SEARCH

SEARCH BY CITATION

Keywords:

  • JAK2;
  • V617F;
  • kinase;
  • simulation;
  • molecular dynamics

Abstract

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS AND DISCUSSION
  5. Acknowledgements
  6. Conflict of Interest Disclosures
  7. References

BACKGROUND:

The tyrosine kinase Janus kinase 2 (JAK2) is important in triggering nuclear translocation and regulation of target genes expression through signal transducer and activator of transcription pathways. The valine-to-phenylalanine mutation at amino acid 617 (V617F), which results in the deregulation of JAK2, has been implicated in the oncogenesis of chronic myeloproliferative disease. However, both the mechanism of JAK2 autoinhibition and the mechanism of V617F constitutive activation remain unclear.

METHOD:

In this work, the authors used molecular dynamics simulation techniques to establish plausible mechanisms of JAK2 autoinhibition and V617F constitutive activation at the atomic level.

RESULTS:

In wild-type JAK2, the activation loop of JAK2-homology domain 1 (JH1) is pulled toward the JH1/JH2 interface through interactions with key residues of JH2, especially S591, F595, and V617, and stabilizes the inactivated form of JH1. In the case of V617F, through the aromatic ring-ring stacking interaction, F617 blocks the interaction of JH1 the activation loop, S591, and F595, thus causing the JH1 activation loop to move back to its activated form.

CONCLUSIONS:

The current results indicated that this simulation-derived mechanism of JAK2 autoregulation is consistent with current available experimental evidence and may lead to a deeper understanding of JAK2 and other kinase systems that are regulated by pseudokinases. Cancer 2009. © 2009 American Cancer Society.

Janus kinase 2 (JAK2) is a member of the Janus kinase (JAK) family and is an essential component of erythropoietin receptor signal transduction.1-7 Through tyrosine phosphorylation, JAK2 activates members of the signal transducers and activators of transcription family (STATs). The phosphorylated STATs then change into their dimer forms and lead to the nuclear translocation and regulation of target gene expression. Normally, JAK2 is in its inactive form until receptor activation occurs; thus, its function is regulated.

Consequently, it has been proposed that the deregulation of JAK2 is associated with hematopoietic disorders and oncogenesis,8-11 especially in patients who suffer from the well known BCR-ABL–negative myeloproliferative disorders (MPDs),12 and, hence, cannot be treated by drugs that were specifically designed for BCR-ABL mutations, such as imatinib. In 2005, 7 independent studies identified the V617F mutation of the JAK-homology domain 2 (JH2) of JAK2 in large numbers of patients with diverse clonal myeloid disorders.9, 13-22

This surprisingly common mutation in patients with BCR-ABL–negative MPDs raises many provocative biologic and clinical questions. It has been suggested that V617 may play important roles in the interaction between JH1 and JH2 to activate JAK2, similar to what has been reported in JAK3.23 In addition, given the finding that it is unusually common, JAK2-V617F may become an important, novel, molecular target for new drug design. The V617F findings suggest that there may be other mutations with similar phenotypes that have been ignored in the past. However, the detailed mechanism of the mutation effect of V617F still remains occult, and this exciting finding of the V617F mutation certainly warrants further relevant research.

JAK2 consists 7 domains, termed JAK-homology domains JH1 through JH7, numbered from the C terminus to the N terminus.2 JH1 is the C-terminal tyrosine kinase domain, JH2 is a pseudokinase domain, JH3 through JH7 may play roles in receptor interactions, and JH4 through JH7 are classified as the FERM domain (band 4.1 ezrin, radixin, and moesin). The relative positions of the 7 domains of JAK2 are shown in Figure 1.

thumbnail image

Figure 1. The relative positions of the major domains of Janus kinase 2 (JAK2) from a snapshot of the wild-type simulations at 60 nanoseconds. Purple indicates JH1; light blue, JH2; blue, JH3; green, JH4 to JH7. Other colored regions: Red indicates the activation loop of JH1 (residues 995 to 1005); yellow, V716F. The activation loop of JH1 (colored red) is very near V617 and clearly is locked by V617 and other surrounding residues.

