PI3Kα oncogenic mutations weaken the interactions that stabilize the p110α/p85α heterodimer

Abstract

Phosphatidylinositol 3-kinase α (PI3Kα) is a heterodimeric lipid kinase that catalyzes the conversion of phosphoinositide-4,5-bisphosphate (PIP₂) to phosphoinositide-3,4,5-trisphosphate (PIP₃). The PI3Kα signaling pathway plays an important role in cell growth, proliferation and survival. This pathway is activated in numerous cancers, where the PI3KCA gene, which encodes for the p110α PI3Kα subunit, is mutated. Its mutation often results in gain of enzymatic activity; however, the mechanism of activation by oncogenic mutations remains unknown. Here, using computational methods, we show that oncogenic mutations that are far from the catalytic site and increase the enzymatic affinity, destabilize the p110α/p85α dimer. By affecting the dynamics of the protein, these mutations favor the conformations that reduce the autoinhibitory effect of the p85α nSH2 domain. For example, we determined that in all the mutants, the nSH2 domain exhibits increased positional heterogeneity compared to the wild type (WT), as evidenced by changes in the fluctuation profiles computed by normal mode analysis (NMA) of coarse-grained elastic network models (ENM). Analysis of the inter-domain interactions of the WT and mutants at the p110α/p85α interface obtained using molecular dynamics (MD) simulations suggest that all the tumor-associated mutations effectively weaken the interactions between the p110α and the p85α subunits by disrupting key stabilizing interactions. These findings have important implications for understanding how oncogenic mutations change the conformational multiplicity of PI3Kα and lead to increased enzymatic activity. This mechanism may apply to other enzymes and/or macromolecular complexes that play a key role in cell signaling.

Graphical abstract

Introduction

Phosphatidylinositol 4,5 bisphosphate 3-kinases (PI3Ks), are a family of lipid kinases that initiate signaling cascades that control a variety of cellular functions including cell growth, proliferation, motility, survival and intracellular trafficking [1]. These enzymes catalyze a simple but important reaction: they phosphorylate the 3-hydroxyl position of the inositol ring of phosphatidylinositol-4,5-bisphosphate (PI(4,5)P₂, or PIP₂) to yield phosphatidylinositol-3,4,5-triphosphate (PIP₃) [2]. PIP₃ is recognized by proteins containing a pleckstrin homology (PH) domain, which are recruited to the cell membrane where they initiate important signaling cascades.

PI3K enzymes belong to different classes that differ in their structure and function [3]. (For a complete description of the PI3K enzymes classification see [4].) PI3Kα, a class IA isoform, is a heterodimer consisting of a catalytic subunit p110α and a regulatory subunit p85α that binds receptor tyrosine kinases (RTK) or other substrates that activate PI3K. Each PI3Kα subunit is composed of five domains (Fig. 1A and 1B). The p110α subunit comprises an adaptor binding domain (ABD), a Ras-binding domain (RBD), a C2 domain, a helical domain, and a kinase domain. The last four domains have significant sequence homology among isoforms. The p85α subunit contains an Src homology 3 (SH3) domain, a GTPase-activating protein (GAP-like or BH) domain and two SH2 domains (nSH2 and cSH2) separated by an inter-SH2 domain (iSH2). PI3Kα forms an obligate heterodimer in which the p110α subunit, which by itself is unstable in cells, is stabilized upon dimerization with p85α [5]. However, the nSH2 domain of the p85α subunit inhibits the basal kinase activity of p110α, suggesting that the activation mechanism of PI3Kα includes a relief of this autoinhibition.

Figure 1. Overview of the p110α/niSH2 heterodimer.

(A) Ribbon diagram of the p110α/niSH2 heterodimer (front view and side view). Arrows indicate the position of the studied mutations, which are shown in sphere representation. (B) Scheme of the domain organization. The same color-coding is used throughout this paper unless specified. Gray regions are linkers between domains. (C) Helical-nSH2 domains interface with the E542R-E545R mutation was modeled. The WT is shown in gray. (D) ABD-kinase domains interface where the R38C-R88Q mutation was modeled. The WT is shown in gray. (E) C2-iSH2 domains interface where the N345K mutation was modeled. The WT is shown in gray.

Physiological activation of PI3Kα is triggered by binding of phosphorylated tyrosine kinase receptors RTK or their accessory proteins, such as the insulin receptor substrate 1, IRS-1 [3,6], that bridge the interaction between RTK and PI3Kα. These phosphorylated activators, by dislodging the nSH2 domain from its inhibitory interaction with the helical domain, relieve the nSH2’s partial inhibition of the PI3Kα activity [6].

