Introduction

Adenosine triphosphate (ATP) is one of the most important molecules on Earth, present in all cells of all living organisms. This high-energy nucleotide powers, in several different ways, most biochemical processes that require energy. One such way is the transfer of a phosphate group to another molecule in a process called phosphorylation. This reaction is carried out by enzymes called kinases. Identification of substrates that are phosphorylated by specific kinases is difficult because of the enormous number of these enzymes, and also because kinases display overlapping substrate specificities [1, 2]. The approach presented by Shah et al. [3] is based on using both mutatated kinases that enlarge the ATP-binding pocket, and ATP analogues, whose specificity allows the kinase substrates to be identified. This method was used successfully to study Rous sarcoma virus tyrosine kinase [3].

The process of designing ATP-analogues complementary to modified kinases has to begin by understanding the conformational behavior of the nucleotide, and assurance that the modification introduced into ATP does not change its conformational properties. In this work, we study the conformation of the ATP molecule and the 12 analogues proposed by Shah et al. [3] bound with magnesium cation (Mg2+) using molecular dynamics (MD) simulation enhanced with simulated annealing (SA). We present a full set of AMBER force-field parameters for each of the ATP analogues, which provides the possibility to use models of these molecules in other computational experiments, such as docking and molecular modeling of the interaction between such analogues and kinases.

Since the discovery of protein kinase activity in 1954 [4], the field of protein kinase drug discovery has advanced dramatically. More and more researchers are involved in the design of new kinase inhibitors, as there is much focus on this subject by the pharmaceutical industry. Molecular modeling is one of the most helpful tools in this field. For example, molecular modeling was used successfully in studies on inhibitors of vascular endothelial growth factor receptor tyrosine kinase [5], the cyclin-dependent kinase family [6, 7], as well as in the case of the serine-threonine kinases p38 [8], Aurora A [9] or checkpoint kinase 1 [10]. The models presented in this work, together with their AMBER force-field parameters, can also be used for modeling kinase inhibitors as well as for designing ATP analogues other than those shown here.

Methods

Initial models

The ATP analogues considered in this work were taken from a set presented by Shah et al. [3]. Models of these ATP analogue molecules were built using MOLDEN [11], using also a model of the ATP molecule from the Structural Cambridge Database (entry ADENTP03 [12]) as a template. Two of the ATP analogue models, namely N6-methoxy ATP (AT1P) and N6-pyrrolidino ATP (AT7P) were built in our previous work [13]. Hybridizations of atom N6 in ATP-derivatives were determined by comparison with molecules having the N-substituent group attached to the aromatic ring. A comparison of crystal structures with the ATP-models is presented in Table 1.The charge of ATP and its analogues was −3, consistent with the models presented by Shah et al. [3]. Parameterization of the ATP analogues to the AMBER force field was performed as recommended in the AMBER [14] manual. Restrained electrostatic potential (RESP) was used to obtain partial atomic charges of ATP and its 12 analogues. The structures of the models were subjected to geometry optimization at the HF/6-31G* level using the ab-initio chemistry package GAMESS [15]. Charges were calculated from optimized geometries using R.E.D. (see http://q4md-forcefieldtools.org/RED/). A complete set of information on the assigned atom types and atomic charges in LEaP Object File Format (OFF) files can be found in the electronic supplementary material. All modifications introduced to the AMBER force field [16], which were used to parameterize ATP analogues, are shown in Table 2.

Table 1 Hybridization of the N6 atom of adenine in 12 ATP analogues, as obtained from similar crystal structures from the Cambridge Crystal Structural Database [27]
Full size table
Table 2 Parameters introduced to the AMBER force field
Full size table

ATP requires the presence of a divalent cation, usually magnesium, which is coordinated by ATP phosphate groups. The presence of the ion is crucial to obtaining proper conformation of ATP but is also very important in catalyzing the phosphotransfer reaction [17], particularly in kinases [18]. One magnesium cation (Mg2+) was added to each model of ATP molecule using the LEaP command addIons. The addIons procedure places the counterion in a shell around one or more ATP analogue anion using a Coulombic potential on a grid.

The reason why only one magnesium cation was added can be found in our previous work [13, 19], where ATP and two ATP-derivatives respectively were docked to the serine-threonine kinase PrkC, and only one magnesium ion was found near the phosphates. An extensive study by Martinez et al. [20] on ATP conformations and ion binding modes in the active site of anthrax edema factor also supported a mechanism of kinase activity based on the presence of one magnesium cation in the active site of the enzyme.