Download figure to PowerPoint

Structure models are vitally important in understanding molecular mechanisms of bioassemblies, such as the autoinhibition/activation mechanism of JAK2. To date, only a crystal structure for the JH1 domain is available.24 A theoretical structure model of the whole JAK2 assembly has been built using homology modeling methods and coevolution data.25, 26 In this early model, indeed, V617 is located at the interface between JH1 and JH2, which is consistent with the clinically observed results of the V617F mutation,13, 27 although other possibilities also exist.28 In this theoretical model, however, other than the finding that V617 is near the JH1/JH2 interface, there are no detailed interactions at the atomic level that can be detailed, and there is no reasonable explanation for the extremely high specificity of the V617F mutation. A homology model that contains only the JH2 domain has been built and used to predict the effect of exon 12 insertion.29 Hence, a high-resolution crystal structure or a refined structure model is needed desperately to further rationalize the detailed mutation mechanism of V617F and lead to the first step of ration drug design.

Here, we report our refined results on the only available structure model through full-scale, long time molecular dynamics (MD) simulations. Simulation results provide models with significantly different details, and the possible mechanisms of JAK2 autoinhibition and the constitutive activation caused by V617F at the atomic level are revealed.

MATERIALS AND METHODS

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS AND DISCUSSION
  5. Acknowledgements
  6. Conflict of Interest Disclosures
  7. References

A homology-based model of the whole JAK2 assembly was used as the initial structure.26 All ionizable residues were considered in the standard ionization state at neutral pH. Then, the whole protein was put into a 130Å × 120Å × 90Å water box filled with TIP3P water molecules.30 The complex was oriented such that its longest axis was aligned with the x-axis of the water box, and its center of mass was placed at the center of the box. Water molecules distanced to any atom of the complex <1.4 Å were removed. Sodium and chloride ions were added randomly in space (but were kept initially at least 4.7 Å away from any solute atoms) to reach the physiologic concentration of sodium ions (0.14 M) and overall electric neutrality. The resulting system contained 138,824 atoms (JAK2, 17,657 atoms; water, 121,062 atoms; sodium ion, 53 atoms; chloride ion, 52 atoms). All preparation steps mentioned were done using the VMD package (version 1.8.6; available at: http://www.ks.uiuc.edu/Research/vmd/ Accessed on May 1, 2008).31

All MD simulations were performed using the NAMD package (version 2.6)32 with the CHARMM27 force field.33, 34 Default parameters and settings were used except as mentioned below. Periodic boundary conditions were used along with the isothermal-isobaric ensemble (NPT) at 1 atmosphere and 298 K using an extended system pressure algorithm35 with an effective mass of 500.0 atomic mass units and Nosé-Hoover thermostat36, 37 with an effective mass of 1000.0 kcal/mol-picoseconds (ps)2, respectively. The smooth particle mesh Ewald method38 was used. A B-spline interpolation order of 4 was used combined with 120, 120, and 75 fast Fourier transform grid points for the lattice directions x, y, and z, respectively. Nonbonded interactions were treated using an atom-based cutoff of 12 Å with switching van der Waals potential beginning at 10 Å. Numeric integration was performed using the leap-frog Verlet algorithm with a 1-femtosecond time step.39 Covalent bond lengths that involved hydrogen were constrained using the SHAKE algorithm.40

The following procedure was used before the data collection (production) MD simulations. All atoms of the protein complex were restrained first at their original homology model structure positions with a force constant of 50 kcal/mol/Å.2 Water and ion molecules were energy-optimized first and then underwent the following simulation annealing to ensure the proper equilibration.

The temperature was increased from 0 K to 600 K at the rate of 1 K per ps and was kept at 600 K for 500 ps. Then, the temperature was decreased from 600 K to 300 K at the rate of 1 K per ps and was kept at 300 K for 500 ps. The system then was kept at 300 K for 4 nanoseconds (ns). The total equilibration time was about 5 ns in this step. The system was brought down to 0 K, and the same 5-ns equilibration was repeated again, resulting in an equilibration simulation of 10 ns for water molecules and ions.

