Department of Medicine and Cancer Center (E.D.S., S.B.H.), and
Department of Pharmacology (J.M.B.), University of California, San
Diego, La Jolla, California
Intrastrand DNA adducts formed by cisplatin and oxaliplatin were
modeled with molecular mechanics minimization and restrained molecular dynamics simulations in a comparative study. A reasonable set
of force field parameters for the Pt atom were refined by using the
available cisplatinated DNA crystal structure as a guide. This crystal
structure was also used as the starting structure for the simulations.
Analysis of the resulting structures indicated that the covalent
effects of oxaliplatin coordination on DNA structure were very similar
to those of cisplatin. The most prominent difference between the two
structures resulted from the presence of the 1,2-diaminocyclohexane ring in the oxaliplatin adduct. The modeling indicated that this ring
protrudes directly outward into, and fills much of, the narrowed major
groove of the bound DNA, forming a markedly altered and less polar
major groove in the area of the adduct. The differences in the
structure of the adducts produced by cisplatin and oxaliplatin are
consistent with the observation that they are differentially recognized
by the DNA mismatch repair system.
 |
Introduction |
It
is generally accepted that the primary cytotoxic mode of action of
cisplatin is the production of adducts in DNA. The most prevalent adduct is the intrastrand linkage of two adjacent guanine bases by the nitrogen atoms at position 7 (the GG adduct).
Experimentally derived structures for the cisplatin GG adduct in
double-stranded DNA are available in the form of an octamer duplex
solved by NMR (Yang et al., 1995
) and a dodecamer duplex solved by
crystallography (Takahara et al., 1996a,b) and NMR (Gelasco and
Lippard, 1998). The three structures display a notable bend toward the
major groove in the double helix of 35-78°, which varies depending
on the structure and the particular method of measurement used. This
bend produces a narrowing of the major groove and a broadening and
flattening of the minor groove similar to that seen in A-DNA.
Furthermore, the crystal structure assumes a more A-DNA-like
conformation on one end of the molecule and a more B-DNA-like
conformation on the other, although this was not observed in the NMR
structures and may be a product of crystal packing forces (Gelasco and
Lippard, 1998).
Oxaliplatin [1R,2R-diaminocyclohexane platinum
(II) oxalate], a platinum compound now in clinical development
(Raymond et al., 1998), differs from cisplatin by the addition of a
cyclohexane ring to the ammines of cisplatin to form a
diaminocyclohexane (DACH) ring (Fig. 1).
Although the postbiotransformation leaving groups of these compounds
are apparently the same (Raymond et al., 1998), and it has been shown
that oxaliplatin has a similar adduct formation profile to that of
cisplatin (Saris et al., 1996; Woynarowski et al., 1998), one would
anticipate the GG adduct to be structurally dissimilar once the drugs
are bound to DNA.
View larger version (9K):
[in this window]
[in a new window]
|
Fig. 1.
Cisplatin and oxaliplatin molecules before DNA
binding. Chemical groups to the right of the Pt are removed during
biotransformation, and both drugs bind to DNA directly via the Pt atom.
Two nitrogen-bound cyclohexane carbons are both in the R
configuration in oxaliplatin.
|
|
It is now known that components of the DNA mismatch repair (MMR) system
bind to the cisplatin GG adduct (Mello et al., 1996; Yamada et al.,
1997) and, by an undetermined mechanism, promote cellular apoptosis.
Thus, cells that lack a functional MMR system are more resistant to
cisplatin injury (Fink et al., 1996). The MMR system may enhance
cisplatin sensitivity through the shielding of adducts from appropriate
repair, direct signaling to the apoptotic machinery, or the initiation
of a futile cycle of attempted repair that eventually produces lethal
DNA strand breaks. In contrast, the loss of functional MMR does not
confer resistance to oxaliplatin (Fink et al., 1996). Thus, oxaliplatin
adducts apparently are not recognized or processed by the MMR system in
the same way as cisplatin adducts. The structural basis for the lack of
oxaliplatin recognition is unclear and could involve direct steric
hindrance by the DACH ring or a distortion in DNA that is
distinct from that produced by cisplatin. Only limited information is
available on the structure of hMSH2 (De las Alas et al., 1998),
the primary DNA binding protein of the MMR system, and no structural
information is available on the other human MMR proteins. However, data
is available for the homologous system in bacteria that suggests that
the MMR proteins contact mismatches at multiple sites, both at the
point of the mismatch and at other atoms up and down the major and
minor grooves (Biswas and Hsieh, 1997).
An experimentally derived structure does not exist for the oxaliplatin
GG adduct. Therefore, we sought to determine a likely structure for the
oxaliplatin adduct to provide insight into the structural features that
might account for differential recognition by the MMR system. We used
molecular mechanics minimization and restrained molecular dynamics to
develop a model of the cisplatin and oxaliplatin GG adducts by using
the known crystal structure of the cisplatin adduct as a guide. A
forcefield was developed that contained a refined set of parameters for
the unusual presence of platinum in the molecule, and this field was
tested against the crystal structure. The oxaliplatin adduct in a
double-stranded DNA molecule was produced through modification of the
cisplatin crystal structure. The modeled cisplatin and oxaliplatin
adducts were then put through identical energy minimization and
molecular dynamics simulations, and the resulting molecules compared to determine the likely effect of the presence of the DACH ring of oxaliplatin on the adduct and overall DNA structure.
An in vacuo modeling environment with implicit consideration of solvent
and some atomic constraints (during dynamics simulations) was used.
This methodology made the computation times tractable during the
multiple, repetitive simulations required in this study. These were
particularly important for the critical development and testing of a
broad range of potential parameters for the platinum atom. Newer
methodologies exist that consider solvent atoms and counterions
explicitly, use the particle mesh Ewald summation for the treatment of
boundary conditions, do not require atomic constraints, and have been
shown to produce more accurate DNA structures than the methods used
here (Young et al., 1997; Duan et al., 1997). However, the extremely
high central processing unit demands and run times of these
methods precluded their use in this study, and the reader is therefore
cautioned to take note of this limitation. Implicit solvent techniques
have been suggested to provide acceptable accuracy for DNA modeling
(Falsafi and Reich, 1993; Kozack and Loechler, 1997), and this initial
work provides a foundation for the eventual exploration of strained,
platinated DNA molecules with the advanced full solvent techniques
mentioned above.
| |
Materials and Methods |
Computing.
