free energy of an individual X-bond varying from 0.8 (Sarwar et al., 2010) up to 30 kJ/mol (Voth et al., 2007).
Halogen bonding (X-bond) has been identified in many crystal structures of low-mass compounds and their supramolecular ensembles (for review see Metrangolo et al., 2008), and more recently in complexes of biomolecules with halogenated ligands (Auffinger et al., 2004;Voth & Ho, 2007;Rendine et al., 2011).Specific interactions between the ligand halogen atoms (Cl, Br, I) and the electron pairs of an oxygen/nitrogen/sulfur/π-electron system have been described, based largely on the observation that the distance between a halogen atom and its electron-donating partner, d X•••Acc , is significantly shorter than the sum of their vdW radii (Fig. 1A).Fluorine, because of its high electronegativity, is a very poor halogen bond donor (Politzer et al., 2007), but it may act as an efficient hydrogen bond acceptor (Howard et al., 1996;Dunitz, 2004) The role of halogenated ligands in biological systems has been widely reviewed, amongst others by Auffinger et al., 2004, Parisini et al., 2011, Rendine et al., 2011, Voth et al., 2007, Voth et al., 2009, Scholfield et al., 2013, Wilcken et al., 2013, Persch et al., 2015 and also by us (Poznanski & Shugar, 2013;Poznanski et al., 2014).These systematic structural studies show numerous examples of halogen bonds formed between a ligand and a protein electron-donating group.The geometry of these halogen bonds has been well described, with a preference for the X•••Acc-C (θ A ) angle of 160°, roughly resembling that of a hydrogen bond (Fig. 1), albeit the distributions of θ X and θ D angles differ significantly.It should however be noted that the distribution of θ A and θ D angles depends on the hybridization of the involved atoms.
Numerous natural drugs (Smit, 2004;Wang et al., 2005;Cabrita et al., 2010) and an increasing number of synthetic drug candidates (Hernandes et al., 2010;Pauletti et al., 2010) are halogenated, comprising approximately 20% of low-mass protein ligands accessible in the Protein Data Bank (PDB), and an even larger number of tested protein kinase inhibitors.The growing number of high-resolution structures of protein kinase-ligand complexes aids in silico development of new inhibitors (Niefind et al., 2009;Ibrahim, 2011;Ibrahim, 2012;Lepsik et al., 2013), many of them halogenated.Understanding the structural requirements for the binding of halogenated ligands, and the estimated contribution of the halogen bonding to the Gibbs free energy of ligand binding is crucial for in silico design of halogenated drugs (Ibrahim, 2012;Jorgensen & Schyman, 2012;Kolar & Hobza, 2012;Wang et al., 2014).
Herein we present a detailed analysis of the geometry and topology of short contacts of halogen atoms identified in all complexes of protein kinases with halogenated ligands accessible in the Protein Data Bank.A statistical approach was applied to estimate, independently for Cl, Br and I as halogen bond donors, their relative contribution to the free energy of halogen bond formation in protein-ligand systems.

