Atomic Level Rendering of DNA-Drug Encounter

Abstract

Computer simulations have been demonstrated to be important for unraveling atomic mechanisms in biological systems. In this study, we show how combining unbiased molecular dynamic simulations with appropriate analysis tools can successfully describe metal-based drug interactions with DNA. To elucidate the noncovalent affinity of cisplatin’s family to DNA, we performed extensive all-atom molecular dynamics simulations (3.7 μs total simulation length). The results show that the parent drug, cisplatin, has less affinity to form noncovalent adducts in the major groove than its aquo complexes. Furthermore, the relative position in which the drugs enter the major groove is dependent on the compound’s net charge. Based on the simulations, we estimated noncovalent binding free energies through the use of Markov state models. In addition, and to overcome the lack of experimental information, we employed two additional methods: Molecular Mechanics Poisson-Boltzmann Surface Area (MMPB-SA) and steered molecular dynamics with the Jarzynski estimator, with an overall good agreement between the three methods. All complexes show interaction energies below 3 kcal/mol with DNA but the charged hydrolysis products have slightly more favorable binding free energies than the parent drug. Moreover, this study sets the precedent for future unbiased DNA-ligand simulations of more complex binders.

Introduction

Independently of the system of interest, an atomic-level understanding of how drugs find the binding site of their targets is crucial. To this aim, specialized software and hardware, together with state-of-the-art analysis algorithms, make in silico techniques important tools when studying a new drug. Biased molecular dynamics simulations have long been used to compute binding free energies (1–4). More recently, breakthrough computational studies have described the unbiased protein-ligand molecular association (5–8). Nevertheless, a nondriven all-atom association mechanism between ligands (in particular metal-based) and DNA, an important chemotherapeutic target, has not been addressed. In the present work we report, for the first time to our knowledge, how cisplatin finds DNA previous to chemical reaction. We analyzed microsecond-long unbiased molecular dynamics simulations and used Markov state models (MSMs) to elucidate the kinetics and thermodynamics of the electrostatic preassociation between cisplatin and DNA. A better understanding of these complex mechanisms, at atomic level, is essential to improve platinum-based therapy.

The drug, clinically known as cisplatin (9,10) (cis-diamminedichloro platinum(II)) and its derivates, are among the most widely used antineoplastic agents (11,12). In addition to testicular treatment (with more than 95% success rate (13)), these platinum (Pt) compounds have worldwide application in many types of human malignancies (14,15). Regardless of the high efficacy of cisplatin, it has many side effects and is innate along with acquired resistance that reduces its medical value. Although this drug was accidentally discovered, newer generations have been designed with characteristics aimed to reduce its undesirable side effects and increase its efficacy toward a larger spectrum of tumors. The curing rates of cisplatin analogs, however, have not been significantly improved after more than 40 years of research (16–18).

It is commonly accepted that the antineoplastic effects produced by cisplatin and other platinum drugs result from covalent attachment between the platinum atom and the electron-rich atoms in DNA bases (19–21). However, before this binding event, a wealth of physiologic reactions occurs. Upon intravenous administration, the high chloro (Cl⁻) concentration, found in blood plasma, hinders the replacement of the labile Cl⁻ ions. When the cell membrane is crossed, however, through passive diffusion or by Cu-transporting proteins (22,23), the Cl⁻ concentration drops from ∼ 100 to 3 mM inducing hydrolysis reactions. Two species are formed: the mono-aquo and the di-aquo complexes resulting from one or two replacements of the Cl⁻ ligands by water molecules, respectively (24,25). In addition to DNA, platinum compounds can react with many biological targets. For example, reaction with proteins, in particular those containing thiol groups (26), have been proposed to be responsible for the severe side effects accompanying platinum drugs intake (27,28). The need for a complete understanding of the different stages of these drugs’ action has prompted innumerous studies at the computational level. There have been studies on the hydrolysis reactions of these drugs using high-level quantum methods (29–34), the interaction with cysteine and methionine models (35), or the covalent binding to DNA (36–38). Also, the effect on DNA’s structure on the binding of platinum drugs has been explored (39–41). However, despite the wealth of computational work various aspects remain unstudied, in many cases, because of the complexity of the processes involved.

