AN ABSTRACT OF THE DISSERTATION OF
Andrea Regier Voth for the degree of Doctor of Philosophy in Biochemistry and
Biophysics presented on September 5, 2007.
Title: Macromolecular Halogen Bonds.
Abstract approved:__________________________________________________
Pui Shing Ho
The halogen bond is a non-covalent, stabilizing interaction analogous to a
hydrogen bond in which an anisotropically polarized halogen atom interacts
electrostatically with a Lewis base. Until very recently, the ability of halogens to form
these stabilizing interactions in biological macromolecules was all but unknown, but
examples of halogen bonding have now been observed in nucleic acids as well as
protein complexes with hormones, drugs and inhibitors. The lack of recognition of
and information about these interactions, however, hinders their utilization in the
design of biological interactions. This thesis deals with work done to elucidate the
capabilities and properties of halogen bonds in the context of biological
macromolecules.
Protein kinases are an important and well-studied class of drug targets for
diseases such as cancer. Despite the prevalence of halogenated inhibitors and drugs
targeted to protein kinases, however, halogen bonds have not generally been
recognized and therefore utilized in the design of ligand binding interactions. The
number of occurrences of halogen bonds between protein kinases and inhibitors
observed in the crystal structures in the Protein Data Bank indicate the potential utility
of the interaction in inhibitor and drug design. Further, their structures suggest a
strategy for targeting halogen bond interaction sites by demonstrating that halogen
bond acceptors offering concave surfaces present a more favorable profile to potential
halogen bond donors.
Halogen bonds are also able to direct the conformation of a biological
molecule. In several competition experiments, halogen bonds were shown to outcompete classical hydrogen bonds to stabilize and direct the conformation of a DNA
Holliday junction. For bromine X-bonds, the energy of stabilization was estimated to
be 2 to 5 kcal/mol more than a classic hydrogen bond. The relative stabilization
provided by interactions with fluorine, bromine, and iodine indicated that polarizable
halogens (such as iodine and bromine) form highly stabilizing halogen bonds, whereas
fluorine does not. The strengths of these interactions follow the order of halogen
polarization (F < Br < I) and specify a range of interaction energies available to the
halogen bond in a macromolecular context. Together, these observations of halogen
bond occurrence and stabilization suggest that halogen bonds can be a powerful tool
for the design of macromolecular interactions.
 Copyright by Andrea Regier Voth
September 5, 2007
All Rights Reserved
Macromolecular Halogen Bonds
by
Andrea Regier Voth
A DISSERTATION
submitted to
Oregon State University
in partial fulfillment of
the requirements for the
degree of
Doctor of Philosophy
Presented September 5, 2007
Commencement June 2008
Doctor of Philosophy dissertation of Andrea Regier Voth presented of September 5,
2007.
APPROVED:
Major Professor, representing Biochemistry and Biophysics
Chair of the Department of Biochemistry and Biophysics
Dean of the Graduate School
I understand that my dissertation with become part of the permanent collection of
Oregon State University libraries. My signature below authorizes release of my
dissertation to any reader upon request.
Andrea Regier Voth, Author
ACKNOWLEDGEMENTS
First, I owe many thanks to my advisor, Dr. P. Shing Ho, for his patient
consideration of my constant stream of questions and misunderstandings. I cannot
imagine a better environment for encouraging discussion and discovery. Also in the
Ho lab, to Dr. Frank Hays (who first informed me that I asked too many questions)
and Dr. Jeff Watson, thank you for your very different and complementary styles of
teaching crystallography. Now that I have taught others many of the things that you
taught me, I can truly appreciate the time and effort you spent on me. And to my
fellow graduate student Trish Khuu, thanks for making life in our computer dungeon
bearable.
Elsewhere in the department, thanks to Dr. P. Andrew Karplus for his
dedicated teaching and ever-present interest in my research and to Drs. Elisar Barbar
and Mike Schimerlik for advice on science and life. Dr. Rick Faber probably deserves
his own page just to thank him for keeping the diffractometer running, and for
answering my questions even when I described them as dumb. To all the graduate
students in my entering class, without whom I wouldn’t have lasted a year, and those
that came before and after me, thank you for making the department a community.
Finally, to my family, old and new, I owe a deep debt for your love and
support during this exciting and challenging time. To my husband, Peter, I give my
thanks for taking this journey together with me. I know that it is just the first of many.
CONTRIBUTION OF AUTHORS
P. Shing Ho was involved in the design, analysis and writing of each experiment and
manuscript. Frank A. Hays played a consultative role in the crystallization, solution,
and refinement of the structures described in Chapter 3.
TABLE OF CONTENTS
Page
Introduction ......................................................................................................... 1
The role of halogen bonding in inhibitor recognition and binding by protein
kinases ................................................................................................................ 9
Summary ............................................................................................... 10
Introduction ........................................................................................... 10
Protein Kinase-Inhibitor Complexes ...................................................... 19
Analysis / Discussion ............................................................................ 40
Conclusions ........................................................................................... 48
Acknowledgements ............................................................................... 50
Directing macromolecular conformation through halogen bonds ....................... 52
Summary ............................................................................................... 53
Introduction ........................................................................................... 53
Results .................................................................................................. 58
Discussion ............................................................................................. 65
Materials and Methods .......................................................................... 70
Acknowledgements ............................................................................... 73
The effect of polarizability on the energy of macromolecular halogen bonds ..... 74
Summary ............................................................................................... 75
Introduction ........................................................................................... 76
Materials and Methods .......................................................................... 78
Results .................................................................................................. 84
TABLE OF CONTENTS (Continued)
Page
Discussion ............................................................................................. 96
Acknowledgements ............................................................................... 99
Conclusion and Discussion ............................................................................. 100
Bibliography ................................................................................................... 104
LIST OF FIGURES
Figure
Page
1. Anisotropic polarization of halogens in model organic halides ................ 3
2. Similarities between hydrogen and halogen bonds ................................... 4
3. The structure of the DNA sequence d(CCAGTACbrUGG) ....................... 5
4. Chemical structures of eleven different kinase inhibitors observed
to form halogen bonds to the protein kinases JNK3, CDK2, CK2,
and MEK1&2 ........................................................................................ 12
5. Structure of the unphosphorylated human JNK3 (PDB code
1JNK) as a representative protein kinase catalytic subunit ..................... 14
6. Comparison of hydrogen bond and halogen bond geometries ................ 16
7. Anisotropic electron distribution of halogen substituents of an
aromatic inhibitor .................................................................................. 17
8. Accessible backbone oxygens at the CK2 ATP binding site ................... 49
9. Structure of the stacked-X DNA Holliday junction ................................ 55
10. Assay for competing X- against H-bonds .............................................. 57
11. Geometries of X-bonds in Br2J and Br1J ................................................ 62
12. Electron densities at the N4 nucleotide positions of the outside
(continuous) and inside (crossing) strands that complement the
N7 nucleotide forming an H- or X-bond in the H2J junction ................... 63
13. Estimating the bromine occupancy at the N7 nucleotide on the
outside strand (H-isomer) of Br1J (labeled as U17, black
triangles) ............................................................................................... 66
14. Electrostatic potentials from ab initio calculations of the bromine
halogen bonds in Br2J and Br1J show the characteristic
anisotropic distribution of charges on the bromine atoms........................ 68
15. Thermodynamic cycle to estimate the free energies of the Xrelative to H-bonds ................................................................................ 69
LIST OF FIGURES (Continued)
Figure
Page
16. DNA Holliday junction numbering and stabilization ............................. 82
17. Schematic of the competition assay using the DNA Holliday
junction ................................................................................................. 86
18. 5σ omit density for the iodine off the C5 of the crossover N7
uracil in the I2J structure indicates that the junction is in the
X-isomer and that the iodine is X-bonded to a phosphate oxygen ........... 88
19. 5σ omit density for the iodine off the C5 of the crossover N7
uracil in the I1J structure indicates that the junction is in the
X-isomer and that the iodine is X-bonded to a phosphate oxygen ........... 90
20. Electron density map around the outside N4 in junction I1J
shows some indication of a missing extracyclic amine group
off C2 .................................................................................................... 92
21. Electron density maps for the crossover (top) and outside
(bottom) N4 bases of the F2J junction ..................................................... 95
22. Overlay of the crossover nucleotides of H2J, F2J, I2J, and I1J ................. 97
LIST OF TABLES
Table
Page
1. Summary of inhibitors that halogen bond to protein kinases ...................... 21
2. Summary of halogen bond interactions between protein kinases
and inhibitors ............................................................................................ 23
3. Constructs and sequences that compete halogen bonds against
hydrogen bonds in DNA junctions ............................................................. 59
4. Crystallographic and geometric parameters for the Holliday
junction constructs Br2J, H2J, and Br1J (see Table 3 for sequences) ........... 60
5. Constructs and sequences that compete fluorine and iodine
X-bonds against H-bonds in DNA junctions .............................................. 79
6. Crystallographic and geometric parameters for the Holliday
junction constructs H2J, F2J, I1J and I2J (for sequences see
Table 5) ..................................................................................................... 81
Macromolecular Halogen Bonds
Chapter 1
Introduction
As early as the 19th century, chemists had observed the formation of stable
complexes between halogens and electron-rich compounds (for example, the stable
complexes between I2 and ammonia (Guthries 1863)). With the development of
spectroscopic and crystallographic techniques in the mid-20th century, such complexes
could be characterized with atomic detail. The Nobel prize winning chemist Odd
Hassel described, for the first time, these complexes as resulting from interatomic
charge-transfer bonding (Hassel 1970). The crystal structure of the 1:1 bromine to
1,4-dioxane adduct showed, for example, “endlessly repeating chains of alternating
dioxane and bromine molecules” with an O to Br distance (2.7 Å) less than the sum of
the two atoms Van der Waals radii, but longer than the sum of their covalent radii
(Hassel and Hvoslef 1954). Despite some controversy over the energetic importance
of the actual charge-transfer component in these charge transfer complexes, the
complexes were generally understood to involve weak electrostatic interactions
involving dispersion and dipole forces (Foster 1969). In the late 1970’s, the term
halogen bonding (X-bonding) was coined (Dumas et al. 1978), and has largely
replaced the earlier “charge transfer” descriptor. Through a variety of quantum
2
mechanical and database studies in the last 30 years (Ramasubbu et al. 1986;
Lommerse et al. 1996), it has become clear that the most important energetic
contributor to the X-bond is the electrostatic interaction between a polarizable halogen
(dihalogen or organic halide) and an electron donor (Lewis base). This electrostatic
interaction results from the anisotropic polarization of the halogen along its X—C
bond (in an organic halide—see Figure 1), which creates an electropositive “crown”
that interacts favorably with the negative electrostatic potential of an electronegative
Lewis base.
The name “halogen bond” invokes this interaction’s similarity to the betterknown hydrogen bond (H-bond) (Metrangolo et al. 2005). Indeed, both are
noncovalent, primarily electrostatic interactions involving a Lewis base acceptor atom
(which we will refer to as the H-bond or X-bond acceptor), though they differ in their
donor atoms. X- and H-bonds have similar geometries (Figure 2) and energies
(estimated to be ~5 kcal/mol for both H-bonds (Baldwin 2003) and X-bonds (Corradi
et al. 2000)). These similarities with the H-bond have made X-bonds a useful tool for
chemists designing intermolecular interactions using small molecules, including the
engineering of crystals and other self-assembling systems for materials research
(Metrangolo et al. 2005).
In 2003, a previous graduate student in Dr. Ho’s lab, Frank Hays, first
happened upon an X-bond interaction by chance. While solving the structure of the
sequence d(CCAGTACbrUGG), where the 5-bromouracil (brU) served as a thymine
analog to help phase the crystallographic data, he observed a 3.0 Å interaction
between the bromine and a phosphate oxygen (Figure 3) (Hays et al. 2003).
3
Figure 1. Anisotropic polarization of halogens in model organic halides (Auffinger et
al. 2004). Ab initio quantum mechanical calculations (calculated by DFT applying
the B3LYP function and the 3-21G* basis set (Schmidt et al. 1993)) show
electropositive (blue), electronegative (red), or neutral (green) potentials. Viewing
each molecule along its C—X bond, the halogens develop electropositive potentials as
they become more polarizable (left to right: F < Cl < Br < I), and as they are bound to
aromatic and differently substituted organic compounds (top to bottom: methane,
uracil, and cytosine). From these calculations, fluorine is seen to be the only halogen
lacking an electropositive crown.
4
Figure 2. Similarities between hydrogen and halogen bonds (figure from (Voth and
Ho 2007)). Schematic hydrogen bond (H-bond) on left and halogen bond (X-bond) on
right. Both the types of atoms involved and their geometries are similar. Θ1 tends to
be linear to maximize the interaction of the acceptor atom with either the hydrogen or
the electropositive crown of the halogen. Θ2 can span a larger range, indicative of the
general electronegative nature of the acceptor atoms (X-bond angular dependencies
from (Lommerse et al. 1996) and (Auffinger et al. 2004)).
5
Figure 3. The structure of the DNA sequence d(CCAGTACbrUGG) (Hays et al.
2003). This sequence crystallized as a Holliday junction with the two bromines (at
N8) on the crossover strands of the junction making stabilizing X-bonding interactions
with phosphate oxygens (electron density map shown in the inset at right). The two
outer strands also have bromines at N8, but they are not in proximity to any Lewis
bases (other than solvent) with which to X-bond.
6
Not only was this 0.4 Å closer than the sum of the two atoms’ van der Waals radii, it
was also in an analogous position to a previously observed H-bond, and seemed to be
stabilizing this sequence in a four-stranded Holliday junction conformation. This
prompted an investigation into the ability of bromines and oxygens to form stabilizing
interactions, and led Dr. Ho to begin working to characterize X-bonds in biological
molecules, where very little work had been done.
A review of the Protein Data Bank (PDB) in 2004 found 113 X-bond
interactions in 66 different protein-ligand and 6 different nucleic acid structures
(Auffinger et al. 2004). In nucleic acid structures, the halogens were usually
incorporated into the molecules via halogenated bases, and interacted primarily with
phosphate oxygens, similar to the interactions seen by Hays, et al. (Hays et al. 2003).
In the protein structures, X-bond interactions most often occurred between halogens
on ligands (Cl—27%, Br—34%, and I—39%) and the backbone carbonyl oxygens of
the polypeptide chain, although interactions with hydroxyl or negatively charged
carboxylic acid sidechains were also sometimes observed. Given the large number of
halogenated drugs and inhibitors (by some estimates, 50% of compounds in highthroughput drug screens are halogenated), it is not surprising that so many X-bonds
were seen in protein-ligand complexes.
Halogens also occur naturally in biological systems. The best-known
examples of this are probably the iodinated thyroid hormones, such as thyroxine. The
crystal structure of the transport protein transthyretin complexed with thyroxine
showed multiple iodine to carbonyl oxygen X-bonds involved in hormone binding
(Muziol et al. 2001). Alternatively, the enzyme eosinophil peroxidase selectively
7
creates brominating oxidants and has been shown to brominate tyrosine residues
incorporated in proteins (Wu et al. 2000), free nucleotides, and double-stranded DNA
(Shen et al. 2001) in vitro. Chlorinating oxidants are also created by the enzyme
myleoperoxidase, and elevated chlorotyrosine levels are associated with chronic
respiratory disease in infants (Buss et al. 2003), while increases in bromotyrosines are
seen in allergen-induced asthma (Wu et al. 2000).
The use of X-bonding to affect intermolecular binding and recognition has so
far been almost entirely serendipitous. In looking for inhibitors to drug-resistant HIV1 reverse transcriptase (RT), for example, researchers came upon a class of iodinated
inhibitors (3-iodo-4-aryloxypyridinones, or IOPY inhibitors) that showed
subnanomolar inhibition against both wildtype and several drug-resistant RT strains.
The crystal structures of these inhibitors with RT showed that this was due to an
iodine interaction with a backbone carbonyl oxygen, whereas previous inhibitors
showed interactions with the sidechain at the same position, leaving them open to
mutation induced drug resistance (Himmel et al. 2005). With this observation, these
research scientists are now working to optimize this class of inhibitors to work against
a range of drug-resistant strains. Although halogens are not generally prevalent atoms
in biological molecules, it is clear that they can and do play important roles in some
biological systems. As we seek to understand and modulate these systems, our
understanding of the role and capabilities of the X-bond will be key to our success in
applying them for biomolecular design.
In this thesis, I describe several studies undertaken to gain insight into the role
X-bonds can play in macromolecular recognition and interaction. The questions that
8
we attempt to address with these studies are: how prevalent are X-bonds in biological
interactions where halogens are present, what types of macromolecular atoms act as
X-bond acceptors and how we might target them, how stabilizing are X-bonds in
macromolecular systems, and how does X-bond stabilization change with different Xbond donors? First, in Chapter 2 we survey the instances of X-bonds in complexes
between ligands and protein kinases, a leading class of drug targets. Despite the fact
that only a handful of these interactions were intentionally designed into the inhibitors,
we see the prevalence of both halogens and X-bonds in inhibitor binding, and we
attempt to draw some lessons about how to target sites for X-bonding interactions.
Then, in Chapter 3 we describe the first experimental estimate of an X-bond’s strength
in a biological molecule, showing that a bromine X-bond can out-compete an H-bond
to control the conformation of a DNA Holliday junction. Finally, in Chapter 4 we
extend the our study from Chapter 3 to examine the strength of the iodine X-bond and
investigate the possibility of a fluorine X-bond in order to observe the range of
stabilization energies afforded by the X-bond in a biological system. From these
studies we conclude that X-bonds are adaptable to a variety of environments and can
be a valuable tool in the design of macromolecular binding and assembly.
9
Chapter 2
The Role of Halogen Bonding in Inhibitor Recognition and Binding by
Protein Kinases
Andrea Regier Voth and P. Shing Ho
Published in Current Topics in Medicinal Chemistry,
Bentham Science Publishers Ltd.
2007, 7 (14), 1336-1348
10
Summary
Halogen bonds are short-range molecular interactions that are analogous to
classical hydrogen bonds, except that a polarized halogen replaces the hydrogen as the
acid in the Lewis acid/base pair. Such interactions occur regularly in the structures of
many ligand-protein complexes, but have only recently been recognized in biological
systems as a distinct class with well-defined physical characteristics. In this review,
we survey twelve of the single crystal structures of protein kinase complexes with
halogenated ligands in order to characterize the role of halogen bonds in conferring
specificity and affinity for halogenated inhibitors in this important class of enzymes.
From this survey, we attempt to identify the properties of halogen bonds that can be
generally applied to bottom-up strategies for designing inhibitors for this and other
enzyme targets.