The water/ion-equilibrated system then was used as the starting point for the production of simulations for both the wild-type and the V617F mutant created by VMD package. Then, for both simulations, the restraint force constant was gradually reduced to 3 kcal/mol/Å2 during a 500-ps period, followed by 2000 steps of energy-minimization for all atoms, and then a 300-ps heating period with the temperature gradually increased from 0 K to 300 K at the rate of 1 K per ps. The production MD simulations of 60 ns were then performed for both the wild-type and the V617F mutant without any constraint or restraint.

RESULTS AND DISCUSSION

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS AND DISCUSSION
  5. Acknowledgements
  6. Conflict of Interest Disclosures
  7. References

To explore the possible mechanisms of the autoinhibition of JAK2 and the constitutive activation because of the V617F mutation, 2 60-ns MD simulations without any constraints/restraints were performed on the wild-type and V617F mutants of the whole JAK2 assembly with 7 domains (JH1 to JH7). Detailed computational setups are described above (see Materials and Methods).

Overall Stability of Simulations

Figure 2 shows the heavy atom and Cα root-mean-squared deviation (RMSD) of the JH1 and JH2 domains (defined as residue 543-1129) from the simulations. RMSD is used commonly as an indicator of stability of MD simulations. MD simulations have been used to optimize homology models and are able to predict structures closer to experimental results.41-43 Figure 2 suggests that the wild-type simulation becomes stable after approximately 20 ns, whereas the V617F simulation does not seem as stable, which suggests that the current simulation may not be long enough to catch the full-scale conformational changes because of the mutational effect in the V617F mutant simulation. However, the V617F simulation does produce significant deviations of important interactions, as discussed below.

thumbnail image

Figure 2. The heavy atom (thick lines) and Cα (thin lines) root-mean-squared deviation (RMSD) (with respect to the starting structures) of JH1 and JH2 (residues 543 to 1129) from simulations. WT indicates wild-type simulation; V617F, simulation of the valine-to-phenylalanine mutation at amino acid 617.

Download figure to PowerPoint

JAK2 Autoinhibition Results From the Interactions Between the JH1 and JH2 Domains

Although it has been demonstrated that JH1/JH2 interaction may be the key to autoinhibition in the JAK3 system,23 and detailed analyses have been reported for the JH1/JH2 interface of JAK2 based on a homology model of JAK2,25, 26 the mechanisms of autoinhibition in pseudokinases remain elusive, as discussed above. Clear interactions between JH1 and JH2 were observed in the wild-type simulation. Figure 1 shows the overall structure of JAK2 at 60 ns. The detailed view of the JH1/JH2 interface is shown in Figure 3 at various simulation time stamps during the MD simulation. At simulation time t = 0 ns, which is the optimized structure after equilibration simulation and has a heavy atom RMSD of 0.452 Å with respect to the original homology modeling structure,25, 26 the JH1 activation loop (colored red in Fig. 3) is near but not in direct contact with JH2. As simulation time proceeds, in the wild-type simulation, the JH1 activation loop moves toward JH2 and forms a relatively strong interaction with JH2 after approximately 5 to 20 ns; whereas, in the V617F simulation, the JH1 activation loop moves away from the JH1/JH2 interface. At 44 ns, the wild-type JAK2 clearly is in its inactivated form, in which the activation loop of JH1 is locked by JH2 residues; whereas the V617F is in an active form, in which the activation loop of JH1 is wide open. Some key distances of the interface between JH1 and JH2 are shown in Figure 4, and their changes along the simulation time are shown in Figure 5. In the specific snapshot shown in Figure 4, the distances from residues L1001, P1002, and F595 to residue V617 are similar. However, only residue L1001 keeps close to V617 during the whole 60-ns wild-type simulation; hence, only the distance between L1001 and V617 is considered important. A second set of simulations (12 ns) with different initial estimated velocities were performed to ensure that differences between the wild-type and V617F simulations are not artifacts because of different initial conditions. The results also are shown in Figure 5.