All modeling was performed on a Silicon Graphics
workstation with the InsightII (95.0) and (97.0) software suites
(Molecular Simulations Inc., San Diego). Molecular mechanics and
dynamics simulations were performed with the Discover module of this
package. The assisted model building with energy refinement
(AMBER) force field (Weiner et al., 1984; equation given below),
as provided with InsightII (95.0), was used with specific modifications
for the presence of a platinum atom as discussed below. Analysis of the
resultant structures was performed with the Analysis module of InsightII.
AMBER Equation.
The AMBER equation used was as follows.
Starting Structure.
The structure on which the
modeling effort was based was the available crystal structure
(Takahara et al., 1996a,b) for the cisplatinated DNA dodecamer
d[CCTCTG*G*TCTCC]/d[GGAGACCAGAGG], where G* denotes an adducted
base. This structure was downloaded from the Protein Data Bank (PDB;
Bernstein et al., 1977; www.pdb.bnl.gov/www.rcsb.org; PDB
code 1AIO) and modified for use in this study. As the PDB file contains
two very similar molecules (one unit cell), we selected the first of
the two in the PDB file for use in our study. We adopted the identical
base numbering pattern (for referring to particular nucleotides) as was
used in the PDB file and by Takahara et al. (1996a,b).
Hydrogens were added to the structure with the hydrogen addition
function available in the InsightII Biopolymer interface. This function
provided hydrogen placement that produced very reasonable nucleotide
structures. Therefore, the hydrogens were not subjected to any form of
optimization after addition to the molecule used in the simulation.
Base 3 in the crystal structure had been substituted in the form of an
isomorphous replacement of 5-bromouridine for thymine. As our goal was
to model the DNA adducts as they would exist in cells, we chose to
modify this base back to a thymine by replacing the bromine atom with a
methyl group. This structure was used as the starting point for all
cisplatin simulations, and all comparisons made against the "crystal
structure" in this article were made to this slightly modified version.
To produce a starting structure for the oxaliplatin GG adduct, a
cyclohexane molecule was bound to the ammine groups of cisplatin via
two adjacent carbon atoms in the ring such that these chiral carbons
were both in the R configuration. The modification was performed through the Builder module of InsightII, and effectively reproduced the DACH structure of oxaliplatin. To optimize the bonds of
the artificially attached ring, the cyclohexane residue was subjected
to conjugate gradient minimization until convergence (a maximum
derivative <0.001 kcal
mol
1Å1) with the rest
of the molecule fixed. All comparisons made to the "starting
structure" of oxaliplatin refer to this structure. It should be noted
that at the beginning of the simulations, the oxaliplatin and cisplatin
adduct structures were completely identical, with the exception of the
added cyclohexane ring of the oxaliplatin structure.
Forcefield Parameterization.
Because cisplatin and
oxaliplatin contain a platinum atom, for which the standard AMBER
forcefield does not include parameters, it was necessary to develop a
modified version of the field that included appropriate values. A set
of forcefield parameters and partial charges has been proposed by Yao
et al. (1994). These parameters integrated a variety of experimental
data and previous molecular modeling studies, and the authors were able
to demonstrate agreement with both NMR and crystallographic data of
cisplatin bound to several small guanine base structures. However, this study was completed before the availability of the full-length, double-stranded NMR (Yang et al., 1995; Gelasco and Lippard, 1998) and
crystal (Takahara et al., 1996a,b) structures now available. It was
deemed reasonable to test these parameters against the double-stranded
crystal structure and make minor modifications if necessary to enhance agreement.
Evaluation of the values suggested by Yao et al. (1994) was conducted
as follows. The values were entered into the AMBER forcefield as
described below. Exploratory energy minimization of the modified cisplatin crystal structure was then initiated in vacuo for 500 steps
of steepest descent and 3000 steps of conjugate gradient minimization.
A distance-dependent dielectric of 
= 4rij was used to mimic solvent effects, an arrangement that has been
successfully used in previous studies (Orozco et al., 1990; Yao et al.,
1994). The 1-4 van der Waals terms were also scaled by a factor of
0.5, as suggested (Discover 95.0 Manual; Yao et al., 1994). For this simulation, we chose to lower the charges on the phosphate groups such
that each nucleotide had a net charge of 
0.2, according to the method
described by Veal and Wilson (1991). This charge reduction mimics the
effects of the counterion condensation that occurs around DNA in
solution. As suggested (Veal and Wilson, 1991; Yao et al, 1994), we did
not reduce the charges on nucleotides bound to cisplatin, so as to
represent the release of counterions on the binding of this cationic ligand.
The molecules that resulted from the above minimization procedure were
superimposed on the crystal structure and compared. Bond stretching and
nonbonded parameters were retained unchanged from the suggested values.
Angle bending and torsional and improper torsional terms were
systematically altered to test for enhanced agreement with the crystal
structure. This was done by both raising and lowering the force
constant for the energy term in question and rerunning the above
simulation. If either an increase or decrease in the force constant
enhanced agreement with the crystal structure, further modification in
the direction suggested was tested. If there was ambiguity as to the
benefits of a particular change, the initial value from Yao et al.
(1994) was favored. The most promising parameter combinations were
subjected to another 3000 iterations of conjugate gradients
minimization and then reevaluated. In the end, it was found that the
suggested parameters (Yao et al., 1994) produced excellent agreement
with the crystal structure, provided a few minor modifications were
made to them.
We adopted the same specialized atom types as Yao et al. (1994; Fig.
2; Table
1): N31 and N32 for the
two ammine/amine nitrogens, NB1 and NB2 for Pt-bound guanine nitrogens,
and PT for the central Pt atom. The ammine/amine hydrogens were treated
as standard H3 hydrogens. Our postevaluation force constants for these
atoms are listed in Table 1. Aside from the values explicitly listed, N31/N32 (referred to as N3X) and NB1/NB2 (referred to as NBX) were
parameterized throughout the forcefield identically to the standard AMBER atom types N3 and NB, respectively.