Introduction
Halogens have many unique chemical properties that make them useful in
designing protein inhibitors and drugs. As good leaving groups, they are important
substituents in the synthesis of organic intermediates. Their electronic withdrawing
properties have been used to alter the electronic properties of molecules. Furthermore,
they have been used to substitute for reactive groups to inhibit the biodegradation of
highly reactive molecules (Hester et al. 2001). This has led to the wide-spread use of
11
halogens for structure based drug design—for example, in a sample of twelve protein
kinase small-molecule inhibitors either in the clinic or in clinical trials in 2004, three
were halogenated (Noble et al. 2004). In addition, approximately half of the
molecules commonly used in high-throughput screening are halogenated. Despite the
prevalence of halogens, however, the precise chemical and structural basis for their
contribution to drug-protein affinity and recognition has, to date, been incompletely
understood and thus has not been fully exploited for rational drug design. In
particular, the role of the recently re-discovered interaction known as the halogen bond
has been largely overlooked. In this review, we will explore the role of this class of
interaction in conferring specificity and affinity to the recognition and binding of
halogenated inhibitors by protein kinases. We start by reviewing the general structural
features of protein kinases and introducing halogen bonds as a distinct molecular
interaction in biomolecular systems. This will be followed by a summary of several
kinase structures in complex with specific inhibitors that form halogen bonds (Figure
4). In these analyses, we identify the particular molecular interactions that are
involved in recognition of these compounds in an attempt to parse out the contribution
of the halogen to the overall effectiveness of the inhibitor.
Structure of Protein Kinases
Kinase enzymes phosphorylate hydroxyl-containing amino acids (usually serine,
threonine, or tyrosine) under sequence-specific contexts. Despite this seemingly
simple function, protein kinases are extremely important to the cell (as evidenced by
the 518 protein kinases that are encoded in the human genome) and are involved in
12
Figure 4. Chemical structures of eleven different kinase inhibitors observed to form
halogen bonds to the protein kinases JNK3, CDK2, CK2, and MEK1&2. Each box
contains inhibitors that bind to the same protein kinase, in the order in which they are
discussed in the text. The first nine are ATP-competitive inhibitors and are oriented
with respect to the kinase hinge region (represented by the curve to the left of each
inhibitor), whereas the last two bind in an adjacent pocket. Polar interactions
(hydrogen or halogen bonds) to the hinge region are shown with a light dash where
applicable. Figure adapted from (Noble et al. 2004).
13
regulating a large number of essential cellular processes, from metabolism and cell
cycle progression to differentiation and apoptosis (Manning et al. 2002). This has
made protein kinases a leading drug target, with applications as immunosuppressants,
and to treat chronic inflammatory diseases, neurodegenerative disorders, viral
infections, unicellular parasites, cardiovascular disease, and, most prominently, cancer
(Cohen 2002; Knockaert et al. 2002; Bogoyevitch 2006).
Given the large variety of cellular functions associated with this class of enzyme,
the catalytic domains of protein kinases, including those discussed here, are
remarkably structurally conserved. The overall structure can be segregated into two
distinct “lobes”—the smaller N-terminal lobe contains a β-sheet and at least one αhelix, while the larger, C-terminal lobe is mostly α-helical (Noble et al. 2004) (see
Figure 5 for a representative structure). The cleft separating the two lobes creates the
ATP-binding pocket, which is invariant among protein kinases. The N- and Cterminal lobes are connected primarily by the hinge region (Figure 5), which also
forms one side of the ATP-binding pocket. Within this pocket, the adenine of the ATP
makes two hydrogen bonds with backbone atoms, and interactions within the hinge
have been also been shown to be very important for inhibitor binding to protein
kinases. Surrounding the highly conserved ATP-binding pocket are a set of smaller
pockets that vary significantly among the kinases, which have served as targets for the
design of protein selective inhibitors and drugs (Noble et al. 2004).
Halogen Bonds
Short, stabilizing interactions of halogens with organic molecules, originally
14
Figure 5. Structure of the unphosphorylated human JNK3 (PDB code 1JNK) as a
representative protein kinase catalytic subunit. The protein backbone is traced to
distinguish the N-terminal lobe (top) and the C-terminal lobe (bottom). The ATPbinding site is sandwiched between the lobes (model of AMP-PNP, shown in black, is
bound in this structure) and the protein hinge region (indicated by the arrow) connects
the two lobes. Figure adapted from (Noble et al. 2004).
15
called “charge-transfer” bonds, were first fully characterized by chemists in the 1950’s
(Metrangolo and Resnati 2001; Metrangolo et al. 2005). These interactions were
renamed “halogen bonds” in the 1980’s to emphasize their similarities to hydrogen
bonds. Both are non-covalent, primarily electrostatic interactions involving similar
atoms, with the hydrogen bond accepting atoms serving an analogous function as
general Lewis bases in the halogen bond. This similarity between interactions extends
to their geometries (Figure 6), and their approximate energies (estimated at ~5
kcal/mol for hydrogen bonds (Baldwin 2003) and, at least in one study, for halogen
bonds (Corradi et al. 2000)). The primary difference is that the halogen bond results
from an anisotropic polarization of a halogen’s electron distribution (Figure 7), rather
than from differences in electronegativity in the D-H bond (D=O, N, S, C). This
polarization creates an electropositive crown (which serves as the Lewis acid in the
interaction) at the tip of the halogen that can vary in size and intensity depending on
the type of halogen and its environment. More polarizable halogens generally have
larger and more electropositive crowns than less polarizable halogens (Cl < Br < I),
while F never develops such a crown and, therefore, does not participate in halogen
bonds. The covalent environment of the halogen also has a significant effect on the
strength of halogen bonds—the intensity of the electropositive crown of a halogen is
increased, for example, in aromatic molecules (Auffinger et al. 2004), consistent with
delocalized π-electron systems withdrawing electrons from the halogen and thus
exaggerating the anisotropy of the electron distribution. To extend the analogy
between halogen and hydrogen bonds even further, we will refer to the halogen here as
the halogen bond donor (equivalent to the hydrogen bond donor) and the Lewis base
16
Figure 6. Comparison of hydrogen bond and halogen bond geometries. There are
analogies between the atoms involved in and the geometric parameters defining
hydrogen and halogen bonds (reviewed in (McDonald and Thornton 1994) and
(Auffinger et al. 2004)). The hydrogen bond donors (D) are electronegative atoms that
polarize the D-H bond, while the halogens (X) that serve as halogen bond donors are
themselves polarized along the C-X bond. The types of atoms that serve as hydrogen
bond acceptors (A) are Lewis bases that also serve as halogen bond acceptors.
Finally, the geometries that define a good hydrogen bond—short distances, linear
alignment of the acceptors towards the donors (Θ1) and alignment of the donor
towards the nonbonding electrons of the acceptor (Θ2)—are similar between the two
interactions.
17
Figure 7. Anisotropic electron distribution of halogen substituents of an aromatic
inhibitor. a. Molecular structure of the CDK2 and CK2 inhibitor TBB with Br(10)
replaced with a chlorine to illustrate the difference in the degree of polarizability
between the halogens. b. Ab initio electrostatic potential surface of the modified TBB
from part a. The bromine (extended to the far right) clearly shows a much more
electropositive crown than the chlorine (pointing out towards the reader), but both are
sufficiently polarized to form a stabilizing halogen bond. Ab initio quantum
mechanical calculations were performed and the results rendered with Spartan ’02 for
Mac running Gaussian (the electrostatic potential scale is given in kcal/mole).
18
partner as the halogen bond acceptor (equivalent to the hydrogen bond acceptor).
Although the properties of halogen bonds are now well characterized in small
molecules, this interaction has been largely unrecognized in the lexicon of biology,
until a recent survey emphasized the prevalence of the interaction in biological
systems (Auffinger et al. 2004). This survey further showed that biological halogen
bonds conform, in general, to the ideal geometry defined in small molecules (Figure
6), although there are features that appear to be specific to biological systems. The
approach of the acceptor relative to the carbon-halogen bond (Θ1) is primarily linear
(similar to hydrogen bonds), and the approach of the halogen relative to the acceptor
(Θ2) clusters around 120˚ (consistent with an interaction to the lone-pair electrons of
the Lewis base acceptor). The Θ2 angle for larger halogens interacting with Lewis
bases that are part of the peptide backbone, however, appears to have a secondary
cluster around ~90˚, which suggests an interaction with π-electron systems, often from
peptide amides. This was borne out by a dihedral angle measurement showing that
such interactions orient the halogen perpendicular to the peptide plane (Auffinger et al.
2004). The greater variation in the geometry of halogen bonds in biological molecules
compared to that seen in small molecule systems (Lommerse et al. 1996) is not
unexpected, considering the increased complexity of biomolecules.
In organic and inorganic small molecule systems, halogen bonds have been
better characterized and have been used extensively in the last twenty-some years to
design self-assembling materials and crystals (Metrangolo et al. 2005). The utility of
halogen bonds in engineering intermolecular interactions, therefore, has already been
established in chemistry. Thus, a more complete understanding of halogen bonding
19
between proteins and inhibitors has the potential to become an invaluable tool for
biomolecular engineering, including structure based drug design. Protein kinases, as
one of the largest classes of drug targets, provide an exciting opportunity to examine
the role that halogen bonding plays in inhibitor recognition and binding. A detailed
analysis of the halogen bonding interactions in these complexes will help us begin to
understand the molecular basis for the specificity and affinities of the inhibitors and, in
the process, initiate the development of halogen bonds as a tool for inhibitor and drug
design.
Protein Kinase—Inhibitor Complexes
The Protein Data Bank (PDB) currently contains sixty structures of protein
kinases complexes with halogenated small molecule inhibitors. Sixteen, or over 25 %,
of these complexes show examples of halogen bonds, as defined by a halogen
approaching an acceptor at a distance closer than the sum of the atoms’ respective van
der Waals radii. Here we will discuss in detail twelve of these sixteen structures (the
four structures not discussed here are PDB codes 2G01 (Liu et al. 2006), 2F2C (Lu
and Schulze-Gahmen 2006), 2BHE (Jautelat et al. 2005), and 2B54 (Markwalder et al.
2004)). These twelve structures represent four different protein kinases, eleven
different inhibitors (nine that are ATP-competitive molecules and two that are not),
and twenty different halogen bonds. The structures of the inhibitors discussed here are
shown in Figure 4, and their properties are summarized in Tables 1 and 2. Although
20
not large, this dataset provides a glimpse of how halogen bonds participate in ligand
specificity and binding, and how they can potentially be applied in rational drug
design. The structures and molecular interactions of each of these four kinases with
each of their respective halogen bonded inhibitors are summarized below.
cJun terminal kinase 3
cJun terminal kinase 3 (JNK3) is a neuronal-specific MAP kinase that is a
potential drug target for the treatment of neurological disorders (Scapin et al. 2003).
As with all MAP kinases, JNK3’s own Ser/Thr kinase activity is triggered by the
phosphorylation of a threonine and a tyrosine on its activation loop, which is required
to form a fully functional active site (Cobb and Goldsmith 1995; Xie et al. 1998). Two
structures of JNK3 that include halogen bonds to inhibitors were found in the PDB. In
both of these structures, the proteins are truncated to the catalytically active domains.
The two inhibitors are nearly identical, both containing two chlorine atoms, except
that the cyclopropyl group of compound 1 is replaced by a cyclohexyl group in
compound 2.
JNK3-compound 1 complex
Compound 1 (cyclopropyl-{4-[5-(3,4-dichlorophenyl)-2-piperidin- 4-yl-3propyl-3H-imidazol-4-yl]-pyrimidin- 2-yl}amine) is a member of the diarylimidazole
family of MAP kinase inhibitors that bind to p38, but not to extracellular regulated
kinase (ERK). Compound 1 has an IC50 for JNK3 of 7 nM but exhibits no JNK
isoform selectivity. The structure of N-terminal truncated (residues 45 through 400),
1P5E
1J91
1ZOE
TBB
TBB
K25
PD318088
PD334581
MEK2
1S9I
1S9J
1ZOH
1FVT
Oxindol16
K44
1CKP
Purvalanol B
1ZOG
1H1R
NU6086
K37
1PMQ
1PMN
Compound 1
Compound 2
PDB
ID
Inhibitor
MEK1
CK2
CDK2
JNK3
Protein
kinase
Not reported
Not reported
740 nM
250 nM
1
1
3
3
2
1
0.5 – 1.6 µMf
140 nM
4
15.6 µMf
2
1 (≈65%)
5
4
0
(1)
(1)
1
(1)g
4
3
1 (≈50%)
1
2
2
# of Hbondsb
1
1
# of Xbondsa
60 nM
6 – 9 µM
2.3 µM
1 nM
7 nM
IC50
Table 1. Summary of inhibitors that halogen bond to protein kinases.
259 Å2
310 Å2
186 Å2
195 Å2
311 Å2e
273 Å2e
227 Å2
371 Å2e
323 Å2e
186 Å2
236 Å2
247 Å2
187 Å2
142 Å2
244 Å2
285 Å2
200 Å2
289 Å2
Inhibitor incompletely
modeled.
359 Å2e
337 Å2
367 Å2
368 Å2e
∆ ASAd
inhibitor
∆ ASAc
protein
21
These IC50 values were obtained using 100 µM ATP.
For structures including ligands other than the inhibitor, those ligands were removed for the ASA calculations.
The number of direct hydrogen bonds between the inhibitor and protein are listed, with those in parentheses
indicating the number of hydrogen bonds that are not direct, but are water-mediated.
g
f
e
∆ASA inhibitor is the difference in the accessible surface area calculated for the inhibitor in the absence and
the presence of the protein.
d
∆ASA protein is the difference in the accessible surface area calculated for the protein in the absence and the
presence of the inhibitor.
c
H-bond is an abbreviation for hydrogen bond.
X-bond is an abbreviation for halogen bond.
b
a
Table 1 (Continued)
22
CK2
CDK2
JNK3
Protein kinase
1CKP
1FVT
Purvalanol B
Oxindol16
K25
1ZOE
1J91
1H1R
NU6086
TBB
1PMQ
Compound 2
1P5E
1PMN
Compound 1
TBB
PDB ID
Inhibitor
3.25 Å
3.05 Å
3.07 Å
3.22 Å
2.90 Å
2.88 Å
3.01 Å
3.05 Å
3.47 Å
3.67 Å
3.24 Å
2.99 Å
3.30 Å
Asp86 OD1e
& OD2
Asp146 OD1
& OD2
Leu83 O – A
–C
Glu81 O – A
–C
Phe80 ringf – A
–C
Ile10 O – A
Arg47 NEg – A
Arg47 O – B
Cl1 (A≈65%)
Br11
Br12
Br13
Br10
Br13 – A
Br13 – B
Val116 O
Glu114 O
Br9
Br1
3.43 Å
3.24 Å
3.16 Å
2.83 Å
Asp86 OD2c – A
– Cd
Cl1 (B≈50%)
Cl1 (A≈50%)
Br4
2.96 Å
2.83 Å
Ala91 Oa
Ala91 O
Length
Lewis base
Cl45
Cl45
Halogenb
Table 2. Summary of halogen bond interactions between protein kinases and inhibitors.
139˚
180˚
165˚
164˚
144˚
154˚
180˚
169˚
172˚
165˚
180˚
127˚
146˚
145˚
134˚
158˚
140˚
160˚
159˚
Θ1
150˚
129˚
100˚
106˚
110˚
74˚
68˚
155˚
151˚
132˚
121˚
87˚
81˚
70˚
79˚
102˚
110˚
90˚
96˚
Θ2
23
PD318088
PD334581
MEK2
3.20 Å
0.20 Å
3.17 Å
0.04 Å
Bromine Halogen Bonds: Average
Standard Deviation
Iodine Halogen Bonds: Average
Standard Deviation
3.25 Å
3.16 Å
Val131 O – A
–B
2.98 Å
0.13 Å
I1
3.13 Å
3.16 Å
3.18 Å
2.89 Å
3.53 Å
3.23 Å
3.11 Å
2.98 Å
3.27 Å
Length
Val127 O
Asp175 OD1
Br10
I1
Glu114 O
Val116 O
Br13
Br12
Phe113 ring
Br13 (B≈40%)
Glu114 O
Glu114 O
Val116 O
Val116 O
Br10 (A≈60%)
Br11 (B≈40%)
Br11 (A≈60%)
Br12 (B≈40%)
Lewis base
Halogenb
Chlorine Halogen Bonds: Averageh
Standard Deviation
1S9I
1S9J
1ZOH
1ZOG
K37
K44
PDB ID
Inhibitor
MEK1
CK2
Protein kinase
Table 2 (Continued)
180˚
0˚
158˚
17˚
152˚
10˚
180˚
180˚
180˚
133˚
145˚
180˚
154˚
144˚
145˚
180˚
180˚
Θ1
125˚
4˚
122˚
31˚
92˚
13˚
123˚
118˚
129˚
128˚
164˚
129˚
69˚
162˚
164˚
130˚
129˚
Θ2
24
O = backbone carbonyl oxygen
A letter in parenthesis indicates an alternate conformation and the percentage following it is the occupancy of that
conformation.
OD 1 or 2 = side-chain oxygen at the delta position
A letter after a residue designates the chain of that residue, indicating that there are two, non-crystallographically
symmetric models in the asymmetric unit which are treated independently.
When a halogen bond is formed with the side-chain of an aspartic or glutamic acid, the halogen often points to the
average position of the two oxygens, in which case the values for both atoms are listed, as here.
Values for halogen bonds to the π-system of Phe are given to the ring midpoint.
NE = side-chain nitrogen at the epsilon position
Averages and standard deviations are done using a mean number for those halogen bonds that are represented twice
via multiple crystallographically non-equivalent models.
a
b
c
d
e
f
g
h
Table 2 (Continued)
25
26
unphosphorylated human JNK3 bound to this inhibitor was solved to 2.2 Å resolution
(PDB code 1PMN) (Scapin et al. 2003). The structure of the protein in the complex
with compound 1 is almost identical to the unbound, unphosphorylated form of human
JNK3 (PDB code 1JNK). Thus, compound 1 binds to the inactive form of JNK3
without inducing any significant structural perturbations. The JNK3-compound 1
complex shows two hydrogen bond contacts, from the main chain N and O of Met149
in the hinge region of the protein to a pyridinyl and a cyclopropyl N on the inhibitor
(3.00 and 3.22 Å, respectively). In addition, there is one short halogen bond from the
Ala91 backbone carbonyl oxygen in hydrophobic pocket I to chlorine45 of the
inhibitor (2.83 Å).
JNK3-compound 2 complex
Compound 2 (cyclohexyl-{4-[5-(3,4-dichlorophenyl)-2-piperidin- 4-yl-3-propyl3H-imidazol-4-yl]-pyrimidin-2-yl}amine) is nearly identical to compound 1, differing
only in the cyclohexyl rather than cyclopropyl substitution off the pyrimidinyl ring.
Being in the same family as compound 1 it also binds p38 and JNK3 (1nM IC50) very
effectively. As with the compound 1 complex, the structure of N-terminal truncated
(residues 45 through 400), unphosphorylated human JNK3 bound to the inhibitor
compound 2 was solved to 2.2 Å resolution (PDB code 1PMQ) (Scapin et al. 2003).
Compound 2 makes the same polar contacts with JNK3 as compound 1, with the
hydrogen bonds from N and O of Met147 to nitrogens on the inhibitor being
somewhat shorter (2.81 Å and 3.05 Å), but the halogen bond from Ala91 O to
27
chlorine45 of the inhibitor is slightly longer (2.96 Å). The most significant difference
in the JNK3-compound 2 structure, as compared to the compound 1 complex, is the
presence of the ATP analog AMP-PNP, which was found to still be a present in this
structure despite having been displaced from the ATP binding site by compound 2.