The magnesium cation was held using distance restraints during the SA simulations. To obtain an input distance restraint, we calculated the distances between the magnesium ion and four oxygen atoms of the phosphate groups beta and gamma (see Fig. 1), from initial structures after using addIons command. The values of distances varied between 2.0 and 4.5 Å. These values were used as boundary values and the force constants for distance restraints were 20 kcal mol−1 Å on each of the four distances (Mg–O1B, Mg–O2B, Mg–O1G and Mg–O2G).

Fig. 1
figure 1

Distance restraints held the magnesium cation (Mg2+) close to the charged oxygen atoms (atom types O1B, O1G, O2B and O2G) of the phosphates (2.0–4.5 Å)

Full size image

Minimization and SA simulations

Initial structures of ATP and its analogues were optimized with SANDER, part of the AMBER 9.0 package [14] using Steepest Descent minimization for 5,000 steps, followed by the 15,000 steps of conjugate gradient minimization. Subsequently, all models were submitted to a 50-ps (50,000 steps with 1 fs time step) run of the SA protocol for 1,000 times. The single SA protocol consisted of four stages (see Fig. 2):

  1. (1)

    0–2 ps Short equilibration

  2. (2)

    2–10 ps heating from 300 K to 1,200 K

  3. (3)

    10–40 ps slow cooling from 1,200 K to 300 K

  4. (4)

    40–50 ps equilibration at 300 K

Fig. 2
figure 2

Time/temperature dependence during single simulated annealing molecular dynamics (MD) run. The simulated annealing (SA) protocol is shown in the box

Full size image

Each set of final coordinates from one SA simulation was used as a starting point for the following simulation. It is worth noting that there is a short equilibration period, in the first stage of the SA run, before the heating stage. Running the heating procedure immediately after minimization, brings instability to the system and causes unusually high energy.

The simulations were carried out using the implicit solvent model, namely the generalized Born (GB) solvation model developed by Onufriev, Bashford and Case (GBOBC) [21]. We are aware that the continuous solvent model is a crude approximation and does not take into account the molecular structure of water. It also neglects several modes of interaction, like hydrogen bond formation between solute atoms and water, but our results suggest that it is good enough to reproduce experimental results. The GB solvent model enabled us to perform quick SA simulations without causing artifacts in the water structure. For temperature regulation we used the Langevin thermostat [22] and a collision rate of 1 ps−1.

Analysis

Analysis of the ψ angle vs the inter-proton distance H8-H1′ and H8-H2′

The calculations of the torsion angle ψ, defined as O4′–C1′–N9′–C4′, and inter-proton distances (H8–H1′ and H8–H2′) were carried out using the PTRAJ module of AMBER 9 [14] based on the final structures from each of the simulation. Collected data are presented in Fig. 3 which was created using the gnuplot program (http://www.gnuplot.info). Points illustrating the dependence between the torsion angle (O4′–C1′–N9–C4) and inter-proton distances were fitted to the function: \( {f(x) = A{\hbox{*sin}}\left( {\left[ {x - o} \right]/B)} \right) + C} \) (Fig. 3a). Parameters obtained after the fitting procedure are presented in Table S1 (see electronic supplementary material). Fitting was done only for calculations concerning ATP, to visualize the shape of the plot of this function.

Fig. 3
figure 3

Dihedral O4′–C1′–N9–C4 angle dependence on inter-proton distances H8–H1′ (crosses) and H8–H2′ (dots) measured for ATP (a) and its analogues (b). This plot can be compared to the plot presented by Jiang et al. [24], the results of which were obtained from crystallography and NMR system (CNS) experiments

Full size image

Results and discussion

Here we report the results of the conformational analysis carried out after MD simulations.

Magnesium cation restraints

The distances between magnesium and oxygen atoms (O) were measured to verify if the magnesium cation was kept in place by the restraints during our simulations. An illustration of what happens during a single SA run is presented in Fig. 1. We show only the first run for ATP, because the remaining runs for the ATP derivatives are very similar and do not show any significant variations.

Figure 1 shows that the distances between the Mg2+ cation and the four charged oxygen atoms vary between 2.0 Å and 4.5 Å. The restraints hold the magnesium cation at the imposed distances. The Mg2+ cation is free to move within the restraints, e.g., around 25 ps, Mg2+ moved from oxygen O2B towards O1B and then went back to its previous position.