The negatively charged DNA molecule, and its high ability for hydrogen bond (Hbond) formation, attracts cations and Hbond donors to its surface. This leads to a very specific microenvironment that will affect chemical reactions. In the case of cisplatin, studies have shown that the binding rates are improved because of weak interactions between the platinum complex and DNA (42). The electrostatic preassociation is expected to affect the reaction rates and positions of platination by increasing the local concentration of the drug in particular sites (42–44). In this work, we study this association before covalent binding and how the process diverges from the parent drug, cisplatin, and its hydrolysis products. Fig. 1 depicts the studied compounds: CPT for cisplatin, CPT1 for the mono-aquo complex, and CPT2 for the di-aquo complex.

Figure 1.

Compounds investigated in this work (left to right): CPT—parent drug (cisplatin with a net charge of 0); CPT1—mono-aquo complex (formal charge +1); CPT2—di-aquo complex (formal charge +2). The central atom in all compounds (dark blue) is a platinum (II) atom. To see this figure in color, go online.

We performed extensive molecular dynamic simulations for these compounds and the results show unique characteristics for each one. The parent drug exhibits extremely low affinity for DNA thus confirming the unlikelihood of direct binding. The higher affinity of the aquo complexes is because of long-range electrostatic interactions that are the main driving force for preassociation to the major groove. Our results provide essential knowledge of the initial noncovalent binding of these compounds, which is valuable information in understanding the covalent addition to DNA and therefore help in designing new compounds.

Finally, this work illustrates how microsecond-long simulations allow for an atomistic description of DNA-ligand interactions. These simulations are now possible in a couple of weeks of computing time (for example, in a small cluster of graphic processing units) making them exceptionally valuable for systematic application.

Material and Methods

MD simulations

We have investigated the noncovalent binding of cisplatin to DNA using unrestrained molecular dynamics (MD) simulation performed with the PMEMD CUDA module within the AMBER11 molecular modeling suit (45). The parm99 force field with the parmbsc0 refinement was used for DNA (46–48) whereas most parameters for the ligands have been derived in this work through quantum mechanical calculations. Some parameters were already available either in the literature or in the general AMBER force field (GAFF) (49,50). Waters were incorporated as TIP3P model (51). The ligands’ partial charges, were derived by fitting the electrostatic potential obtained at HF/6-31G(d) level (calculated with Gaussian 03) through the restrained electrostatic potential (RESP) method and are available in the Supporting Material (50,52). The initial DNA structure was taken from the protein database (PDB) entry 2K0V (53) corresponding to an undamaged sequence ((CCTCTGGTCTCC)·(GGAGACCAGAGG)). This structure has been chosen because it has an identical sequence to an available crystal structure of a DNA strand containing a cisplatin cross-link (3LPV) (54).

We prepared all systems following the same procedure. The DNA + Pt-complexes were neutralized (because we wished to simulate the cell environment with Cl^- concentration of ∼ 3 mM, no additional salt was added) by addition of the convenient number of Na⁺ ions and then surrounded by a 15 Å layer of preequilibrated water molecules in a truncated octahedron box containing ∼ 45,000 atoms. First, the system was minimized through 10,000 steps: 5000 for ions and water minimization followed by 5000 for the entire system. Then the system’s temperature was progressively raised to 300 K using a weak-coupling algorithm during 200 ps of constant pressure dynamics. A time step of 0.5 fs was used throughout the simulations in combination with the SHAKE algorithm to constrain bond lengths involving hydrogen atoms (55,56). Nonbonded interactions were explicitly evaluated for distances below 9 Å. The particle mesh Ewald method was employed to treat long-range electrostatic interactions (57). Constant pressure and temperature (NPT ensemble) were maintained by weakly coupling the system to an external bath at 1 bar and 298 K, using the Berendsen barostat and thermostat, respectively (58). To improve the extraction of statistical data from the ensemble produced using the Berendsen thermostat we have used a relaxation time of 5 ps (59). Keeping in mind that errors can be introduced by the usage of this thermostat, previous work has shown that binding free energies (computed with MSM) in accordance with experimental values can be estimated despite the thermostat being used (60). Simulations were considered equilibrated after ∼ 1 ns by inspection of convergence of total energy, temperature, and pressure. All computed times presented in this study have as time 0 the beginning of the production process. Total production times are 1200 ns for CPT, 1400 ns for CPT1, and 1100 ns for CPT2, and structures were saved every 20 ps. The Supporting Material provides further information as well as all needed parameters (bonding, nonbonding (in Fig. S3 in the Supporting Material), and charges) to reproduce the results presented in this paper.