AMP-PNP binds in a groove on the exterior of the protein leading towards the site
where the substrate protein is expected to bind, and forms a hydrogen bond to the N-4piperidyl nitrogen of compound 2 at the edge of the ATP-binding site. This suggests
that the additional binding site could be used to develop bidentate inhibitors targeted
against both the primary and this secondary ATP-binding site to create extremely
specific drugs.
In these JNK3 complexes, the length of the halogen bond present in each
inhibitor apparently does not correlate with the affinity of the enzyme for the
respective inhibitor (as judged from the IC50). The higher affinity of compound 2 for
JNK3 likely stems from the increased hydrophobic contacts of the protein with the
larger cyclohexyl group. This is reflected in the additional 27 Å2 of buried surface in
compound 2 as compared to compound 1 (Table 1). The authors conclude from
comparisons of these and two other JNK3 structures, however, that the potency of
these inhibitors could best be affected by altering the specific interactions between the
inhibitor and the ATP binding site of the protein, suggesting that the halogen bond
needs to be taken into account, along with all the other interactions, when trying to
refine JNK3 inhibitors.
28
Cyclin dependant kinase 2
Cyclin-dependant kinase 2 (CDK2) is a Ser/Thr protein kinase that plays a
major role in cell cycle control, especially the regulation of the G1/S transition, S
phase, and G2 phase (Knockaert et al. 2002). Like all cyclin dependant kinases,
CDK2 is activated by two separate events, with binding of a cyclin initiating a low
level of activity, and phosphorylation of a conserved threonine on its activation loop
creating the fully active protein. There has been significant effort in developing CDK
inhibitors because of the variety of CDKs that are deregulated in cancers. Of the over
eighty structures of CDK2 available in the PDB (most in complex with different
inhibitors), six inhibitors were observed to form halogen bonds with this protein. Here
we will discuss two of the three chlorinated inhibitors (NU6086 and PVB) and two of
the three brominated inhibitors (oxindole16 and TBB) observed that form halogen
bonds with CDK2.
CDK2-NU6086 complex
NU6086 (6-cyclohexylmethoxy-2-(3'-chloroanilino) purine) is a purine-based
inhibitor with an IC50 for CDK2-cyclinA3 of 2.3 ± 0.3 µM. The structure of the fully
active Thr160-phosphorylated CDK2-cyclinA complex bound to the inhibitor NU6086
was solved to 2.0 Å resolution (PDB code 1H1R) (Davies et al. 2002). The
asymmetric unit consisted of two crystallographically non-equivalent complexes,
allowing two separate observations of the inhibitor-protein interactions. Each
complex showed a triplet of hydrogen bonds, where three backbone atoms in the hinge
29
region of the protein (Glu81 O, Leu83 N, and Leu83 O) interact with three nitrogens
(N1, N2, and N9 nitrogens of the purine group—Figure 4) of the inhibitor (at distances
of 2.67Å/2.81 Å, 2.93Å/3.10 Å, and 2.46Å/2.57 Å, respectively, where each pair are
for two nonequivalent complexes). The chlorine that can potentially form a halogen
bond is found at the meta-position of the anilino (phenyl) ring, which is seen in each
complex to have two equally populated rotamers. One rotamer forms a halogen bond
to the side-chain oxygen (OD2) of Asp86 in the ATP-binding pocket (2.83 / 3.16 Å)
whereas the other rotamer points the chlorine out of the pocket, precluding halogen
bond formation.
This inhibitor was derived via structure-based design by addition of a metachloroanilino group in an attempt to improve the affinity and selectivity of the smaller
inhibitor NU2058 (O6-cyclohexylmethylguanine). However, the failure of the chlorine
(as well as the bromine and fluorine at the same position) to improve the affinity
relative to the parent inhibitor may be more due to the meta-position of the substituent
rather than the actual substituent. For example, a sulfonamide group added at the
para-position of the anilino-ring of NU6102 increased the affinity 150-fold for both
CDK1 & 2.
CDK2-Purvalanol B complex
Purvalanol B (2-(1R-siopropyl-2-hydroxyethylamino)-6-(3-chloro-4carboxyanilino)-9-isopropylpurine) was selected as an inhibitor for CDK2 from a
combinatorial library of 2,6,9-trisubstituted purines, which were themselves designed
based on the inhibitor olomoucine. The inhibitor has an IC50 of between 6 and 9 nM
30
for CDK2s, depending on whether the protein is complexed with cyclin A or E. The
structure of unphosphorylated human CDK2 bound to the inhibitor Purvalanol B was
determined to 2.05 Å resolution (PDB code 1CKP) (Gray et al. 1998). The inhibitor
makes two hydrogen bonds between backbone atoms in the protein’s hinge region
(Leu83 N and O) and inhibitor nitrogens (at distances of 3.18 Å and 2.54 Å,
respectively). A halogen bond is formed between the chlorine of the carboxyanilino
group and the carboxylate of Asp86, but only ~65% of the time (the other 35% has the
anilino ring rotated to orient the chlorine away from the protein). In this case, the near
equal distances (3.25 Å and 3.05 Å) between the chlorine and the two carboxylate
oxygens suggest that the interaction is to the center of the delocalized π-system. The
authors concluded that the increased affinity and selectivity of this inhibitor relative to
other related compounds in the combinatorial library is attributable to the region of the
inhibitor containing the chlorine. They did not, however, explicitly identify the polar
halogen bond interaction as the primary determinant for the higher affinity of
Purvalanol B relative to olomoucine, but suggest that the purine and chlorinated
aniline rings reduce the conformational entropy relative to the parent inhibitor by
allowing the 3-chloroanilino group to pack tightly against the side chains of Ile10 and
Phe82 of the protein.
CDK2-Oxindol16 complex
Oxindol16 (4-(5-bromo-2-oxo-2H-indol-3-ylazo)-benzenesulfonamide) is an
analog of the 3-(benzylidene)indolin-2-ones, a class of compounds that inhibit
receptor tyrosine kinases, but was found to selectively inhibit CDK2 with an IC50 of
31
60 nM. The crystal structure of unphosphorylated human CDK2 bound to the
inhibitor oxindol16 was solved to 2.2 Å resolution (PDB code 1FVT) (Davis et al.
2001), with the hope of using this crystal structure as a basis for designing more
effective inhibitors against CDK2. The inhibitor was seen in this structure to form
three hydrogen bonds to the main-chain atoms in the hinge region (Glu81 O and
Leu83 N and O, 2.73 Å, 2.76 Å, and 3.51 Å, respectively) and one to the OD2 oxygen
of Asp86 sidechain (2.91 Å). It also makes one halogen bond, this time from a
bromine to an average position between the two Asp145 side-chain oxygens (3.07 Å
and 3.22 Å). The authors had postulated, however, that replacing the bromine with a
Lewis base to accept a hydrogen bond from a neighboring Lys33 side-chain would
create a more effective inhibitor, which it did indeed do, demonstrating that the
halogen and hydrogen bond are essentially interchangeable in this system.
CDK2-TBB complex
TBB (4,5,6,7-Tetrabromobenzotriazole) was derived from the compound 1-(β D-ribofuranosyl)-5,6-dichlorobenzimidazole as an inhibitor for the CK2 protein kinase
(see below)(Sarno et al. 2001), but was found to also inhibit CDK2 with an IC50 of
15.6 µM (at 100 µM ATP concentration). The structure of the fully active Thr160phosphorylated CDK2-cyclinA complex bound to the inhibitor TBB was solved to
2.22 Å resolution, and, as with the NU6086 inhibitor, contained two
crystallographically non-equivalent complexes in the asymmetric unit, thereby
allowing two separate observations of the inhibitor-protein interactions (PDB code
1P5E) (De Moliner et al. 2003). Interestingly, this crystal structure shows that TBB
32
makes no direct hydrogen bonds with active CDK2, although there is one watermediated hydrogen bond that is conserved between the two non-equivalent models in
the structure—the NZ of Lys33 hydrogen bonds to a water (2.75 / 2.91 Å) which then
hydrogen bonds to N8 nitrogen of TBB (2.38 / 2.94 Å). Instead TBB relies on
halogen bonds from three or all four bromines of the inhibitor for specificity and
affinity. There are three halogen bonds that are conserved between the two complexes
of the asymmetric unit. Two halogen bonds involve the protein backbone oxygens in
the hinge region (Glu81 and Leu83, at distances of 3.01 Å / 3.05 Å and 2.90 Å / 2.88
Å, respectively) and one previously unrecognized interaction involving the side-chain
π-system of Phe80 to a bromine (3.26 Å / 3.24 Å to CD2 of Phe80 or 3.47 / 3.67 Å to
the center of the ring). A fourth halogen bond (to the backbone oxygen of Ile10 at a
distance of 3.24 Å) is seen in only one of the two models because the large size of the
hydrophobic pocket adjacent to the ATP-binding site in CDK2 allows the plane of the
inhibitor to shift ~13˚ between models. Since this inhibitor was designed for CK2, the
differences in modes of binding for the two kinases could be compared. This led to
the authors concluding that some halogen bond acceptors can direct specific
interactions with bromines (presumably similar to the three conserved halogen bonds
in this complex), while others (which involve primarily hydrophobic contacts) are
much less specific.
CK2
The function of this serine/threonine kinase, previously known as casein kinase
II, is still being unraveled. It is considered to be the most pleiotropic of all protein
33
kinases, with more than 308 substrate proteins known as of 2003 (Meggio and Pinna
2003). The enzyme activity is constitutively on and is not affected by
phosphorylation; however, there is likely some form of regulation of its function via a
regulatory subunit or other mechanism (Pinna 2003). What is clear is that CK2 plays
an important role in the cell as an anti-apoptotic factor, and, from a drug discovery
perspective, may be an important target since the levels of this kinase have been found
to be elevated in all human cancers and experimental tumors examined to date (Unger
et al. 2004). Four examples of CK2 halogen bonded to related brominated inhibitors
were found in the PDB, possibly creating the opportunity to parse out the effects of the
halogen bond in this family of inhibitors.
CK2-TBB complex
TBB (4,5,6,7-Tetrabromobenzotriazole) was designed originally as a specific
inhibitor of CK2 (as discussed above) and has an IC50 of approximately 0.5 to 1.6 µM
for this kinase (depending on the source of the enzyme, but both with 100µM ATP
present).(Sarno et al. 2001; Sarno et al. 2003). The crystal structure of the Zea mays
CK2α catalytic subunit complexed with the inhibitor TBB was determined to 2.2 Å
resolution, with two crystallographically non-equivalent complexes observed in the
structure, allowing for two separate observations of the inhibitor-protein interactions
(PDB code 1J91) (Battistutta et al. 2001). The complex between the kinase and TBB
shows one direct hydrogen bond (conserved in both non-equivalent models) between
Lys68 NZ and TBB N9 (3.29 / 3.44 Å). There is also one halogen bond to the Arg47
residue of CK2; however, in each model the bromine of the TBB inhibitor interacts
34
with different atoms of this residue (in one case to the epsilon-nitrogen of the side
chain and in the other to the carbonyl-oxygen of Arg47, at distances of 2.99 Å and
3.30 Å, respectively). Despite the small number of polar contacts seen in these
structure, the inhibitor’s increased affinity for CK2, as compared to CDK2, can be
attributed to the complimentarity between the size of the ATP-binding and peripheral
pockets and the size of the inhibitor, with a higher total number of nonpolar contacts
from TBB to CK2 than to CDK2 (as reflected in the ~37 Å2 more buried surface for
the inhibitor in the CK2 complex as compared to CDK2). The greatly decreased
affinity and selectivity for CK2 when an only slightly smaller chlorine atom was
substituted for the bromines of the TBB inhibitor was, indeed, attributed to the
decrease in nonpolar contacts (Szyszka et al. 1995). We note, however, that this effect
may also be associated with the dramatically reduced polarizability of chlorine
compared to bromine (Figure 7).
CK2-K25 complex
K25 (4,5,6,7-tetrabromo-N,N-dimethyl-1H-benximidazol-2-amine) is an
amino derivative of TBB which has an IC50 of 140 nM and a Ki of 40 nM for CK2, the
lowest values reported for any CK2 inhibitor. The crystal structure of the Zea mays
CK2α catalytic subunit complexed with the inhibitor K25 was determined to 1.8 Å
resolution (PDB code 1ZOE) (Battistutta et al. 2005). This structure shows no direct
and only one solvent chloride-mediated hydrogen bond between the protein and the
inhibitor, from the zeta-nitrogen of Lys68 CK2 to a solvent Cl- to K25 N14 (3.18 Å
and 3.37 Å, respectively). There are two halogen bonds present, however, both from
35
hinge region backbone oxygens (Glu114 and Val116) to K25 bromines (3.43 Å and
3.24 Å, respectively). This inhibitor, although similar to TBB on the edge that makes
contacts with the hinge region, penetrates more deeply into the pocket than TBB,
pushing back both the Glu114 and Val116 residues. The authors argue that because of
this deeper penetration in the ATP-pocket, K25 buries more accessible surface area
upon binding than TBB, giving it a higher affinity for CK2; we can find no indication
of this, however, in our own surface area calculations, either in terms of the buried
inhibitor or the buried protein surfaces (Table 1).
CK2-K37 complex
K37 (4,5,6,7-tetrabromo-2-(methylsulfanyl)-1H-benximidazole) is a
methylsulfanyl derivative of TBB, and has an IC50 of 250 nM and a Ki of 70 nM for
CK2. The crystal structure of the Zea mays CK2α catalytic subunit complexed with
the inhibitor K37 was determined to 2.3 Å resolution (PDB code 1ZOG) (Battistutta et
al. 2005). In this crystal structure, K37 takes on two different orientations in the ATPbinding pocket, one of which is similar to that of that seen with K44 (see below) and is
seen at ~60% occupancy (orientation A), while the other is similar to that of K25 and
is ~40% occupied (orientation B). K37 makes no direct hydrogen bonds to CK2 and
only orientation B makes one water-mediated hydrogen bond (from N8 nitrogen of
K37 to water to the amide-nitrogen of Asp175, with distances of 3.14 Å and 3.06 Å,
respectively). Each orientation makes at least two halogen bonds with the protein
backbone oxygens of Glu114 and Val116 in the hinge region, but to different K37
bromine atoms and with different Br – O distances (3.23 Å and 2.98 Å in orientation
36
A, and 3.11 Å and 3.27 Å in orientation B). Orientation B also shows a halogen bond
between the π-system of Phe113 side chain and an inhibitor bromine (Br to CD2
distance of 3.23 Å, or 3.53 Å to the center of the ring of Phe113). As with K25, K37
in both orientations binds further into CK2’s ATP-binding pocket than does TBB,
which could account for the increase in the affinity relative to the parent compound.
CK2-K44 complex
K44 (N1,N2-ethylene-2-methylamino-4,5,6,7-tetrabromo-benzimidazole) is a
derivative of TBB that includes an additional extracyclic ring, and has an IC50 of 740
nM and a Ki of 100 nM for CK2. The crystal structure of the Zea mays CK2α
catalytic subunit complexed with the inhibitor K44, determined to 1.8 Å resolution
(PDB code 1ZOH) (Battistutta et al. 2005), shows no direct or water-mediated
hydrogen bonds between the inhibitor and protein. There were three halogen bonds
observed from the bromines of the inhibitor to the protein, two to backbone oxygens
of Glu114 (3.18 Å distance) and Val116 (2.89 Å distance) in the protein hinge region,
and one to the side-chain oxygen OD1 of Asp175 (3.16 Å). This inhibitor also
penetrates further into the pocket than TBB.
In comparing the relative affinities for the highly related K25, K37, and K44
inhibitors, the authors of these studies indicated that there was no direct correlation
between the number and type of polar interactions observed and their relative
affinities. Indeed, simply counting the hydrogen and halogen bonds for these
inhibitors do not account for the ~5-fold differences in affinities (neither do the buried
surface areas, from our analyses). They go on to note, however, that in this regard,
37
these three inhibitors are quite different from the majority of protein kinase inhibitors
seen to date, where affinity can be improved by increasing the number and quality of
polar interactions. The authors do suggest, though, that polar interactions help to
orient each individual ligand within the binding pocket.
MAP Kinase- / ERK- activating kinase 1 & 2
MAPK- / ERK- activating kinase 1 & 2 (MEK1 & 2) are MAPK kinases of the
extracellular signal regulated kinase 1 & 2 cascade (also known as the
Ras/Raf/MEK/ERK pathway). MEK1 & 2 are rare, dual specificity kinases
responsible for the activation of ERK 1 & 2 via dual phosphorylation of the threonineX-tyrosine segment of their activation loops (Kolch 2000). A dual phosphorylation of
two serines is also required for the activation of MEK1 & 2 themselves, with the
primary MEK activator in most cell types being the Raf kinases (Schaeffer and Weber
1999). The Ras/Raf/MEK/ERK pathway governs extracellularly stimulated cell
growth and differentiation and some 30% of all human tumors have this pathway
constitutively activated (Ohren et al. 2004). This has made this pathway an attractive
target for anti-cancer therapeutics, and the MEK proteins especially useful given their
specificity for the Ras/Raf/MEK/ERK pathway (as opposed to Ras and Raf which also
function in other pathways) (Herrera and Sebolt-Leopold 2002). Two examples of
MEK proteins halogen bonded to related iodine-containing inhibitors were found in
the PDB, and are quite unusual in that they are noncompetitive for both ATP and
MAPK (Sebolt-Leopold et al. 1999).
38
MEK1-PD318088 complex
PD318088 was developed from the novel, non-ATP-competitive inhibitor
PD184352 which was shown to be both a specific and effective (IC50 of 17 nM)
inhibitor for MEK1 in an earlier study (Sebolt-Leopold et al. 1999). No IC50 was
reported for the PD318088 inhibitor with MEK1, but the crystal structure of human,
unphosphorylated, N-terminally truncated MEK1 complexed with PD318088 and
MgATP was determined to 2.4 Å (PDB code 1S9J) (Ohren et al. 2004). The structure
of the complex shows the non-ATP-competitive binding site to be a pocket adjacent to
the ATP-binding site, and close enough to allow interaction with the ATP phosphates
but separate enough to enable simultaneous binding in both. Four hydrogen bonds are
formed between PD318088 and the rest of the complex, three to the protein and one to
the O3 of the ATP’s γ-phosphate (inhibitor O3 to ATP O3G – 2.66 Å). Of the
hydrogen bonds to MEK1, two are from inhibitor oxygens (O1 and O4) to the NZ of
Lys97 (2.65 and 2.87 Å, respectively), while one is from an inhibitor fluorine (F1) to
the hydrogen of the backbone amide nitrogen of Ser212 (3.53 Å). There is also one
important edge-to-face aromatic interaction between the fluorine and iodine
substituted phenyl ring of the inhibitor and the sidechain phenyl ring of Phe209
(inhibitor C5 to Phe209 ring center distance – 4.08 Å). The halogen bond formed
between the I1 iodine of the PD318088 and the Val127 backbone carbonyl oxygen of
MEK1 is 3.13 Å, quite close for the large iodine atom.
