ReviewStructural Biology

Structural Basis for Activation and Inhibition of Class I Phosphoinositide 3-Kinases

See allHide authors and affiliations

Science Signaling  18 Oct 2011:
Vol. 4, Issue 195, pp. re2
DOI: 10.1126/scisignal.2002165

Abstract

Phosphoinositide 3-kinases (PI3Ks) are implicated in a broad spectrum of cellular activities, such as growth, proliferation, differentiation, migration, and metabolism. Activation of class I PI3Ks by mutation or overexpression correlates with the development and maintenance of various human cancers. These PI3Ks are heterodimers, and the activity of the catalytic subunits is tightly controlled by the associated regulatory subunits. Although the same p85 regulatory subunits associate with all class IA PI3Ks, the functional outcome depends on the isotype of the catalytic subunit. New PI3K partners that affect the signaling by the PI3K heterodimers have been uncovered, including phosphate and tensin homolog (PTEN), cyclic adenosine monophosphate–dependent protein kinase (PKA), and nonstructural protein 1. Interactions with PI3K regulators modulate the intrinsic membrane affinity and either the rate of phosphoryl transfer or product release. Crystal structures for the class I and class III PI3Ks in complexes with associated regulators and inhibitors have contributed to developing isoform-specific inhibitors and have shed light on the numerous regulatory mechanisms controlling PI3K activation and inhibition.

Introduction

Phosphoinositide 3-kinases (PI3Ks) are lipid kinases that phosphorylate the D3 hydroxyl on the inositol ring of phosphatidylinositol (PI) and its phosphorylated derivatives phosphatidylinositol 4,5-bisphosphate (PIP2) and phosphatidylinositol 4-monophosphate (PI4P) (1). PI3Ks receive inputs from activated receptor tyrosine kinases and heterotrimeric guanine nucleotide–binding protein (G protein)–coupled receptors (GPCRs) and produce lipid second messengers that affect cell proliferation, growth, metabolism, motility, and intracellular trafficking (Fig. 1). On the basis of sequence similarity of the catalytic subunits and lipid substrate preference, the PI3Ks can be grouped in three distinct classes. The best characterized is the class I PI3Ks, which are heterodimers composed of a catalytic (p110) and a regulatory subunit that produces the second messenger, phosphatidylinositol 3,4,5-trisphosphate (PIP3). This class is further subdivided into class IA (p110α, p110β, and p110δ) and class IB (p110γ) on the basis of their different regulatory subunits. The catalytic subunits share a common domain organization composed of an N-terminal adaptor binding domain (ABD), a Ras binding domain (RBD), a C2 domain (C2), a helical domain, and a C-terminal kinase domain with a bilobal fold (consisting of N and C lobes), which shares some similarities with the catalytic domain of protein kinases (Fig. 2) (2). Class IA catalytic subunits are activated by and associate with p85-like regulatory subunits, of which there are five variants (p85α, p55α, p50α, p85β, and p55γ). The p85-like regulatory subunits contain two Src homology 2 domains, nSH2 and cSH2, separated by an intervening coiled-coiled domain (iSH2) that mediates binding to p110. In p85α and p85β, these domains are preceded by an SH3 domain, a Bar cluster region homology domain (BH), and two proline-rich regions (Fig. 2). The class IB p110γ associates with either p84 or p101 regulatory subunits, which do not have recognizable domains, and is activated exclusively downstream of GPCRs. The class IA p110β is also activated downstream of GPCRs through Gβγ heterodimers (Fig. 1) (35).