View larger version (20K):
[in this window]
[in a new window]
|
Fig. 2.
MOLSCRIPT (Kraulis, 1991) image depicting the atom
potential types used in modeling the GG adducts. The view faces the
major groove and is a close-up of the two bound guanines and the
cisplatin adduct (partner strand cytosines are not shown). The image
has a similar overall orientation to that described for Fig. 5.
Hydrogens are only shown for the cisplatin adduct and two adducted
guanines. Oxaliplatin adducts were arranged with identical atom types
to those shown, except that one H3 atom on both N31 and N32 was
replaced by a CT atom of the cyclohexane ring.
|
|
View this table:
[in this window]
[in a new window]
|
TABLE 1
Atom potential types and force constants used in modeling the cisplatin
and oxaliplatin adducts
Positions of atom types shown in Fig. 2.
|
|
Charges for the specialized atom types and the bound guanine groups
were arranged exactly as suggested (Yao et al., 1994; Figs.
3 and 4;
Table 2), as were the values for bond
stretching. The suggested values for angle bend deformation were used,
with one exception: Yao et al. (1994) had suggested that the
CB-NB/NBX-CK bond be adjusted to 104.1° from the standard
value of 103.8°. The bond angles given in the crystal
structure were closer to the 103.8° value, and as our
intent was to model the crystal structure as closely as possible, the
standard value was retained. Yao et al. (1994) did not provide an angle
or force constant for the PT-N3X-H3 bond, so we treated PT as an sp3
carbon (CT) to achieve an angle of 109.5° and force
constant of 35 kcal mol1
rad2 (Table 1).
View larger version (20K):
[in this window]
[in a new window]
|
Fig. 3.
MOLSCRIPT (Kraulis, 1991) image depicting the names
for atoms in the cisplatin adduct and bound guanines. The orientation
is identical with that in Fig. 2. Atoms in the oxaliplatin adducts were
similar, except that H22 and H12 were replaced by the cyclohexane
carbons.
|
|
View larger version (16K):
[in this window]
[in a new window]
|
Fig. 4.
Two-dimensional drawing of the adducts depicting
partial charge assignments used for each by atom name. Charge values
and their sources are also provided in Table 2 and explained in the
text. Atoms shown in bold differed between the two adducts in charge
value or in presence/absence. Charges shown in italics are unchanged
from the standard AMBER values. Hydrogens are not shown for the
cyclohexane ring of oxaliplatin (all had the same partial charge, as
provided in Table 2). G7 and G6 refer to the bound guanine bases.
Charges for these were arranged identically for the two adducts and are
listed in Table 2.
|
|
View this table:
[in this window]
[in a new window]
|
TABLE 2
Partial charges on cisplatin, oxaliplatin, bound guanines, and
phosphates by atom name
Positions of named atoms shown in Figs. 3 and 4.
|
|
The values for torsional deformation suggested by Yao et al. (1994)
were expanded and modified slightly. Although the authors provided
values for most of the torsions involving NBX, PT, and N3X, they did
not suggest values for CB/CK-NB2-PT-N32 and CB/CK-NB1-PT-N31. It was
reasoned that the considerable torsional values already suggested for
these atoms, although in different combinations of bonds, would be
sufficient for the assumption of the proper structure, and that further
values would be redundant. Furthermore, torsions about these bonds
would most likely have a very low resistance because of the nearly
linear nature of the NB2-PT-N32 and NB1-PT-N31 bonds (Fig. 2)
Therefore, the torsions for the above groups were given a force
constant of 0.
Yao et al. (1994) also did not suggest torsional parameters for
rotations around the PT-N3X bond. As above, we used the CT atom type as
a stand-in for PT. Rotations around the CT-N3 bond are treated in the
standard AMBER field (Weiner et al., 1984) with a *-CT-N3-* term, where
* denotes any atom. This arrangement was duplicated with a *-PT-N3X-*
term, with a periodicity of 3 and a force constant of 1.40 kcal
mol1 (Table 1).
Torsions without platinum as a central atom also required adjustment.
In the normal AMBER field, torsions with the atoms CB-NB at the center
are arranged with the term *-CB-NB-*, where * denotes any atom. This
"wildcard" term is assigned a high force constant, with a 
angle
of 180°, which promotes planarity of the base. If the
torsions for NBX were simply copied from this arrangement, the result
was terms such as C-CB-NBX-PT and CB-CB-NBX-PT with extremely high
planar force constants. On exploratory energy minimization, a severe
pucker of the base in the direction of the platinum atom occurred as
the molecule strained to achieve planarity. An identical situation
existed for *-CK-NBX-*. In both cases, it was found that setting the
torsional force constants to 0 (where PT was involved) corrected the
problem. However, the planar geometry of the base was somewhat
destabilized, so it was decided to retain a CB-CB-NBX-PT force constant
of 5 kcal/mol (Table 1).
Yao et al. (1994) also included an improper torsional value in their
forcefield to help prevent base plane pucker, and our evaluation
suggested that this term was very helpful in achieving that result. As
such, it was included in our forcefield.
After 500 steepest descent and 6000 conjugate gradient minimization
steps, the listed parameter set (Tables 1 and 2) produced reasonable
agreement with the crystal structure, with a heavy atom root mean
squared (RMS) deviation of 0.84Å. If only the platinum, ammine
nitrogens, and heavy atoms of the two bound guanine bases were
considered, the RMS deviation was 0.34Å, a value close to that
achieved in similar measurements of complexes in Yao et al. (1994).
These parameters were used for the remainder of the study.
Oxaliplatin Parameterization.
A small number of additional
parameters were necessary for the oxaliplatin adduct as a result of the
presence of the cyclohexane ring in this compound. The standard AMBER
atom types of CT for the carbons and HC for the hydrogens were used for
the cyclohexane ring. As with the cisplatin parameterization, many of
the parameters for interaction of N3X with the cyclohexane ring were
arranged through the use of the standard AMBER parameters for N3. This method was sufficient for parameterization of all interactions not
involving the PT atom, and therefore explicit values are not listed in
Table 1.