39
MEK2-PD334581 complex
PD334581 was also developed from PD184352, and is quite similar to both its
predecessor and PD318088 (Figure 4). In order to look for a non-ATP-competitive
inhibitor binding site similar to that of MEK1 in the highly homologous MEK2
protein, the crystal structure of human, unphosphorylated, N-terminally truncated
MEK2 complexed with PD334581 and MgATP was determined to 3.2 Å (PDB code
1S9I) (Ohren et al. 2004). The asymmetric unit of this structure consisted of two
crystallographically non-equivalent complexes, and each complex showed three
similar hydrogen bonds and one equivalent halogen bond, and one of the complexes
also has two extra hydrogen bonds. The hydrogen bonds that are similar between the
two complexes are from the inhibitor fluorine F17 to the hydrogen of the backbone
amide nitrogen of Ser216 (2.63 and 2.65 Å, for complexes A and B, respectively).
Also, hydrogen bonds are formed from inhibitor nitrogen N14 to the NZ of Lys101
(3.06 and 2.61 Å) as well as from inhibitor nitrogen N8 to the sidechain OD1 of
Asp212 (3.25 and 3.01 Å). Although the last two hydrogen bonds were longer in
complex A, it compensates by having an extra hydrogen bond similar to one seen in
the MEK1 structure: inhibitor nitrogen N18 interacts with the O2 oxygen on the γphosphate of ATP at a distance of 3.13 Å. The aromatic edge-to-face interaction seen
in the MEK1 structure is also observed here and is, in fact, stronger here. The fluorine
and iodine substituted phenyl ring of the inhibitor again interacts with a phenylalanine
sidechain (Phe213), this time at a distance of only 3.62 and 3.91 Å from the inhibitor
C21 to the center of the Phe213 ring. This interaction in complex A may even be
40
short enough to be considered an unusual hydrogen bond, further compensating for the
length of some of its other hydrogen bonds. The halogen bond in both complexes is
much like that in the MEK1 structure: PD334581 iodine I1 interacts with the backbone
carbonyl oxygen of Val131 at distances of 3.25 and 3.16 Å, respectively.
PD318088 and PD334581 bind MEK1 & 2 in quite similar fashion, and
provide the first structural examples of noncompetitive, small molecule inhibitors for
protein kinases. The binding site that these inhibitors occupy is not present in many
other protein kinases—in fact sequence homology to other protein kinases in this area
is quite low (other than MEK5), so it may be unique to MEK1 & 2, making it an
extremely specific drug target. The halogen bond between iodine and the protein
oxygen in this binding site is exhibited by both of these complexes and has
exceptionally good geometric parameters (see Table 2), indicating that it is likely to be
a strong interaction and may be quite important for the inhibitor binding. This effect
is difficult to quantify, however, since no similar inhibitor lacking the iodine has, to
our knowledge, been described.
Analysis / Discussion
Halogens involved in halogen bonding to protein kinases
Of the twelve examples here of inhibitors halogen bonded to protein kinases, six
involve bromines, four involve chlorines, and two involve iodines, and this ratio
becomes 7:7:2 when the other four known structures including halogen bonds to
41
protein kinases are factored in. The small size of this dataset may over-represent the
contribution of bromine, given that two different kinases were bound to the same
brominated inhibitor (TBB), and all four inhibitors of CK2 were TBB or derivatives of
TBB. The small number of iodine type halogen bonds in this set of kinase structures
was somewhat surprising from the perspective of the favorability of the interaction.
Iodines, being more polarizable than either chlorine of bromine, have the potential to
form even stronger halogen bonds, but the low relative representation of iodines in
drugs and inhibitors in general may explain why only two were observed.
Regardless of the halogen involved, in every case, the halogen was a substituent
of an aromatic ring. This can be attributed to the fact that the inhibitors were all
designed to target the ATP-binding pocket of protein kinases by mimicking the planar
purine base of ATP. Given that π-systems enhance the anisotropic distribution of
electrons in halogens (Auffinger et al. 2004), this makes halogen bonding an excellent
potential tool for the design of small molecule inhibitors that target the ATP-binding
sites of protein kinases.
Lewis bases that serve as halogen bond acceptors in protein kinases
Oxygens from the peptide backbone and, to a lesser extent, the side chains of
aspartic acid residues serve as the primary halogen bond acceptors in the protein
kinases surveyed here. In addition, there was one nitrogen (the epsilon-nitrogen of an
arginine side chain) as well as two examples of aromatic rings acting as halogen bond
acceptors. In both these cases, the halogen bonds are oriented towards the π-systems
of the respective amino acid side chains.
42
The possibility of π-systems aromatic rings acting as halogen bond partners has
not been reported in the small molecule systems, where halogen bonding was initially
characterized (the only reference on the subject focused on fluorine intramolecular
interactions with aromatic rings (Prasanna and Guru Row 2000)). Consequently, we
look instead to analogous hydrogen bonds to π-systems to better understand this type
of interaction. Hydrogen bonds to π-systems are generally weaker and longer than
those to the more formally electronegative atoms such as oxygen and nitrogen. The
energies of such hydrogen bonds range from -0.5 to -1.5 kcal/mol and lengths from 3.2
to 3.8 Å (donor to acceptor distance), as compared -5 to -7 kcal/mol and 2.5 to 3.5 Å
for the more standard hydrogen bonded partners (reviewed in (Meyer et al. 2003) and
(McDonald and Thornton 1994)). The two examples of halogen bonds to aromatic πelectrons found here (the bromines of TBB and K37 to Phe80 of CDK2 and Phe113 of
CK2, respectively) also show longer distances (3.47 Å to 3.67 Å) as compared to
halogen bonds to oxygen or nitrogen, but these interaction distances still represent an
encroachment of the van der Waals radius of the halogen into the thickness of the
phenyl ring (the aromatic ring of benzene is 1.9 Å thick and, therefore, the minimum
distance for a noninteraction is predicted to be 3.7 Å for Br to benzene) (Prasanna and
Guru Row 2000).
To determine whether these halogen bonds to aromatic rings are potentially
favorable interactions, we applied ab initio quantum mechanics calculations to
estimate the ground state energies of a bromobenzene model with its bromine oriented
towards the center of a second benzene molecule. The WebMO interface was used to
construct these models and GAMESS used for the calculations (applying density
43
function theory with the B3LYP function and the 6-31G(d) basis set (Schmidt et al.
1993; Schmidt and Polik 2005)) on the model system. By varying the interaction
distance between the bromine substituent of the bromobenzene and interacting
benzene, we observe a minimum energy of nearly –1 kcal/mol at ~3.6 Å separating the
bromine from the center of the aromatic ring. This optimum distance is comparable to
those we observe in the kinase-inhibitor complexes, indicating that these are
legitimate, albeit weaker, halogen bonding interactions.
The Lewis bases that serve as halogen bond acceptors are primarily oxygens
from the peptide bonds of the protein backbone, while fewer are oxygens from the
side-chains of aspartic acid (approximately 2:1). This distribution may again be
skewed by the overrepresentation of TBB and its derivatives in this dataset—nearly all
of the halogen bonds from this class of inhibitors interact with similar residues. With
the exception of the TBB interactions with CK2 and one interaction of K37 with the
aromatic ring of Phe113, the remainder are halogen bonds to the backbone carbonyl
oxygens of Glu114 and Val116, both of which also serve as hydrogen bond acceptors
to the purine of ATP. This preference for peptide oxygens as halogen bond acceptors,
however, is seen generally in the more complete dataset of halogen bond interactions
in biological molecules (Auffinger et al. 2004) and, therefore, it appears that the
kinases simply mirror the preferences seen in other protein-ligand complexes.
Geometries of halogen bond interactions in protein kinases
The geometries of the twenty different halogen bonds described here fit very
well with the general features for all biological halogen bonds surveyed in the PDB
44
(Auffinger et al. 2004). The key feature of the halogen bond is that these are short
range interactions, with distances shorter than the sum of the respective atoms’ van der
Waals radii. The average chlorine to halogen bond acceptor distance of 2.98 Å ± 0.13
Å in the protein kinase structures is approximately 0.3 Å shorter than the sum of the
van der Waals radii for the respective atoms, and falls within the average distance of
3.06 Å seen in the PDB survey (oxygens were the only halogen bond acceptors present
in either set for Cl-type halogen bonds). This is similar to the bromine and iodine type
halogen bonds, where the average distance in the kinase structures (Br—3.20 Å ± 0.20
Å and I—3.17 Å ± 0.04 Å, or approximately 0.2 Å and 0.4 Å shorter than Σrvdw) are
comparable to the larger dataset (average 3.15 Å and 3.24 Å). One set of halogen
bonds seen here that were not included in the original survey are the short bromine
interactions with the side chain rings of phenylalanine described above. The thickness
of phenyl rings compared to other halogen bond acceptor atoms (mostly oxygens and
one to a nitrogen) make these longer range interactions (average distance of 3.56 Å,
Table 2), which may skew the average Br – acceptor distances seen here to be slightly
longer than those generally seen in the PDB (in the absence of the Br•••Phe
interactions, the average bromine halogen bond in the current kinase dataset is 3.11
Å).
The other important geometric features of halogen bonds, analogous to hydrogen
bonds (Figure 6), are the orientations of the interacting atoms, with the acceptor
aligned approximately linearly with the polarization of the halogen (near linear Θ1angle), and halogens aligned primarily towards the nonbonding orbitals of the Lewis
base (<Θ2-angle> ≈ 120°). The primary feature that distinguishes biological from
45
small molecule interactions, however, is that halogen bonds to the peptide backbone of
proteins tended to interact also to the delocalized π-electrons of the amide. Values of
Θ1 = 152˚ ± 10˚ for chlorine and 158˚ ± 17˚ for bromine fit the overall averages for all
biological halogen bonds (151˚ for chlorine and 154˚ for bromine). This simple
analysis of the averages can be somewhat misleading, though, given that Θ1 for all
biological halogen bonds showed a bimodal distribution with maxima around 145˚ and
165˚ with a minimum between them at about 155˚ (Auffinger et al. 2004). The kinase
structures, however, also showed a bimodal distribution, with two distinct modes at Θ1
= 145° and 180°. In these structures, it was clear that most of the peptide carbonyl
oxygens oriented linearly towards the halogen polarization (11/16 such interactions
had Θ1 > 150°), while the side chain oxygens are in the nonlinear mode (6/7
interactions with Θ1 < 150°). In contrast, the “perfect” values seen for the two iodine
halogen bonds (Θ1 = 180˚ ± 0˚) must be considered the happenstance product of an
extremely small dataset, since the larger survey showed a similarly bimodal Θ1
distribution, centered around 157˚.
Bromines in the complete dataset of biological halogen bonds (Auffinger et al.
2004) were most often observed interacting with non-bonded electrons (Θ2 ≈ 120˚),
but also showed a subpopulation interacting with oxygen π-system electrons (Θ2 ≈
90˚). This is reiterated in the current kinase dataset, with all but two halogen bonds to
oxygens or nitrogens with Θ2 ≥ 100°. All of the interactions to side chains of
phenylalanine are, as described above, perpendicular to the phenyl-ring. The average
Θ2 values for chlorine (92˚ ± 13˚) and iodine (125˚ ± 4˚), however, are probably
attributable to the small size of both samples, given that chlorine was observed to be
46
the least likely of the three polarizable halogens to interact with π-electrons according
to the full PDB halogen bond dataset, while iodine was observed to do so most often
(likely due to its larger size and greater steric restraints) (Auffinger et al. 2004).
Surface preferences of halogen bonds
Finally, we also observed a preference in the type of surface that the halogen
bond acceptors presented to the inhibitors of protein kinases. We plotted the solvent
accessible surface area using probes approximating the size of chlorine, bromine, and
iodine (1.70 – 2.12 Å radii) for each of the acceptors to determine the shape of the
surfaces that are “seen” by the halogen of the inhibitors. For comparison, we probed
the surfaces of a random sampling of 33 oxygens in the 10 kinase structures that do
not participate in halogen bonding, but are found in the ATP-binding and surrounding
pockets. The basic surface shapes could be grouped into three classes: concave (so
that the ligand sees basically a round hole), convex (so that the ligand sees a round
bump), and a saddle (so that the ligand sees a bump with a furrow in it). For those
Lewis bases involved in halogen bonds, only three of the eighteen presented a convex
surface, whereas nine were concave and eight were saddles. The sampling of nonhalogen bonding oxygens showed a very random distribution of shapes, nine convex,
ten concave, and fourteen saddles, indicating that the halogen bonding population of
acceptors is biased away from convex surfaces. We further reclassified the saddle
type surfaces as being mostly convex or mostly concave (although five were too
ambiguous to assign). With these definitions, there were almost four times as many
concave-type surfaces as convex-type among the halogen bond acceptors (15:4) while
47
the non-halogen bonding oxygens were, again, fairly evenly split (13:16) between the
surface types. This supports the concept that shape complementarity is an important
component in defining the effectiveness of an interaction, even at the atomic level.
CK2 as a case study for identifying carbonyl oxygens as halogen bond targets
To explore the idea that the shape of the halogen bond acceptor surface
presented to the halogen influences the propensity to form a halogen bond, we
analyzed in greater detail the structures of CK2, which had the most representations of
halogen bonds to inhibitors in this dataset. The crystal structures of apo-CK2 and
CK2 bound to the ATP analog AMPPMP are both available (PDB codes 1JAM and
1DAW, respectively), so we analyzed the accessible surface area of the ATP-binding
pocket in both structures as well as the halogen bonded structures for comparison.
The apo-CK2 structure showed several accessible backbone carbonyl oxygens (four to
six, depending on the probe size) in the ATP-bonding pocket, but the carbonyl
oxygens of Glu114 and Val116, which are observed to be involved in halogen bonds
in three of the four CK2 structures, were seen to be accessible only to probes with
radii ≤ 1.70 Å and 1.65 Å, respectively (i.e. chlorine sized or smaller). The structure
of CK2 with the ATP analog has a more open ATP binding pocket, as expected, and
presents six accessible backbone carbonyl oxygens to a probe of 1.9 Å radius
(approximately the size of bromine). These were to Arg47, Glu114, Val116, Asn118,
His160, and Val145. Glu114 and Val116 both present concave accessible surfaces to
probes of radius ≤ 1.95 Å and 2.40 Å, respectively. The other four oxygens all present
48
convex or mostly convex saddle surfaces to probes the size of halogens or greater
(Figure 8). Of these, only the carbonyl of Arg47 is seen to form a halogen bond, and a
very long one at that (at 3.30 Å, the second longest bromine halogen bond in this
dataset). Thus, if we were to apply shape complementarity to structure-based target
identification for potential halogen bonds, it would be more effective if the structure
being analyzed is one with a ligand bound. This very simple study shows that it is
possible to start to define explicit features of the protein structure (such as the shape of
pockets) to guide the incorporation of halogen bonds in the design of more specific
inhibitors to enzymes such as protein kinases.
Conclusions
Clearly halogen bonds have been designed into protein kinase inhibitors, if
somewhat unintentionally. The geometric properties of halogen bonds in the current
dataset of protein kinase structures mirror those seen in the larger dataset of all
biological halogen bonds. A detailed analysis of this one class of inhibitor interactions
shows qualitatively that halogen bonds play important roles in ligand recognition and
binding, but, at least with protein kinases, it is not possible, with our current limited
understanding of the interaction in biological systems, to quantitatively correlate the
affinity or specificity of these inhibitors to any particular feature or property of the
halogen bond. Thus, the field is open for further exploration. Despite this limited
understanding, there is sufficient evidence to indicate that halogen bonds will become
49
Figure 8. Accessible backbone oxygens at the CK2 ATP binding site. As a case
study, we analyzed the ATP binding pocket of the structure of CK2 (in blue) bound to
the ATP analog AMPPMP (colored by atom). For these accessible surface area
calculations, we removed the AMPPMP ligand and probed the open pocket using a
bromine sized probe (1.9 Å) (shown in the inset on the right). There are six accessible
backbone oxygens (red) in the pocket, two of which present concave surfaces to the
ligand (Val116 and Glu114, which are also the backbone oxygens in CK2 that have
most often been seen to participate in halogen bonds).
50
an important tool in the field of rational drug design. For example, we can now define
a favorable profile for a halogen bond acceptor as one which presents a concave or
mostly concave surface to the halogen. This provides a structural criterion to identify
Lewis bases in proteins as acceptors for halogen bond interactions in structure based
drug design.
The strong selectivity of halogen bonds for the carbonyl oxygens along the
polypeptide backbone could also be exploited to help design inhibitors that are less
susceptible to adaptive drug resistance. Inhibitors that interact primarily with specific
amino acid side chains are highly susceptible to the protein acquiring resistance
through simple missense mutations. However, it would be much more difficult to
mutate away interactions to targets (such as the carbonyl oxygen) that are essential
components of the polypeptide backbone without dramatically affecting the protein’s
structure and/or function. Our observation from this and previous studies that a
variety of backbone carbonyl oxygens can act as effective halogen bond acceptors
suggests that inhibitors designed to exploit such interactions would be less susceptible
to this problem of acquired resistance.
Acknowledgements
We thank Dr. Rick Faber for his help with data mining and Keita Oishi for his
work with the ab initio quantum calculations. The work from the PSH laboratory was
funded by grants from the National Institutes of Health (R1GM62957A), the National
51
Science Foundation (MCB0090615, the National Institutes of Environmental Health
Sciences (ES00210), the Franco-American Fulbright Commission, and a grant from
the Institut Universitaire de France.
52
Chapter 3
Directing Macromolecular Conformation Through Halogen Bonds
Andrea Regier Voth, Franklin A. Hays, and P. Shing Ho
Published in Proceedings of the National Academy of Sciences, U.S.A.,
The National Academy of Sciences, Washington, D.C., USA
2007, 104 (15), 6188-6193
53
Summary
The halogen bond, a non-covalent interaction involving polarizable chlorine,
bromine, or iodine molecular substituents, is now being exploited to control the
assembly of small molecules in the design of supramolecular complexes and new
materials. We demonstrate that a halogen bond formed between a brominated-uracil
and phosphate oxygen can be engineered to direct the conformation of a biological
molecule, in this case to define the conformational isomer of a four-stranded DNA
junction when placed in direct competition against a classic hydrogen bond. As a
result, this bromine interaction is estimated to be at ~2 to 5 kcal/mol stronger than the
analogous hydrogen bond in this environment, depending on the geometry of the
halogen bond. This helps to establish halogen bonding as a potential tool for the
rational design and construction of molecular materials using DNA and other
biological macromolecules.
Introduction
Halogen bonds have recently seen a resurgence of interest as a tool for
“bottom-up” molecular design. Chlorines, bromines, and iodines in organic and
inorganic compounds are known to polarize along their covalent bonds to generate an
electropositive crown; the halogen thus acts as a Lewis acid to pair with Lewis bases,
including oxygens and nitrogens. These electrostatic pairs, originally called charge-
54
transfer bonds (Hassel 1972), are now known as halogen bonds (X-bonds),
recognizing their similarities to hydrogen bonds (H-bonds) in their strength and
directionality (Metrangolo and Resnati 2001). In chemistry, X-bonds are being
exploited in the design and engineering of supramolecular assemblies (Brisdon 2002)
and molecular crystals (reviewed by Metrangolo, et al, 2005 (Metrangolo et al. 2005)),
with an iodine X-bond estimated to be ~3.5 kcal/mol more stable than an O-H···O Hbond in organic crystals (Corradi et al. 2000).
The X-bond, however, has not generally been a part of the biologist’s lexicon.