Binding free energies

The EMMA software package (61,62) was used to estimate binding free energies using MSMs. This method is able to describe the dynamics of complex systems through transitions between different states. The complete simulation data for each compound (coordinates printed every 20 ps) were initially aligned to a reference structure of the DNA backbone strand. The MSM was built by following several steps (62) briefly described here: 1), extract the Cartesian coordinates of the central platinum atom; 2), build 500 microstates using the k-means clustering (different number of cluster were tested); 3), assign all trajectories frame into discrete microstates by a Voronoi discretization; 4), certify the connectivity of these microstates and determine the largest set of microstates; 5), ensure that the implied timescale becomes constant after certain lag times (τ), (see plateau in Fig. S5, i.e., 100, 200, and 50 ns are chosen for estimating transition matrices in CPT, CPT1, and CPT2, respectively). Once transition matrices are defined, the stationary distribution of the microstate can be calculated as π = π T(τ), which is the computing eigenvector of the transition matrix with eigenvalue 1. Then, the potential mean force (PMF) profile, Gi, was obtained by Boltzmann inversion of the stationary distribution, $G_i=-k_{B}Tlog{\pi }_i$. After constructing the three-dimensional PMF, the binding free energy through $\Delta G_0=-k_{B}T\mathit{log}\left({\mathit{v}}_{\mathit{b}}/{\mathit{v}}_0\right)-\Delta W$ is computed. Here k_B is the Boltzmann constant, T = 300 K, V₀ = 1661 Å³ (for 1 M concentration), V_b is the bound volume of the PMF, and ΔW is the difference from the minimum (bounded state) and the bulk (unbounded state) PMF value. Furthermore, the PCCA+ method has been used to determine the metastable sets and validation of the MSM established by the Chapman-Kolmogorov tests (see Fig. S6).

Results

Ligand-DNA dynamics

MD simulations using AMBER11 (45) software with CUDA acceleration were performed for the species CPT, CPT1, and CPT2 with more than 3,7 μs of accumulated simulation time, with individual trajectories ranging from 80 to 200 ns. Table S1 presents a summary of all simulations. Simulations were initiated by positioning the ligand randomly an average 20 Å away from the DNA double helix. The DNA structure used in the present work corresponds to an undamaged B-DNA sequence (d(CCTCTGGTCTCC)·d(GGAGACCAGAGG)) from the PDB with entry 2K0V (53) identical to a cisplatin-DNA complex where the ligand is complexed in a GG (G6 and G7 in bold in the above representation and shown in the left panel of Fig. 2) site in the major groove. For simplicity purposes, we will refer to this G6G7 site as the active site (AS), despite the fact that other less common platination sites exist (14,19,63–65). The results from all simulations are shown in Fig. 2.

Figure 2.

Relative position of the central platinum atom (yellow dots) of CPT (left), CPT1 (center), and CPT2 (right) through the full extent (1200 ns for CPT; 1400 ns for CPT1; 1100 ns for CPT2) of the simulations. The nucleic bases are represented in different colors: pink—guanine (G); green—cytosine (C); blue—adenine (A); gray—thymine (T). The two guanines indicated in the left panel correspond to the most common lesion region in the major groove. The identical positions can be seen also in pink for the other two compounds. To see this figure in color, go online.

In Fig. 2 we can see the initial position of the DNA molecule along with all the relative positions occupied by the central Pt atom through all simulations (only the initial position of the DNA molecule is displayed but all frames are aligned to this structure). For better visualization, Fig. 3 shows the distance between the Pt atom and the N7 atom in G6, with a red horizontal line identifying visits to the AS with distances inferior to 5 Å. The data includes all the individual trajectories with 1.2 μs for CPT, 1.4 μs for CPT1, and 1.1 μs for CPT2.

Figure 3.

Variation of the distance between the platinum atom and the N7 atom from G6 for the three complexes under study. Within each compound, individual simulations are shown separately. To see this figure in color, go online.