PIK3CA, the gene that encodes the p110α subunit, has been found to be mutated in 12% of all tumor sequences deposited in the catalog of mutations in cancers [7]. Mutations of the PIK3CA gene in human cancers [8,9] are associated with increased enzymatic PI3Kα activity [10,11]. Cancer associated mutations [12–14] have been found throughout the sequence of p110α [15,16]. However, “hot-spot” mutations are located mostly on the helical (residues E542 and E545) and the kinase domain (residue H1047). For example, of the 7,548 unique mutations cataloged at the COSMIC database 1687 (22%) correspond to the H1047 residue, 1153 (15%) to the E545 residue and 758 (10%) to the E542 residue¹. These mutations are known to activate PI3Kα by different mechanisms [16–18]. Mutations of E542 and E545 are located in the loop of the helical domain that contacts the nSH2 domain [16]. It has been suggested that in the WT the interactions between the nSH2 and the helical domains are such that they “lock” PI3K in its inactive conformation, and that oncogenic mutations activate PI3Kα by weakening the interaction between the helical and the nSH2 domains [10,11,16,18]. In contrast, the structure of the H1047R mutant [16] shows that R1047 adopts a conformation perpendicular to the orientation of H1047 in the WT enzyme and two loops of the kinase domain contact the cell membrane in the oncogenic mutant. Biochemical assays revealed that the enzymatic activity of the p110α H1047R mutant is more sensitive than the WT to the lipid membrane composition [16]. These findings suggest that the functional effect of the H1047R mutation is due to the change in the interaction between PI3Kα and the cellular membrane [19,20], increasing the lipid kinase activity by allowing easier access to the membrane-bound PIP₂ substrate [16]. Other common mutations, also associated with enzyme activation, are located in the ABD and C2 domains [7], and their activation mechanism remains unknown [18,21,22].

Currently, several PI3K inhibitors are under development as possible cancer therapeutic agents [23–25]. However, as different PI3K isoforms have different biological functions, an effective PI3K inhibitor has to be isoform selective to avoid side effects [26–28]. Furthermore, depending on the activation mechanism of the associated oncogenic mutation, it may be necessary to design mutant specific inhibitors. Understanding the functional effects of somatic mutations in the activation and the activity of PI3K is necessary to guide the design of these inhibitors.

A recent experimental study of the dynamics of WT and cancer-linked mutants of the p110α/p85α complex revealed important information for understanding the dynamical effects of common oncogenic mutations. Using hydrogen/deuterium exchange mass spectroscopy (HDX-MS), Williams and co-workers determined that four distinct conformational events are associated with the activation of PI3Kα [22]: 1) disruption of the nSH2–helical interface, 2) disruption of the iSH2–C2 interface, 3) movement of the ABD domain relative to the kinase domain, and 4) interaction of the kinase domain with lipid. One or more of these four distinct events, detected in the activation of WT p110α/p85α, are thought to also activate the enzyme in cancer related mutants. However, the HDX-MS studies can determine neither the order of these events, nor whether they are spatially or temporally correlated.

The molecular basis by which the structural and dynamic effects of mutations in different domains of PI3Kα are propagated throughout the structure and affect the activity cannot be directly inferred from structural and/or HDX-MS studies. Based on available evidence we hypothesize that PI3Kα becomes activated through dynamic changes in the relative conformations of the two subunits. If this is the case, the allosteric effects of PI3Kα mutations far from the active site that increase the enzymatic activity are associated with changes in the large-scale dynamic behavior of the protein: changes in the thermal fluctuation would facilitate allosteric changes in the kinase domain without large changes in the average structure of the other domains [29–32]. To probe this mechanism we used computational methods to characterize changes in the inter-domain interactions and their effects on the dynamics of the WT and mutant PI3K. We explored the effects of five different oncogenic mutations to elucidate the molecular mechanism by which oncogenic mutations lead to enzyme activation. This comprehensive computational investigation addresses the flexibility and dynamics of WT and mutant PI3Kα, and provides the basis for understanding how these properties affect PI3K activation. Normal-mode analysis (NMA) using coarse-grained elastic network models (ENM) and molecular dynamics (MD) simulations allowed assessing the conformational fluctuations of the system over a range of timescales. With these results, we propose a model for the activation of PI3Kα by oncogenic mutations that involves allosteric communication between different parts of the proteins mediated by the destabilization of the p110α/p85α dimer.