Although halogens are widely used in drug design and to probe molecular interactions,
X-bonds have only recently been recognized as a distinct interaction in ligand
recognition and molecular folding, and in the assembly of proteins and nucleic acids
(Muzet et al. 2003; Auffinger et al. 2004). With the growing application of biological
molecules (biomolecules), particularly nucleic acids (reviewed by Seeman (Seeman
2005)), in the design of nanomechanical devices, we ask here whether specific Xbonds can be engineered to direct conformational switching in a biomolecule.
To compare X- and H-bonds in the complex environment of a biomolecule, we
have designed a crystallographic assay to determine whether an intramolecular Xbond could be engineered to direct the conformational isomerization of a DNA
Holliday junction by competing an X-bond against a classic H-bond and,
consequently, are able to compare the stabilization energies afforded by these two
types of interactions. The stacked-X form of the DNA Holliday junction (Figure 9),
seen in high salt solutions (Duckett et al. 1988) and in crystal structures (OrtizLombardía et al. 1999; Eichman et al. 2000; Hays et al. 2005), is a simple and well
55
Figure 9. Structure of the stacked-X DNA Holliday junction. The structure of
d(CCGGTACCGG) (ACC-J) as a four-stranded junction (Eichman et al. 2000) is
shown with the inside cross-over strands colored in yellow and green and the outside
non-crossing strands in blue and red. The pairs of stacked duplex arms are highlighted
with cylinders. Details of the molecular interactions that stabilize junctions are in
crystals are shown, with the essential H-bond from the C8 cytosine to the phosphate of
the cross-over C7 nucleotide in the blue box, and the weaker H-bond from C7 to A6 in
the ACC-J or the weak electrostatic interaction from the methyl of T7 to A6 in
d(CCGATATCGG) (ATC-J) in the red boxes (Hays et al. 2003; Hays et al. 2005).
56
controlled biomolecular assay system that can isomerize between two nearly
isoenergetic and structurally similar conformers (Figure 10) (Miick et al. 1997;
Grainger et al. 1998; McKinney et al. 2003). The presence of an X- or H-bond at the
junction cross-over will be reflected, in this assay, by the preference for one of the two
isomer forms.
A cytosine to phosphate H-bond at the N7 nucleotide position has been shown
to help stabilize the junction in the sequence d(CCGGTACCGG) (ACC-J) in crystals
(Eichman et al. 2000; Hays et al. 2003; Hays et al. 2005) and in solution (Hays et al.
2006). The stability of this junction, formed by a single inverted repeat sequence, is
dependent entirely on this set of intramolecular interactions. The brominated
junctions of the current competition assay can adopt a conformer (the H-isomer) that is
stabilized by a similar cytosine to phosphate H-bond. Alternatively, they can adopt a
conformer (the X-isomer) that is stabilized by an X-bond from a 5-bromouracil (BrU)
that replaces the cytosine (Figure 10). Thus, the isomer form of the junction will be
explicitly identified by the position of the bromine: if the BrU sits at the inside
crossing strands, then the junction is the X-isomer, but if it sits at the outside strand, it
is H-isomer. Furthermore, the distribution of X- and H-isomers observed in the crystal
will reflect the differences in the overall energies and, by extension, the differences in
energies of the intramolecular interactions that stabilize these isomers, assuming
identical crystal lattice energies for the conformers.
57
Figure 10. Assay for competing X- against H-bonds. The isomeric forms of the
stacked-X junction, resulting from restacking of the arms and migration of the junction
(top), places an H-bond (H-isomer) or X-bond (X-isomer) at the junction cross-over.
The crystal structures of H2J in the H-isomer with its H-bonding interaction and Br2J
in the X-isomer with its X-bond are shown below. Bromines have been modeled at the
outside strand of the H2J structure to show their positions, if present, in the H-isomer
of Br2J.
58
Results
We have designed three DNA constructs (Table 3) in which bromine X-bonds
compete against H-bonds in ratios of 2X:2H (Br2J), 1X:2H (Br1J), and 0X:2H (H2J) at
the cross-over to drive the isomeric form of the resulting stacked-X junction. The
single-crystal structure directly identifies the isomeric form and the stabilizing
interactions for each conformer (Figure 10). For the Br2J and Br1J constructs, a
bromine at the inside cross-over strand would indicate formation of the X-isomer,
while a bromine at the outside strand would be indicative of the H-isomer. Since the
crystals of the three constructs are isomorphous (Table 4) and the sites of stabilizing
interactions distant from any direct intermolecular DNA lattice contacts, we can
assume that the isomer preference is not defined by the crystal lattice. Therefore, the
ratio of H- to X-isomers in the crystals reflects the differences in energy between the
H- and X-bonding interactions. The sequence dependent formation of DNA junctions
in solution has been directly correlated to the structure and molecular interactions seen
in crystals (Hays et al. 2006), indicating that the properties of the DNA junction in
crystals closely mirror those in solution and are not explicitly induced by the crystal—
the crystal structure, therefore, is a means by which we can directly identify the type
of stabilizing interaction for a particular conformer form of each DNA construct.
The Br2J construct shows that two X-bonds compete effectively against two classic Hbonds in the DNA junction. This structure (Figure 10) was initially refined without
specifying the nucleotides at the N7 or complementary N4 nucleotides of the strands.
The strong positive difference density adjacent to the C5 carbon of the N7 pyrimidine
59
Table 3. Constructs and sequences that compete halogen bonds against hydrogen
bonds in DNA junctions. The ratios of halogen vs hydrogen bonds that potentially
compete in each construct are listed as X:H.
Construct
Sequences (brU = 5-bromouracil)
X:H
Br2J
d(CCGGTAbrUCGG)2/d(CCGATACCGG)2
2:2
Br1J
H2J
d(CCGGTAbrUCGG)/d(CCGGTAUCGG)/
d(CCGATACCGG)2
d(CCGGTAUCGG)2/d(CCGATACCGG)2
1:2
0:2
60
Table 4. Crystallographic and geometric parameters for the Holliday junction
constructs Br2J, H2J, and Br1J (see Table 3 for sequences). Listed are the
crystallographic and refinement parameters, and the geometric parameters that
describe the rotation of the stacked duplex arms across the junction (Jtwist), and the
rotation (Jroll) and translation (Jslide) of each set of stacked arms along the respective
helical axes(Vargason and Ho 2002; Watson et al. 2004) for each junction construct.
Junction:
Space group
Unit cell
a (Å)
b (Å)
c (Å)
β-angle
Unique reflections
(for refinement)
Resolution (Å)
Completeness1
I/sig(I)1
Rmerge1
Rcryst (Rfree)
Number of atoms:
DNA (Solvent)
<B-factor> DNA
(Solvent)
RMSD bond lengths
(Å)
RMSD bond angles
PDB Accession Codes
Br2J
H2 J
Crystallographic Parameters
C2
C2
Br1J
C2
65.78
24.41
37.32
111.07°
65.61
24.17
36.96
110.01°
65.89
24.21
37.29
111.02°
3,362
(3,200)
4,176
(3,992)
4,350
(4,247)
2.00
1.90
82.8% (49.2%) 89.6% (63.4%)
13.70 (2.31)
20.41 (3.76)
4.9% (28.1%)
5.1% (25.2%)
Refinement Statistics
21.8% (27.2%) 21.8% (26.8%)
404
403
(106)
(101)
1.85
87.6% (61.7%)
21.18 (7.73)
3.2% (9.3%)
22.4% (25.2%)
405
(102)
13.1
(15.6)
16.3
(20.9)
15.4
(17.5)
0.004
0.007
0.009
0.8°
2ORG
1.1°
2ORH
1.4°
2ORF
Junction Geometry
40.2°
39.6°
142.5°
137.9°
0.21 Å
1.13 Å
Jtwist
42.4°
Jroll
138.6°
Jslide
1.10 Å
RMSD relative
0.00 Å
0.49 Å
0.71 Å
to Br2J
1
Values for the highest resolution shell are shown in parentheses
61
showed that the bromines sit at the junction crossovers, leaving the H-bonding
cytosine at the outside strands (Figure 11a). The short distance (2.87 Å) and
approximately linear alignment between the bromines and the phosphate oxygens
between T5 and A6 of the crossing strands, and the orientation of the halogens towards
the nonbonding electrons of the oxygens are all hallmarks of a strong X-bond (Muzet
et al. 2003; Auffinger et al. 2004; Metrangolo et al. 2005), indicating that Br2J adopts
the X-isomer induced by an X-bond.
The H2J structure (Figure 10) shows that the H-isomer can be accommodated by
this crystal lattice. A detailed analysis showed a guanine at the N4 position of the
inside strand and an adenine at the outside strand (Figure 12). Thus, the N7 cytosines
of this structure are located at the cross-over strands, with the uracil bases located
exclusively at the outside strands, indicating that the junction adopts the H-isomer,
opposite of Br2J. The uracils of this construct cannot form X- or H-bonds to the
phosphate backbone; therefore, in the absence of an X-bond, the junction is stabilized
by H-bonding interactions similar to those originally seen in the ACC-J junction
(Eichman et al. 2000).
The Br1J construct is midway between Br2J and H2J, competing a single Xbond against two H-bonds. The resulting structure showed two bromines (~half
occupancy for each of the two symmetry related strands, essentially accounting for the
single bromine of the construct) at the junction cross-over (Figure 11a). This structure
can be interpreted as either a homogeneous population where half the uracils are
brominated and half are unbrominated in each junction, or a heterogeneous mixture of
equal populations of the fully brominated Br2J and unbrominated H2J junctions. The
62
Figure 11. Geometries of X-bonds in Br2J and Br1J. a. Omit electron density maps
contoured at 5σ comparing geometries at the tight U-turns of the Br2J and Br1J
junctions. Closest distances from the bromines to the X-bonded phosphate oxygens
are labeled. b. Overlay of all common DNA atoms for nucleotides N5, N6, and N7 at
the core of the junctions of Br2J (red), Br1J (yellow), H2J (blue) and the previously
published structure of ATC-J (green). Conformational rearrangements are seen at the
N5 nucleotide to allow rotation of the phosphate to form a weak electrostatic
interaction (green arrow) with the methyl group of T7 in ATC-J, halogen bonds
(magenta arrows) to the bromines in Br2J and Br1J, and a hydrogen bond (blue arrow)
to the amino group of C7 in H2J.
63
Figure 12. Electron densities at the N4 nucleotide positions of the outside (continuous)
and inside (crossing) strands that complement the N7 nucleotide forming an H- or Xbond in the H2J junction. The 2Fo-Fc electron density maps calculated at 1σ and 3σ
levels are shown in blue and red, respectively, and the Fo-Fc difference map
calculated at 2σ and 3σ levels shown in yellow and green, respectively. The left-hand
panel shows the N4 nucleotide base of the outside strand of the asymmetric unit
modeled as an adenine, with the N7 nucleotide base of the complementary strand
modeled as a uracil. The 2Fo-Fc map extends away from the C2 carbon towards the
minor groove of the purine base, while the Fo-Fc map shows a strong, spherical
difference density located in the minor groove at a distance equivalent to a C-N
covalent bond from the C2 carbon (arrow). Together, the two sets of electron
densities indicate the presence of an extracyclic N2 amino nitrogen, consistent with
this base being a guanine and its complement at nucleotide N7 being a cytosine. The
right-hand panel shows the base for the same position of the inside stand also modeled
as an adenine. In this case, the calculated electron densities show no evidence for an
N2 amino group, indicating that it is properly modeled as an adenine.
64
Br1J structure is seen to be structurally unique, with no evidence for multiple
conformations; therefore, the crystal represents a nearly homogenous population of
singly-brominated four-stranded DNA complexes in the X-isomer form with
insignificant amounts of Br2J or H2J.
The three junctions are nearly identical in structure, with significant structural
perturbations localized at the core of the junction, consistent with the intramolecular
interactions being the primary determinant of the stability (Eichman et al. 2000; Hays
et al. 2005) and the isomeric form of the junction. Interestingly, the X and H-bonding
pyrimidine bases of N7 overlay almost exactly in all three structures (Figure 11b).
When compared to the analogous junction of d(CCGATATCGG) (ATC-J, with a
thymine replacing BrU), the bromines of Br2J and Br1J do not appear to significantly
affect the base-pair geometry at N7, indicating that the substituents do not perturb the
base pairing or stacking of the DNA arms in the junction. The most dramatic
perturbations are seen as rotations about the β-angle of N6, which subsequently defines
the interaction distance of the phosphate to the N7 nucleotide. We note, however, that
these angles are not unusual for B-DNA duplexes. The Br···O distances for Br2J and
Br1J are closer than the sums of their respective van der Waals radii indicating the
formation of X-bonds, and are closer than the analogous methyl to phosphate distance
in the ATC-J junction. ATC-J is amphimorphic (crystallizing as both junction and BDNA duplex) (Hays et al. 2005), suggesting that the thymine to phosphate
electrostatic interaction is significantly weaker than either the X- or H-bonds. In Br2J
and Br1J, the phosphate oxygens are drawn towards the bromines to establish the
respective X-bonds, similar to the H-bond in H2J.
65
The energies of the X- vs H-bond could be estimated by quantifying the
bromines at nucleotide N7 on the inside and outside strands of Br1J. The estimated
occupancy of bromine was 0.42Br/uracil (± 0.02) (or 0.84 bromines for the two
uracils) at the junction center and 0.05 to 0.15 Br/uracil at the outside strands (Figure
13). This translates to ~1 kcal/mol (range from 0.6 to 1.3 kcal/mol) greater stability for
the X- vs H-isomer. Since Br1J has 0.5 X-bonds/H-bond, the normalized single
bromine X-bond is extrapolated to be twice this, or ~2 kcal/mol more stable than the
N-H···O- H-bond, assuming that there is no entropy difference between the two
conformers of this construct.
Discussion
In the current study, we have shown that bromine type X-bonds compete
effectively against a classic N-H···O- type H-bond in defining the isomeric form of
DNA junctions. This intramolecular assay system is thus an elegant method to
directly compare the ability of X- and H-bonds to direct biomolecular conformation.
Using the distributions of X- and H-isomer forms observed in the Br1J crystals, we
estimate an energy difference between X- and H-bonds in this system to be ~2
kcal/mol. One may ask whether thermodynamic quantities can be derived from
crystallographic assays, where the conformers are not in equilibrium, but locked into a
defined lattice. We had previously applied a similar competitive crystallographic
assay to estimate the relative energies of reverse base pairs in DNA, taking advantage
66
Figure 13. Estimating the bromine occupancy at the N7 nucleotide on the outside
strand (H-isomer) of Br1J (labeled as U17, black triangles). The amount of bromine
located at the N7 position of the Br1J construct was quantified by adding increments of
bromine (in 1%, 2%, 3%, 4%, and 5% increments for five separate studies, closed
triangles) and monitoring the effect on the crystallographic Rfree parameter. This set of
data was fit to a second-order polynomial equation (R of fit = 0.91). The minimum for
the curve and the average data falls at ~0.08 Br. As controls, increments of bromine
atoms were added to pyrimidine bases that should have no bromines in either isomer
form (C2 of the inside strand, open squares, and C18 of the outside strand, grey circles),
and are on the same helical arm as U17.
67
of the contributions of such base pairs to the crystal lattice (Mooers et al. 1997). In the
current system, however, the distribution of isomer forms of the junction in crystals
are defined by the intramolecular interactions that stabilize the conformers, and not by
any differences in the crystal lattice; therefore, we can consider the crystals as
sampling a pre-established equilibrium in solution. Thus, this ~2 kcal/mol is a valid
estimate for the difference in energy of the bromine X-bond vs N-H···O- H-bond in the
Br1J crystals.
The X-bond in the Br1J junction is longer and, consequently, expected to be
weaker than that of Br2J (Figure 11), as might be predicted with only half the
interactions present. Ab initio calculations on atomic models derived from the
coordinates of the Br1J and Br2J crystal structures showed the interaction between the
bromouracil and phosphate groups to be ~3 kcal/mol more favorable in the shorter
Br2J geometry (Figure 14). As with H-bonds, the stabilizing potential of an X-bond
may also depend on competing solvent effects, but an analysis of the solvent
accessible surfaces showed no significant differences in free energies of hydration for
the X- and H-isomeric forms. Thus, each X-bond in Br2J can be estimated (through a
simple thermodynamic cycle) to be ~5 kcal/mol more stable than the H-bond of H2J
(Figure 15). This magnitude of difference in the energies is consistent with the lack of
any observable quantity of the competing H-isomer in the Br2J crystals.
DNA junctions have been applied to a number of different biomolecule based
nanodevices, including molecular arrays and lattices, computers and translational
devices or walkers (Seeman 2005). The ability to direct the isomer form extends the
design aspects of junctions from simple sequence complementarity to true three-
68
Figure 14. Electrostatic potentials from ab initio calculations of the bromine halogen
bonds in Br2J and Br1J show the characteristic anisotropic distribution of charges on
the bromine atoms. Charge potentials were rendered using SPARTAN (Wavefunction,
Inc., Irvine, CA), with negative electrostatic charges shown in red, positive charges in
blue, and neutral charges in green. The difference in energy of the X-bond in the Br2J
geometry is estimated to be ~3 kcal/mol more stable than the one in the Br1J
geometry.
69
Figure 15. Thermodynamic cycle to estimate the free energies of the X- relative to Hbonds. A free energy difference of ~1 kcal/mol was estimated between the X- and Hisomers from the occupancies of bromine at the inside and outside strands of Br1J
(Br1J-X and Br1J-H) (Figure 13), or ~2kcal/mol for 1 X-bond vs 1 H-bond. Since
there is very little contribution of hydration free energy to placing the bromine in the
H- or X-isomeric forms (∆G°hydration ≈ 0), we can assume that the primary effect on the
energy of the X-bonds are electrostatic. Finally, if we assume that there is no effect of
placing a bromine on the outside strand (H-isomer) on the energy of the bromine or
the energy of the hydrogen bond in either the Br1J or Br2J constructs (Br1J-H and Br2JH, bottom of cycle), then the energy of the X-bond in the Br2J construct (∆G˚Br2J-X)
can be estimated as the sum the Br1J X-bond and the difference in electrostatic energy
(∆E) estimated from ab initio calculations (Figure 14).
70
dimensional control of such devices. We can imagine, for example, that cycling
between brominated and nonbrominated strands of a junction would allow a DNA
junction to scissor between isomeric forms where the DNA arms pair and repair in a
well controlled manner.
X-bonds as a general tool in biomolecular engineering, as opposed to nonhydrated organic systems (Corradi et al. 2000), would typically see contributions from
both electrostatic effects from polarization and hydrophobic effects from the
sequestering of the hydrophobic halogen from water. Unlike H-bonds (Baldwin
2003), however, both effects should help to stabilize the X-bond in a buried
environment. By extension from the DNA junction described here, we would thus
expect the stabilizing effects of X-bonds to be even greater in molecular systems
where the halogen is buried, for example within protein folds or at protein-protein or
protein-nucleic acid interfaces, making this an attractive general interaction for
controlling and manipulating biomolecule structures and complexes.