We find that for CPT the AS was seldom visited by the ligand that preferred to explore the full extent of the minor groove (narrower region on the opposite side of the major groove) where it is seen “jumping” from site to site (left panel in Fig. 2). The ligand only visits the AS three times, remaining there for very short periods of time, i.e., ∼ 0.6% of the total simulation time. Although a large amount of points for CPT are >25 Å (away from DNA’s surface, Fig. 3), CPT1 and CPT2 spend a large amount of time in the surface of DNA and less time in the bulk of the solvent.

The pattern of binding sites, along the minor groove, for the CPT1 molecule is similar to the parent drug. In addition to these, the ligand now finds the AS much more frequently than in the case of CPT. According to our simulations, the ligand requires an average of 21 ns (for situations where the ligand actually reaches the AS) to find the AS where it remains 10% (with a slight preference for G7 (10%) vs. G6 (8%)) of the total simulation time. The convergence of the AS residence time percentage for CPT1 and CPT2 along the entire simulation is shown in Fig. S1.

The di-aquo complex, CPT2, spends most of the time close to the DNA double helices. It encounters the AS in very short periods of time (in average less than 7 ns) where it remains 39% of the total simulation time. In contrast to CPT and CPT1, this compound locates the AS in all individual trajectories. We see that in the case of CPT2, there is a clear preference for proximity to the G7 site where it is seen 39% of the time in contrast to 24% observed for G6. The sites in the minor groove, frequently visited by CPT and CPT1 complexes, are rarely seen for this complex.

We have also examined the variations in Hbond patterns along all the trajectories. Analyses show, as expected, very different arrays for the three compounds (see Fig. S2). CPT forms Hbonds essentially along the phosphate backbone and in the minor groove, with only 1% of Hbonds to the major grove. In CPT1 both the major and minor grooves form Hbonds. The most remarked difference, in agreement with the populations, is seen for CPT2 where the main interactions occur in the major groove.

In addition to the different binding sites observed for the three compounds, the manner in which they approach the AS diverges. In the case of the parent drug (in the few instances that it reaches the AS), the compound approaches the guanine site with the amine groups facing the N7 atoms of guanines 6 and 7. This is clear from the left panel of Fig. 4 (where the blue color corresponds to nitrogen atoms in the amine ligands and green to Cl^-). We have computed the productive orientations1 and found that only 4.6% of all frames found in the AS are in a suitable position for reaction. In all other cases the negatively charged chloro ligands remain in the outer part, away from the N7 atoms, in an unfavorable orientation for posterior ligand replacement by guanine. In the case of the CPT1, in addition to the amine groups, we observe the approximation of the water ligand, in particular toward the G6 (Fig. 4, center; the red corresponds to the oxygen atom from the water ligands). We find 27% productive orientations for G6 and 22% for G7, the uneven orientation being consequence of CPT1 asymmetry. Finally, CPT2 has 27% productive orientations equally distributed over the two guanines (Fig. 4, right).

Figure 4.

Relative position of the ligands: CPT (left), CPT1 (center), and CPT2 (right) to the active site base guanines 6 and 7. The color scheme is as follows: nitrogen (blue), chloride (green), and oxygen (red). To see this figure in color, go online.

Although CPT is neutral, CPT1 and CPT2 have +1 and +2 formal charges, respectively. We have also found that the two aquo complexes have a higher capability for Hbond formation (given the extra water ligands). Knowing these structural differences, we wanted to determine which of these factors is the driving force for the different behavior. For this, we artificially modeled a new compound, CPT_model, with hybrid characteristics. This complex has the same structure as CPT but the partial charges in each atom have been altered to emulate the point charges of CPT2 (+2 net charge). The CPT_model behaves as the di-aquo complex with identical sites in the DNA surface being visited and rapid introduction in the AS where it remains most of the simulation. In fact, even the orientation in the AS is altered with both the “pseudo” chloro ions (now positively charged as a result of the summation of the water molecule’s partial charges—charges available in the Supporting Material) and the amine groups facing the guanines 6 and 7 alike the behavior seen for CPT2.