Fig. 1

Class I PI3K signaling pathway and downstream effects on cellular functions. PIP3 generated by PI3Ks activates the kinases PDK1 and Akt [also known as protein kinase B (PKB)] (98100). Akt is also activated by mTORC2 (101). Akt promotes cell survival by inhibiting the ubiquitin E3 ligase MDM2 and the proapoptotic factor BAD (102). Akt promotes growth, metabolism, and tumorigenesis by inhibiting the Forkhead box (FOXO) family of transcription factors (102). Akt promotes cell cycle progression by inhibiting glycogen synthase kinase 3 (GSK3). Akt promotes mTORC1 activity by phosphorylating and inhibiting the tuberous sclerosis proteins 1 and 2 (TSC1 and 2), thereby enabling the GTPase Rheb to activate mTORC1 (103). Two proteins, S6 kinase (S6K) and the growth factor receptor-bound protein 10 (Grb10), are phosphorylated by mTORC1 and act in a feedback loop that inhibits signaling by the insulin receptor, insulin-like growth factor (IGF) receptor, and the adaptor protein IRS-1 (104). mTORC1 promotes protein synthesis by phophorylating translational regulators S6K and eIF4E binding proteins 4EBP1 and 4EBP2 (103). RTK, receptor tyrosine kinase.

CREDIT: Y. HAMMOND/SCIENCE SIGNALING
Fig. 2

Class I PI3Ks interactome. Schematic representation of class I PI3K heterodimers at the center with their regulatory partners. Proteins represented in green stimulate PI3K activity, whereas those in orange have inhibitory effects. Although PTEN does not directly inhibit PI3K activity, it is represented in orange because it counteracts PI3K action. RTK or pY represents receptor tyrosine kinases, adaptors, or phosphotyrosine-containing motifs. Gray double-sided arrows represents protein-protein interactions. Dashed lines represent putative or unknown interactions.

CREDIT: Y. HAMMOND/SCIENCE SIGNALING

With their diverse roles in cell signaling, it is not surprising that reduced or excessive activity of class I PI3Ks is associated with various disorders. Both p110α and p110β transduce signals downstream of the insulin receptor, making diabetes a potential side effect of PI3K inhibitors (6, 7). The p110δ and p110γ isoforms, which are abundant in leukocytes, have functions in chronic inflammation and allergy (811). Because all class I PI3Ks can induce cellular transformation and mutations in p110α are associated with numerous human cancers, these enzymes constitute a major target for cancer therapy (12, 13).

Class II PI3Ks consists of three enzymes (C2α, C2β, and C2γ) that share sequence similarity with class I PI3Ks, but they have no known regulatory partners (14). Some isoforms have been linked to insulin secretion, clathrin-mediated exocytosis, and smooth muscle contraction (1), but their full cellular roles are not yet clear. Structures are not available for any of the class II PI3Ks, but sequence analysis shows that they have the core domain organization of the class I enzymes with an extended N-terminal proline-rich region and a C-terminal extension with Phox homology (PX) and C2 domains. Because of the limited structural information available for these enzymes, we will not cover them further in this review.

The only class III PI3K, vacuolar protein sorting 34 (Vps34), is found in all clades of eukaryotes (15). The catalytic subunit is structurally related to class I enzymes, but it lacks the ABD and RBD. Its regulatory subunit, Vps15 (also known as p150) does not share any similarity with p85-like subunits. Instead, it includes a Ser/Thr kinase domain, a HEAT domain, and a WD domain. Vps34 uses only PI as a substrate, generating phosphatidylinositol 3-phosphate (PI3P), which is important in autophagy, phagocytosis, lysosomal sorting, and cell signaling. On the basis of the crystal structure of the catalytic core of Vps34, it appears that the regulation of Vps34 resembles that of class I PI3Ks in some aspects (16). In contrast to the class I enzymes, misregulation of class II nor III PI3Ks have not yet been linked to human diseases.

Stimulation of PI3Ks

Activation by receptor tyrosine kinases

Receptor tyrosine kinases and their adaptor proteins can activate class IA PI3Ks through binding of the SH2 domains of the regulatory subunit to tyrosine-phosphorylated pYXXM (where Y is Tyr, X is any amino acid, and M is Met) motifs (Fig. 2) (8, 17). Phosphopeptides mimicking phosphorylated tyrosine motifs activate p110-p85 heterodimers in vitro (18, 19). In the cytosol, catalytic subunits are stabilized and maintained in an inhibited state by their regulatory subunit (20). Early studies demonstrated that a construct encompassing the nSH2 and iSH2 domains of p85α inhibited p110α activity and that this inhibition could be released by pY-containing peptides (21). However, studies on p110β and p110δ have also revealed an inhibitory function for the cSH2 domain (22, 23).