MATERIAL AND METHODS
Structural data.The Protein Data Bank (PDB) was searched to identify all entries of protein kinases (EC 2.7.10,2.7.11 and 2.7.12), while histidine protein kinases (2.7.13) were omitted.
Structural analysis.All analyses were performed with the aid of the Yasara Model package (Krieger et al., 2009).For each halogen atom type, all intermolecular ligand-protein contacts were identified, using 4 Å as a threshold for the distance between a halogen atom and a putative halogen bond acceptor.The analysis was further restricted to interactions characterized by the d X•••Acc distance between a halogen atom and a potential halogen bond acceptor shorter than the sum of their vdW radii.The contacts for which the C-X•••Acc angle exceeded 140º (Fig. 1) were annotated as halogen bonds.Multiple protein molecules in the unit cell, as well as objects displaying multiple partially occupied forms (i.e.side-chain rotamers or ligand locations) were analyzed separately.
Structure validation.The analysis was done with the aid of Coot (Emsley & Cowtan, 2004;Emsley et al., 2010) and figures with the PyMol program (DeLano & Lam, 2005).Reliability of the presence, position and identity of solvent molecules in the vicinity of the halogen atoms was assessed in several ways.First, we eliminated all structures with resolution lower than 2 Å and structures with muliple conformations of the halogenated part of the ligand.Next, we manually inspected EDS- (Kleywegt et al., 2004) and PDB REDO- (Joosten et al., 2014) generated F o -F c (difference maps indicating disagreement between the observed, F o , and calculated, F c , electron densities) as well as 2F o -F c electron density maps (maps calculated with model phases and experimental structure factors, with an additional F o -F c correction that counteracts the model bias).Finally we analyzed B-factors, coordination geometry and topology of the solvent molecules in question.Since there were only a few molecules fulfilling all selection criteria, we restricted solvent analysis to high resolution structures (< 2.5 Å) with deposited electron density maps, and excluded all solvent molecules in extremely short contacts with the halogen atoms (< 2.5 Å).In all analyzed cases there were ions of molecular weight comparable to water present in the crystallization buffer (Na + , Mg 2+ or NH 4 + ).While metal ions should, in principle, be distinguishable from water on the basis of the coordination sphere, it is very hard to tell apart the ammonium ion and water with crystallographic methods, and thus we cannot absolutely exclude polar character of the identified interactions.X-ray radiation induced partial ligand decomposition also cannot be excluded.
Statistical analysis.To overcome the categorization issue, all distributions are presented in a cumulative manner as a CDF (cumulative distribution function), which is the integral of a commonly used distribution function.This form of presentation helps in visual comparison of various distributions of samples of a limited size.Since, according to the Anderson-Darling test (Anderson & Darling, 1952), most distributions were found to be non-Gaussian (data not shown), the statistical significance of the observed differences was estimated according to the Mann-Whitney U test (Mann & Whitney, 1947) for comparison of two datasets, and the Kruskal-Wallis H test (Kruskal & Wallis, 1952) for 3 or more groups.When the above tests did not show statistically significant differences in the location of analyzed distributions, the Kolmogorov-Smirnov two-sample test, much more sensitive for the distribution shape, was applied (Massey, 1951).All analyses were performed using Statistica 10 (StatSoft, 2011).Null hypotheses that given distributions do not differ from each other were tested at a significance level of α = 0.05, and those with p-values below 0.05 were rejected.The p-values listed in the text are indexed according to the applied method: p MW , p KW and p KS for Mann-Whitney, Kruskal-Wallis and Kolmogorov-Smirnov test, respectively.In general, the distributions of d X•••Acc distance and C-X•••Acc angle (judged by the smaller-larger principle), were preferably analyzed using Mann-Whitney or Kruskal-Wallis tests, and the X•••Acc-C angle distribution (interpreted in the wider-narrower terms), with the Kolmogorov-Smirnov two-sample test.

Preferred topology of short contacts between a ligand halogen atom and a protein kinase
A total number of 424 short contacts between halogen atoms and potential X-bond acceptors was found in 320 of 664 structures of protein kinases in complexes with halogenated ligands.This includes 151 PDB records for protein tyrosine kinases (Enzyme Classification 2.7.10), 386 for protein serine/threonine kinases (EC 2.7.11) and 127 for dual-specificity protein kinases (EC 2.7.12).Short contacts were identified using thresholds calculated individually according to X-bond donor and acceptor types as the sum of their van der Waals (vdW) radii of 1.52, 1.55, 1.70, 1.75, 1.80, 1.85 and 1.98 Å for oxygen, nitrogen, carbon, chlorine, sulfur, bromine and iodine, respectively (Bondi, 1964).Overall, 223, 148 and 53 short contacts were identified for chlorine, bromine and iodine atoms attached to a carbon atom (halide ions were excluded from the analysis).This includes, respectively, 102, 88 and 46 interactions fulfilling the distance and angle requirements for a halogen bond (Desiraju et al., 2013).The numbers of identified short contacts and halogen bonds are presented in Table 1.Only three of the highest populated X-bond acceptor types were pre-sent in sufficient numbers to assess the statistical significance of the observed halogen-dependent differences in the parameters describing the halogen bond geometry.
The most targeted protein kinase regions are the β-sheets of the N-terminal lobes, for which putative acceptors include carbonyl oxygen and/or π-electrons of a peptide bond; and carbonyls of residues located in the hinge region that are involved in the ATP binding (see Fig. 2