Binding free energies

Computing binding free energies is not a simple task because large amounts of data are necessary to attain statistical convergence. In recent years, however, long molecular dynamics simulations, through special purpose MD machines (such as Anton, Pittsburgh, PA (66)), running on graphical processing units (GPU) (67) or grid computing (68) are now possible. MSMs provide a systematic way to decompose the data into meaningful substates and estimating the transition probabilities between these states. In this way it is possible, for example, to identify long-living species, transition pathways, and reaction rates (62). Recent studies have computed binding free energies for protein-ligand interactions with excellent agreement with experimental values (5,6). Thus, force fields together with accurate parameterization of ligands appear to be reliable for a quantitative description of binding free energies as long as sufficient sampling is provided. This opens the possibility for obtaining binding energies in situations where experimental measurements are challenging. In this study, we estimated the DNA-ligand binding free energies for the electrostatic preassociation process and present a kinetic model for the process. Using MSMs, we produced potential mean force (PMF) surfaces for each compound following the procedure explained in the Methods section (two-dimensional PMF contour plots of Fig. S4). Then we have measured the total binding energy for the ligand bound in any region of the DNA molecule (considering all possible binding sites) and the local binding energy in the AS. All energies are summarized in Table 1.

Table 1.

Binding free energies calculated with MSM and MM-PBSA

Binding free energy	CPT	CPT1	CPT2
MSM active site	–	–0.8	–1.2
MSM global	–1.4	–2.1	–2.8
MM-PBSA global	–2.4	–3.3	–3.8
SMD	–1.6	–2.6	–2.8

All values in this table are in kcal/mol.

The overall binding free energy for cisplatin’s preassociation in the complete DNA’s surface is –1.45 (standard deviation ± 0.10) kcal/mol. This low value reflects the weak affinity this ligand has for DNA. The binding free energy has been computed to be –2.08 (± 0.17) kcal/mol for CPT1, and –2.77 (± 0.10) kcal/mol for CPT2. As discussed earlier, CPT finds the AS in very few occasions, and it was not possible to compute the binding free energy for it, which we considered negligible. CPT1 on the other hand is seen visiting the AS quite often, which translates in a local binding free energy of –0.78 (± 0.19) kcal/mol. In the case of CPT2 where the main binding site is located in the AS the binding free energy is –1.19 (± 0.23) kcal/mol. Compared with most intercalators, these are quite weak affinities. Recent MD studies with daunomycin, a common anticancer drug, show a binding free energy of about –10 kcal/mol (1). However, this drug exerts its action solely by stacking with DNA whereas the platinum-based drugs irreversibly bind to DNA.

We have also used the more traditional MM-PBSA method to compute the total binding energy (method's details can be found in the Supporting Material) (69). The calculations summarized in Table 1 retrieved –2.4 kcal/mol for CPT, –3.3 kcal/mol for CPT1, and –3.8 kcal/mol for CPT2. MM-PSBA is known for overestimating binding free energies partially because of incorrect entropic contributions (70). Having this in mind, despite the ∼ 1 kcal/mol bias in the MM-PBSA results, the relative interactions are in good agreement between the two methods. Again CPT presents a lower binding affinity than its aquation products. Steered molecular dynamics with the Jarzynski estimator (71) were also performed to compute the binding free energies for the three compounds. More than 450 independent simulations were performed to adequately estimate these energies. For more details on the method please consult the Supporting Material. The results show, in agreement with the other two methods, a lower binding affinity for CPT (–1.6 kcal/mol) whereas CPT1 and CPT2 show –2.6 and –2.8 kcal/mol, respectively. These results encourage us to believe that the MSM binding energies are a good approximation to the absolute values.

Kinetics and cisplatin binding mechanism

MSMs are a useful tool for extracting kinetic information from atomistic simulations. For this, we have identified the most populated states for each simulation and how these were connected. In Fig. 5 we show the most populated states for each compound. In the case of CPT, six metastable states were identified: A (green), B (blue), C (purple), D (red), E (mauve), and the bulk solvent (S) in yellow. In the lower part of the left panel we see the relative rate constants for each process with solvent exchange rate constants corresponding to the arrows pointing in and out of each state. The size of the arrows correlates with the relative rate constants (values are available in Table S2). Fig. 5 illustrates that the fastest processes in CPT correspond to exit from states A, B, D, and E so it would seem that the dissociation rate is larger than the association. This is in agreement with the low affinity observed for CPT. The location of these sites are as follows. The A state is located in one extremity of the DNA double helix close to the N3 atoms of A22 and G23. The B state is close to A20 and G21. The C state is “behind” the major groove GG site close to the two guanine N3 atoms. The D state is close to A17 and C18. Finally, the E state is in DNA’s 5′ terminal of the second strand close to A15 and G16. It is interesting to observe that in the minor groove for both CPT and CPT1 there is a clear preference for binding in the AG sequences.