To better understand what is happening in the same time scale, Fig. 2 illustrates the time/temperature dependency during the same, single SA run. Significant changes in the four monitored distances were observed at the slow cooling stage of the SA run, namely between 10 ps and 40 ps. This is understandable, because the slow cooling stage allows the system to obtain the most ordered or lowest energy arrangement of molecules, and stay close to the equilibrium state [23].

To verify if the magnesium cation is kept close to the initial position during all 1,000 SA MD runs, we measured the distances between Mg2+ and the charged oxygen atoms from the beta and gamma phosphate groups (O1B, O2B, O1G and O2G) for each snapshot taken at the end of every SA MD run.

Figure 4 presents the frequencies of occurrence of the distance values around 2.0 Å and 4.5 Å for ATP (ATP-analogues showed similar results). The frequency of occurrence of values at 2.0 Å is higher than the frequency of occurrence of values at 4.5 Å. We observed that when the cation is bound within a close distance (2.0 Å) between two oxygen atoms it does not change its position, which is why we observe the Mg2+ cation at almost the same distances. Close distance also represents stronger ion interaction between the magnesium cation and the negatively charged oxygen atoms. Occurrences of distance values in the range between 2 Å and 3.5 Å are very rare because this position is energetically unfavorable.

Fig. 4
figure 4

Distances between the magnesium cation (Mg2+) and oxygen atoms from beta and gamma phosphate groups (O1B, O1G, O2B and O2G) in each of the MD runs, calculated for ATP

Full size image

Analysis of the distribution of the values of torsion angle O4′–C1′–N9–C4

ATP conformation can be partially defined by value of the torsion angle between atoms O4′–C1′–N9–C4 (see Fig. S1). If the value varies between 0 ± 90 degrees, then the conformation is denoted as “syn”, and if the value varies between 180 ± 90 degrees than the conformation is denoted as “anti”. We checked the distribution of values in conformations of our models, which were submitted to SA MD. Values of the dihedral angle (O4′–C1′–N9–C4) were measured for each of the snapshots taken after the equilibration stage. Table 3 presents the measured values of the torsion angle for each of the ATP-analogues. As an example of the distribution of the analyzed dihedral angle, Fig. 5 shows the distribution of values of the O4′–C1′–N9–C4 angle in ATP.

Table 3 Occurrences of dihedral O4′–C1′–N9–C4 angle values within small, 30° degree ranges
Full size table
Fig. 5
figure 5

Distribution of values of the dihedral angle O4′–C1′–N9–C4, partially defining the conformation of ATP obtained in 1,000 MD runs. There are two maxima, first present around −120 degrees (conformation anti) and the second around 60 degrees (conformation syn)

Full size image

Both Table 3 and Fig. 5 clearly show that there are two maxima in the frequency of occurrence of the torsion angle value. The first maximum is present around −120 degrees (conformation anti) and the second maximum is around 60 degrees (conformation syn). The ATP molecule or its analogue frequently adopts the anti conformation when it binds inside the ATP-binding pocket. The results show that the first maximum is broad and the other is rather sharp. These two maxima represent two major sets of conformations of ATP and its analogues, and the transitions between them may be essential for kinase activation.

Dependence between O4′–C1′–N9–C4 angle and inter-proton distances H8–H1′/H8–H2′

The distribution of conformations of ATP can also be partially described using analysis of the inter-proton distances H8–H1′ and H8–H2′ versus the torsion angle ψ (C4–N9–C1′–O4′). This analysis was previously applied by Jiang and Mao [24] to the interpretation of combined NMR and molecular modeling experiments on unmodified ATP molecules. Comparison of our results and those obtained by Jiang and Mao shows that we observe the same structural behavior. From Fig. 3, we can clearly conclude that there are two main conformations of ATP model, the first, which is mostly in the anti conformation with ψ angle value of −120 ± 60 degrees ,and the second (syn) with ψ angle value of 45 ± 30 degrees. The two least frequent conformations are characterized by ψ angle values of −30 ± 30 and 120 ± 30, respectively.

Dihedral angle C4–N9–C1′–O4′ analysis shows that, in ATP as well as in the 12 analogues, there are two gaps in the distribution of angle values ranges, the first between −30 and 30 degrees described by cis-conformation, and the second between 90 and 175 degrees, which mostly covers region of anti conformation, as presented on the right side of Fig. 3a. The cis-conformation is forbidden only for some of the ATP analogue models, namely 3, 7, 10 and 12 (see Fig. 3b). This is a transitional conformation between the “inactive conformation” (ψ angle values around 60) and the “active conformation” (ψ angle values range around 160). By the “active conformation” we mean a conformation of ATP that is able to form a complex with protein. If the molecular conformation falls into the inactive state and the cis-conformation is forbidden, then there is no possibility to change the conformation into the active state. This conformational “gap” can possibly decrease the binding affinity of such a ligand. This hypothesis explains the results of biological experiments, namely the low values of inhibition for ATP-analogues 3, 7 and 10 found by Shokat and coworkers [3, 25].