Materials and Methods
Deoxyoligonucleotide sequences (Table 3) were synthesized with the tritylprotecting group attached, and subsequently purified by high performance liquid
chromatography, followed by size exclusion chromatography on a Sephadex G-25
column after detritylation. Crystals were grown by the sitting drop vapor diffusion
method, using solutions of 0.7 mM DNA in 25 mM sodium cacodylate pH 7.0 buffer
71
with 10 to 20 mM calcium chloride and 1.0 to 1.2 mM spermine, and equilibrated
against reservoirs of 26 to 30% aqueous 2-methyl-2,4-pentanediol. Data for the
crystals were collected at liquid nitrogen temperatures using CuKa radiation from a
Rigaku (Tokyo) RU-H3R rotating anode generator with an Raxis-IV image plate
detector, and processed using DENZO and SCALEPACK from the HKL suite of
programs (Otwinowski and Minor 1997). Structures were solved by molecular
replacement using EPMR (Kissinger et al. 1999) with d(CCAGTACbrUGG) (UD0030
PDB code) as the starting model. Distinct solutions with correlation coefficients of
~79% and Rcryst from 39.4% to 45.1% were obtained with two unique strands in the
asymmetric unit—the other pairs of strands of the full four-stranded junctions were
generated from the unique crossing and outside strands of the asymmetric units by the
crystallographic 2-fold symmetry axis running through the center of the fours-stranded
DNA junction. Subsequent refinement of the initial junction structures in CNS
(Brünger et al. 1998) with rigid body refinement, simulated annealing, several rounds
of positional and individual B-factor refinement, and addition of solvent resulted in the
final models for subsequent analyses (Table 4). The brominated structures were
initially refined in the absence of bromines, and the halogens subsequently located
from difference maps. The bromine occupancies on the inside crossing and outside
strands of Br1J were initially refined in CNS, and the errors estimated by monitoring
the range of occupancy values that had little or no effect on the minimum Rfree value of
the refinement (Figure 13).
Ab initio calculations were performed with the program GAMESS (Schmidt et
al. 1993) using the WebMO interface (Schmidt and Polik 2005) for importing and
72
constructing models. The models of the X-bonding interactions were constructed using
the BrU base and the oxygen and phosphorus atoms of the interacting phosphate
groups from the Br1J and Br2J crystal structures. Methyl groups were added to the
equivalent O3’ and O5’ oxygens of the phosphates (with an overall charge of –1) and
to the N1 carbon of the uracil base, and hydrogens were added to complete the valence
states of the appropriate atoms. Ground-state energies for these models were
calculated by density function theory applying the B3LYP function and the 6-31G(d)
basis set (the highest set applicable to row 4 elements). The infinite interaction
distance energy for each complex was approximated as the sum of the energies for the
BrU and phosphate dimethylester components.
To estimate the effects of solvation on the relative free energies of the X- and
H-isomers, we first calculated the solvent accessible surfaces (SAS) of the bromine of
the X-bonding BrU base and the H-bonding N2 amino group of the competing
cytosine base at the outside and inside strands of the respective conformers. The SAS
were translated to free energies of hydration (∆G˚hydroation) using an atomic solvation
parameter (ASP) of 36.8 cal/mol/Å2 for a bromine atom (derived from the difference
in the partition coefficient of bromobenzene versus benzene) and -63.0 cal/mol/Å2 for
the extracyclic amino group according to the following relationship (Kagawa et al.
1989): ∆G˚hydroation = SAS x ASP. The exposed surfaces for the bromine atoms were
calculated to be 12.89 Å2 at the inside strand (X-isomer) and 21.46 Å2 on the outside
strand (H-isomer), while the exposure of the amino group was 10.42 Å2 for the outside
strand (X-isomer) and 14.67 Å2 at the inside strand (H-isomer) of the Br1J junction.
73
This translates to a difference in ∆G˚hydration ≈ –0.2 kcal/mol for the X- versus Hisomers, which is negligible compared to the electrostatic effects.
Acknowledgements
We thank P. A. Karplus and his group for helpful discussions. This work was
funded by a grant from the National Institutes of Health (R1GM62957A). The X-ray
diffraction facilities are supported by the Proteins and Nucleic Acids Facility Core of
the Environmental Health Sciences Center at OSU (NIEHS ES00210) and a grant
from the Murdock Charitable Trust.
74
Chapter 4
The Effect of Polarizability on the Energy of Macromolecular Halogen Bonds
Andrea Regier Voth and P. Shing Ho
Formatted for Submission
75
Summary
Halogen bonds are favorable electrostatic interactions between a polarizable
halogen and a Lewis base such as oxygen or nitrogen. They occur in carbon-bonded
halogens, where an asymmetrical distribution of the halogen’s electrons leaves an
electropositive region at the crown of the atom. These interactions have already been
incorporated into small molecules to design networks of repeated interactions for selfassembling materials and are of interest in the design of biological molecules and
ligands with new properties. We have designed a DNA Holliday junction in which
halogen bonds and classical hydrogen bonds compete to stabilize the junction and
define its conformational isomer. Using this junction in a competitive crystallographic
study, we demonstrate that an iodine halogen bond confers 2 to 9 kcal/mole greater
stability to the junction than a classical hydrogen bond, whereas a fluorine is
significantly less stabilizing than a hydrogen bond. This, along with a previous study
on bromine halogen bonds, allows us to generate the first experimental comparison of
halogen bond stability in biological molecules, showing than F < Br < I and, therefore,
that halogen bond strength can be fine-tuned, according to the type and numbers of
halogens, to match different biomolecular design needs.
76
Introduction
Biological systems utilize a variety of weak intermolecular interactions to
create structures of exquisite stability and specificity. Most of these interactions, such
as hydrogen bonding (H-bonding) and van der Waals interactions, are well known, but
recently we have begun to appreciate that the lesser known interaction of the halogen
bond (X-bond) has also been utilized by nature to control biological interactions, and
that we may also be able to exploit the X-bond in our own design of macromolecular
interactions. The X-bond is a short, primarily electrostatic interaction between a
polarizable halogen and an electron-rich Lewis base (usually an oxygen or nitrogen in
biological molecules). The halogen, usually an organic halide, is polarized along its
σ-covalent bond, leaving an electropositive crown at the tip of the atom, sometimes
referred to as a “σ-hole” (Politzer et al. 2007). This σ-hole acts as a Lewis acid to
interact electrostatically with a Lewis base, and these interactions, which are closer
than the sum of the atoms’ van der Waals radii but longer than covalent bonds, are
now known as X-bonds to emphasize their similarity to the more familiar H-bond.
Better known in the chemical community than in biology, X-bonds have been
used to design self-assembling materials from crystals to fluorinated coatings for over
a decade now (Metrangolo et al. 2005). Indeed, the use of X-bonds to control selfassembly has been exploited to estimate the energy associated with the interaction. A
study in 2000 compared the ability of an X-bond and an H-bond to control the
assembly of a small molecule crystal and found that the iodine X-bond they used was
more stabilizing than an H-bond to either an oxygen or a nitrogen, and estimated the
77
X-bond energy at about 5 kcal/mol (Corradi et al. 2000). In biology, though halogens
are not as common, molecules such as the thyroid hormones utilize iodine X-bonds for
binding to transport and target proteins (Wojtczak et al. 2001). A survey of the
Protein Data Bank in 2004 found almost a hundred X-bonding interactions between
protein and ligands such as drugs and inhibitors (Auffinger et al. 2004).
Recently we reported the first estimate of the stabilization energy provided by
an X-bond in a biological molecule, using the DNA Holliday junction as our
experimental system (Voth et al. 2007). Using a crystallographic competition assay
analogous to the small molecule studies of Corradi et al, we found that a bromine to
oxygen X-bond could out-compete either one or two H-bonds to stabilize the junction,
leading to the estimate that the X-bond is 2-5 kcal/mol more stabilizing than the Hbond in this context. Given the lack of experimental data on X-bonds in
macromolecules and our success in using the Holliday junction to probe these
interactions, we were interested in following up our earlier studies by testing the
strength of other halogens against the same H-bond in the junction. These studies
would create the first self-consistent set of experimental energies for X-bonds and
allow us to elucidate the extent to which halogen polarizability determines X-bond
strength. In this work we show that fluorine, as the least polarizable of the halogens is
generally less able than an H-bond to stabilize the Holliday junction, while iodine, the
most polarizable halogen studied to date also formed the strongest X-bond seen in a
biological molecule.
78
Materials and Methods
For the current studies we designed three deoxynucleotide sequences which
can pair and form four-stranded DNA Holliday junctions, analogous to our previous
brominated constructs (Table 5) (Voth et al. 2007). I2J is a junction construct
designed to compete two iodine X-bonds against two H-bonds, whereas I1J competes
only one iodine X-bond against two H-bonds and F2J competes two fluorine X-bonds
against two H-bonds.
Deoxyoligonucleotides, synthesized with the trityl-protecting
group attached, were purified by high performance liquid chromatography, followed
by size exclusion chromatography on a Sephadex G-25 column after detritylation. I2J
crystals were grown by the sitting drop vapor diffusion method, using a solution of 0.7
mM DNA in 25 mM sodium cacodylate pH 7.0 buffer with 1.0 mM spermine, and
equilibrated against a reservoir of 34% aqueous 2-methyl-2,4-pentanediol (MPD).
Crystals of I1J were grown in the same way in a solution of 0.7 mM DNA in 25 mM
sodium cacodylate pH 7.0 buffer with 1.0 mM spermine and 2 mM CaCl2, equilibrated
against a reservoir of 30% aqueous MPD, and F2J crystals similarly, with 0.2 mM
spermine, 40 mM CaCl2, and a reservoir of 30% MPD.
Data for the I1J and F2J crystals were collected at liquid nitrogen temperatures
using CuKα radiation from a Rigaku (Tokyo) RU-H3R rotating anode generator
operating at 50kV and 100mA with an Raxis-IV image plate detector. Data for the I2J
crystal was collected at liquid nitrogen temperatures at the 14-BM-C station of the
BioCARS CAT at the Advanced Photon Source (Argonne National Laboratory,
Argonne, IL) using an ADSC Quantum-315 detector. All data was processed using
79
Table 5. Constructs and sequences that compete fluorine and iodine X-bonds against
H-bonds in DNA junctions. Shown are the name, sequence and stoichiometries, and
the ratio and type of X-bonds to H-Bonds (X:H) that are designed to compete in each
construct.
Construct
H2J
Sequences
d(CCGGTAUCGG)2 / d(CCGATACCGG)2
X:H
0:2
F2 J
d(CCGGTAflUCGG)2 / d(CCGATACCGG)2
2(F):2
I1 J
d(CCGGTAiUCGG) / d(CCGGTAUCGG) /
d(CCGATACCGG)2
1(I):2
I2 J
d(CCGGTAiUCGG)2 / d(CCGATACCGG)2
2(I):2
80
DENZO and SCALEPACK from the HKL suite of programs (Otwinowski and Minor
1997). Structures were solved by molecular replacement using EPMR (Kissinger et al.
1999) with 2ORG (the Br2J structure) as the starting model. For I2J, a distinct solution
with a correlation coefficient (cc) of 72.4% and Rcryst of 45.4% was obtained.
Similarly, a solution for I1J with a cc of 80.0% and Rcryst of 28.8%, and a solution for
F2J with a cc of 78.5% and Rcryst of 33.2% were found. All three solutions have two
unique strands in the asymmetric unit, with the full four-stranded junction generated
by the crystallographic 2-fold symmetry axis. Subsequent refinement of the initial
junction structures in CNS (Brünger et al. 1998) using rigid body refinement,
simulated annealing, several rounds of positional and individual B-factor refinement,
as well as addition of solvent resulted in the final models for subsequent analyses
(Table 6).
All structures were initially refined in the absence of halogens (at the crossover
and outside strand N7 positions) and with generic purine bases at both N4 positions
(their base-pairing partners—Figure 16). In I2J and I1J, the iodines were subsequently
identified from difference maps, while in F2J a substituent was identified off the C2
atom of the outside N4, indicating that it was a guanine. The halogen occupancies on
the inside crossing and outside strands of all three junctions were initially refined in
CNS, and for I1J their errors were estimated by monitoring the range of occupancy
values that had little or no effect on the Rfree value of the refinement.
Ab initio calculations were performed with the program GAMESS (Schmidt et
al. 1993) with the WebMO interface (WebMO, version 8.0; www.webmo.net )
(Schmidt and Polik 2005) for importing and building models. The models of
81
Table 6. Crystallographic and geometric parameters for the Holliday junction
constructs H2J, F2J, I1J and I2J (for sequences see Table 5). The geometric parameters
describe the rotation of the stacked duplex arms across the junction (Jtwist) and the
rotation (Jroll) and translation (Jslide) of the duplex arms along their respective helical
axes (Watson et al. 2004).
Junction:
H2 J
F2J
Crystallographic Parameters
C2
C2
I1J
I2J
C2
C2
Space group
Unit cell
a (Å)
b (Å)
c (Å)
β-angle
65.61
24.17
36.96
110.01°
65.46
24.35
39.86
118.57°
65.74
25.19
37.17
110.88°
64.96
24.77
37.62
111.59°
Unique reflections
(for refinement)
4,176
(3,992)
2666
(2529)
3562
(3423)
6266
(5772)
Resolution (Å)
1.9
2.1
1.9
1.7
Completeness1
89.6%
(63.4%)
79.4%
(35.1%)
76.4%
(69.9%)
95.4%
(90.3%)
I/sig(I)1
20.41 (3.76)
14.44 (3.58)
11.31 (2.20)
14.14 (3.45)
Rmerge1
5.1%
(25.2%)
6.7%
(22.8%)
4.2%
(24.4%)
6.9%
(25.4%)
Refinement Statistics
Rcryst
(Rfree)
21.8%
(26.8%)
23.7%
(27.6%)
23.6%
(26.3%)
23.9%
(25.5%)
Number of atoms:
DNA (Solvent)
403
(101)
405
(52)
404
(73)
404
(74)
<B-factor> :
DNA (Solvent)
16.3
(20.9)
17.3
(13.6)
30.0
(35.1)
17.3
(21.7)
RMSD bond
lengths (Å)
0.007
0.006
0.005
0.007
RMSD bond angles
1.1°
0.9˚
1.1˚
1.1˚
Junction Geometry
39.6°
37.4˚
137.9°
135.1˚
1.13 Å
0.76 Å
Jtwist
43.2˚
Jroll
143.6˚
Jslide
0.60 Å
RMSD relative
0.0 Å
0.62 Å
0.68 Å
to H2J
1
Values for the highest resolution shell are shown in parentheses
42.8˚
150.3˚
0.55 Å
0.93 Å
82
Figure 16. DNA Holliday junction numbering and stabilization. a. The junction is
composed of 4 strands with nucleotides numbered as shown here. The outer strands
are in blue and pink while the crossover strands are in orange and green. b. The
compact stacked-X form of the junction, shown here, is stabilized by two interactions
on each crossover strand (only one of each is shown here). Figure adapted from
(Khuu et al. 2006).
83
the halogen bonding interactions were constructed by importing the halogenated base
and the oxygen and phosphorus atoms of the interacting phosphate groups from each
crystal structure. Methyl esters were added to the equivalent O3’ and O5’ oxygens of
the phosphate groups, and hydrogens added to complete the valence states of the
atoms. The complexes were assigned overall charges of –1. Ground-state energies for
these models were calculated by density function theory applying the B3LYP function
and a 3-21G basis set (the highest set applicable to row 5 elements) for the I2J and I1J
models, while the 6-311+G(d,p) basis set was used for the F2J model. The infinite
interaction distance energy for each complex was approximated as the sum of the
energies for the halogenated uracil and phosphate dimethylester components.
To estimate the effects of solvation on the relative free energies of the XB- and
HB-isomers, we first calculated the solvent-accessible surfaces (SAS) of the halogens
on the uracil bases and the N4 amino groups on the cytosines that trade positions on
the outside and crossover strands of the respective conformers. The SAS were
translated to free energies of hydration (ΔG0hydration) by using atomic solvation
parameters (ASP) of 163.0 J/mol/Å2 for iodine, 41.1 J/mol/Å2 for fluorine, and -263.8
J/mol/Å2 for the extracyclic amino group according to the following relationship
(Kagawa et al. 1989): ΔG0hydration = SAS*ASP. The ASP values for extracyclic iodine,
fluorine and amino groups were derived from the difference in the octanol/water
partition coefficients of iodobenzene and benzene, fluorobenzene and benzene, and
aniline and benzene, respectively.
84
Results
Experimental Design
A DNA Holliday junction was used as the experimental system for measuring
the difference in stabilization provided by a X-bond and a H-bond. The Holliday
junction has been seen to be stabilized in its compact, stacked-X form by two
intramolecular interactions that take place on each crossover strand of the junction
(Figure 16). These interactions keep the DNA in its four-stranded junction form rather
than allowing the junction to migrate off the ends of the DNA to resolve into two BDNA duplexes. Typically these interactions are H-bonds between a phosphate oxygen
and a base substituent (such as an extracyclic amine on a cytosine) (Hays et al. 2003).
In cases where a halogenated base is present at certain positions, however, X-bonds
between a phosphate oxygen and the halogen substituent have been observed to
replace H-bonds in stabilizing the junction (Hays et al. 2003).
For this and a previous study (Voth et al. 2007), we have designed a junction
where the essential H-bonding interaction at the N8 cytosine is left intact, but a second,
more variable interaction at nucleotide 7, N7, is designed so that half the strands have
the potential to form a classical H-bond, and the other half have the potential to form
an X-bond. The sequences for the constructs are designed so that each junction must
have two of each type of strand, and thus the junction becomes an arena for competing
X-bonds against H-bonds. If the H-bond is stronger and more stabilizing, it will be
found at the crossover strands of the junction (called the H-isomer) while the strands
with the halogens will be partitioned to the outside of the junction where they cannot
85
participate in any intramolecular interactions. If, on the other hand, the X-bond is
stronger and more stabilizing, it will be found at the crossover strands and the junction
will be in the X-isomer with the H-bonding nucleotide partitioned to the outside strand
(Figure 17). To determine which interaction is occurring at the crossover, we
crystallize the junction and solve its structure to observe the position of the halogen,
making this a crystallographic competition assay between the X- and H-bond. Since
the outside of the junction where the crystal contacts occur is identical no matter
which interaction is at the center of the junction, the crystal lattice itself should impose
no particular preference for either the X- or H-isomeric form. This is borne out by the
observation that all the junctions of this general construct type solved to date are in
essentially isomorphous unit cells, which implies that the crystal lattice does not affect
the isomeric form observed in the crystal (see Table 6). Therefore, the isomer or
mixture of isomers seen in the crystal structure should solely reflect the equilibrium
distribution of isomers in the crystallization solution.
The reference construct for this crystallographic competition assay, to which
all other constructs will be compared, is the H2J junction (see Table 5) which
competes two H-bonds against zero X-bonds. H2J, which was solved previously
(Voth et al. 2007), was observed to adopt the H-isomer with the cytosine N7 at the
junction crossover forming an H-bond to the phosphate oxygen of the N6 position
(Figure 16b). The alternative X-isomeric form would place a uracil at the competing
position of the crossover stands, but uracil (which can only serve as an H-bond
acceptor at this position) cannot directly H-bond to a phosphate oxygen. By adding
iodine or fluorine to this uracil, however, we have created constructs that can
86
Figure 17. Schematic of the competition assay using the DNA Holliday junction.