Figure 5.

Network of the most relevant transition pathways for the association and dissociation processes and binding sites. The clusters represented by the different colors and letters correspond to independent occupation sites. The yellow beads match the clusters coinciding with the bulk solution (S state). The orange spheres in CPT1 and CPT2 match the expected binding site in the major groove. A putative kinetic mechanism for ligand binding for each of the three figures is depicted (lower panel). The size of the arrows indicates the relative reaction rate (by orders of magnitude) for each individual process. To see this figure in color, go online.

In the case of CPT1, we find the minor groove states already seen in CPT and two additional states: F and G. The F (pink) state in Fig 5 is seen in the entrance of the major groove and is connected to the G (orange) state (in the major groove) and the D state (in the minor groove). The G state, located in the AS, is only connected to the solvent and to the F state. The largest rate constant for this system is the binding from the solvent in the G (AS) state and exit from the F state. For CPT2, we had already seen that the ligand binds very quickly to the AS and does not visit the minor groove often. By observation of the right panel of Fig. 5 we can confirm that the binding process for CPT2 is different from the other two compounds. The states in the minor groove seen in the other two complexes are not found here. The binding process flows both from the bulk directly to the AS (G state) or passing by both site E and F to the G state.

Specific DNA analyses

The conformational modifications that occur on the DNA molecule upon the formation of Pt-DNA adducts have been extensively studied through x-ray and NMR structures as well as computational work (54,64,72). The large flexibility of the DNA molecules allows it to accommodate intrastrand cross-links by adopting a bent structure. In this study we were interested in observing whether the noncovalent binding of the ligands studied would have any effect on DNA’s structure. For this, we employed the Curves+ software (73). We analyzed the axis bending of the three systems along the complete simulation time. The results show that for all three compounds significant axis bending occurs that is not necessarily associated to the binding process. We do not observe any particular trend and assent that covalent binding is required for significant bending (in the direction of the major groove) of the molecule’s backbone. In Fig. S7 we compare the results for the two cases: CPT and CPT2.

The presence of a spine of hydration in B-DNA, which is essential to maintain its native conformation, is well known (74,75). Visual inspection of the spine of hydration along the simulations confirms several permanent hydration shells in the minor groove, which are disrupted with the introduction of ligands in this region. The movie S1 simulates one such event.

Discussion

It has been theorized that cisplatin’s cytotoxicity is attributable to its ability to bind, cross-link, and structurally distort DNA. When DNA is cross-linked, complex repair mechanisms are activated ultimately leading to cell death. Despite the limitations of cisplatin-based therapy (i.e., its side effects and its intrinsic and acquired resistance), its amazing cure rates for some malignancies, in particular in testicular cancer, make it a promising model in oncology research. Strategies today include modified versions of cisplatin with improved single and multinuclear platinum complexes, Pt(IV) prodrugs as well as specific delivery through, for example, nanoparticules (43,76–78). Current and future developments could benefit from a complete understanding of the dynamical interactions of these compounds especially on how they locate their cytotoxic targets. In this work, we sought to elucidate the preassociation mechanism of cisplatin and its hydrolysis products to DNA. Furthermore, we sought to show that routine calculations are capable in such a study.

The results previously described indicate that the replacement of the chloro ions by the aquo ligands has a drastic effect on the way the drug approaches DNA. The binding sites differ considerably and also the orientation in which the compounds approach the nucleic bases is divergent. The replacement of the first chloro ligand by a water molecule not only increases the amount of time the ligand spends in the AS (from 0.6% to 10%) but also the number of productive orientations (4.6% to 27%). Of these productive orientations in CPT1, 27% are closer to the G6 whereas 22% are closer to the G7. The fact that the water ligand is, on average, closer to the G6 site could favor this nucleic base as preferential site for the first nucleophilic substitution. In fact, the preference for first platination in the 5′ position is well known (36,79,80). In the case of the second chloro replacement, the increase in time spent in the AS is even more significant. CPT2 spends 39% of the simulation time with 27% productive orientations equally distributed between the two guanines. Another interesting aspect from this study is the type of potential adducts that can be formed according to the positions observed along the DNA strands. The main adduct for CPT1 and CPT2 are found in the AS (G state in Fig. 5). In addition, other potential binding sites have been observed, especially in AG sequences (close to adenine and guanine N3 atoms) in the minor groove. These results confirm that diverse platination sites can be expected. Experimentally the main adducts are seen in 1,2-intrastrand cross-link between the N7 atoms of two adjacent purine bases GG (60% to 65%) or AG (20% to 25%) (14,19,63–65).