Results

Biochemical and cellular studies have shown that the oncogenic mutations studied here are associated with gain of enzymatic activity of PI3Kα [10,17]. It is also known that mutations in different p110α domains may activate PI3Kα by different mechanisms [16,17]. Nevertheless, these mutations have a common characteristic: they are located far from the catalytic site at the interface between different PI3Kα domains (Fig. 1A). For example, the helical domain mutations E542K and E545K are located at the interface between the helical of p110α and the nSH2 domain of p85α. These mutations produce the displacement of one of the nSH2 domains β-sheet (residues 377 to 383), changing the interaction between the nSH2 domain and the helical domain (Fig. 1C). The ABD mutations R38 and R88, which are located at the interface of the ABD domain and the kinase domain (Fig. 1D) disrupt the interactions of residues R38 and R88 with residues D743 and D746 in the kinase domain. The C2 N345K mutation, located at the interface of the C2 and iSH2 domains (Fig. 1E), changes the conformation of the loop region where this residue is located. Overall, our results show that the analyzed mutations induce local perturbations at interfaces between PI3Kα domains. These models are in agreement with the experimental evidence that shows that oncogenic mutations results in local structural changes [16], which we propose change the dynamics of the protein and result in the release of auto inhibition.

Oncogenic mutations increase the fluctuations of the nSH2 domain

The effects of the oncogenic mutations on the positional fluctuations of the different domains were studied by computing the elastic network normal modes of the WT and the mutant proteins. Elastic network models are a powerful tool to study the inherent flexibility of the protein, from which conformational transitions can be inferred by identifying and characterizing the collective motions described by vibrational modes [33]. Fig. 2 shows the characteristic fluctuation profile of every PI3Kα domain. ENM analysis of the WT and mutant PI3Kαs shows that residue fluctuations and overall collective motions of the protein are affected by point mutations in the ABD, C2 and helical domain. Remarkably, all these mutations significantly affect the fluctuation profile of the nSH2 domain of the p85α subunit. Specifically, for all mutants, the nSH2 domain showed significantly larger amplitude of movement (i.e. up to 3-fold increase of the average fluctuations) than the WT model (Fig. 2F), particularly in the region spanning residues 377 to 410. This segment includes the β-sheet and α-helix that face the interface between the nSH2 domain and the helical domain (Fig. 1C). In the case of the N345K mutant (C2 domain), the ABD and iSH2 domains become more rigid, particularly residues 506–524 of the iSH2 domain (Fig. 2G). These residues correspond to the coil region in the coiled-coil turn at the interface between the ABD and iSH2 domain. No significant changes in the fluctuation profiles were observed for the C2 (Fig. 2C), helical (Fig. 2D) and kinase (Fig. 2E) domains. However, dynamical effects that are not detected by ENM may be present. These results show that, regardless of where the mutation is located, they all ultimately result in releasing the auto inhibitory effect of the nSH2 domain. The molecular mechanism of how such effect is transmitted along the protein was investigated using MD simulations.

Figure 2. Fluctuations of the p110α and p85α domains.

The fluctuation profiles were obtained by generating random linear combinations of the first six vibrational modes, shown in arbitrary units (a.u.). Fluctuations of the WT protein are in grey in all plots for comparison. (A) Fluctuation of the ABD domain. (B) Fluctuations of the RBD domain. (C) Fluctuation of the C2 domain. (D) Fluctuations of the helical domain. (E) Fluctuations of the kinase domain. (F) Fluctuation of the nSH2 domain. (G) Fluctuations of the iSH2 domain. Shaded areas correspond to one standard deviation on the fluctuations.

Oncogenic mutations disrupt the interactions at the p110α-p85α interface

As already mentioned, all the oncogenic mutations studied here are located at interfaces between domains. We studied the dynamical effects of these mutations by analyzing the residue-to-residue distances at the interfaces between the helical and nSH2, C2 and iSH2 and ABD and kinase domains in MD trajectories (see Methods section). All mutations significantly change how the nSH2 and helical domains interact (Fig. 3). For example, the interaction between residues E542 (helical domain) and L380 (nSH2 domain) is conserved in all cases, even when residue 542 is mutated to lysine. On the other hand, the interaction between residues N542 (helical domain) and S361 (nSH2 domain) is lost in all mutants. The distance distribution for this interaction has two well-defined peaks, which correspond to direct protein-protein interactions and water-mediated interactions. The E542K-E545K mutation also disrupts the interaction between residues E545 (helical domain) and K379 (nSH2 domain). Additionally, mutants N345K and R38C-R88C significantly change the distance distribution for helical domain residues K379, I381 and K382. These results support the notion that even distant mutations (R38C-R88Q and N345K) effectively weaken the interaction between the nSH2 and helical domain, thus increasing the population of molecules in which the nSH2 domain is not inhibitory.

Figure 3. Stabilizing interactions at the helical-nSH2 domain interface.

Probability distribution of the distance between residues along the MD trajectories (see Methods section). Different mutants: WT (grey), E542K-E545K (orange), N345K (blue) and R38C-R88Q (green). In panels with a star the distance involves mutated residues.

