Hydrogen bond geometry in DNA-minor groove binding drug complexes
Hydrogen bond geometry in DNA-minor groove binding drug complexes
Lydia
Tabernero*
,
Jordi
Bella
and
Carlos
Alemán
1
Department of Biological Sciences, Lilly Hall of Life Sciences, Purdue
University,
West Lafayette
, IN 47907,
USA
and
1
Departament d'Enginyeria Química, ETSEIB, Universitat Politècnica de Catalunya, Diagonal 647, E-08028
Barcelona
,
Spain
Received February 8, 1996;
Revised and Accepted July 2, 1996
ABSTRACT
The geometry of the hydrogen bonding interaction between DNA and minor-groove binding drugs has been analyzed from a sample of 22 crystal
structures of DNA-drug complexes, retrieved from the Nucleic Acid Database. Seventy-seven interactions between the drugs and acceptor groups in the
nucleotide bases can be classified as hydrogen bonds. Their geometry departs
significantly from linearity since, in most instances, the interactions can be
described as three-center or multiple hydrogen bonds. Results also show that there is no
preference for hydrogen bonds involving positively charged groups in the drugs.
Relationships between hydrogen bond geometry and positioning of the drug along
the minor groove are also discussed. The information presented may be useful in
the design of new specific minor groove binding drugs.
INTRODUCTION
Netropsin, distamycin, Hoechst 33258 or berenil are examples of low molecular
weight compounds with antibiotic, antiviral and antitumour activities, which
are known to bind specifically to the DNA minor groove. These non-intercalating drugs have a binding preference for stretches of AT-rich over GC sequences (
1
). Different experimental and theoretical analyses have brought a large amount
of information about sequence specificity, binding energies and stability of the several DNA-drug complexes (
2
-
15
).
From the very beginning it was postulated that minor groove binding drugs as
netropsin could recognize specific DNA sequences by selective hydrogen bonds to
the DNA bases (
16
). X-ray crystallographic studies of oligonucleotide-drug complexes showed soon that the mode of interaction between drugs and DNA
could be much more subtle (
8
). The typical drug in this category is a flat, crescent-shaped molecule, which accommodates itself into the minor groove of the DNA double helix establishing a
complex interaction that simultaneously involves electrostatic, hydrogen
bonding and van der Waals effects. The relative significance of these terms for
the stability of the DNA-drug complexes is still a matter of discussion, but hydrogen bonding
between the drug and the DNA bases seems to keep its role as a main responsible
for the observed specificity.
Many X-ray structures of different DNA sequences complexed with minor groove
binders have been reported in the past years, and hydrogen bonding interactions
have been described in all of them (Table
1
; see also ref.
17
for a recent review). A comparative analysis of the different observed interactions may provide valuable information for a better understanding of the binding mechanism and
specificity.
We present here a detailed analysis of the hydrogen bonding geometries found in
the crystal structures of several DNA-minor groove drug complexes, whose coordinates have been retrieved from
the Nucleic Acid Database (
18
). Along with the obvious classification by acceptor types, we have analyzed the
geometry of the interaction based on the type of donor group involved. We will discuss the high correlation observed between drug positioning into the minor group and hydrogen bonding geometry. Finally, previous theoretical results (
19
) seem to suggest that the positively charged donor groups may have a greater stabilizing role than the neutral ones,
and therefore our analysis will also extend into the search for a structural
correlation with these theoretical results.
MATERIALS AND METHODS
The crystal structures
Coordinates for the crystal structures of oligonucleotide-minor groove binder complexes were retrieved from the Nucleic Acid
Database (NDB) [(
18
), release of December 1994]. Table
1
lists the NDB entries used in this paper.
Classification of hydrogen bonding groups
In all the structures analyzed in this paper, a basic asymmetry arises from the
fact that the drug provides the donor hydrogen bonding groups, whereas the oligonucleotide molecule participates with its acceptor groups located at the bottom and walls of the DNA minor
groove. To compare hydrogen bonding geometries of charged with non-charged groups, we classified the drug donor groups into three categories:
Type A, in which the NH
2
donor group is part of an amidinium or guanidinium moiety; Type B, in which the NH donor group is part of an amide moiety; and Type C, in which the donor NH group is part of a benzimidazole ring. Type A donors, when present, are located at the ends of the drug, with formal charge
+1. Type B and C groups are usually internal in the drug molecule and have formal charge zero.