below).
A carbonyl oxygen, in accordance with the PDB screenings (Auffinger et al., 2004;Lu et al., 2009;Hardegger et al., 2011;Parisini et al., 2011), is the most abundant putative acceptor of a halogen bond, contributing, together with an amide nitrogen, to over 50% of the identified short intermolecular contacts.Due to geometrical reasons, most of the contacts between a halogen atom and a backbone nitrogen accompany interaction(s) with a proximal carbonyl group.There are, however, some structures strictly representing the concept of orthogonal halogen bonds to π-electrons of the amide group, originally identified by Voth (Voth et al., 2009).In the complex of human CDK2 with a brominated triazole-pyrimidine inhibitor (pdb2c69; Richardson et al., 2006), the separate X-bond to the backbone nitrogen of Glu12 could be identified (Fig. 2A), while in the complex of epidermal growth factor receptor variant with PD168393 (pdb4lrm; Yasuda et al., 2013), the bromine atom makes numerous short orthogonal contacts with X-bond acceptors located in the proximal β-sheets (Fig. 2B).

Halogen bonding with the π-electron system of an aromatic residue
Protein kinase sequences include many aromatic residues, some of which are involved in catalysis, either by direct binding of ATP or transfer of the phosphate group.Those located in the vicinity of the ATP-binding site may form short contacts with halogenated ATPcompetitive ligands.The conserved aromatic residues in protein kinases are generally found in the glycinerich loop (Y50 in CK2α/Y15 in CDK2), hinge region (F113/F80 and Y115/F82), catalytic loop (H154/H125 and H160/Q131), and the DFG-motif (W176/F146).Locations of these residues for protein kinase CK2α are shown in Supplementary Fig. 1.Interestingly, the aromatic residues are not strictly conserved in protein kinases and thus may be targeted by X-bonding to enhance ligand specificity.
Halogen bonding to π-electron systems is well documented in the Cambridge Structural Database (CSD), but depicted by only a few structures demonstrating the interactions between halogen atoms in organic systems and aromatic groups, shorter than the sum of their van der Waals radii (Reddy et al., 1996;Hubig et al., 2000).Halogen bonds to π-electron systems have also been identified in complexes of halogenated ligands with proteins, e.g.serine protease Xa (Nazare et al., 2005), farnesyltransferase (Tong et al., 2003), or HIV-1 reverse transcriptase (Das et al., 2004).
Three modes of interaction between a halogen atom and an aromatic system have been identified, based on the orientation of the aromatic ring with respect to a C-X bond, which can be positioned either perpendicular or parallel to the normal vector defined by the plane of the aromatic ring.When a C-X bond is perpendicular to the plane of the aromatic ring (i.e.parallel to the normal vector), the halogen atom may interact either with the center of the π-electron system (Fig. 3A) or with its rim (Fig. 3B).The mode in which the C-X bond lies over the plane of the aromatic ring (Fig. 3C), does not fulfill the formal geometrical requirements for halogen bonding, since the σ-hole, located along the C-X axis (Clark et al., 2007), is not directed toward the potential halogen bond acceptor.When a halogen atom is attached to an aromatic moiety, possible π-π stacking interactions additionally compete with a halogen atom for a proximal π-electron system (Li et al., 2012).
Screening of structures of protein kinases in complexes with halogenated ligands has evidenced numerous close contacts between halogen atoms and π-electron systems.Their topology varies, but several groups can be identified.The most abundant short contacts with halogen atoms involve the phenylalanine residue of the hinge region (F113 in CK2α), next is a tyrosine residue located in the glycine-rich loop (Y50), and an aromatic residue from the DFG motif (W176).All structures that display short contacts between a halogen atom and an aromatic ring are collected in Table 2.The representative geometries are shown in Fig. 2E-L.It should be noted, however, that the orientation of a C-X bond with respect to a proximal aromatic ring for numerous identified systems disagrees with the idealized geometry of a halogen bond and instead resembles the parallel arrangement (see Fig. 3C).Thus, the halogenated ligand may be involved in a canonical π-π interaction with the protein aromatic residue, as shown in Fig. 2E, F for chlorinated and brominated ligands (1RU with hepatocyte growth factor receptor in pdb4knb (Steinig et al., 2013) and TV4 with serine/threonine-protein kinase B-Raf in pdb3tv4 (Wenglowsky et al., 2011), respectively).