Only one additional parameterization was required that involved the
platinum atom, and once again modified forms of the parameters for the
CT atom type were used to determine values for PT. For the CT-N3X-PT
bond angle deformation, an angle of 109.5° (the same as
that for H3-N3X-PT) was chosen to promote a tetrahedral geometry around
the N3X atoms. However, the CT-N3-CT force constant of 50 kcal
mol1 rad2 was used to
reflect the presence of CT (as opposed to H3) in the CT-N3X-PT angle group.
Charges for the oxaliplatin adduct also had to be developed, and they
were obtained from those used for the cisplatin adduct, with the
exception of the charges on the N3X atoms. It was necessary to adjust
the charges on these atoms to reflect attachment to the two CT atoms of
cyclohexane as opposed to H3 hydrogens. To estimate the appropriate
charge shift, the Pt atom was substituted with a carbon, and the
automatic forcefield charge assignments were examined for the ammine
nitrogens, a cyclohexane ring, and amine nitrogens bound to cyclohexane
to produce a pseudo "oxaliplatin" molecule. The AMBER field did not
have a full set of built-in charge assignments for this interaction, so
the consistent valence force field available through the
InsightII interface was used. The relative charge shifts observed were
adapted to the actual oxaliplatin adduct, based on the cisplatin
charges provided by Yao et al. (1994), and the cyclohexane charges as
assigned by AMBER. The resultant oxaliplatin charges are provided in
Table 2 and Fig. 4. The partial positive charge residing on hydrogens in the unbound molecules (a total of +0.344) was relocated to the
amine-bound CT atom, which previously retained a partial charge of
0.132. An additional charge of 0.05 was also withdrawn from CT by
N3X, resulting in a final CT charge of +0.262 and a final N3X charge of
0.698.
Simulations.
This study consisted of both energy
minimization and molecular dynamics simulations. In both cases, the aim
was to establish valid simulation conditions by ensuring agreement of
the modeled cisplatin adduct structure with that of the crystal
structure. Once the conditions were validated, they were applied to the
oxaliplatin adduct.
Energy Minimization.
Energy minimization of the
parameterized molecules was performed in vacuo with the same dielectric
and 1-4 nonbond adjustments as described above. The phosphate charges
were reduced to 0.2 per nucleotide to account for counterion
condensation. Both adducted DNA molecules were minimized for 500 steepest descent and 6000 conjugate gradient steps, at which point the
structures had RMS derivatives of ~0.03 kcal
mol1Å1 for cisplatin
and ~0.008 kcal mol1Å
1 for oxaliplatin. The energy-minimized cisplatin adduct
retained a structure very similar to the crystal structure (heavy atom RMS deviation for the whole molecule was 0.84Å). However, the oxaliplatin-adducted DNA exhibited multiple structural deviations, most
notably, severe narrowing of the major groove and broadening of the
minor groove. Although these differences suggested an alternate overall
DNA structure for the oxaliplatin adduct (Scheeff and Howell, 1998),
severe narrowing of the major groove requires tight packing of the
phosphate groups, a configuration that seemed unlikely.
Exploratory molecular dynamics trajectories of the cisplatin adduct
displayed a rapid and irreversible narrowing of the major groove
similar to that seen in the energy-minimized oxaliplatin structure. As
this groove narrowing is inconsistent with the crystal structure, it
was concluded that this was most likely an artifact of our phosphate
charge-reduction scheme. Presumably, lowering the charges on the
phosphate groups to 0.2 provides an insufficient repulsive force to
prevent the close packing of the phosphates in a bent DNA molecule,
leading to collapse of the structure. These results may indicate a
limitation to this often-used technique.
Therefore, experiments were conducted with different levels of
phosphate charge: 0.4, 0.6, 0.8, and the standard level of 1.
An energy minimization of the cisplatin molecule was performed as
described above and then an exploratory dynamics trajectory as
described below was undertaken with these different reduction levels. A
charge of 0.6 per nucleotide was found to be the minimum requirement
for maintenance of the basic major groove geometry observed in the
crystal structure. Thus, a charge reduction of 0.6 was used for the
remainder of the simulations, with specific point charges arranged in
the manner described previously (Veal and Wilson, 1991; Table 2).
The cisplatin and oxaliplatin adduct starting structures were subjected
to 1000 steps of steepest descent and 7000 steps of conjugate gradient
energy minimization in the in vacuo conditions described above. At the
conclusion of the minimization, the cisplatin structure had a RMS
derivative of ~0.05 kcal
mol1Å1. The
oxaliplatin structure achieved convergence (maximum derivative <0.001
kcal mol1Å1) before
the completion of the full 7000 conjugate gradient steps. The resultant
structures were analyzed and also were used as starting structures for
the molecular dynamics simulations.
As our cisplatin parameters had been initially been refined in a
molecule with phosphate charges of 0.2, it was important to ensure
that the changes made to the phosphate charges had not compromised the
reliability of the forcefield values. A cisplatin molecule with
phosphate charges of 0.2 was subjected to an identical minimization
as described above, and the minimized forms of 0.2 and 0.6
phosphate-charged molecules compared. The shift in charge had a minimal
effect on the structures, with a whole-molecule (heavy atom) RMS
deviation of only 0.23Å. If only the platinum, amine nitrogens, and
heavy atoms of the two bound guanine bases were considered, RMS
deviation was only 0.03Å.
Checked against the crystal structure, the energy-minimized cisplatin
structure (with a 0.6 charge per nucleotide) still provided a
reasonable model, with a whole-molecule (heavy atom) RMS deviation of
0.92Å. If only the platinum, amine nitrogens, and heavy atoms of the
two bound guanine bases were considered, the RMS deviation was only
0.34Å, a value equivalent to that achieved in the initial parameter
development stage. Thus, no adjustments were made to the platinum
parameters from the values determined earlier in this study.
Molecular Dynamics.
To search a larger conformational space
and provide a structure less subject to the local minima often
achieved during energy minimization, molecular dynamics simulations of
the cisplatin- and oxaliplatin-adducted DNA models were performed with
the Discover module of InsightII. All dynamics simulations were
performed in the same in vacuo conditions and with the same parameters
and charges as described above. The starting structures used were the
energy-minimized forms described in the previous section. Exploratory
dynamics trajectories were run at a constant temperature of 298°K for
200 ps after a 20-ps equilibration. A time step of 0.001 ps was used,
and conformers were sampled every 0.05 ps. No cutoffs were used for
nonbonded interactions.