Top left: In the I2J and F2J constructs, equimolar ratios of the H-bonding and Xbonding strands are mixed. Top right: If the X-bond is more stabilizing to the
junction, it will be present at the crossover, directing the junction to form the Xisomer. Bottom left: If the H-bond is more stabilizing, it will direct the junction to
form the H-isomer. Bottom right: Due to base pairing constraints, the H- and Xisomers are the only two forms the stacked-X junction can take, but they can freely
interconvert via the unstacked, open-X conformation shown here.
87
potentially compete different types and numbers of X-bonds against the H-bond of
H2J, giving us the ability to directly compare the X-bonds against each other (and
against the bromine X-bond that we had previously characterized). Here we report the
results of three new constructs: I2J, I1J, and F2J, which compete two or one iodine Xbonds and two potential fluorine X-bonds against the two H2J H-bonds.
The I2J Structure
The I2J structure shows that iodine X-bonds at this position are more effective
(in a head-to-head, 1:1 competition) at stabilizing than H-bonds. The electron density
maps of this junction showed very strong positive difference density off the C5 of the
generic N7 base at the crossover, and no difference density off the N7 on the outside
strand, indicating the iodine was at the crossover and the junction was entirely in the
X-isomer (Figure 18). The position of the iodine after refinement showed a strong Xbond with almost ideal geometry—the interaction is characterized by a short iodine to
oxygen distance (3.01Å) and close to linear approach of the iodine electropositive
crown to the phosphate oxygen’s non-bonding electrons (170.7˚ Θ1 and 113.1˚ Θ2).
Final occupancy refinement of the iodine indicated that it resided entirely (within the
error of the method) at the crossover of the junction in the X-isomer.
In the I2J construct, the only difference between the junction in the X- and Hisomer would be the switching of the X-bonding 5-iodouracil base and the H-bonding
cytosine base at the N7 positions of the crossover and outside strands, a difference that
could impart different solvation energies to the two isomers. We quantified the
potential hydrophobic effect using the accessible surface areas of the iodines of the 5-
88
Figure 18. 5σ omit density for the iodine off the C5 of the crossover N7 uracil in the
I2J structure indicates that the junction is in the X-isomer and that the iodine is Xbonded to a phosphate oxygen.
89
iodouracil and the amino groups of cytosine bases in both the XB- and the putative Hisomers. Using atomic solvation parameters derived from partition coefficients, we
estimate that the free energy of hydration adds approximately 0.5 kcal/mol of
stabilization to the X-isomer compared to the H-isomer due to the increased burial of
the hydrophobic iodine. If this were the only energy difference between the two forms
of the junction, however, we would expect to observe about 30% of the junction in the
H-isomer. Since only the X-isomer is observed in this construct, we can conclude that
an additional stabilizing interaction in the form of an X-bond is contributing
significant energy to the stabilization of the X-isomer.
The I1J Structure
The I1J construct represents the halfway titration between the I2J and H2J
constructs, competing one X-bond against two H-bonds. Since the I2J and H2J are
apparently true end-points, adopting only the X- and H-isomers, respectively, in their
crystals, the I1J construct was designed to potentially help us to quantify the difference
in the energy of the iodine X-bond over the H-bond. In this structure, as in the I2J
structure, there is strong positive difference density observed off the C5 of the N7 base
at the crossover (Figure 19), and none off the N7 base on the outside strand, indicating
the iodine is at the crossover and the junction is in the X-isomer. The iodine X-bond
in this structure is not quite as close or well aligned (3.15 Å, 167.9˚ Θ1 and 97.6˚ Θ2)
as that in the I2J structure, but at 3.15 Å is still shorter than the sum of the atoms’ van
der Waals radii (3.4 Å). Refinement of the occupancies of the iodine at both N7
positions indicates that there is 50% (or slightly more) iodine at each crossover strand
90
Figure 19. 5σ omit density for the iodine off the C5 of the crossover N7 uracil in the
I1J structure indicates that the junction is in the X-isomer and that the iodine is Xbonded to a phosphate oxygen.
91
(or one iodine split between two crossover strands) and, at most, 1% at each outside
strand. Further manual occupancy refinement was accomplished by monitoring how
Rfree changed as occupancy was added or taken away from each iodine. These
calculations confirm that there is at least 50% iodine occupancy at each crossover
strand, and while they suggest that there may also be some amount of iodine at each
outer strand, the amount is below the quantifiable threshold.
Because I1J is the halfway titration between the I2J and H2J constructs, the
three DNA strands in this construct have the ability to base pair into a 1:1 mixture of
I2J and H2J junctions, or to form a hybrid junction with one potential X-bond and two
potential H-bonds. The crystal structure of I1J showed no indications of alternate
backbone conformations, which would be expected in a high resolution structure such
as this if it were a 50%/50% mixture between two such different structures (see Table
6 – Junction Geometry). The geometry of the refined I1J structure itself was also not a
simple average of the I2J and H2J geometries as one would expect in a 50%/50%
mixture. There were indications, however, of some amount of an alternate
conformation being present. The base pairing partners of the crossover strand and
outside strand N7 bases (the N4 bases) were initially refined as generic purines, and the
crossover strand N4 showed clear difference density off its C4 (indicating it was a
guanine, consistent with it base-paring to a cytosine at the outer strand N7 in the Xisomer). The purine at the outer strand N4 also showed a smaller amount of difference
density near its C4 (Figure 20). The outer N4 would be a guanine if the I1J junction
were in the H-isomer, but the lack of difference density for the iodine on the
complementary pyrimidine at the outside strand makes this unlikely. It is more likely
92
Figure 20. Electron density map around the outside N4 in junction I1J shows some
indication of a missing extracyclic amine group off C2. An amine in this position
would make this base a guanine and indicate the presence of some H2J junction in the
crystal. Blue and red density are 1σ and 3σ 2Fo-Fc contours, respectively. Yellow
and green are +2σ and +3σ Fo-Fc contours.
93
that there would be some guanine present at the outer N4 because there is some small
but observable quantity of H2J in the H-isomer present in the crystal. Together, these
observations lead us to conclude that the I1J crystal is neither entirely made up of the
hybrid I1J junction, nor of a 50%/50% mixture of the I2J and H2J junctions, but is
instead closer to what would be expected for a statistical distribution of 25% I2J, 50%
I1J, and 25% H2J.
We have previously shown that with similar quality data we could detect an
approximately 5 to 15% occupied bromine, or about 1.75 electrons. By analogy, an
iodine, which has significantly more electrons than bromine, could therefore be
reliably quantified down to about 3% occupancy. Since we don’t observe any iodine
on the outside strand, the upper limit to the amount of H-isomer of I1J that can be
present is 12% (6% occupied iodine on each outside strand of the 50% I1J in the
crystal leading to an observation of 3% occupied iodine). This places the lower limit
to the difference in Gibbs free energy between the X- and H-isomers at -1.2 kcal/mol.
Once again the increased solvent accessibility of the iodine in the H-isomer provides
part of the X-isomer’s stabilization, quantified as 0.4 kcal/mol. This leaves the one Xbond to provide a lower limit of approximately 1 kcal/mol stabilization over the two
H-bonds present in the H-isomer. Normalizing this to a one-to-one comparison, this
structure shows that this iodine X-bond has a lower limit of 2 kcal/mol greater stability
than one H-bond.
94
The F2J Structure
The F2J construct was designed to test the ability of fluorine to X-bond in a
macromolecular context. Given that fluorine is similar in scattering power to oxygen
and that this dataset is at lower resolution than the I1J and I2J datasets, it is not
surprising that no large difference density peaks were observed in the initial maps to
indicate the position of the halogen. The generic purine bases at the outside N4
positions instead showed difference density off the C4 of the ring, indicating that the
outside N4 should have an extracyclic amine, making it a guanine. This guanine
would base-pair to a cytosine on the crossover strand, making F2J primarily the Hisomer (see Figure 21). Occupancy refinement of both an extracyclic amine on each
N4 and a fluorine on each N7 gave a ratio of 20-25% X-isomer to 75-80% H-isomer.
The H-bond is therefore more stabilizing to the junction than whatever interaction the
fluorine is able to make with the phosphate oxygen. Indeed the difference in solvation
energies between the two isomers favors the X-isomer slightly (0.2 kcal/mol) and the
observed ratios of the isomers give an energy difference of about 0.8 kcal/mol (in
favor of the H-isomer), giving a rough estimate for the interaction between the
fluorine and the phosphate oxygen present in the X-isomer as about 1 kcal/mol less
stabilizing than the H-bond present in the H-isomer.
The low occupancy of the X-isomer and the lower resolution of this structure
mean that we cannot unambiguously discern where the fluorine is in relation to the
phosphate oxygen when it is in the X-isomer. If we simply model it in on top of the
cytosine that is at the N7 position in the H-isomer we see an extremely close
interaction (2.4 Å) similar to an X-bond, but quantum mechanical calculations on this
95
Figure 21. Electron density maps for the crossover (top) and outside (bottom) N4
bases of the F2J junction. The crossover N4 fits correctly as an adenine, but the
outside N4 has both 2Fo-Fc and Fo-Fc density off its C2 indicating that it should be a
guanine, meaning the junction is in the HB-isomer. Blue and red density are 1σ and
3σ 2Fo-Fc contours, respectively, while yellow is the +2σ Fo-Fc contour.
96
model indicate that this distance should be very destabilizing (by ca. 10 kcal/mol),
making this a somewhat unlikely possible location. Instead it seems more likely that
the fluorine in the X-isomer is positioned further away from the base but that it is at
too low an occupancy to observe it in this structure. Despite these complexities, F2J is
mostly found to exist in the H-isomer and its structure is found to be most similar to
that of H2J, the only other junction of this construct to date to be solved in the Hisomeric form (Table 5 and Figure 22). Therefore F2J shows that fluorine is unlikely
to form X-bond in a macromolecular environment.
Discussion
We have shown here that the iodine X-bond effectively out-competes both an
equal or double number of H-bonds to stabilize the X-isomer form of a Holliday
junction, with the energy of the X-bond estimated to be at least 2 kcal/mol stronger
than the H-bond. How much more favorable might it be? By comparing the I1J and
I2J structures we can explore how geometry may affect the energetics of the iodine Xbond. The I2J X-bond is closer and has a more idealized geometry for an X-bond than
the I1J X-bond (Figure 22), which should, therefore, make it a stronger interaction. Ab
initio quantum mechanical energy calculations on models of the two X-bonds taken
from the geometries of the crystal structures suggest that there is a 5 kcal/mol increase
in the stabilization provided by the improved geometry of the I2J X-bond over the I1J
X-bond. Although the absolute values given by these calculations may not yet be very
97
Figure 22. Overlay of the crossover nucleotides of H2J, F2J, I2J, and I1J. The H2J
(blue) and F2J (green) structures overlay almost exactly, indicating that the F2J
structure is primarily in the H-isomer. In the very similar I2J (red) and I1J (yellow)
structures, both the N6 phosphate and the N7 base rearrange slightly to accommodate
the iodine substituent, aligning it with a N6 phosphate oxygen. Fitting of all the
common DNA atoms for these structures was done in the program Profit V2.5 (Martin
2005) using the McLachlan algorithm (McLachlan 1982).
98
accurate, the relative difference between the two should be more robust. Assuming,
then, that other differences between the two constructs are minimal (due to their
highly similar sequences and structures), each I2J X-bond is then seen, using a
thermodynamic cycle, to be at least 5 kcal/mol more favorable than an H-bond. This
gives a range of iodine X-bond strengths from 2 to 7 kcal/mol more favorable than the
H-bond at this position of the junction. At the other extreme, the F2J structure shows a
fluorine to oxygen interaction in this system to be roughly 1 kcal/mol less stabilizing
than an H-bond, being the first halogenated base that we have observed partitioned to
the outside of the junction (H-isomer) using this construct.
In this and one earlier study, we have shown that the Holliday junction is able
to accommodate a variety of halogens. The junction partitions fluorine to the outside
strand in favor of stabilization from an H-bond, but prefers the extra 2-6 kcal/mol of
stabilization provided by a bromine X-bond and the greater than 2-7 kcal/mol of
stabilization from an iodine X-bond over an H-bond (here using the same basis set for
the ab initio calculations for direct comparison between the bromine and iodine Xbond strengths). Thus, we can place the stability of X-bonds in this system as F < Br
< I. This trend is contrary to the overall substituent effects of the halogens as
indicated by their Hammett substituent constants (Streitwieser and Heathcock 1976),
supporting our interpretation that the difference in stabilization between the isomeric
forms of the junction is indeed due to the strength of the halogen bond rather than
substituent effects on the uracil. Therefore, in this first systematic study of X-bond
strength, the results support the explanation of halogen polarization being the largest
contributor to the formation and strength of the X-bond. Moreover, the range of X-
99
bonds strengths so far observed is a promising sign for the utility of X-bonds in
macromolecular design. For some applications, such as drug design, the ability to
balance halogen hydrophobicity with X-bond strength is most important, while for the
design of molecular machines that must straddle two different conformations it will be
the ability to fine-tune the X-bond energy that will eventually make X-bonds a useful
tool. By selecting a different halogen, we can therefore specify the strength of an Xbond relative to a competing H-bond and, in this way, control the conformational
equilibrium of a biological macromolecule.
Acknowledgements
We thank Karolyn Luger and Mark van der Woerd for computing facilities and
support. This work was supported by grants from the National Institutes of Health
(R1GM62957A) and the Medical Research Foundation of Oregon. The x-ray
diffraction facilities at OSU are supported by the Proteins and Nucleic Acids Facility
Core of the Environmental Health Sciences Center at OSU (National Institute on
Environmental Health Sciences Grant ES00210) and by a grant from the Murdock
Charitable Trust. Use of the Advanced Photon Source was supported by the U.S.
Department of Energy, Basic Energy Sciences, Office of Science, under Contract No.
W-31-109-Eng-38. Use of the BioCARS Sector 14 was supported by the National
Institutes of Health, National Center for Research Resources, under grant number
RR07707.
100
Chapter 5
Conclusion and Discussion
Until quite recently, the ability of halogens to form stabilizing interactions with
electron donors has been largely overlooked in biology. Once aware of this ability,
however, the potential for utilizing this interaction in a variety of biological contexts
becomes clear. We have pointed out in the studies described here, the role of X-bonds
in the binding of ligands such as drugs to protein targets and in driving the
conformation of flexible macromolecular structures such as the Holliday junction.
From these examples of using and exploring the X-bond, three basic conclusions can
be drawn.
X-bonds can be stabilizing inter- and intra-molecular interactions in biological
macromolecules
We have provided examples of inhibitors with nanomolar inhibition constants
whose only polar interactions are X-bonds (Battistutta et al. 2005), as well as X-bonds
determining the conformation of macromolecular complexes (Voth et al. 2007), so this
conclusion may seem self-evident. Nonetheless, the different environments provided
by most of the small molecule systems in which X-bonds have been used and that of a
biological system make this seemingly simple conclusion necessary. Specifically, the
101
aqueous nature of biological systems complicates our understanding and prediction of
X-bond stabilization, and there is still work to be done to clarify the role of halogen –
solvent X-bonding interactions. Our studies show, however, that solvent interactions
are often outweighed by the intrinsic strength of the X-bond, as in our estimation of
the iodine X-bond strength in Chapter 4, where we subtracted out our best estimate of
the stabilization provided by shielding the iodine, and still retained a minimum of 2 to
9 kcal/mol greater stabilization than the H-bond purely due to X-bond formation. So
not only are X-bonds a stabilizing interaction in biological molecules, they are also
often made even more stabilizing by the hydrophobicity of the halogen itself which
tends to push it towards the core of a macromolecule where X-bond acceptors are
present.
X-bonds occur in a variety of strengths and can be significantly stronger than Hbonds
Like H-bonds, X-bonds occur in a variety of geometries which ab initio
quantum mechanical calculations such as those in Chapters 3 and 4 indicate should
lead to different interaction energies. X-bonds also occur with a variety of acceptors,
all the way from the partially negatively charged phosphate oxygen to the delocalized
π-system of a phenyl ring (first characterized in Chapter 2) which, by analogy to the
H-bond, should have some of the strongest and weakest interaction energies,
respectively. In neither case (interaction geometry or X-bond acceptor) have we been
able to experimentally test this range of interaction energies, but in terms of the
halogen itself, we have. The iodine X-bond in the I1J junction stabilized the junction
102
more effectively than the bromine X-bond of the Br1J junction, giving the first
experimental measure of the range of X-bond strengths in macromolecules. And in
both cases these X-bonds were significantly stronger than the competing H-bond,
indicating that, in the correct geometry and with the correct acceptor atoms, X-bonds
have an energy range useful for the design and control of macromolecular interactions.
The ability of a halogen to form a stabilizing X-bond is best described by its
polarizability
The experiments described in Chapters 3 and 4 are the first X-bond energy
estimates done on multiple halogens using the same experimental construct, and thus
provide the first series of energy estimates that can be directly compared to one
another. Using this comparison to the same H-bond in the Holliday junction, we see
that a fluorine interaction (though probably not an X-bond) is less stabilizing than the
H-bond, while a bromine X-bond is 2-6 kcal/mol more stabilizing than the H-bond and
the iodine X-bond is over 2-7 kcal/mol more stabilizing than the H-bond. This order
mirrors several characteristics of the halogens, including their size, polarizability, and,
by some measures, their hydrophobicity. It does not seem logical that size would be
the determining factor in this series, given that it is the larger halogens that are
partitioned to the more crowded crossover of the junction in these studies, so we will
not attempt to make an argument for it. The hydrophobicity of the halogens seems a
more reasonable explanation, but our best estimates of the energy differences
associated with solvation for each halogen being at the crossover of the junction
instead of the outside do not account for more than a fraction of the preference we see
103
(and for fluorine are even contrary to the observed preference of the junction).
Therefore the polarizability of the halogens, as suggested by ab initio quantum
mechanical calculations (Lommerse et al. 1996; Auffinger et al. 2004), is the best
explanation for this trend of F < Br < I X-bond stabilization in biological
macromolecules.
Together, these observations point to the suitability of X-bonds for
biomolecular engineering applications. X-bonds in macromolecules are seen to be
well-behaved interactions, tunable according to the acceptor and donor atom types and
the solvation environment and sufficiently stabilizing to make important contributions
to biomolecular recognition and interactions. Hopefully continued study in this area
will elucidate further the role that X-bonds can and have played in a macromolecular
context.
104
Bibliography
Auffinger, P., F. A. Hays, E. Westhof and P. S. Ho (2004). "Halogen bonds in
biological molecules." Proc Natl Acad Sci U S A 101(48): 16789-94.
Baldwin, R. L. (2003). "In search of the energetic role of peptide hydrogen bonds." J
Biol Chem 278(20): 17581-8.
Battistutta, R., E. De Moliner, S. Sarno, G. Zanotti and L. A. Pinna (2001). "Structural
features underlying selective inhibition of protein kinase CK2 by ATP site-directed
tetrabromo-2-benzotriazole." Protein Sci 10(11): 2200-6.
Battistutta, R., M. Mazzorana, S. Sarno, Z. Kazimierczuk, G. Zanotti and L. A. Pinna
(2005). "Inspecting the structure-activity relationship of protein kinase CK2 inhibitors
derived from tetrabromo-benzimidazole." Chem Biol 12(11): 1211-9.