In numerous other structures, the C-X•••Acc angle also differs substantially from the range of 160-180° found optimal in CSD (Rosokha & Kochi, 2008), as shown in Fig. 2G, H for RTX with serine/threonine-protein kinase pim-1 (pdb4med) and Z21 with subunit alpha of cAMP-dependent protein kinase (pd-b4c37; Couty et al., 2013).The largest number of short contacts with strongly perturbed geometry is observed for chlorinated ligands (> 60% of all identified), in contrast to brominated and iodinated ones, for which the geometry close to optimal is preserved in the majority of analyzed cases (95 and 100%, respectively).Finally, a total number of 24 halogen bonds to π-electron aromatic systems, for which all geometrical requirements for efficient halogen bonding are fulfilled, have been identified (Table 2 and Fig. 2I-L).
The most frequently observed halogen bond to the aromatic residue engages the phenylalanine of the hinge region.It is often accompanied by two parallel hydrogen bonds formed with the backbone of the downstream residue (Phe+3), that for polyhalogenated ligands may be substituted by halogen bond(s) to the carbonyl groups of residues (+1) and (+3) (Fig. 2J, L).Alternatively, a halogen-π interaction may involve an aromatic residue of the glycine-rich loop, as observed in the complexes of mitogen-activated protein kinase 1 with E57 (pdb4fv6) or VTX-11e (pdb4qte; Chaikuad et al., 2014) , 2013).All these ligands also make hydrogen bond(s) with the (+3) residue of the hinge region.Interestingly, a C-X•••π halogen bond is frequently accompanied by a parallel interaction of the halogen atom with a proximal solvent molecule, identified in over 50% of the analyzed structures (see Table 2).For ligands forming an X-bond with the hinge region, location of this solvent molecule is highly conserved, and the distance between the halogen atom and the oxygen atom (in case this solvent molecule is interpreted as water) is substantially shorter than the vdW limit.This may represent a possible example of a bifurcated halogen bond identified in crystals of small organic compounds (Lu et al., 2006;Carlsson et al., 2015;Novak et al., 2015).The observed C-X•••O wat angle of approximately 120° (see Table 2) strictly corresponds to a minute maximum identified in the distribution of the C-X•••O angles by Scholfield (Scholfield et al., 2013), however it seems to be too far from the optimal 160-180° found for a plausible halogen bond in previous screenings of the PDB Table 2. Short contacts between a halogen atom and an aromatic ring identified in complexes of protein kinases with halogenated ligands together with accompanying short contacts between halogen atoms and solvent molecules.All contacts marked in bold fulfill the formal definition of an X•••π halogen bond.The interaction classes are indicated: cen -approximately perpendicular orientation of a C-X bond and the plane of an aromatic ring, and halogen directed towards the center of the ring (see Fig. 3A); rim -as above but halogen facing the rim of the aromatic ring (Fig. 3B); π-π -parallel orientation (Fig. 3C).Res -resolution, molprotein chain ID in the PDB file, dist.-interatomic distance, q -a projection of the halogen atom on the aromatic ring (closest point), cat loop -catalytic loop., E, H). (Auffinger et al., 2004;Parisini et al., 2011;Poznanski & Shugar, 2013;Scholfield et al., 2013) and CSD (Metrangolo et al., 2005).Solvent molecules proximal to an X-bond have also been identified in other protein-ligand complexes (Beale et al., 2013), but this type of three-body interaction has to date not been listed in the IUPAC definition of a halogen bond (Desiraju et al., 2013).Moreover, a water molecule itself does not fulfill the actual IUPAC definition of an X-bond acceptor.