.
Crystallographic entries from the NDB (18) used in this work
NDB code
DNA sequence-drug name
R value
Max. res.
Reference
gdl001
CGCGATATCGCG-Netropsin
20.2
2.40
(30)
gdl002
CGCGAATTCGCG-Hoechst 33258
15.7
2.25
(31)
gdl003
CGCAAATTTGCG-Distamycin
20.2
2.20
(25)
gdl004
CGCGATATCGCG-Netropsin
20.1
2.40
(30)
gdlb05
CGCGAATT
Br
CGCG-Netropsin
21.1
2.21
(8)
gdl006
CGCGAATTCGCG-Hoechst 33258
14.0
2.20
(32)
gdl008
CGCGAATTCGCG-DAPI
21.5
2.40
(33)
gdl009
CGCGAATTCGCG-Berenil
16.7
2.50
(20)
gdl010
CGCGAATTCGCG-Hoechst 33258
15.7
2.00
(26)
gdl011
CGCGAATTCGCG-Hoechst 33258
15.7
2.00
(26)
gdl012
CGCGAATTCGCG-Hoechst 33258
15.2
1.90
(26)
gdl013
CGCGAATTCGCG-Hoechst 33258
14.9
2.00
(26)
gdl014
CGCAAATTTGCG-Netropsin
19.8
2.20
(27)
gdl015
CGCGAATTCGCG-Pentamidine
19.4
2.10
(34)
gdl016
CGCAAATTTGCG-Berenil
18.3
2.00
(35)
gdlb17
CGC
(E)
GAATTCGCG-Netropsin
15.6
2.50
(36)
gdl018
CGCGAATTCGCG-Netropsin
16.4
2.20
(36)
gdlb19
CGC
(E)
GAATTCGCG-Hoechst 33258
14.5
2.50
(37)
gdlb20
CGC
(E)
GAATTCGCG-Hoechst 33342
15.7
2.50
(37)
gdl021
CGCGAATTCGCG-Hoechst 33342
16.8
2.25
(37)
gdl022
CGCGAATTCGCG-Hoechst 33258
17.2
2.00
(37)
gdl023
CGCGAATTCGCG-Propamidine
17.4
2.10
(38)
On the DNA side, we considered two major acceptor types: carbonyl groups and
aromatic nitrogen atoms, since they impose different geometry requirements on
the hydrogen bonding interaction. We also included less frequent interactions
involving O' atoms from deoxyribose rings, but we did not consider solvent-mediated interactions in this study. Water bridges are not a common
feature in most of the structures, although they may be of some importance in
helping the binding of drugs like berenil to DNA (
20
).
Most of the hydrogen bonding groups discussed above are also present in
proteins, for which exhaustive hydrogen bonding statistical analysis are
available (
21
), and can be used for comparison. The -N= aromatic groups are genuine of DNA in macromolecules, and their
hydrogen bonding geometry has been analyzed in small molecule crystal
structures (
22
).
Metric analysis and selection criteria for hydrogen bonding interactions
Ideal positions for the hydrogen atoms were calculated and built up for each set
of coordinates. Pertinent planar geometry was adopted for all three types of
donor groups. Occasionally the original coordinates showed significant
deviations from planarity on guanidinium, amidinium or amide groups. In these
cases hydrogen atoms were built as closely as possible to their ideal position
in a planar group. No attempt was made to correct the positions of the non-hydrogen atoms.
Donor-acceptor pairs were first selected from the 22 structures in Table
1
, using a distance cutoff of 3.5 Å between the two non-hydrogen atoms. This yielded a total set of 164 putative hydrogen
bonding interactions. Of these, only contacts with distances
d
(H[middot][middot][middot]A) < 2.9 Å (A being the acceptor atom), and angles [alpha](N-H[middot][middot][middot]A) > 90o were
retained in succeeding geometry calculations. This selection criterion resulted
in a total of 77 hydrogen bonds. If classified by acceptor type, 39 cases
involve the O2 atom from thymines or cytosines, 26 cases involve the N3 atom
from adenines and eight cases are hydrogen bonds to the O4' or O3' atoms of sugar rings (the standard nucleotide nomenclature is
adopted for all atoms). The remaining four cases correspond to an unusual
geometry in which the NH
2
groups of guanine residues are in better disposition as to be considered the
hydrogen bonding donors in their interaction with the drug. These interactions
will be discussed separately.