Further studies with the CPT_model compound (identical to CPT but with a total charge of +2) show that it quickly moves into the AS like the di-aquo complex. It is interesting to see that even though this compound is unable to form the Hbond network of CPT2 it behaves in an identical manner showing that the fast diffusion of the ligand to the major groove is mainly driven by electrostatics. Recent molecular dynamics studies by Guéroult et al. (81), which investigated the interaction between Mg²⁺ and DNA, showed that divalent cations preferentially bind in DNA’s major groove. The authors proposed that the electrostatic potential in the major groove is particularly attractive for cations. The results found in the present study confirm the strong electrostatic nature of the preassociation process for the positively charged compounds. Furthermore, our results indicate no meaningful distortion of the DNA structure (bending and unwinding of the double helix) (54,64,72,82), associated with this electrostatic preassociation.

Microsecond time-scale simulations appear to be sufficient to describe the cisplatin-DNA binding process and to derive free energies through the transition probabilities in the MSMs. Indeed one of the main advantages of the use of MSM consists in determining just how much data is needed to compute the binding free energies (and avoid extending simulations beyond what is necessary). The results indicate lower global interaction (–1.4 kcal/mol) between the CPT compound and DNA but slightly more favorable for the aquo complexes (–2.1 and –2.8 kcal/mol for CPT1 and CPT2, respectively). Most importantly, whereas for CPT the binding free energy in the AS is negligible, CPT1 and CPT2 show a local binding free energy of –0.8 and –1.2 kcal/mol, respectively evidencing their higher affinity to bind in this site. Since cisplatin binds covalently to DNA, no experimental information exists on the electrostatic preassociation. As mentioned, recent developments in computational modeling sampling and analysis techniques can fill this gap. We have used different theoretical methods ranging from inexpensive MM-PBSA to unbiased molecular dynamics simulations associated with MSM. We see that, despite small differences between the methods, all indicate that the parent drug is the one exhibiting the lower affinity toward DNA whereas the charged aquo products interact with DNA more strongly.

For many years, new platinum-based drugs have been proposed based on structural assumptions. Most of the new compounds put forward followed basic rules, such as cis disposition of the two amines, a minimum of one N-H group on the amine, leaving groups with weaker trans effect than the amine, etc. However, it has become clear that more information is needed to effectively improve the clinical value of this family of compounds. For this, it is essential to understand all steps involved. With the improvement of computational techniques, details have been revealed at the molecular level. We now know that differences exist in the hydrolysis rate of cisplatin, carboplatin, oxaliplatin, and nedaplatin. Variations in the first and second hydrolysis rates have been suggested to be responsible for differences in activity (32). Several authors have also been interested in how these drugs covalently bind to guanine and adenine bases using small models and DFT methods. However, it is important to know exactly how these drugs approach the binding sites in DNA and for this reason a complete understanding of the preassociation process is fundamental. Not only the studies presented here can establish differences in the affinity of different drugs to DNA but most importantly this type of work will allow for an adequate description of the covalent binding by knowing the correct orientation of the drug in the binding site before reaction.

Overall, we show that the simplicity of the system when compared with protein-ligand interactions permit gathering sufficient data to map all-atom association/dissociation mechanisms in DNA-ligand interactions. The computational cost, both in CPU time and hardware, is quite modest: 1 wk for ligand parameterization (if needed), some weeks for running the MD simulations, and ∼ $5,000 in hardware units. This in turn, allows for the extraction of thermodynamic and kinetic information, which are of high value. This study sets the precedent for unbiased DNA-ligand simulations of more complex molecules such as intercalators, and its extension to diverse DNA sequences.