We also investigated the interaction between the C2 and the iSH2 domains. In the WT, residue N345 (C2 domain) is within hydrogen bonding distance of residues E560 and E564 (iSH2 domain). These interactions are lost in the N345K mutant (Fig. 4). Surprisingly, these interactions are also lost in the structures with mutations in the helical and ABD domains (Fig. 4). The fact that the E542K-E545K mutation (helical domain) has a significant effect in the C2-iSH2 interface suggests that weakening the helical-nSH2 interface affects the overall dynamics of the subunit interaction in the PI3K heterodimer, that result in significant effects far from the mutation sites. The R38C-R88Q mutations (ABD domain) also significantly disrupt the C2-iSH2 interface, for which we monitor the H450 to Y467 contact. All our simulations show that residues H450 (C2 domain) and Y467 (iSH2 domain) are ~3.4 Å apart (Fig. 4). H450 is located in a loop region of the C2 domain and interacts via stacking interactions with residue Y467 of the iSH2 domain. This interaction is more prominent in the E542K-E545K (helical domain) and R38C-R88Q (ABD domain) mutants. Previous studies have suggested that the disruption of the C2-iSH2 domain leads to enzyme activation [34]. These structural changes have also been detected in hydrogen/deuterium exchange mass spectroscopy (HDX-MS) experiments, which suggest that the disruption of this interface may occur upon membrane binding as part of the enzyme’s catalytic cycle [19]. We also evaluated the interface between the ABD and kinase domains. In the WT, residues R38 and R88 are at hydrogen bonding distance from residues D743 and D746, which are located in consecutive turns of the helices of the kinase domain. Upon mutation of residues R38 and R88 these interactions are disturbed (Fig. 5), and may affect the relation between the ABD and iSH2 domains and the p110α core, as suggested by [35]. The other mutants do not exhibit significant changes in these interactions (Fig. 5). Consequently, the R38C-R88Q mutation disrupts all three interfaces considered, while the E542K-E545K and N345K mutants only disrupt the helical-nSH2 and C2-iSH2 interfaces. This suggests that the localized structural and dynamical changes induced by the R38C-R88Q mutation are transmitted to the nSH2 domain through the rigid coiled-coil iSH2 domain. These changes reduce the auto inhibitory interaction between nSH2 and the P110 subunit.

Figure 4. Stabilizing interactions at the C2-iSH2 domain interface.

Probability distribution of the distance between residues along the MD trajectories (see Methods section). Different mutants: WT (grey), E542K-E545K (orange), N345K (blue) and R38C-R88Q (green). In panels with a star the distance involves mutated residues.

Figure 5. Stabilizing interactions at the ABD-kinase domain interface.

Different mutants: WT (grey), E542K-E545K (orange), N345K (blue) and R38C-R88Q (green). In panels with a star the distance involves mutated residues.

Oncogenic mutations increase the conformational heterogeneity of the p85α subunit

To further characterize the dynamical effects of the studied oncogenic mutations, we computed the Solvent Accessible Surface Area (SASA) for all residues of the p85α subunit, as a function of time. First, we identified all the nSH2 residues that are buried in the WT by defining a cutoff of nSASA<0.2 (normalized SASA; see Methods section). We determined that those residues become more exposed upon mutations on the helical (E542K-E545K) and C2 (N345K) domains (Fig. 6A), as shown by the increased probability of higher nSASA. Similarly, comparison of the nSASA of buried residues in the iSH2 domain shows that these residues become more exposed to solvent in all the mutants (Fig. 6B).

Figure 6. nSASA of the nSH2 and iSH2 residues.

A) nSASA probability distribution for residues at the interface between the helical and nSH2 domains. B) Ribbon representation of the helical-p85α interface. Residues shown as sticks are those for which we measured large changes in the nSASA with respect to the WT. C) nSASA probability distribution for residues at the interface between the p110α subunit and the iSH2 domain. D) Ribbon representation of the p110α-iSH2 interface. Residues shown as yellow sticks are those for which we measured large changes in the nSASA with respect to the WT. We used the same domain color-coding as Fig. 1.

Among the nSH2 domain residues that become more exposed upon mutation are residues 344 to 346, located in the helix at the interface of the C2, iSH2 and nSH2 domains. The nSASAs of these residues show large variations over time, especially in the N345K C2 mutant. Residue 369, located in the interface between the helical and nSH2 and facing the loop where the E542K-E545K mutation is located, becomes slightly more exposed in the helical domain mutant, but not in the other mutants. The nSASA of residues 414–420, located in a loop facing the helical domain, also shows large fluctuations, suggesting that the nSH2 domain exhibits larger fluctuations in the mutants.

In the iSH2 domain, residues 481, 564 and 568, located at the C2-iSH2 interface, have on average larger nSASA in the mutants than in the WT (data not shown). Residues 574–577 and 447–454, present in the helices of the coiled-coil iSH2 domain facing the nSH2 domain, show either larger nSASA, or larger fluctuations in the nSASA in the mutants with respect to the WT. Similarly, residues 505 and 523–526, found at the interface between the iSH2 domain and the ABD domain, also exhibit larger fluctuations in the mutants. These results suggest that mutations in all three interfaces studied have a significant effect on the dynamics of the iSH2 domain.