Atomic subsets of coordinates were created for every hydrogen bonding
interaction by selecting only those atoms that were relevant for describing the
geometry of the interaction. Every subset of coordinates was inspected and then
reduced to a common reference system using the graphics program CHAIN (
23
).
As it will be discussed later, many of the interactions turned out to be parts
of multiple hydrogen bonds, mainly those in which one hydrogen bonding donor
group is shared between two hydrogen bonding acceptor groups. We will designate
these interactions as three-center hydrogen bonds, and will use the term `bifurcated' hydrogen bonds
for those cases in which one single donor group uses two protons to interact
with a single acceptor atom (
24
).
RESULTS
Score of hydrogen bonding
Table
2
presents the scoring of success (X) and failure (0) for the different
interactions in each DNA-drug complex according to the criterion stated above. Those interactions
with unusual geometry in which the donor group would correspond to the
oligonucleotide part are designated as *. For those drugs with only terminal
charged groups (Berenil, Pentamidine, Propamidine), the percentage of putative
interactions fulfilled is 87.5% (considering at least one hydrogen bond per
donor group). For drugs with only neutral, Type C donor groups (Hoechst), the
rate of success is 85%. Finally, on those drugs that have both charged and
neutral donor groups (Netropsin, Distamycin, DAPI), the scoring percentage
drops to 53% for charged groups and 52% for neutral ones. One of the complexes
in this category shows no hydrogen bonding interactions.
X indicates that at least one hydrogen bond is formed between the particular
donor group in the drug and the DNA molecule; 0 indicates no hydrogen bonds; *
corresponds to unusual geometries (see text). Non-charged donor groups along the drug backbone are designated as A
1-3
(internal groups) or A
0
(distamycin terminal amide group). G
+
and A
+
correspond to guanidinium or amidinium groups at the ends of the drug molecule.
.
Statistics of the hydrogen bond interactions: average values for typical
hydrogen bonding parameters
Donor/acceptor
Number
N[middot][middot][middot]Y (Å)
H[middot][middot][middot]Y (Å)
N-H[middot][middot][middot]Y (o)
H[middot][middot][middot]O=C (o)
Type A/O
9
2.84 (0.29)
2.10 (0.37)
128 (15)
145 (31)
Type A/N
11
3.12 (0.26)
2.39 (0.28)
129 (16)
Type A/O4', O3'
8
2.92 (0.38)
2.37 (0.28)
115 (24)
Average Type A
28
2.94 (0.31)
2.26 (0.33)
124 (18)
Type B/O
8
2.95 (0.32)
2.25 (0.41)
125 (10)
143 (13)
Type B/N
7
3.16 (0.12)
2.50 (0.21)
127 (27)
Average Type B
15
3.07 (0.26)
2.39 (0.34)
126 (19)
Type C/O
22
2.96 (0.24)
2.38 (0.30)
116 (14)
144 (18)
Type C/N
8
3.06 (0.19)
2.23 (0.27)
138 (13)
Average Type C
30
2.99 (0.23)
2.35 (0.29)
122 (19)
Average ABC
/
O
39
2.93 (0.26)
2.29 (0.35)
121 (14)
Average ABC
/
N
26
3.11 (0.21)
2.37 (0.27)
131 (19)
Type A/Gua N2*
4
2.96 (0.22)
2.24 (0.37)
130 (23)
135 (19)
Average
77
2.99 (0.27)
2.32 (0.32)
124 (18)
143 (20)
Standard deviations are shown in parentheses.
Hydrogen bonding geometry at the donor side
The two parameters analyzed on the donor part are the distance H[middot][middot][middot]A and the angle N-H[middot][middot][middot]A (A meaning the acceptor
atom). Table
3
summarizes the global statistics for these and other hydrogen bonding
parameters. Histograms showing the distribution of selected hydrogen bonding
parameters are plotted in Figure
1
. Figures
2
and
3
show the distribution of these parameters for the interactions included in the
above defined criterion, classified by donor and acceptor types.