thumbnail image

Figure 3. Selected snapshots of the JH1/JH2 interface from the wild-type simulations (top) and the V617F simulations (bottom). The numbers in the middle are simulation times in nanoseconds (ns). The JH1 activation loop is colored red. Key residues are marked by text in the same color (R541, white; S591, yellow; F595/F617, purple; V617, tan; P1002, orange-red; Q1003, orange; K1030, blue). For each snapshot, JH1 is on the left, and JH2 is on the right. Note that, at simulation time t = 0 ns, the 2 simulations, except V/F617 side chains, have identical structures.

Download figure to PowerPoint

thumbnail image

Figure 4. Some key distances of the JH1/JH2 interface are indicated by green lines. I-2, the interface between the JH1 activation loop and residues F595 and S591 of JH2; I-3, the hydrophobic core-like interface between 2 loops of JH1 and JH2.

Download figure to PowerPoint

thumbnail image

Figure 5. Key distances of the JH1/JH2 interface (defined in Fig. 4) versus simulation time in the wild-type (WT) simulation (top) and the V617F simulation (bottom). Note that the 2 simulations have identical distance values at simulation time t = 0 nanoseconds (ns). The 60-ns simulations are illustrated on the left, and the shorter (12-ns) simulations, which were started from a different initial estimated velocity, are shown on the right. These 2 simulations demonstrated similar behavior.

Download figure to PowerPoint

Detailed Interactions in the JH1/JH2 Interface

To our knowledge, the detailed interaction pattern between JH1 and JH2 has never been revealed experimentally. To further understand the details of the JH1-JH2 interface from our simulations, a closer look at interacting residues is demonstrated in Figures 4 and 6. There are 3 subinterfaces in the JH1/JH2 interface: 1) the interface between the 2 helices from JH1 and JH2 (defined as I-1) (Fig. 6), 2) the interface between the JH1 activation loop and residues F595 and S591 of JH2 (defined as I-2) (Fig. 4), and 3) the hydrophobic core-like interface between 2 loops of JH1 and JH2 (defined as I-3, Fig. 4).

thumbnail image

Figure 6. The interface between 2 helices, defined as I-1 in the text, of Janus kinase-homology domain 1 (JH1) (residues E890 to S904) and JH2 (residues R588 to K603). Hydrogen bonds of the interfaces are indicated by green lines.

Download figure to PowerPoint

The I-1 interface, which consists of 2 helices (residues R588 to K603 of JH2 and residues E890 to S904 of JH1) (Fig. 6), is similar to the Interface I in a previously reported, homology-based analysis25; however, the hydrogen-bonding patterns are quite different between our simulations and the homology-based model, although the homology-based model was used as the starting point for our simulations. In our simulations, K603 of JH2 hydrogen bonded to E900 and S904 of JH1, and N589 of JH2 bonded to R893 of JH1. Not all of these bonds were observed in the original homology model. In addition, in the original homology model, the side chain of R897 of JH1 is pointing toward to the JH1/JH2 interface and binds to E596 of JH2. This hydrogen bond was not observed in our simulations.

The I-2 interface, which, to our knowledge, has never been reported previously, has 2 surprising interactions: the hydrophobic interaction between F595 of JH2 and the JH1 activation loop, especially P1002, and a possible polar interaction between Q1003 and S591. These 2 interactions stayed consistently after approximately 15 ns in the wild-type simulation. In the original homology-based model, F595 and S591 do not interact with JH1 at all. Conversely, in our MD-optimized model, these 2 interactions seem to be critical to maintain the JH1/JH2 interface.