Although the adjustment of the phosphate charges to 0.6 prevented
closure of the major groove in the cisplatin molecule, the exploratory
simulations were somewhat unstable. Irreversible base pair-opening
events were observed in most simulations within the first 100 ps,
leading to considerable degradation of the DNA structure, especially at
the ends of the molecule. This sort of structure degradation can be
dealt with by using force constraints to maintain hydrogen bond
distances between base pairs in the molecule (Kozack and Loechler,
1997). However, this will enforce a set of base-pairing
distances and geometries, and as our priority was to study the adduct
at the center of the molecule, where some base pairings are distorted,
a different, simpler scheme was chosen. The simulation was constrained
by fixing the position of the terminal phosphorous atoms at the 5' and
3' ends of both strands. Although this arrangement reduced the amount
of conformational space that could be sampled, it was sufficient to
prevent base opening and provided insurance against degradation of the
overall structure of the molecule. Furthermore, because only two atoms
at each end of the molecule were restrained, the adducts (located in
the center of a reasonably long and flexible DNA molecule) were
relatively free to move throughout the simulation. However, it should
be noted that this arrangement, although permitting reasonable
exploration of local adduct conformation, reduces the ability of the
simulation to make predictions about global DNA changes brought on by
the adducts.
Final dynamics trajectories were run at a constant temperature of
298°K for 500 ps after a 20-ps equilibration. A time step of 0.001 ps
was used, and conformers were sampled every 0.05 ps. No cutoffs were
used for nonbonded interactions. Among the resulting pool of 10,000 conformers, every fourth structure was sampled to produce a group of
2500 conformers. These structures were averaged through the Analysis
module of InsightII. The resulting average structure was
energy-minimized for 100 steepest descent steps without restraints to
remove artifacts of the structure averaging procedure. The resultant
structures were analyzed as the "final" cisplatin and oxaliplatin
adduct structures.
| |
Results |
The Cisplatin Adduct Model.
The final cisplatin structure
predicted by the modeling process was compared with the crystal
starting structure to validate the chosen parameters. The parameters
proved to be a reasonable predictor of cisplatin complex behavior. If
the overall structures were considered, RMS deviation (of heavy atoms)
between the final cisplatin structure and the crystal structure was
1.37Å. If only the platinum, ammine nitrogens, and heavy atoms of the
two bound guanine bases were considered, RMS deviation was 0.28Å, a
value similar to those observed in the validation steps of Yao et al. (1994), and superior to the 0.34Å value that was achieved in the prior
energy minimization step.
The final cisplatin adduct structure is shown superimposed over the
crystal structure in Fig. 5. Although the
structures exhibited a large degree of agreement, there were some
differences between the modeled and crystal structures. The modeled
structure exhibited an adduct with a square planar structure across the
N32-PT-NB2 and N31-PT-NB1 bonds, whereas the crystal structure had a
mildly bent geometry. However, the planar geometry is expected for
cisplatin molecules, and was also observed in Yao et al. (1994), and
other modeling studies of cisplatin (Kozelka and Chottard, 1990; Herman et al., 1990). As the 2.6Å resolution of the crystal structure (Takahara et al., 1996a,b) was not adequate to determine the exact position of the ammine and Pt atoms, the planar geometry observed in
the current study is certainly not unreasonable.
View larger version (49K):
[in this window]
[in a new window]
|
Fig. 5.
View of the final modeled cisplatin adduct structure
superimposed over the cisplatin adduct crystal structure. RMS deviation
of heavy atoms was 1.37Å. The view faces the major groove, and is a
close-up of the middle four base pairs of the dodecamer, with the
cisplatin-bound guanine bases at the center. The ammine groups of
cisplatin project outward toward the viewer. The cisplatin bound
strand, on the right-hand side of the image, progresses from 3' to 5'
from the top to the bottom of the image. This orientation was used for
the adducted strand throughout this work to provide enhanced visual
clarity. The bases on the adducted stand are labeled with type and
sequence number. Hydrogens are only shown for the cisplatin adduct and
two adducted guanines. Blue, overall cisplatin crystal
structure. Cyan, two bound guanines and cisplatin adduct
of crystal structure. Red, overall cisplatin final
structure. Orange, two bound guanines and cisplatin
adduct of final structure. Gray, platinum atom of
cisplatin in both structures. Magenta, oxygen involved
in H-bond with an ammine group. White, hydrogen involved in H-bond with
oxygen. O6, guanine oxygen H-bonded to ammine hydrogen in final
structure; OP, phosphate oxygen H-bonded to ammine hydrogen in crystal
structure.
|
|
The predicted hydrogen bonding pattern was also different. The crystal
structure displayed an H-bond between the 5' ammine (N31) hydrogen and
one of the 5' phosphate pendant oxygens. No H-bonds were present
involving the 5' ammine group in the final modeled structure. However,
formation of such a bond is clearly not precluded in the modeled
arrangement, as it was maintained in the energy minimized form before
the molecular dynamics (MD) simulation. As the final modeled structure
is an average of MD conformers, it will not necessarily display H-bond
interactions that were achieved in particular sections of the
simulation. Furthermore, such an H-bond was also not observed in the
NMR solution structures (Yang et al., 1995; Gelasco and Lippard, 1998),
suggesting that this interaction is not an essential characteristic of
cisplatin adducts.
Conversely to the phosphate H-bond above, final modeled structure also
displayed an H-bond that the crystal structure lacks. The 3' ammine
group (N32) hydrogen formed an H-bond with the O6 atom of G7. In the
energy-minimized modeled structure (pre-MD), this ammine group was
H-bonded to the O4 atom of T8. The variety of bonds seen in both ammine
groups suggests that their H-bonds may be dynamic in a real system,
interacting with multiple internal and solvent atoms.