Halogen bond between a ligand and the backbone carbonyl oxygen
The shortest distances between a halogen atom and the carbonyl oxygen are observed for bromine (median of 3.19 Å calculated for all halogen-oxygen contacts shorter than 3.5 Å), whereas, despite the large difference in vdW radii, the distributions for chlorine and iodine donors are almost identical (medians of 3.25 and 3.28 Å, respectively).This, in view of the vdW radii (1.52, 1.75, 1.85 and 1.98 Å for O, Cl, Br and I, respectively), denotes that medians for Br and I are smaller than the sum of the corresponding vdW radii by approximately 0.2 Å, indicative of a halogen bond formation (Desiraju et al., 2013).Figure 4A shows the distributions of what we refer to as the "void" distance Δd X•••O , i.e. shortening of halogen to oxygen distance relative to the vdW radii sum, calculated for all structures for which Δd X•••O < 0.
The distribution of halogen-to-oxygen distances shows that the interaction between a chlorine and a carbonyl oxygen is substantially weaker than for bromine and iodine, i.e.Δd Cl•••O is less negative than Δd Br•••O and Δd I•••O (p MW = 0.003 and 0.03, respectively), which do not differ from each other significantly (p KS > 0.1, solid lines in Fig. 4A).Correspondingly, θ X (C-X•••Acc) and θ A (X•••Acc-C) angles, which define the geometry of an X-bond, differ qualitatively between the three halogen types (Fig. 4B, C).For iodine, the distribution of θ X is indicative of the halogen bond formation, while for chlorine and bromine it is much more broadly distributed (p MW = 3 • 10 -11 and 9 • 10 -7 ), with minimal difference between the two (p MW > 0.3; p KS < 0.1).Moreover, in contrast to iodine, for chlorine and bromine in approximately 25% of structures the θ X angle is smaller than the assumed limit of 140º, and in only about a half of all cases falls within the optimal range of 160-180º (Fig. 4B).
The sharp maximum in the θ A (X•••Acc-C) distribution for iodine, observed as the upcast in the cumulative distribution at 126º (Fig. 4C), coincides with the optimal halogen-oxygen orientation, provided the spatial distribution of electron density of oxygen in sp 2 hybridization.The same effect can also be observed for 40% of carbonyl-bromine contacts (θ A ~ 133º), and less evidently for 20% of carbonyl-chlorine interactions.It should be noted that, despite the minute differences in location, these distributions differ significantly in the shape (p KS < 0.03 for Cl vs. Br, and p KS < 0.001 for I vs. Br/Cl).Consequently, much narrower distributions are indicative of stronger halogen-carbonyl interaction, i.e. iodine is significantly better X-bond donor than bromine, while virtually no preferences are observed for chlorine.This is better visible, when the restricted set of contacts with θ X > 140º is analyzed (as evidenced by chopped lines in Fig. 4A-C), however due to decreased number of cases, the differences in distributions are less significant.
Statistically, a halogen bond between a carbonyl oxygen and iodine is stronger than that between a carbonyl and bromine, geometry of which is less restricted to values optimal for a halogen bond (θ X ≈ 160-180º and θ A ≈ 120-160º), while virtually no propensity for halogen bonding is observed for chlorine.