Oncogenic mutations lead to PI3Kα activation by releasing the p85α nSH2 inhibitory effect

Fig. 7 shows the model proposed for the effect of the studied oncogenic mutations and how they contribute the PI3Kα’s activations [18]. In the WT the p85α nSH2 domain has an auto-inhibitory effect on the PI3Kα activity (Fig. 7A). Studied mutations E542K-E545K, R38C-R88Q and N345K are located at interfaces between the helical and nSH2, C2 and iSH2 and ABD and kinase domains (Fig. 7B). Mutations E542K and E545K are located in the loop region of the helical domain that contacts the nSH2 domain (Fig. 7B). Structural studies have shown that these residues interact with residues K379 and R340 of nSH2 domain [16] “locking” the inhibited conformation of PI3Kα. Therefore, to bind the phosphorylated tyrosine-containing loop of the effector protein, the nSH2 domain must spend part of the time away from its inhibitory position (Fig. 7C). The C2 and ABD domain mutations increase this effect by weakening the interaction between the helical and nSH2 domains and allowing the nSH2 spend a larger fraction of the time away from the nSH2 domain, resulting in PI3Kα activation. These results were observed in the ENM and MD simulations: ENM shows that the mutants’ nSH2 domains exhibit larger fluctuations, suggesting an increased in positional heterogeneity; MD simulations show that all the mutations weaken the inter-domain interactions at the helical and nSH2 interface, where residues also exhibit larger fluctuations in the nSASA.

Figure 7. Schematic representation of the mechanism of activation by oncogenic mutations at domain interfaces.

A) The physiological activation of PI3Kα represents the reversal of the autoinhibition by the nSH2 domain. B) The studied oncogenic mutations are located at the helical-nSH2, C2-iSH2 and ABD-kinase domain interfaces. C) All studied mutations effectively weaken the helical-nSH2 and the C2-iSH2 domain interfaces [18] D) Schematic free energy profile that exemplifies that, in the WT, the p110α nSH2 has a small set well defined conformations (single well). In these conformations the nSH2 inhibits the activity of PI3Kα. E) Oncogenic mutations change the dynamics of the nSH2 domain. In these cases, the average conformation is the same, but now the nSH2 domain exhibits a larger positional heterogeneity. As a result, the inhibitory conformation is less populated.

We observed, in addition, that the mutations examined have significant effects on the domain-domain interactions at the C2-iSH2 interface, and that for all mutants the iSH2 domain has an increased positional heterogeneity (Fig. 7D). This is reflected in the increase of nSASA fluctuations showed by residues located at the interface between p110α and p85α subunit. The ABD-kinase interface is only weakened by mutations in the ABD domain. However, the nSASA of residues at the interface between the ABD-kinase-iSH2 domains, also show larger fluctuations in the mutants than in the WT. These results support the proposed mechanism that, given the rigidity of the coiled-coil, the iSH2 domain acts as a rod that transmits the conformational and dynamical effects of the different mutations throughout the protein [18]. The proposed mechanism explains how mutations that are so far away from the nSH2 domain, can still act by relieving the nSH2 inhibition. Moreover, this mechanism exemplifies how the activation of PI3Kα corresponds to allosteric activation: changes in the stability between different components of the system induce structural rearrangements [29,30,36]. Here, changes in protein dynamics, especially due to the destabilization of the p110α/p85α interactions, result in allosteric communication between distant sites of the protein that increase the population of the kinase active conformations [37–39].

In its resting state, PI3Kα is autoinhibited by the interaction of a loop of the helical domain of p110α with a groove in the nSH2 domain of p85α [40,41]. Despite this inhibition, the enzyme has a significant basal activity [10]. Activation of PI3Kα by its physiological effectors—the phosphotyrosine (pY) residue of an activated receptor tyrosine kinase (RTK) or of an RTK substrate—results in a modest 2 to 4 fold increase in the rate of the reaction [10,40]. Activation is the result of pY binding to the same groove of the nSH2 domain as the helical domain loop, dislodging the nSH2 domain from its inhibitory position. For the pY to have access to bind to the nSH2 groove, the groove must be empty at least part of the time, indicating a dynamic interaction. The fraction of the time that the nSH2 spends away from its inhibitory position may account for the PI3Kα basal activity. All these considerations point to a highly dynamic autoinhibited ground state that can access through fluctuations non-inhibited states of highly similar energies.