Three-center hydrogen bonds are much more common in biological molecules than
previously thought (
24
). Up to 20% of the common N-H[middot][middot][middot]O=C hydrogen bonds are actually three-centered, as shown by a survey on X-ray and neutron crystal structures of
biological small molecules (
28
). Three-center hydrogen bonds can be symmetrical, with comparable values for
hydrogen bonding distances H[middot][middot][middot]A, H[middot][middot][middot]A', and angles N-H[middot][middot][middot]A, N-H[middot][middot][middot]A'. Most often they exhibit unsymmetrical geometry, with a major and a minor component, whose H[middot][middot][middot]A distances can differ by as much as 1.0 Å
(
24
). Therefore, it is not uncommon for the minor components of three-center hydrogen bonds to show H[middot][middot][middot]A' distances of >= 2.9 Å, and N-H[middot][middot][middot]A'
angles of 90o. This is the situation for many of the Type B and C hydrogen bonds
analyzed in this work. Figure
4
b shows the average hydrogen bonding parameters for Types B and C three-center hydrogen bonds when grouped according to major and minor
components. The average values are consistent with an unsymmetrical, three-center hydrogen bond description of these interactions, even for those cases in which the
minor component geometry falls clearly out of the range initially used as a
selection criterion for this work.
The attractive character of the three-centered hydrogen bonds is related to the coplanarity of the hydrogen atom
with the plane defined by the atom donor N and its two acceptors A, A'. The sum of the angles N-H-A, N-H-A' and A-H-A' should be near
360o for a three-center hydrogen bond with good geometry, and the hydrogen atom should
be between 0.0 and 0.2 Å from the plane A-N-A'. For the interactions shown in Figure
4
a-b the average deviation of the planarity is 20o +- 15o, which corresponds to an out-of-plane component of 0.4 Å for the hydrogen atoms.
Further analysis by donor types does not show significant differences between
the geometry of three-center hydrogen bonds involving Type B or Type C groups, suggesting that
they behave similarly.
The last class of donors, Type A, have a broader variety of interactions, given
their situation at the terminal regions of the drugs. For example, all N-H[middot][middot][middot]O' drug-sugar interactions belong to this
class. Usually, Type A donors do not participate in three-center hydrogen bonds as those already discussed (a notable exception is
the drug berenil in the gdl016 structure), and in several instances they seem
to be involved in multiple interactions using both hydrogens from their NH
2
groups. The N-H[middot][middot][middot]A angles for Type A hydrogen bonds are not more
linear than those of the three-centered interactions (Table
3
). Additionally, a common motif is observed in 10 cases, for which the NH
2
group is placed with its two hydrogen atoms around one acceptor atom. The
primary interaction shows an average H[middot][middot][middot]A distance of 2.35 (+-0.28) Å, and an N-H[middot][middot][middot]A angle of
121o (+-15o). The secondary interaction H'[middot][middot][middot]A has much worse geometry, 3.01
(+-0.30) Å and 79o (+-8o) respectively, and cannot be considered a
hydrogen bond on itself. However it may be responsible for the deviation from
linearity of the primary H[middot][middot][middot]A hydrogen bond.
Hydrogen bonding geometry at the acceptor side
Table
3
shows the average values of the H[middot][middot][middot]O=C angle for the drug-DNA hydrogen bonds in this sample. They are very
similar to those observed for peptide N-H[middot][middot][middot]O=C hydrogen bonds in protein crystal structures:
147o for [alpha]-helices or 151o for [beta]-sheets (
21
). The distribution of donor hydrogen atoms around the C=O group (Fig.
5
b) is also similar to that observed around carbonyl groups in proteins. This
distribution is very broad and does not follow the ideal orientation expected
for the lone pairs on the oxygen atom, which would be coplanar with the
aromatic ring, at +-120o from the axis of the C=O bond. Approaching the C=O bond from the
plane of the ring is not possible for a minor-groove binding drug without deforming the DNA double helix: either the
sugar ring from the same nucleotide or the nitrogenous base from the
complementary chain would pose unsurmountable steric obstacles. Instead, the
mode of approach for all drugs involved in minor-groove binding is at a certain angle with the plane of the ring (Fig.