Short contacts between a halogen atom and a sidechain oxygen
There are no halogen-type specific differences in either distance or angle distributions of short contacts between a halogen atom and a side-chain oxygen (in all cases p KS > 0.1, Fig. 4D-F).Moreover, the angular preferences of such contacts differ, for each halogen type tested, from that made with a backbone carbonyl (in all cases p MW < 0.03, Fig. 4B, C vs. 4E, F), suggesting that the backbone carbonyl might be a stronger X-bond acceptor than a side-chain oxygen.These differences cannot be explained by heterogenic hybridization of sidechain oxygen atoms (sp 2 for Asn, Asp, Gln and Glu, and sp 3 for Ser, Thr, Tyr).However, the differences observed for distance distributions are not significant (only for chlorine p MW < 0.05, Fig. 4A vs. 4D).

Short contacts between a halogen atom and a solvent molecule
The distance distribution between a halogen atom and a proximal solvent molecule resembles trends found for a carbonyl oxygen acting as an X-bond acceptor (p KS > 0.1, Fig. 4G vs. 4A), however the distribution of the θ X (C-X•••O) angle is visibly broader (for each halogen type p MW < 0.0003, Fig. 4H vs. 4B).Contrary to the conclusions for contacts with side-chain oxygen, the distance preferences for θ X angle depend on the halogen type, and are weaker for chlorine than for the other ones (p MW = 0.02 and 0.10 for Cl vs. Br and I, respectively).It follows that, if solvent molecules were correctly identified as water, some of them may be regarded as a weak, but noticeable, X-bond acceptors (Fig. 2R-T).

Halogen vs. hydrogen bonding in protein kinase-ligand complexes
The most known example of a replacement of an H-bond by an X-bond is observed in the recurring pattern of two halogen bonds with backbone carbonyls in the hinge region, which resembles the common mode of ATP-recognition by a protein kinase (pdb1hck; Schulze-Gahmen et al., 1996, Fig. 2Q vs. 2M).Interestingly, the geometry of a halogen bond involving a backbone carbonyl of a protein kinase visibly differs from that observed for a hydrogen bond formed between a backbone carbonyl of a protein kinase and a nitrogen of ei- ther halogenated or non-halogenated ligand.The largest differences concern distance between a halogen atom and a carbonyl acceptor, which is approximately 0.3 Å larger than the nitrogen to oxygen distance of 2.87 Å observed for a hydrogen bond, which precisely corresponds to the difference in radii between N and X (see Fig. 4I).Broad distributions of the θ A angle for halogen bonds are shifted toward the idealized value of 120°, significantly differing from that observed for an H-bond (p MW < 10 -8 , Fig. 4C), clearly indicating that the geometry of an X-bond is much more restricted.Moreover, the θ X (C-X•••O) angle qualitatively differs from the θ D (C-N•••O) (see Fig. 1 for definitions).While θ X approaches the expected linear configuration, for θ D with a ligand nitrogen (found mostly in sp 2 hybridization) acting as an H-bond donor, the distribution is shifted towards 120° (Fig. 4B).
In summary, despite the general topological similarity of a halogen and hydrogen bond, geometrical requirements for both are visibly different, so they may not be equivalent when ligands are tightly packed inside the ATP-binding cavity of a protein kinase.A significant contribution of vdW interactions between atoms neighboring donor and acceptor sites, results in systematic deviation of θ A from its optimal value of 120°, expected for the sp 2 hybridization of the carbonyl oxygen.

Hydrogen bonds formed by halogenated ligands
More detailed analysis shows that the presence of a halogen atom in the ligand affects the geometry of hydrogen bonds that it forms.The effect is more pronounced for the cases, when a ligand nitrogen forms an H-bond with a protein backbone carbonyl than for those, in which backbone nitrogen acts as an H-bond donor (Fig. 4J-L and 4M-O, red and blue lines vs. black ones).Small differences are observed for N•••O distance distributions (Fig. 4J and 4M, red vs. black lines), but variations in θ D (Fig. 4K, O) and θ A (Fig. 4L, N) are even more remarkable.All these differences are indicative of enhancement of the strength of a hydrogen bond.They are statistically significant when a nitrogen of a halogenated ligand acts as an H-bond donor (Fig. 4J-L; p KW < 0.05), but not for those in which a ligand oxygen acts as an H-bond acceptor (Fig. 4M-O; p KW < 0.05 only for N•••O-C angle).The foregoing supports the trend of H-bond strengthening for halogenated ligands carrying a nitrogen H-bond donor, identified in a larger set of PDB structures (Poznanski et al., 2014), however it is worth noting that the geometry of an H-bond, in which a ligand oxygen is the acceptor, is closer to the idealized geometry than that when a ligand donates an H-bond (Fig. 4N vs. 4L, p KW < 0.01).