To date, the coordinates of eleven structures of large PI3Kα fragments have been deposited in the Protein Data Bank (PDB) [15,16,40–43], including five reported by us [15,16,41]. The proteins were crystallized under different conditions in several different crystal forms with different crystal contacts. Nevertheless, the rms deviation for over 870 aligned alpha carbons varies only between 0.55 and 0.98 Å. In some of the structures the density for the nSH2 domain was reported to be weak, indicating positional disorder for this p85α domain. In addition, some of the largest deviations were observed at the ends of the iSH2 coiled-coil. These observations can be interpreted as indicating that the nSH2 and the ends of the iSH2 are conformationally less constrained than the p110 domains. Nevertheless, one area of the p110 shows conformational heterogeneity: in most of the structures no density is observed for the activation loop (residues 933–957 of p110). When it is observed in the structures of these and other kinases, it adopts different conformations [40,41]. This is particularly important because a particular conformation of this loop is one of the determinants of activation of the enzyme [37,41]. These local diverse conformations—those of the activation loop and those reflecting weak interactions between p110 and p85—do not amount to a defined global conformational change but rather to dynamically accessible local conformations [44]. We propose that oncogenic mutations in the ABD and the C2 exert their activation effect by weakening the interactions that modulate the dynamics of the global motions of the nSH2 and the iSH2 domains, including the position of the nSH2 relative to the rest of the molecule. Mutations at positions 542 and 545 do this directly by disrupting the quaternary interaction between the nSH2 and helical domains [40]. For example, if for simplicity we assume that the energetics of the motions of the nSH2 along a coordinate that changes its interaction with the helical domain are approximated by a harmonic potential

$$P(x)=\mathit{exp}-[\mathit{1}/\mathit{2}·k·{(x-x_o)}^{\mathit{2}}]$$

where P(x) is the probability of adopting the position x away from the equilibrium position x_o and k is the force restoring constant, we can estimate the effect of weakening the interactions of the nSH2 with p110. For a state that is, for example, 1.5 Šfrom the ground state, if the value of an originally soft constant of 2 Kcal/Ų is change to a softer constant (1 Kcal/Ų) the probability of the state increases by a factor of over 3 without a change in the structure of the ground state (Fig. 7D, 7E). A simple weakening of the interaction between p110 and p85 can by itself explain the levels of activation observed in PI3K.

Visualization of the active site shed light of the effects of these changes in the activation of PI3Kα. For example, Fig. 8B and 8C show the conformation of the kinase activation loop of the WT and that of the E542K-E545K mutant. Fig. 8A shows that the activation loop in the WT MD simulations has a conformation that resembles that of the crystal structure, which is locked in the inactive conformation by interactions between the activation loop and iSH2 domain. In contrast, in the simulation of the E542K-E545K mutant, the interaction between the iSH2 domain and the activation loop is disrupted, displacing the activation loop from the iSH2 domain. Further research should address how these changes relate to the active conformation of the activation loop.

Figure 8. Structural insights into the active site.

A) Model of the active site based on the alignment of the crystal structures of PI3Kα in complex with its lipid substrate, diC4-PIP₂ (PDB ID 4OVV) and the p110α-ATP complex (PDB ID 1E8X). In the active site, the lipid substrate (yellow) binding site is adjacent to the ATP-binding site, which is located between the P-loop (residues 772–778 of p110α), the activation loop (residues 933–957 of p110α) and the p85α iSH2 domain. In this model, the activation loop is locked into the inactive conformation by a stacking interaction between F945 (p110α) and L598 (p85α) and interactions between K948 (p110α) and Q591 (p85α). B) Most representative structure obtained from the WT MD simulations (colored). The model described in A is shown in grey. In this conformation, the activation loop adopts a similar structure that the one shown in A, locked in the inactive conformations. C) Most representative structure obtained from the E542K-E545K mutant MD simulations (colored). The model described in A is shown in grey. This conformation shows how in the mutant the interactions between the iSH2 domain and the activation loop are disrupted, and the activation loop is displaced from the iSH2 domain.

Discussion

We investigated the effects of five p110α oncogenic mutations on the conformational heterogeneity and dynamics of PI3Kα using different computational methods. Experimental evidence has linked all these oncogenic mutations with the gain of enzymatic activity. We explored the hypothesis that the gain of enzymatic activity is brought about by the local increase in the populations of PI3Kα with larger separation in the inhibitory interface between the p110α helical domain and the p85α nSH2 domain. As a consequence, the relative occupancies of the active and inactive conformations of the kinase domain are shifted, making the activated conformation more populated in the mutated PI3Kα. Overall, we observe that all oncogenic mutations weaken the p110α-p85α interactions, both at the nSH2 and iSH2 interfaces, which results on a higher positional heterogeneity of the nSH2 domain. This was evidenced in ENM calculations by the changes in the fluctuation profiles of the nSH2 domain of the mutants compared to those of the WT. Closer examination of these profiles reveals that in all studied mutations the nSH2 region that faces the helical domain exhibits increased positional heterogeneity. Furthermore, by analyzing the changes in the surface accessible area obtained from MD simulations, we determined that residues located at the interface between the nSH2 and helical domains become more exposed in the mutants, in agreement with the HDX-MS data [22]. Most importantly, these changes were observed even when the mutations were located distant from the nSH2-helical interface. Finally, analysis of the molecular contacts between residues at the interface between the p110α and p85α subunits shows the different effects of each of the studied oncogenic mutants. For example, some of the stabilizing contacts between the helical and nSH2 domains and the C2 and iSH2 domains are lost in all the mutants, suggesting that the overall effect of the oncogenic mutations is to increase the conformational heterogeneity of the PI3K dimer. The stabilizing contacts between the ABD and iSH2 domains were only weaken in the R38C-R88Q mutant.