One more notable divergence between the crystal and modeled structures
was the position of thymine 8 (T8). In the crystal structure, this base
opened outward into the major groove, resulting in irregular hydrogen
bonds with its pairing base, adenine 17 (A17). In the final modeled
structure, as well as in the energy-minimized structure, this base was
rotated back into the helix to establish normal Watson-Crick base
pairing (Fig. 5).
The above findings indicate that the platinum parameters and
experimental environment selected produced a reasonable model of the
cisplatin adduct structure. Although some differences were apparent,
many were in interactions that were variable and not essential
components of cisplatin adduct structure. The most important aspect of
the model, the specific geometry of the cisplatin adduct and the bound
guanine bases, was rendered with a high degree of fidelity by the
developed parameters. Thus, we compared the modeled oxaliplatin and
cisplatin structures to look for structural differences that might
account for differential recognition of these adducts.
The Oxaliplatin Adduct Model.
Overall, the cisplatin and
oxaliplatin final structures exhibited a great degree of similarity. If
the cyclohexane ring was disregarded and RMS deviation between the two
molecules calculated, the differences were minimal. The RMS deviation
between the heavy atoms of the entire molecules was 0.98Å, a value
less than that for the cisplatin final modeled structure versus the
crystal structure. When only the Pt, ammine/amine nitrogen, and heavy
guanine atoms were considered, the RMS deviation was 0.07Å.
Although the adduct structures were very similar, the presence of the
oxaliplatin cyclohexane ring, which maintained the expected "chair"
conformation, did have a mild effect on the adduct geometry. The
N31-PT-N32 and NB1-PT-NB2 angles were decreased, and the N31-PT-NB2 and
N32-PT-NB1 angles were increased, as shown in Table
3. Furthermore, base roll relative to the
platinum atom was increased as measured by the N*-NBX-PT (N9-N7-PT) and
CB-NBX-PT (C4-N7-PT) angles. The oxaliplatin structure exhibited the
same square-planar geometry as the cisplatin structure, although the
cyclohexane residue appeared to reduce the planarity of the complex
mildly. The NB1-PT-N31 and NB2-PT-N32 angles were reduced from complete
planarity (180°) to a greater degree than those in cisplatin (Table
3).
View this table:
[in this window]
[in a new window]
|
TABLE 3
Measurements of adduct angles in the final cisplatin and oxaliplatin
modeled structures, defined by atom potential types as shown in Fig. 2
|
|
Other features described above for the final modeled cisplatin
structure, such as the lack of H-bonding between the 5' amine (N31) and
a 5' phosphate oxygen and the restoration of Watson-Crick base pairing
at T8, were maintained in the final modeled oxaliplatin structure.
However, the oxaliplatin adduct displayed a different hydrogen bonding
pattern at the 3' amine (N32). A hydrogen from this group was
H-bonded to the O4 atom of thymine 8, an arrangement seen in
both the cisplatin and oxaliplatin energy-minimized (pre-MD) structures. It is possible that the cyclohexane residue promoted retention of this bond during the MD simulation through the limitation it imposed on the rotation of the amine groups.
It is unclear whether the mild deviations in the immediate adduct
structure between the cisplatin and oxaliplatin models are a result of
direct effects of the cyclohexane residue or indirect (nonbonded)
effects that have been translated through the DNA strand to the two
bound guanine bases. The fact that the RMS deviation was so low between
the two immediate adduct structures (0.07Å), yet much higher between
the overall structures (0.98Å), suggested that the primary mode of the
differential effect of oxaliplatin on the DNA structure might be
nonbonded interactions between the cyclohexane residue and the DNA strands.
As explained in the discussion of the molecular dynamics methodology,
restraints placed on the terminal phosphates reduce the capability of
the simulation to comment on the relative global DNA effects produced
by oxaliplatin and cisplatin. However, as the molecules were completely
unrestrained during the energy minimization step, and only partially
restrained during the dynamics simulation, several interesting
observations about global structural differences can be made. As shown
in Fig. 6, the overall DNA configuration was similar for cisplatin and oxaliplatin. Although the structures displayed minimal divergence, there was a narrowing of the major groove
in the oxaliplatin final structure relative to the cisplatin final
structure. Following a method previously described (Falsafi and Reich,
1993), the major groove width was measured as the distance between the
phosphate group of one nucleotide and the phosphate group of its
closest neighbor across the groove (this nucleotide was five to six
steps down the helix from the first in our molecules). Measured in this
way, the average major groove width was 14.2 ± 2.3 Å in the
cisplatin molecule but only 12.2 ± 0.8 Å in the oxaliplatin
molecule (similar to the crystal structure width of 11.7 ± 1.4 Å) The narrowed major groove suggests the possibility of a more
A-DNA-like conformation in the case of oxaliplatin.
View larger version (45K):
[in this window]
[in a new window]
|
Fig. 6.
MOLSCRIPT (Kraulis, 1991) image of the cisplatin and
oxaliplatin final modeled structures shown in the same orientation
(sideways with respect to the plane of the adducts). The adducted
strand proceeds 3' to 5' from the top to the bottom of the image. A
ribbon traces the phosphate backbone of each DNA strand. Both adducts
(the darkened portions of the molecules) can be seen to protrude
outward into the major groove. The platinum atom is rendered as a small
sphere. Hydrogens are only shown for the drug adducts.
|
|
Minor groove width was measured in a similar manner, with the closest
neighbor phosphate across the minor groove being three steps down the
helix. As demonstrated by the data presented in Table
4, the width was very similar, averaging
13.1 ± 1.5 Å in the cisplatin molecule and 13.2 ± 1.4 Å in the oxaliplatin molecule. Average width in the crystal structure was
larger at 15.2 ± 1.0 Å. Both models retained a similar minor
groove width to that of the crystal structure in the region of the
helix containing the platinum-bound G6 and G7. However, the groove
quickly narrows as one proceeds away from the adduct. This suggests a
more B-DNA-like character to the modeled cisplatin and oxaliplatin
molecules relative to the cisplatin crystal structure.