Electrostatic contribution to ligand binding
Structure-activity screening of halogenated benzimidazole derivative inhibitors revealed a reasonably good correlation between the inhibitory activity and the change of ligand solvent-accessible surface upon binding (Battistutta et al., 2007), which is indicative of predominance of hydrophobic interactions.However, comparison of binding modes of tetrabromobenzotriazole (TBBt) by two closely related protein kinases: CDK2 (pdb1p5e; De Moliner et al., 2003) and CK2α (pdb1j91; Battistutta et al., 2001) clearly shows that small differences in charge distribution may result in an alternative mode of TBBt binding (Fig. 2M, N).Similarly, three structurally related ligands: TBBt, tetrabromobenzimidazole (K17, TBBz, pdb2oxy; Battistutta et al., 2007) and pentabromoinda-zole (K64, pdb3kxg; Sarno et al., 2011) bind to CK2α in different orientations (Fig. 2N-P).Nonetheless, the poses for TBBt with CDK2 and TBBz with CK2α are almost identical (Fig. 2M vs. 2O).Altogether, the analysis of protein kinase complexes with halogenated benzimidazoles suggests that subtle electrostatic interactions contribute substantially to ligand binding.
We have systematically explored electrostatic contribution to ligand binding by analyzing the structure-activity relationship for a series of TBBt derivatives (Wasik et al., 2010), in which the Br at C(5) of TBBt is replaced by various groups differing in their physicochemical properties, and also for a series of nine bromobenzotriazoles representing all possible patterns of halogenation on the benzene ring (Wasik et al., 2012a).Overall, the hydrophobicity of the monoanionic form of the ligand appeared to be the principal factor governing its inhibitory activity against CK2α (Wasik et al., 2010;Wasik et al., 2012b).Furthermore, the moderate inhibitory activity exhibited by 4,5,6,7-tetramethylbenzotriazole (Zien et al., 2003), which in contrast to TBBt is in the neutral form at physiological pH (Poznanski et al., 2007), again points to a balance of electrostatic and hydrophobic interactions as an important factor contributing to CK2α inhibition.Accordingly, recent DSC-derived thermodynamic data for binding of TBBt, TBBz and their close structural analogues to CK2α (Winiewska et al., 2015a;Winiewska et al., 2015b) confirm the predominant contribution of electrostatic and hydrophobic interactions.For ligands that are mostly dissociated (i.e.pK a < 6.5), the aqueous solubility and pK a for dissociation of the triazole proton together account for more than 95% of the variance of the free energy of binding determined with the aid of Microscale Thermophoresis (Fig. 5).Three remaining, less dissociated ligands, 4-bromobenzotriazole, 5-bromobenzotriazole and 5,6-dibromobenzotriazole are most probably differently oriented in the ATP binding site, as qualitatively confirmed by tyrosine quenching (Winiewska et al., 2015a).