5
b). This angle can be derived from the dihedral angle H[middot][middot][middot]O2=C2-N1, which should be 0o or 180o for an on-plane approach. For the hydrogen
atoms shown in Figure
5
b, the average H[middot][middot][middot]O2=C2-N1 dihedral angle is 115o, with a very broad distribution.
Unusual geometry: guanine NH
2
groups acting as donors?
An unexpected hydrogen bonding geometry is observed in four cases, all them
involving the NH
2
groups from guanine residues in the fourth position and different nitrogen
atoms from netropsin molecules: N1 in gdl001, N10 in gdl004, N1 in gdl018 and
N3 in gdl018. Nitrogen-nitrogen distances for these pairs are 3.04, 3.03, 2.64 and 3.14 Å respectively, which might suggest a drug-DNA hydrogen bonding interaction. However, when standard
hydrogen positions are built for these groups, the resulting geometry suggests
in fact that the hydrogen bonds go in the other direction, that is, with the NH
2
groups from guanines acting as donors and nitrogen atoms on guanidinium or amidinium groups acting as acceptors. Only two complexes show this unexpected type of interaction, since gdl001 and gdl004 correspond to the same netropsin-DNA complex with the drug refined in two different orientations. In other
complexes the distances from guanine NH
2
groups to netropsin nitrogens are usually >4 Å, with the NH
2
hydrogen atoms pointing away from the drug.
DISCUSSION
Position of the drugs along the minor groove and hydrogen bonding geometry
The positioning of the drug along the minor groove has some variability among
all the structures. Even for the same drug, different complexes show slightly shifted positions. The structures of netropsin bound to three different dodecanucleotide sequences can be grouped
in two main categories, as has been recently suggested by Goodsell
et al.
(
29
). Class I complexes include gdlb05, gdl014 and gdl003 (Distamycin complex). In
them, pyrrole rings are opposed to the A-T base pairs and amide moieties are located between two successive base
pairs (Fig.
6
). With this disposition, most amide nitrogens form three-center hydrogen bonds to adenine or thymine bases on both strands of the
DNA fragment, although in some cases the three-center hydrogen bond can be rather unsymmetrical. In class II complexes, gdl001, gdl004, gdlb17 and gdl018, the amide moieties lay in the plane of
the base pairs and the pyrroles are positioned midway between two successive
steps (Fig.
6
). Many of the putative DNA-drug interactions in the class II complexes do not fulfill our selection
criterion for a hydrogen bond: either the donor-acceptor distances are >3.5 Å or the X-H[middot][middot][middot]Y angles are too closed (<90o) (Table
2
). Additionally, those are the only complexes that show the unusual interaction
involving guanine NH
2
groups as donors (Table
2
, indicated by the asterisk).
Hydrogen bonding geometry and formal charge of the donor group
We have not observed any significant correlation between nature of the hydrogen
bonding donor and goodness of the hydrogen bonding geometry. For all three
types of hydrogen bonding donors we observe a significant deviation from the ideal hydrogen bond parameters, which can be at least partially rationalized as a result of
the formation of different types of multiple hydrogen bonds.
Charged and non-charged groups do exhibit different strategies because of their
positioning along the drug. The arrangement observed in the crystal structures
from Table
1
seems to indicate that the terminal, charged groups, have more options at hand
to fulfill hydrogen bonding interactions with the DNA molecule (for example
with oxygen atoms from sugar rings), whereas the neutral, internal groups are
somehow restricted by the more rigid conformation of the pyrrole or
benzimidazole rings and the available acceptor groups at the bottom of the
minor groove. The intrinsic curvature of long drug molecules like Netropsin or
Distamycin does not match perfectly the curvature of the bottom of the minor
groove. This precludes the formation of all possible hydrogen bonds for a flat,
completely extended Netropsin-like drug. Indeed, all Netropsin or Distamycin molecules in Table
1
complexes show a rotation of the plane from their charged groups with respect
to the mean plane of the central amide groups (see for example Fig.