In summary, our studies shed light on the mechanism by which some oncogenic mutations found in different PI3Kα domains may lead to constitutively active enzymes. As described, oncogenic mutations at the allosteric sites destabilize the p110α/p85α dimer, shifting the relative population of nSH2 domain from being mostly in its auto-inhibitory conformation to an ensemble of conformation with higher conformational heterogeneity, in which the non-inhibited conformations are more highly populated. This allosteric change in the conformation the nSH2 domain changes the relative stability of the active and inactive conformations of the kinase domain, making the active conformation more accessible. It remains to be determined how these dynamical changes affect the conformational heterogeneity of the activation and catalytic loops in the kinase domain and, how this is linked to constitutive activation of PI3Kα. We propose that the observed dynamical changes shift the relative occupancy of the active and inactive conformations of the kinase domain, making the active conformation more populated in the mutant PI3Kαs. This system exemplifies a case of allostery, where changes in the thermal fluctuations due to destabilization of its dimeric state result in an increased enzymatic activity. This mechanism may apply to other enzymes and/or macromolecular complexes that play a key role in cell signaling [39,45].

From a modeling perspective, we have shown that, when studying large multimeric proteins or macromolecular complexes, different computational approaches have different strengths and that a comprehensive approach can provide new insights into answering biologically relevant questions. We used ENM to obtain significant dynamic information by focusing in the general movements present in a subset (i.e. a low frequency subset) of normal modes rather than characterizing individual modes. This was accomplished by generating random linear combinations of displacement vectors of atoms to study the flexibility of different domains of the protein. Even though the results of these combinations not necessarily represent the actual movements of the protein, they are useful tools for assessing the flexibility of different regions of the molecule. MD simulations, on the other hand, provided detailed information about the molecular contacts that define the stability of the p110α/p85α dimer.

Materials and Methods

Initial Models

The crystallographic structures of the WT p110α in complex with the iSH2 domain of p85α [15] (pdb accession code 2RD0) and the H1047R mutant of p110α in complex with niSH2 domain of p85α [16] (pdb accession code 3HIZ) were used to build the different models. Related structures of the free WT and in complex with the lipid substrate have been recently published (pdb accession code 4OVU) [41]. All crystal structures lack some loops regions which were built using the loop-building option of MODELLER 9v8 [46]. One loop of critical importance that needed to be modeled was the activation loop (p110α residues 933 to 957). This loop was modeled in the inactive conformation using as a template the activation loop of the PI3Kβ isoform [47]. We modeled the wild type and five mutants of p110α in complex with the nSH2 and iSH2 domains of p85α. Starting with the WT model, each mutant was modeled in silico using MODELLER 9v8 [46]. We studied the helical domain mutations E542R/E545R and E542K/E545K, the ABD domain mutations R38C/R88Q and R38D/R88D and the C2 domain N345K mutation. In the models of the helical and ABD mutants we combined both commonly observed mutations to enhance the structural and dynamical effects observed by computational methods.

The potential energy of each of these structures was minimized using the program NAMD with the CHARMM27/CMAP force field [48,49], using the conjugate gradient and adapted basis Newton Raphson method (ABNR) algorithms. After minimization, each of these models was used for normal modes analysis (NMA) and molecular dynamics (MD) simulations.

Elastic network models

Internal motions of proteins are important for their biological function, especially large amplitude (and/or low frequency) motions that may be necessary for activation or for allosteric transitions. Normal mode analysis (NMA) has been shown to be a successful and direct method for identifying these large amplitude motions [50,51]. In general terms, NMA is the study of coupled harmonic potential wells by analytical means [33,52]. In particular, we focused in the normal modes analysis of coarse-grained elastic network models (ENM) of the WT and mutant PI3Kαs. Low-frequency modes obtained by this method have been shown to correspond to the wide conformational fluctuations observed experimentally around a given stable conformation [53–55].