CONCLUSIONS
The foregoing analysis clearly shows that in the tightly packed ATP binding pocket of a protein kinase, due to observed significant differences in geometrical preferences, a pattern of H-bonds cannot a priori be replaced by X-bonds.However, the ATP-like H-bonding pattern to the hinge region may be replaced by two parallel Xbonds formed between backbone carbonyl groups and two halogen atoms attached to vicinal carbons of the benzene ring.This interaction with the hinge region (either via halogen or hydrogen bonds), when accompanied by an X-bond formed with the aromatic residue located upstream of the hinge region, may possibly be used to strengthen ligand binding or to enhance ligand selectivity.
Apart from direct effects of halogenation: increased ligand hydrophobicity and possible X-bonding, there are additional effects.These include modulation of the electron density, pK a changes of a dissociable group, or strengthening of H-bonds formed between a halogenated ligand and a protein.All these factors affect the binding mode, so that closely related ligands may bind in different orientations, as a result of a subtle balance of electrostatic, hydrogen-bonding and halogen-bonding interactions, with the hydrophobic and electrostatic components predominating.This makes computer-aided drug design for protein kinases extremely challenging.

Figure 1 .
Figure 1.The structural analogy between a halogen (A) and a hydrogen (B) bond.

Figure 2 .
Figure 2. Representative structures of protein kinases in complexes with halogenated ligands that display short contacts between a halogen atom and a protein: (A, B) contacts orthogonal to the peptide bond; (C, D) unusual interaction between a halogen atom and a proximal aspartate side-chain carboxyl; (E, F) π-π interaction between aromatic rings; (G, H) parallel orientation of the C-X bond and the aromatic ring; (I-L) halogen bonds between the ligand and an aromatic ring (optimal perpendicular configuration); (M, N) alternative binding modes of TBBt by two closely related protein kinases, and (N-P) closely related halogenated ligands that substantially differ in their location at the ATP-binding site of protein kinase CK2α; (Q) hydrogen bonding pattern with ATP; (R-T) short contacts between a halogen atom and a solvent molecule.The original pdb codes and protein acronyms are denoted for each structure.The corresponding residues in CK2α are indicated in brackets.The figure includes EDS generated 2F o -F c (grey) and F o -F c (red -negative, green -positive) electron density maps contoured at given rmsd levels (inaccessible for 1j91).The halogen atoms and their short contacts are colored green and the hydrogen bonds and metal-ligand interactions grey.The glycine-rich loop, hinge region, catalytic loop and DFG motif are denoted in magenta, yellow, red and blue, respectively.Panels M and S are sligtly rotated with respect to the others and some side chains are omitted for clarity.

Figure 3 .
Figure 3. Schematic representation of the perpendicular "over the center" (A), "over the rim" (B) and parallel (C) orientations of bromomethane relative to a proximal benzene aromatic ring.

Figure 4 .
Figure 4. Cumulative distributions of the parameters describing the geometry of an interaction between a halogenated ligand and a backbone carbonyl (A-C, I), side-chain oxygen (D-F) and a water molecule (G-H), determined separately for each halogen type.As a reference, the distributions for an H-bond between a non-halogenated ligand and a backbone carbonyl are presented as black lines in (B, C, I), and additionally shown for non-halogenated (HL), fluorinated (FL) and otherwise halogenated (XL) ligands acting either as donors (J-L) or acceptors (M-O) of an H-bond formed with a protein backbone (only those between nitrogen and oxygen).Chopped lines in (A-H) represent cumulative distributions obtained for θ X restricted to the range of 140-180º, indicative of X-bond formation, which is denoted by vertical arrows (B, E, H).

Figure 5 .
Figure 5. Correlation between the aqueous solubility (Cw) and pKa for dissociation with binding affinity to protein kinase CK2α for a series of nine benzotriazoles halogenated on the benzene ring.Data for three ligands (open circles), pKa for which is close to the physiological pH, disagree with the general trend.log (C ) w

Table 1 . Short intermolecular contacts between the halogen atom of a ligand and various types of potential X-bond acceptors identi- fied in 320 PDB structures of protein kinases with halogenated ligands.
The second numbers reported in each cell represent values determined for X••Acc interactions with C-X••Acc angle > 140°.The identity of the solvent molecule cannot be deduced with 100% certainty from X-ray crystallographic data , and cAMP-dependent protein kinase with H-89 (pdb3vqh; Pflug et al., 2012) or CCT196539 (pdb4c37; Couty et *