View this table:
[in this window]
[in a new window]
|
TABLE 4
Minor groove widths, measured as the distance between one phosphate and
its nearest neighbor across the groove three phosphates down the helix
|
|
Sugar puckers in both molecules were primarily of the unusual O4'-endo
conformation, most likely as an artifact of the averaged MD structure.
angles of the DNA backbone were almost entirely in the range of
105-130°, higher than the angle for the C3'-endo pucker of canonical
A-DNA (~85°) and lower than the angle for the C2'-endo pucker of
canonical B-DNA (~155°; canonical angles as given by Takahara et
al., 1996b). This prevented a meaningful comparison of the sugar-pucker
types (and their A-DNA- or B-DNA-like character) in the modeled
cisplatin and oxaliplatin adducts. The degree of bend of the
oxaliplatin and cisplatin final structures was not quantified, but by
visual inspection, there was very little divergence from the bend
observed in the crystal structure for either cisplatin or oxaliplatin.
As shown in Figs. 6 and 7, the most
distinctive difference between the final modeled structures of
oxaliplatin and cisplatin remained the cyclohexane ring present in the
oxaliplatin adduct. The ring protruded an additional 3.7Å into the
major groove relative to cisplatin. The cyclohexane "chair"
conformation was apparently accommodated well by the square planar
geometry of the platinum and nitrogen groups, and it protruded directly
outward in the plane described by these atoms. Angle measurements
across the NB1/N32/CT (cyclohexane) and NB2/N31/CT angles yielded
near-planar results, within the limits of the uneven cyclohexane ring,
of ~170°. In the space-filling model presented in Fig. 7, the
cyclohexane ring can be seen to fill much of the narrowed major groove.
View larger version (75K):
[in this window]
[in a new window]
|
Fig. 7.
Space-filling view of the cisplatin (left) and
oxaliplatin (right) final modeled structures in the same orientation as
in Fig. 6. The DACH ring of oxaliplatin can be seen to fill much of the
major groove. A cyan ribbon traces the phosphate backbone of each DNA
strand. Green, carbon. Purple, nitrogen.
Red, oxygen. White, hydrogen.
Magenta, phosphorous. Gray, platinum.
Orange, ammine/amine nitrogen and hydrogens.
Yellow, cyclohexane ring of oxaliplatin.
|
|
| |
Discussion |
The molecular modeling results presented here suggest that
cisplatin and oxaliplatin produce similar effects on DNA structure. This is particularly true for the covalent effects of the two drugs.
The marked similarity in the geometry of the shared cisplatin and
oxaliplatin adduct bonds in our models indicates that the cisplatin
adduct geometry is likely to be capable of absorbing the stresses
introduced by the cyclohexane residue of oxaliplatin without
significant distortion.
However, the results do suggest a mild effect of the cyclohexane
residue of oxaliplatin on the major groove structure. Our preliminary
work, with a phosphate-charge reduction to 0.2 per nucleotide,
demonstrated a severe tendency for major groove narrowing relative to
cisplatin on energy minimization. Although this simulation was
unstable, work with a more stable system (with a phosphate charge
reduction to 0.6 per nucleotide) still displayed a more narrow major
groove in the oxaliplatin molecule than in the cisplatin molecule. This
may suggest oxaliplatin has different effects on the larger DNA
structure, perhaps a tendency toward a more A-DNA-like helix. Although
a broadened range of conformational space was searched with the
molecular dynamics simulations, the fact that constraining of the
terminal phosphorous atoms was necessary to stabilize the helix
prevented a complete survey of potential helical conformations.
Therefore, the possibility that the DACH ring of oxaliplatin produces a
more dramatic shift in overall DNA conformation through nonbonded
interactions cannot be excluded.
Because the adducted portions of the molecules were subject to the
greatest degree of conformational search, the refined parameters for
the Pt and other adduct atoms were also maximally tested. The
parameters for Pt and its interacting atoms were quite successful at
reproducing the cisplatin adduct within the context of a full-length double- stranded molecule. Although some aspects of the overall molecule and its interactions with the adduct atoms, such as H-bonding, differed from those of the crystal structure, many of these
interactions appear to be highly dynamic. Others, such as the rotation
of T8 to establish standard H-bonding, may be more a result of the
features of the AMBER field than the introduced parameters and may
represent corrections of peculiarities in the crystal structure brought on by packing forces.
As this study was nearing completion, the NMR solution structure of an
essentially identical cisplatinated DNA molecule to the one solved
crystallographically (and used here) became available (Gelasco and
Lippard, 1998). Although this structure displayed the same overall
features as the crystal structure, it also differed with respect to
specific aspects of the geometry, most notably a larger helical bend
and some increase in base roll at the site of the adduct (Gelasco and
Lippard, 1998). When compared with the crystal structure in the form
used in this study, the NMR structure displays a whole-molecule,
heavy-atom RMS deviation of 5.27 Å. If only the platinum, ammine
nitrogens, and heavy atoms of the two bound guanine bases are
considered, RMS deviation is still a notable 1.22 Å. The final
cisplatin model presented here yields essentially identical results
when compared with the NMR structure, with RMS deviation
measurements of 5.29 and 1.20 Å, respective to the above. As the
crystal structure was used as the basis for the cisplatin model, it is
not surprising that both structures compare similarly to the NMR
results. However, it is interesting to note that relative to the
changes seen in these two experimentally derived cisplatin structures,
the changes observed here between the cisplatin and oxaliplatin models
can be considered to be extremely small.
A complete examination of the overall effects on DNA structure of
oxaliplatin adduct formation may become possible in the future through
the use of full solvent, explicit counter-ions, and the particle mesh
Ewald summation for the treatment of boundary conditions. Recently,
this method has been shown to provide simulations in which the
stability of the DNA helix was retained, and very reasonable structures
produced, in DNA molecules (Young et al., 1997; Duan et al., 1997). As
these studies are expanded into the area of platinum drug adducts, we
offer the cisplatin and oxaliplatin parameters refined and developed in
this study.
Although cisplatin and oxaliplatin produced only small differences in
the structure of the DNA itself, the presence of the DACH ring in
oxaliplatin resulted in a large difference in adduct structure. The
modeling indicates that the DACH ring projects directly outward into
the major groove, filling much of the available space near the two
bound guanines. Not only is the DACH group bulky relative to ammines of
cisplatin (Fig. 7), Fig. 8 shows that it
is electrostatically distinct due to its nonpolar character. The
adducts of both drugs reside in a portion of the DNA molecule populated
by polar groups and hydrogen bond donors/acceptors. The polar
ammine groups of the cisplatin adduct match this environment nicely,
but the nonpolar cyclohexane group does not. The two amine groups of
oxaliplatin and portions of the other polar groups are partially hidden
behind the cyclohexane residue in the oxaliplatin adduct. This
configuration would surely present a very distinct "binding pocket"
to proteins of the mismatch repair system and any other DNA-binding
proteins that might require interactions with the major groove.