The second region is forbidden for all the ligands, ATP and its derivatives. To understand why the second region is forbidden for all the ATP models, we calculated the energy profile of the ATP model, changing the torsion angle by rotating the adenine ring. We observed an increase in Van der Waals (VDW) energy in the second forbidden-conformational state (data not shown). Close distances between the adenine nitrogen atom N9 and hydrogen H2′ and H3′ atoms from ribose ring were revealed by visual inspection of the conformation.

To summarize, the anti-conformation is disallowed due to steric effects in all of the models, while in contrast the presence of the cis-conformation is crucial for biological activity.

How does ATP conformation depend on the conformation of sugar?

The conformational change between the active and inactive stage is possible only through the cis-conformation. The role of ribose conformation might also be significant in obtaining the active/inactive stage. To verify this hypothesis, we calculated the relationship of ψ angle and torsion angle between atoms C1′–C2′–C3′–C4′, which defines the sugar conformation (see Fig. S1 and Fig. 6). The most commonly populated sugar conformations [26] were the C3′-endo region, corresponding to a torsion angle around 35°, and the C2′-endo region with torsion angle values around −35°.

Fig. 6
figure 6

Dihedral O4′–C1′–N9–C4 angle dependence on sugar conformation, described by dihedral C1′–C2′–C3′–C4′angle measured for ATP (a) and its analogues (b). In a, the schematic conformational transition from inactive to active state is also shown. The conformations were not clustered but we indicate only those that were close to each other (black boxes in a)

Full size image

We observed that rotation of the adenine depends on the conformation of the sugar. Figure 6 presents four main conformations, two in the active state and two in the inactive state. One of the conformations in the inactive state is preferable and reflects the C2′-endo sugar conformation and the torsion ψ angle around 45 degrees. For most of the analogues it is not possible to change the adenine position from the inactive to the active state having the C2′-endo conformation. Furthermore, the switch between the active and inactive state is possible with the C3′-endo sugar conformation present. The same conformational behavior was observed in the case of the ATP molecule. Focusing on the active conformation, we observed that it is also favorable when the sugar is in the C2′-endo conformation.

We speculate that changes in sugar conformation are involved in the transition from the inactive to the active state of ATP analogues, and that the sequence of changes includes the following stages (Fig. 6a):

  1. (5)

    C2′-endo cis-conformation inactive

  2. (1)

    C3′-endo cis-conformation

  3. (2)

    C3′-endo anti-conformation

  4. (3)

    C2′-endo anti-conformation active

Summary

In this work we present parameters introduced into the AMBER force field and MD calculations for ATP and 12 ATP analogues proposed by Shah et al. [3] with one magnesium cation, Mg2+. The ion was found close (2 Å) to the charged oxygen atoms of the beta and gamma phosphate groups of the ATP analogue molecule. Analyzing the O4′–C1′–N9–C4 dihedral angle values, which partially describe the conformation of ATP (and its N6-substituted derivatives), we discovered the existence of two maxima. One sharp maximum located near the +60° value of the angle, which we call the “inactive-state”; and a second, broad maximum located close to −120°, called the “active-state”. The active-state conformation is frequently found when ATP is bound in the ATP-binding pocket of kinases. Change between the states is possible only through cis-conformation (0 ± 30°), which is not allowed for all ATP-analogues, namely AT3P, AT7P, AT10P and AT12P. These results correspond to the findings of Shah et al. [3], which explained the low values of inhibition activity of those ATP-derivatives with wild-type and mutant Rous sarcoma tyrosine kinase. Additionally, we determined the character of the contribution of the conformation of the ribose ring to the transition between the inactive/active state. The path from the inactive state to the active state leads through changes between the C2′-endo and C3′-endo of the sugar.

Substitutions proposed by Shah et al. on the N6 adenine atom of ATP do not influence the main conformational properties of the nucleoside. Parameters introduced by us to the AMBER force field tested by MD simulations gave reasonable results regarding conformation, which are comparable to experimental values [24].