The I-3 interface (partially shown in Fig. 4), which seems to be another surprise, consists of the loop (V617 to E621) that connects 2 beta-stands of the JH2 N-lobe, a JH1 loop (E1028 to S1032) that interacts with the activation loop, and part of the activation loop itself. This interface is unexpected from the original homology model in inactivated form,27 in which V617 appears to make close contact with the JH1 activation loop and presumably stabilizes the loop. Our simulations, conversely, suggest that the 3 groups (V617 to E621, E1028 to S1032, and part of the activation loop) form a stable core, as indicated by the stable distances from V671 to K1030 and L1001 in Figure 5, and maintain the local conformation of the JH1/JH2 interface.

V617F Not Only Changes the Interaction Between the JH1 Activation Loop and V617F: More Important, It Also Causes Different Binding Modes Between JH1/JH2

Although our MD-optimized results indicate significant differences in the JH1/JH2 interface compared with the original homology model, our results provide a reasonable explanation for the constitutive activation of V617F, as supported by the MD simulation of V617F. Figure 3 compares the MD results between the wild-type and V617F mutant simulations. The JH1 activation loop moves toward the JH2 domain in the wild-type simulation and forms 3 key interfaces, as discussed above. In the V617 mutant simulation, the JH1 activation loop moves away from the JH2 domain, which is surprisingly consistent with what would be expected. From Figure 5, where some key distances along the simulation trajectories are shown, it is clear that the V617F mutation causes the changes in the interfaces of I-2 and I-3 described above.

This mutation breaks down the I-3 interface, although F617 still forms a weak hydrophobic interaction with K1030 in some simulation periods. It causes the total loss of the I-2 interface: First, F617 forms a strong interaction with F595 through an aromatic ring-ring stacking interaction (shown in Fig. 3; 10-ns and 44-ns snapshots). This interaction clearly blocks the hydrophobic interaction between P1002 and F595 as well as the polar interaction between Q1003 and S591 (Fig. 3). Losing these 2 key interactions, the JH1 activation loop moves away from JH2 domain and changes JH1 to its activated form. Thus, the reason for the constitutive activation of JAK2 because of V617F is that F617 destroys the I-2 interface by blocking the interaction between the JH1 activation loop and 2 key residues, F595 and S591, of JH2. Figure 7, which was taken from the snapshot at 60 ns of the V617F simulation, summarizes the overall structure of JAK2-V617F and the relative position of the activation loop of JH1 and F617.

thumbnail image

Figure 7. The relative positions of the major domains of Janus kinase 2 (JAK2) from a snapshot of the V617F simulations at 60 ns. Purple indicates JAK-homology domain 1 (JH1); light blue, JH2; blue, JH3; green, JH4 to JH7. Other colored regions: Red indicates the activation loop of JH1 (995 to 1005); yellow, the residue F617. The activation loop of JH1 (colored red) is open widely and is not interacting with residues of JH2.

Download figure to PowerPoint

Simulations Are Consistent With Available Experimental and Clinical Evidence

The simulations reported here are consistent with currently available experimental and clinical evidence. It has been very puzzling that the V617F mutation is so unique.7 Our results suggest that formation of the aromatic ring-ring stacking interaction between F595 and F617 is the key to breaking down the critical I-2 interface and causes the activation of JH1, which implies that only V617W and V617Y could lead to similar results. Although V617W is unlikely to be observed because of the requirement of 3 DNA base changes (GUC to UGG); V617Y may have different mechanism because of the polar nature of its −OH end group. Hence, V617F would be the uniquely observed mutation from activated JAK2 assays. However, we also would expect to observe some mutational effects in large hydrophobic group substitutions, because they also may block F595 to some degree.

Similar mutations have been observed in other systems but with different results. The corresponding mutations in JAK1 (V658F) and tyrosine kinase 2 (Tyk2) (V678F) also lead to constitutive activation of the kinase domain, but not in JAK3 (M592F).44 From the sequence alignment of the JH1/JH2 region between the JAK family,25 F595, the proposed interacting partner of F617 in JAK2, in fact, corresponds to an aromatic residue in JAK1 (F636) and Tyk2 (Y656) but to a nonaromatic residue in JAK3 (L570). This suggests that, in JAK1 and Tyk2, a mutation from V to F at the V658/V768 position will cause constitutive activation through a similar mechanism; whereas, for JAK3/F592, there is no corresponding aromatic residue interacting with the mutated residue (F592) and, thus, no constitutive activation has been observed.