Starting from a stable structure of a protein, assumed to be a minimum of the potential energy surface, a harmonic approximation is used to calculate vibrations around this minimum. In this model each residue of the protein is represented by the Cα and the springs connecting the nodes represent the bonded and non-bonded interactions between the pairs of residues that are within a distance range defined by a cutoff r_c. Furthermore, in this representation it is assumed that the fluctuations are isotropic and Gaussian, such that the conformational potential, for a system of N residues, is given by:

$$V=\frac{\gamma }{2}\left[{\sum }_{i,j}^N[{(\mathrm{\Delta }{X}_i-\mathrm{\Delta }{X}_j)}^2+{(\mathrm{\Delta }{Y}_i-\mathrm{\Delta }{Y}_j)}^2]+{(\mathrm{\Delta }{Z}_i-\mathrm{\Delta }{Z}_j)}^2\right]$$ (1)

where γ is a single force constant and ΔX_i, ΔY_i and ΔZ_i are the N-dimensional vectors whose elements are the fluctuations of residue i. To perform the NMA it is necessary to diagonalize the Hessian matrix, whose elements are the second derivative of the potential energy with respect to the coordinates (Eq. 1). As a result, one obtains a set of orthonormal basis vectors that describes the N-dimensional configurational space of the protein.

Structural Analysis of the Normal Modes

To characterize the dynamical effects of the mutations, we analyzed the eigenvectors obtained by NMA for every PI3Kα model. Every eigenvector ΔR_k(i) is normalized such that:

$${\sum }_{k=1}^N\mathrm{\Delta }{R}_k{(i)}^2=1,$$ (2)

where k indicates the normal mode and the i identifies the residue in the structure. The set the N-1 eigenvectors form an orthogonal basis (ΔR_K·ΔR_l = 0 for k≠ l), that can be used to model the configurational space of the protein near the initial conformation. Previous studies have shown that functional movements associated with conformational changes are encoded in low-frequencies modes [50,51,56].

We investigated the fluctuations of each of the protein’s domains by analyzing the information encoded in the eigenvectors. For example, for residues i_n to i_m the sum $s={\sum }_{i=i_n}^{i_m}\mathrm{\Delta }{R}_k(i)$ can be interpreted as the contributions of the segment between residues i to i_m to the motion described the k-th eigenvector.

Since comparing individual normal modes may not provide straightforward information we generated random linear combinations of the first six vibrational normal modes for different domains. We computed the fluctuations vector (q) by, first, extracting from the k-th eigenvector the portion of the normal mode that corresponds to the domain (d) being studied ( $\mathrm{\Delta }{R}_k^d$) and then generating the linear combination as:

$$q={\sum }_{k=1}^6{\alpha }_k\mathrm{\Delta }{R}_k^d$$ (3)

where the coefficient α correspond to random numbers that satisfy the condition Σ_kα_k = 1. We determined that 500 linear combinations ( $q_{\text{final}}={\sum }_{m=1}^{500}{q}_m$) are sufficient to obtain convergent fluctuation profiles for the protein’s domains. We compared fluctuation profiles of the WT with those of the mutants.

Molecular dynamics simulations

Explicit solvent simulations were carried out using the NAMD 2.9 software [57], the CHARMM27/CMAP force field [58,59] and the TIP3P water model. The initial models were solvated in a box of water molecules of dimension 125 Å × 117 Å × 102 Å. The system was minimized using, first, the conjugate gradient and adapted basis Newton Raphson method (ABNR) methods, for 4000 steps in each case. The system was then gradually heated from 100 to 300 K over a 4 ns simulation, followed equilibration at 300 K (for 2 ns) in the NPT ensemble. Periodic boundary conditions were used in all our calculations, and long-range electrostatics were treated with the particle mesh Ewald method [60]. The cutoff distance for non-bonded Coulomb and Lennard-Jones interactions was set to 12 Å cutoff, with a switching function at 10 Å. Simulations were carried out with an integration time step of 2 fs using the SETTLE algorithm, while keeping all bonds to hydrogen atoms rigid. Production simulations were performed at a constant temperature (of 310 K) and pressure by using the Langevin dynamics. Production runs were 50 ns long. For analysis we considered the last 40 ns of each production run.

Structural analysis of the MD simulations

To identify inter-domain contacts at the interface between subunits, two residues were considered to be in contact if the distance between their Cβs (Cα for Gly) was shorter than 9 Å. The dynamics of each interaction was estimated by recording the smallest distance between heavy atoms within every 20 ps time frame along each trajectory. The Solvent accessible surface area (SASA) for all residues of the p85α subunit was calculated using the NACCESS program [61], which was modified to read different frames of the trajectory and measure de SASA per residue as function of simulation time. A probe of 1.4 Å was used. We defined the normalized SASA (nSASA) as the measured SASA divided by the maximum possible SASA for each residue [62], such that the nSASA ranges between 0 and 1.

Abbreviations

ENM

elastic network model

NMA

normal modes analysis

MD

molecular dynamics

SASA

solvent accessible surface area