View larger version (75K):
[in this window]
[in a new window]
|
Fig. 8.
Space-filling view of the cisplatin (left) and
oxaliplatin (right) final modeled structures colored by atom according
to partial charge. Coloration is on a sliding scale, with shades of
purple denoting positive charge, white denoting neutral, and shades of
red denoting negative charge. Dark purple indicates charge  0.5, and
dark red indicates charge  0.5. Atoms with charges between these
extremes are shaded the appropriate lightened color as they approach
neutrality (white). The molecules are rotated approximately 90°
around the helical axis from the image presented in Fig. 7, such that
the viewer is facing the major groove and the platinum adducts. The two
adducts sit slightly to the right of the middle of the two molecules.
The platinum atom is colored cyan (it would be dark purple if
colored on the basis of its charge) and is visible in the cisplatin
molecule but only slightly visible in the oxaliplatin molecule. The
phosphate backbone is traced by a cyan ribbon to clarify the groove
structure. The area between the two ribbons, in which the adducts are
situated, is the major groove. The area above and below the ribbons is
the minor groove. The mild major groove narrowing in the oxaliplatin
molecule can be discerned visually. The polar cisplatin adduct and the
polar DNA region around it can be clearly discerned as an area of red
and purple atoms. The oxaliplatin cyclohexane ring can be seen to sit
within this region, a distinctly nonpolar (white and pink) molecule.
|
|
The presence of the DACH ring in the major groove may be sufficient to
explain the differential treatment cisplatin and oxaliplatin adducts
apparently receive from the MMR system. Both hMSH2 (Mello et al., 1996)
and the heterodimer of hMSH2 and hMSH6 (Yamada et al., 1997) bind to
cisplatin adducts. The normal function of the hMSH2/hMSH6 dimer is to
bind to mispaired bases and single base loops. Single or double base
mispairs produce very slight perturbations in DNA structure (Hunter et
al., 1986; Prive et al., 1991) that may be mimicked by the projecting
ammine groups of cisplatin. Additionally, some single base loops can
produce helical bends in DNA (Turner, 1992) and the bend produced by
the cisplatin adduct may also mimic this structure. This mimicry is
thought to be the basis for recognition of cisplatin adducts by the MMR
system. The cisplatin adducts may be sufficiently similar to true
mismatches to attract the MMR proteins, but sufficiently unique to be
improperly processed by the system or bound so tightly that appropriate
repair systems cannot be engaged. In any event, the cyclohexane residue present in oxaliplatin adducts is well positioned to prevent this binding through both its bulkiness and nonpolar character. Figure 9 provides a stereoview of the modeled
oxaliplatin structure.
View larger version (47K):
[in this window]
[in a new window]
|
Fig. 9.
MOLSCRIPT (Kraulis, 1991) stereo view of the
oxaliplatin adduct final modeled structure, with the numbering
scheme of the nucleotides. The molecule is in approximately the same
orientation as in Fig. 8 with the DACH ring projecting outward toward
the viewer from the major groove.
|
|
Even if, despite steric hindrance by the DACH ring, the hMSH2/hMSH6
heterodimer could bind to oxaliplatin adducts, the presence of the ring
might alter the ability of the MMR complex to process the adduct due to
structural differences in the DNA itself. The modeling suggested that
the DACH ring promotes narrowing of the major groove, and perhaps a
more A-DNA-like conformation to the helix. This conformation change,
alone or when coupled to the bulky DACH adduct, might provide a
distortion that is distinct enough to alter processing.
The lack of a crystal structure for hMSH2 or any other MMR protein
limits speculation on the specific role of the DACH ring of oxaliplatin
in the differential MMR response to oxaliplatin adducts relative to
cisplatin adducts. However, information on the binding of the bacterial
MutS protein (Biswas and Hsieh, 1997) suggests that MMR proteins
contact mismatches at multiple points along the major and minor grooves
near the adduct, as well as contacting the adduct directly. Thus, it is
possible that both adduct structure and the drug-induced changes in
overall DNA conformation modulate MMR protein binding. We offer the
modeled oxaliplatin adduct structure as a tool with which to
investigate this issue further.
We thank Molecular Simulations, Inc. (MSI) for providing
software licenses to the San Diego Supercomputer Center in support of
molecular modeling classes in which this research was initiated, and
the San Diego Supercomputer Center for computer time and
facilities support for this work. We also thank Roberto Lins for
helpful discussions and suggestions, Celeste Lashley for assistance
with computing issues, and John Tate for help in preparing the
MOLSCRIPT figures.
This work was supported by a search contract from Sanofi
Research (Malvern, PA). This work was conducted in part by Clayton Foundation for Research
California Division. Dr. Howell is a Clayton Foundation investigator. A preliminary account of this work was presented at the 1998 Annual Meeting of the American Association for
Cancer Research [Abstract no. 1082, Proc Am Assoc Cancer
Res 39:158]
DACH, diaminocyclohexane;
AMBER, assisted model
building with energy refinement;
MD, molecular dynamics;
MMR, DNA
mismatch repair;
PDB, Protein Data Bank;
RMS, root mean squared.
![[Feedback]](/icons/banner/feedbackACT.gif){kind=icon}
![[For Subscribers]](/icons/banner/subscriberACT.gif){kind=icon}
![[Archive]](/icons/banner/archiveACT.gif){kind=icon}
![[Search]](/icons/banner/searchACT.gif){kind=icon}
![[Contents]](/icons/banner/tocACT.gif){kind=icon}
Top
Summary
Introduction
](/content/vol56/issue3/fulltext/633/img002.gif)
](/content/vol56/issue3/fulltext/633/img004.gif)
](/content/vol56/issue3/fulltext/633/img005.gif)