Several Predictions Can Be Made Based on the Simulation Results

A few key residues are critical in the interfaces described. In particular, S591 of JH2 forms a polar interaction with Q1003 of JH1. The disturbance of this interacting pair would reduce the driving force to lock the JH1 activation loop to JH2 interfaces. Hence, nonconservative mutations of these 2 residues could cause activation of JH1.

The current results demonstrate that L1001 and K1030 also are important in the I-3 interface. However, they interact with V617 with nonpolar van der Waals interactions; thus, substitutions at these positions probably will not cause significant effects, unless the local electrostatic environment is altered significantly.

For the I-1 interface, breaking of the hydrogen bonding network shown in Figure 6 also could weaken the JH1/JH2 interface. Nevertheless, this hydrogen bonding network consists of several strong hydrogen bonds, and a strong perturbation of the hydrogen binding pattern probably is needed to observe the mutational effect.

Long Time Simulations Are Needed to Explore the Possible Conformations

From the RMSD plot in Figure 2, it is apparent that the wild-type simulation needs more than 15 ns to reach a steady state, whereas V617F probably has not yet reached such a stable state in our simulation time (60 ns). Although the simulation results reported here are long enough to derive a possible autoregulation mechanism of JAK2, longer simulations may be necessary to catch overall molecular motions of the JAK2 system. In addition, as shown in Figure 5, the key distances of the JH1/JH2 interfaces (wild-type simulation) fluctuate over the simulation period; for example, the distance between P1002:CG and F595:CZ is ≈5 Å at ≈55 ns and ≈7 Å at ≈58 ns. This is because of the flexibility of the side chain of F595. Such flexibility suggests that a long time simulation is needed if quantitatively analysis on the interface is desired. In the same figure, the V617F mutation shows long-time scale fluctuation of the key distances, which implies a possible breathing motion, with a time scale of approximately 20 ns, of the JH1/JH2 interface entry (near the I-3 interface). Longer simulations or enhanced sampling techniques may be needed to capture this motion fully.

In conclusion, detailed mechanisms of JAK2 autoinhibition and V617F constitutive activation are revealed from MD simulations. The activation loop of the JH1 domain is pulled toward the JH1/JH2 interface through interactions with key residues of JH2, especially S591, F595, and V617, and stabilizes the inactivated form of JH1. In the case of V617F, through the aromatic ring-ring stacking interaction, F617 blocks the interaction of the JH1 activation loop and of residues S591 and F595 and, hence, causes the JH1 activation loop to change back to its activated form. This picture of JAK2 autoregulation is consistent with current available experimental evidence and may lead to a deeper understanding of JAK2 and other kinase systems that are regulated by pseudokinases.

Acknowledgements

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS AND DISCUSSION
  5. Acknowledgements
  6. Conflict of Interest Disclosures
  7. References

We thank Dr. Romano Kroemer and Dr. Thomas Loerting for providing the atomic coordinates of their homology model of Janus kinase 2.

Conflict of Interest Disclosures

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS AND DISCUSSION
  5. Acknowledgements
  6. Conflict of Interest Disclosures
  7. References

Supported in part by the computational resources of Minnesota Supercomputing Institute, and the National Center for Supercomputing Applications under grants MCB050050N and MCB070003T and used the SGI Altix (cobalt) system of the National Center for Supercomputing Applications (NCSA) and the Dell Linux Cluster (Longstar) of Texas Advanced Computing Center, the University of Texas at Austin, through NCSA's TeraGrid mechanism.

References

  1. Top of page
  2. Abstract
  3. MATERIALS AND METHODS
  4. RESULTS AND DISCUSSION
  5. Acknowledgements
  6. Conflict of Interest Disclosures
  7. References