6
). Thus, flexible ends in different DNA-drug complexes adopt alternative orientations as to find different
hydrogen bond mates, either at the DNA bases or at the sugar rings. Some of
these interactions may involve multiple or bifurcated hydrogen bonds, although
the observed geometry in the crystal structures of this sample does not provide
definitive evidence for the later case. Interestingly, in gdl015 and gdl023
complexes, Propamidine and Pentamidine terminal groups attach themselves to
only one of the DNA strands instead of adopting a three-centered geometry. Three out of four of the hydrogen bonding NH
2
groups in these drugs are positioned as to orient both protons towards the
acceptor atom in a bifurcated-like geometry.
Amide and benzimidazole groups are often kept farther away from the base pairs
and therefore they adopt a different strategy. In the most favourable case
these internal donor groups are positioned between successive base pair steps
and form three-center hydrogen bonds to both strands of the DNA double helix. These
interactions define a characteristic pattern that is repeated for either amide
or benzaimidazole donors (Fig.
6
and ref.
29
). Deviations of this pattern by sliding the position of the drug along the
minor groove result in a worse geometry for the interactions and loss of
several hydrogen bonds to the bases. Thus, appropriate spacing between donor
groups remains as one of the essential characteristics for a successful minor-binding drug.
It is possible that the inclusion of charged groups at the flanking regions of a
minor groove-binding drug may increase its anchorage capabilities during and after the
initial steps of the complex formation, compensating for any lack of
interaction at other points in the drug. Charge itself may be critical at the
initial steps of the binding process, as the drug must compete with and replace
the pre-existing water molecules at the minor groove (
8
,
10
). In a related study, we have carried out
ab initio
quantum mechanical calculations of model compounds for the hydrogen bonding
interactions between amide and thymine carbonyl groups. These calculations
predict a much tighter interaction of charged nitrogen groups with thymine
carbonyl acceptors (
19
). Furthermore, after introduction of solvent effects, the calculated
interaction energy between a donor group and a thymine carbonyl acceptor, is
negative for positively charged amide groups, and positive for neutral amides (
19
). This would suggest a leading role for the positively charged moieties at the
initial steps of the formation of the DNA-drug complex, whereas interactions between neutral groups would be
established later. However, as we mentioned above, no correlation is observed
between charged donor groups and better hydrogen bonding geometry in the
crystal structures studied. This probably occurs because the crystal structures
represent the final stage in the complex formation where a combination of different interactions takes place.
Hydrogen bonding geometry and DNA flexibility
The conformation of the DNA double helix in these complexes is mainly restricted
by the local geometry of the specific sequence. A certain degree of flexibility
can be achieved by means of propeller twist variations and the capacity of the
base pairs to adopt a more favorable disposition in optimizing their hydrogen bonding interactions with the particular drug. Sequences with alternating ATAT steps have a lower degree of propeller twist in complexes
with drugs that sequences with successive AATT or AAATTT steps (
27
). This indicates that tracts of A-T base pairs are able to adapt their local conformation to improve the hydrogen bonding geometry with the internal donor groups from the drugs and therefore increase their specificity.
Final remarks: effect of the resolution of the crystal structures
One of the main drawbacks of the sample of the crystal structures analyzed in
this work is their relative low resolution (Table
1
), that introduces two undesirable effects. First, the final geometries of the
crystallographic model can be rather poor, mainly on the non-covalent interactions, for which very soft or no restraints are applied
during the crystallographic refinement. Second, the interpretation of the
electronic density maps at the early stages of the refinement can be very
ambiguous as to where and how to model the drug molecule on the DNA minor
groove (
29
). Still, meaningful information can be deduced from comparison between the
available structures and identification of common, repetitive motifs. Thus, the
description of a particular hydrogen bond in one crystal structure can be quite
inaccurate, but the average values over several structures will probably
represent a better description of the actual interactions taking place between
DNA fragments and minor groove binding drugs. We think that the geometric
features described in this work provide a useful framework to which future
structural determinations of DNA complexes can be compared.
REFERENCES
1 Zimmer, C. and Wahnert, U. (1986) Proc. Biophys. Mol. Biol., 47, 31-112.
27 Tabernero, L., Verdaguer, N., Coll, M., Fita, I., van der Marel, G.A., van Boom, J.H., Rich, A. and Aymami, J. (1993) Biochemistry, 32, 8403-8410.MEDLINE Abstract
28 Taylor, R., Kennard, O. and Versichel, W. (1984) Acta Crystallogr., B40, 280-288.