Bogoyevitch, M. A. (2006). "The isoform-specific functions of the c-Jun N-terminal
Kinases (JNKs): differences revealed by gene targeting." Bioessays 28(9): 923-34.
Brisdon, A. (2002). "Halogens and noble gases." Annu. Rep. Prog. Chem., Sect. A 98:
107-114.
Brünger, A. T., P. D. Adams, G. M. Clore, W. L. DeLano, P. Gros, R. W. GrosseKunstleve, J. S. Jiang, J. Kuszewski, M. Nilges, N. S. Pannu, R. J. Read, L. M. Rice,
T. Simonson and G. L. Warren (1998). "Crystallography & NMR system: A new
software suite for macromolecular structure determination." Acta Crystallogr D Biol
Crystallogr 54 ( Pt 5): 905-21.
Buss, I. H., R. Senthilmohan, B. A. Darlow, N. Mogridge, A. J. Kettle and C. C.
Winterbourn (2003). "3-Chlorotyrosine as a marker of protein damage by
myeloperoxidase in tracheal aspirates from preterm infants: association with adverse
respiratory outcome." Pediatr Res 53(3): 455-62.
Cobb, M. H. and E. J. Goldsmith (1995). "How MAP kinases are regulated." J Biol
Chem 270(25): 14843-6.
Cohen, P. (2002). "Protein kinases--the major drug targets of the twenty-first
century?" Nat Rev Drug Discov 1(4): 309-15.
Corradi, E., S. V. Meille, M. T. Messina, P. Metrangolo and G. Resnati (2000).
"Halogen bonding versus hydrogen bonding in driving self-assembly processes."
Angew. Chem. Int. Ed. 112: 1852-1856.
105
Davies, T. G., J. Bentley, C. E. Arris, F. T. Boyle, N. J. Curtin, J. A. Endicott, A. E.
Gibson, B. T. Golding, R. J. Griffin, I. R. Hardcastle, P. Jewsbury, L. N. Johnson, V.
Mesguiche, D. R. Newell, M. E. Noble, J. A. Tucker, L. Wang and H. J. Whitfield
(2002). "Structure-based design of a potent purine-based cyclin-dependent kinase
inhibitor." Nat Struct Biol 9(10): 745-9.
Davis, S. T., B. G. Benson, H. N. Bramson, D. E. Chapman, S. H. Dickerson, K. M.
Dold, D. J. Eberwein, M. Edelstein, S. V. Frye, R. T. Gampe Jr, R. J. Griffin, P. A.
Harris, A. M. Hassell, W. D. Holmes, R. N. Hunter, V. B. Knick, K. Lackey, B.
Lovejoy, M. J. Luzzio, D. Murray, P. Parker, W. J. Rocque, L. Shewchuk, J. M. Veal,
D. H. Walker and L. F. Kuyper (2001). "Prevention of chemotherapy-induced alopecia
in rats by CDK inhibitors." Science 291(5501): 134-7.
De Moliner, E., N. R. Brown and L. N. Johnson (2003). "Alternative binding modes of
an inhibitor to two different kinases." Eur J Biochem 270(15): 3174-81.
Duckett, D. R., A. I. H. Murchie, S. Diekmann, E. von Kitzing, B. Kemper and D. M.
J. Lilley (1988). "The structure of the Holliday junction, and its resolution." Cell 55:
79-89.
Dumas, J.-M., H. Peurichard and M. J. Gomel (1978). Chem. Res. (S): 54-57.
Eichman, B. F., J. M. Vargason, B. H. M. Mooers and P. S. Ho (2000). "The Holliday
junction in an inverted repeat sequence: sequence effects on the structure of four-way
junctions." Proc. Natl. Acad. Sci. USA 97: 3971-3976.
Foster, R. (1969). Organic Charge-Transfer Complexes. London, Academic Press.
Grainger, R. J., A. I. H. Murchie and D. M. J. Lilley (1998). "Exchange between
stacking conformers in a four-way DNA junction." Biochemistry 37: 23-32.
Gray, N. S., L. Wodicka, A. M. Thunnissen, T. C. Norman, S. Kwon, F. H. Espinoza,
D. O. Morgan, G. Barnes, S. LeClerc, L. Meijer, S. H. Kim, D. J. Lockhart and P. G.
Schultz (1998). "Exploiting chemical libraries, structure, and genomics in the search
for kinase inhibitors." Science 281(5376): 533-8.
Guthries, F. (1863). "On the iodide of iodammonium." J. Am. Chem. Soc. 16: 239244.
Hassel, O. (1970). "Structural aspects of interatomic charge-transfer bonding." Science
170: 497-502.
Hassel, O. (1972). Structural aspects of interatomic charge-transfer bonding. Nobel
lectures, Chemistry 1963-1970. Amsterdam, Elsevier publishing company.
106
Hassel, O. and J. Hvoslef (1954). Acta chemica Scandinavica 8: 873.
Hays, F. A., V. Schirf, P. S. Ho and B. Demeler (2006). "Solution formation of
Holliday junctions in inverted-repeat DNA sequences." Biochemistry 45(8): 2467-71.
Hays, F. A., A. Teegarden, Z. J. Jones, M. Harms, D. Raup, J. Watson, E. Cavaliere
and P. S. Ho (2005). "How sequence defines structure: a crystallographic map of DNA
structure and conformation." Proc Natl Acad Sci U S A 102(20): 7157-62.
Hays, F. A., J. M. Vargason and P. S. Ho (2003). "Effect of sequence on the
conformation of DNA holliday junctions." Biochemistry 42(32): 9586-97.
Hays, F. A., J. Watson and P. S. Ho (2003). "Caution! DNA crossing: crystal
structures of Holliday junctions." J Biol Chem 278(50): 49663-6.
Herrera, R. and J. S. Sebolt-Leopold (2002). "Unraveling the complexities of the
Raf/MAP kinase pathway for pharmacological intervention." Trends Mol Med 8(4
Suppl): S27-31.
Hester, J. B., J. K. Gibson, L. V. Buchanan, M. G. Cimini, M. A. Clark, D. E. Emmert,
M. A. Glavanovich, R. J. Imbordino, R. J. LeMay, M. W. McMillan, S. C. Perricone,
D. M. Squires and R. R. Walters (2001). "Progress toward the development of a safe
and effective agent for treating reentrant cardiac arrhythmias: synthesis and evaluation
of ibutilide analogues with enhanced metabolic stability and diminished proarrhythmic
potential." J Med Chem 44(7): 1099-115.
Himmel, D. M., K. Das, A. D. Clark, Jr., S. H. Hughes, A. Benjahad, S. Oumouch, J.
Guillemont, S. Coupa, A. Poncelet, I. Csoka, C. Meyer, K. Andries, C. H. Nguyen, D.
S. Grierson and E. Arnold (2005). "Crystal structures for HIV-1 reverse transcriptase
in complexes with three pyridinone derivatives: a new class of non-nucleoside
inhibitors effective against a broad range of drug-resistant strains." J Med Chem
48(24): 7582-91.
Jautelat, R., T. Brumby, M. Schafer, H. Briem, G. Eisenbrand, S. Schwahn, M.
Kruger, U. Lucking, O. Prien and G. Siemeister (2005). "From the insoluble dye
indirubin towards highly active, soluble CDK2-inhibitors." Chembiochem 6(3): 53140.
Kagawa, T. F., D. Stoddard, G. W. Zhou and P. S. Ho (1989). "Quantitative analysis
of DNA secondary structure from solvent-accessible surfaces: the B- to Z-DNA
transition as a model." Biochemistry 28(16): 6642-51.
Khuu, P. A., A. R. Voth, F. A. Hays and P. S. Ho (2006). "The stacked-X DNA
Holliday junction and protein recognition." J. Mol. Recognit. 19: 1-9.
107
Kissinger, C. R., D. K. Gehlhaar and D. B. Fogel (1999). "Rapid automated molecular
replacement by evolutionary search." Acta Crystallogr. D 55: 484-491.
Knockaert, M., P. Greengard and L. Meijer (2002). "Pharmacological inhibitors of
cyclin-dependent kinases." Trends Pharmacol Sci 23(9): 417-25.
Kolch, W. (2000). "Meaningful relationships: the regulation of the
Ras/Raf/MEK/ERK pathway by protein interactions." Biochem J 351 Pt 2: 289-305.
Liu, M., Z. Xin, J. E. Clampit, S. Wang, R. J. Gum, D. L. Haasch, J. M. Trevillyan, C.
Abad-Zapatero, E. H. Fry, H. L. Sham and G. Liu (2006). "Synthesis and SAR of 1,9dihydro-9-hydroxypyrazolo[3,4-b]quinolin-4-ones as novel, selective c-Jun N-terminal
kinase inhibitors." Bioorg Med Chem Lett 16(10): 2590-4.
Lommerse, J. P. M., A. J. Stone, R. Taylor and F. H. Allen (1996). "The nature and
geometry of intramolecular interactions between halogens and oxygen or nitrogen." J.
Am. Chem. Soc. 118: 3108-3116.
Lu, H. and U. Schulze-Gahmen (2006). "Toward understanding the structural basis of
cyclin-dependent kinase 6 specific inhibition." J Med Chem 49(13): 3826-31.
Manning, G., D. B. Whyte, R. Martinez, T. Hunter and S. Sudarsanam (2002). "The
protein kinase complement of the human genome." Science 298(5600): 1912-34.
Markwalder, J. A., M. R. Arnone, P. A. Benfield, M. Boisclair, C. R. Burton, C. H.
Chang, S. S. Cox, P. M. Czerniak, C. L. Dean, D. Doleniak, R. Grafstrom, B. A.
Harrison, R. F. Kaltenbach, 3rd, D. A. Nugiel, K. A. Rossi, S. R. Sherk, L. M. Sisk, P.
Stouten, G. L. Trainor, P. Worland and S. P. Seitz (2004). "Synthesis and biological
evaluation of 1-aryl-4,5-dihydro-1H-pyrazolo[3,4-d]pyrimidin-4-one inhibitors of
cyclin-dependent kinases." J Med Chem 47(24): 5894-911.
Martin, A. C. R. (2005). "ProFit V2.5." from
http://www.bioinf.org.uk/software/profit/.
McDonald, I. K. and J. M. Thornton (1994). "Satisfying hydrogen bonding potential in
proteins." J Mol Biol 238(5): 777-93.
McKinney, S. A., A. C. Declais, D. M. Lilley and T. Ha (2003). "Structural dynamics
of individual Holliday junctions." Nat Struct Biol 10(2): 93-7.
McLachlan, A. D. (1982). "Rapid Comparison of Protein Structures." Acta Cryst A
38: 871-873.
Meggio, F. and L. A. Pinna (2003). "One-thousand-and-one substrates of protein
kinase CK2?" Faseb J 17(3): 349-68.
108
Metrangolo, P., H. Neukirch, T. Pilati and G. Resnati (2005). "Halogen bonding based
recognition processes: A world parallel to hydrogen bonding." Acc Chem Res 38(5):
386-395.
Metrangolo, P. and G. Resnati (2001). "Halogen bonding: a paradigm in
supramolecular chemistry." Chemistry 7(12): 2511-9.
Meyer, E. A., R. K. Castellano and F. Diederich (2003). "Interactions with aromatic
rings in chemical and biological recognition." Angew Chem Int Ed Engl 42(11): 121050.
Miick, S. M., R. S. Fee, D. P. Millar and W. J. Chazin (1997). "Crossover isomer bias
is the primary sequence-dependent property of immobilized Holliday junctions." Proc.
Natl. Acad. Sci. USA 94: 9080-9084.
Mooers, B. H., B. F. Eichman and P. S. Ho (1997). "The structures and relative
stabilities of d(G x G) reverse Hoogsteen, d(G x T) reverse wobble, and d(G x C)
reverse Watson-Crick base-pairs in DNA crystals." J Mol Biol 269(5): 796-810.
Muzet, N., B. Guillot, C. Jelsch, E. Howard and C. Lecomte (2003). "Electrostatic
complementarity in an aldose reductase complex from ultra-high-resolution
crystallography and first-principles calculations." Proc. Natl. Acad. Sci. 100(15):
8742-7.
Muziol, T., V. Cody, J. R. Luft, W. Pangborn and A. Wojtczak (2001). "Complex of
rat transthyretin with tetraiodothyroacetic acid refined at 2.1 and 1.8 A resolution."
Acta Biochim Pol 48(4): 877-84.
Noble, M. E., J. A. Endicott and L. N. Johnson (2004). "Protein kinase inhibitors:
insights into drug design from structure." Science 303(5665): 1800-5.
Ohren, J. F., H. Chen, A. Pavlovsky, C. Whitehead, E. Zhang, P. Kuffa, C. Yan, P.
McConnell, C. Spessard, C. Banotai, W. T. Mueller, A. Delaney, C. Omer, J. SeboltLeopold, D. T. Dudley, I. K. Leung, C. Flamme, J. Warmus, M. Kaufman, S. Barrett,
H. Tecle and C. A. Hasemann (2004). "Structures of human MAP kinase kinase 1
(MEK1) and MEK2 describe novel noncompetitive kinase inhibition." Nat Struct Mol
Biol 11(12): 1192-7.
Ortiz-Lombardía, M., A. González, R. Eritja, J. Aymamí, F. Azorín and M. Coll
(1999). "Crystal structure of a DNA Holliday junction." Nat. Struct. Biol. 6: 913-917.
Otwinowski, Z. and W. Minor (1997). "Processing of x-ray diffraction data collected
in oscillation mode." Methods Enzymol. 276: 307-326.
109
Pinna, L. A. (2003). "The raison d'etre of constitutively active protein kinases: the
lesson of CK2." Acc Chem Res 36(6): 378-84.
Politzer, P., P. Lane, M. C. Concha, Y. Ma and J. S. Murray (2007). "An overview of
halogen bonding." J Mol Model 13(2): 305-11.
Prasanna, M. D. and T. N. Guru Row (2000). "C-halogen to π interactions and their
influence on molecular conformation and crystal packing: a database study." Crystal
Engineering 3: 135-154.
Ramasubbu, N., R. Parthasarathy and P. Murray-Rust (1986). "Angular preferences of
intermolecular forces around halogen centers: preferred directions of approach of
electrophiles and nucleophiles around the carbon-halogen bond." Journal of the
American Chemical Society 108(15): 4308-4314.
Sarno, S., E. de Moliner, M. Ruzzene, M. A. Pagano, R. Battistutta, J. Bain, D.
Fabbro, J. Schoepfer, M. Elliott, P. Furet, F. Meggio, G. Zanotti and L. A. Pinna
(2003). "Biochemical and three-dimensional-structural study of the specific inhibition
of protein kinase CK2 by [5-oxo-5,6-dihydroindolo-(1,2-a)quinazolin-7-yl]acetic acid
(IQA)." Biochem J 374(Pt 3): 639-46.
Sarno, S., H. Reddy, F. Meggio, M. Ruzzene, S. P. Davies, A. Donella-Deana, D.
Shugar and L. A. Pinna (2001). "Selectivity of 4,5,6,7-tetrabromobenzotriazole, an
ATP site-directed inhibitor of protein kinase CK2 ('casein kinase-2')." FEBS Lett
496(1): 44-8.
Scapin, G., S. B. Patel, J. Lisnock, J. W. Becker and P. V. LoGrasso (2003). "The
structure of JNK3 in complex with small molecule inhibitors: structural basis for
potency and selectivity." Chem Biol 10(8): 705-12.
Schaeffer, H. J. and M. J. Weber (1999). "Mitogen-activated protein kinases: specific
messages from ubiquitous messengers." Mol Cell Biol 19(4): 2435-44.
Schmidt, J. R. and W. F. Polik. (2005). "WebMO." 6.0. from http://www.webmo.net.
Schmidt, M. W. K., K. Baldridge, J. A. Boatz, S. T. Elbert, M. S. Gordon, J. J. Jensen,
S. Koseki, N. Matsunaga, K. A. Nguyen, S. Su, T. L. Windus, M. Dupuis and J. A.
Montgomery (1993). "General atomic and molecular electronic structure system." J.
Comput. Chem. 14: 1347-1363.
Sebolt-Leopold, J. S., D. T. Dudley, R. Herrera, K. Van Becelaere, A. Wiland, R. C.
Gowan, H. Tecle, S. D. Barrett, A. Bridges, S. Przybranowski, W. R. Leopold and A.
R. Saltiel (1999). "Blockade of the MAP kinase pathway suppresses growth of colon
tumors in vivo." Nat Med 5(7): 810-6.
110
Seeman, N. C. (2005). "From genes to machines: DNA nanomechanical devices."
Trends Biochem Sci 30(3): 119-25.
Shen, Z., S. N. Mitra, W. Wu, Y. Chen, Y. Yang, J. Qin and S. L. Hazen (2001).
"Eosinophil peroxidase catalyzes bromination of free nucleosides and double-stranded
DNA." Biochemistry 40(7): 2041-51.
Streitwieser, A. and C. H. Heathcock (1976). Organic Charge-Transfer Complexes.
New York, Macmillian Publishing Co., Inc.
Szyszka, R., N. Grankowski, K. Felczak and D. Shugar (1995). "Halogenated
benzimidazoles and benzotriazoles as selective inhibitors of protein kinases CK I and
CK II from Saccharomyces cerevisiae and other sources." Biochem Biophys Res
Commun 208(1): 418-24.
Unger, G. M., A. T. Davis, J. W. Slaton and K. Ahmed (2004). "Protein kinase CK2 as
regulator of cell survival: implications for cancer therapy." Curr Cancer Drug Targets
4(1): 77-84.
Vargason, J. M. and P. S. Ho (2002). "The effect of cytosine methylation on the
structure and geometry of the Holliday junction: the structure of
d(CCGGTACm5CGG) at 1.5 A resolution." J Biol Chem 277(23): 21041-9.
Voth, A. R., F. A. Hays and P. S. Ho (2007). "Directing macromolecular conformation
through halogen bonds." Proc Natl Acad Sci U S A 104(15): 6188-93.
Voth, A. R. and P. S. Ho (2007). "The role of halogen bonding in inhibitor recognition
and binding by protein kinases." Curr Top Med Chem 7(14): 1336-48.
Watson, J., F. A. Hays and P. S. Ho (2004). "Definitions and analysis of DNA
Holliday junction geometry." Nucleic Acids Res 32(10): 3017-27.
Wojtczak, A., V. Cody, J. R. Luft and W. Pangborn (2001). "Structure of rat
transthyretin (rTTR) complex with thyroxine at 2.5 A resolution: first non-biased
insight into thyroxine binding reveals different hormone orientation in two binding
sites." Acta Crystallogr D Biol Crystallogr 57(Pt 8): 1061-70.
Wu, W., M. K. Samoszuk, S. A. Comhair, M. J. Thomassen, C. F. Farver, R. A.
Dweik, M. S. Kavuru, S. C. Erzurum and S. L. Hazen (2000). "Eosinophils generate
brominating oxidants in allergen-induced asthma." J Clin Invest 105(10): 1455-63.
Xie, X., Y. Gu, T. Fox, J. T. Coll, M. A. Fleming, W. Markland, P. R. Caron, K. P.
Wilson and M. S. Su (1998). "Crystal structure of JNK3: a kinase implicated in
neuronal apoptosis." Structure 6(8): 983-91.