The first crystal structure of a p110α in complex with the iSH2 domain of p85α showed that the iSH2 domain sits in a crevasse formed by the catalytic subunit, contacting the ABD and C2 domains of p110 (24). The iSH2-ABD contact is responsible for the high-affinity interaction between the two subunits, whereas contact with the C2 domain is part of the inhibitory function of p85. The loops in the C2 domain that contact the iSH2 domain show high sequence divergence among the class I PI3Ks, which could contribute to the differential inhibitions of p110α, p110β, and p110δ by p85 (25). The crystal structure of p110α with the nSH2 and iSH2 domains of p85α shows that the nSH2 domain contacts three regions of p110α: (i) the helical domain around the mutational hotspot residue Glu545; (ii) three loops in the C2 domain (β1-β2, β5-β6, and β7-β8); and (iii) residues in helix kα10, part of the “regulatory arch” in the kinase domain (Figs. 2 and 3) (26). The cSH2 domain interacts only with the kinase domain, and this involves a critical contact with the “elbow” formed between helices kα11 and kα12 at the C terminus of p110 (Fig. 3). His1047 in p110α, a residue that is often mutated in cancers, is located at this elbow, pointing at a crucial role of this region for lipid kinase activity. The cSH2 domain also contacts the kα7-kα8 loop in p110β, which contributes to the differential inhibitory functions by p85 on the p110 isotypes. The nSH2 domain has an inhibitory effect on p110 that can be released by addition of receptor tyrosine kinase–derived phosphopeptides or by point mutations in either the nSH2 domain or the catalytic subunit (Fig. 4) (27). The cSH2-mediated inhibition can also be released by phosphopeptides derived from receptor tyrosine kinases or point mutations but only affects p110β and p110δ, not p110α (22, 23).

Fig. 3

PI3K catalytic elements and the regulatory arch. (A) Model of a PI3K heterodimer with ATP and a regulatory construct encompassing the nSH2, iSH2, and cSH2 domains of the regulatory subunit, modeled from Protein Data Bank (PDB) identification codes (IDs) 1E8X (p110γ), 2Y3A (p110β), and 3HHM (p110α). The regulatory arch of p110β is shown in gray, and somatic mutations in the catalytic domain of p110α found in at least two different tumor samples are depicted as balls (purple). A frequently occurring mutation in p110α, H1047R, is shown as a red sphere. (B) The regulatory arch for class I and III enzymes, colored by isotype. The arch rests on a second layer of helices (kα8 and kα9) (white). The C-terminal helix (kα12) varies considerably in conformation and extent of order among the structures. For wild-type p110α and p110δ, the region is not ordered beyond the elbow. For the H1047R mutant of p110α, there is more order, but the C-terminal region is not helical. For Vps34, the C-terminal helix is turned away, and the catalytic loop takes on an active conformation. The phosphorylation site at Thr1024 in p110γ is shown with a red sphere. (Bottom) Representations of the inhibited and activated states of PI3K emphasize the conformations adopted by the helix kα12.

To view interactive versions of the structural images, see the Interactive Figures.

CREDIT: O. VADAS, X. ZHANG, R. L. WILLIAMS/MEDICAL RESEARCH COUNCIL, LABORATORY OF MOLECULAR BIOLOGY, UK, AND Y. HAMMOND/SCIENCE SIGNALING
Fig. 4

Somatic cancer-associated mutations in p110α. (A) Histogram showing all mutations in p110α currently reported in the COSMIC database (http://www.sanger.ac.uk/genetics/CGP/cosmic). The locations of the different mutations are indicated by their colors, which represent domain locations of the mutations. (B) A structural model of p110 and p85 interactions was generated with the crystal structures of p110α in complex with the nSH2 and iSH2 domains of p85α (PDB ID 3HHM) and p110β with the iSH2 and cSH2 domains of p85β (PDB ID 2Y3A). The catalytic subunit and nSH2 and iSH2 domains from 3HHM were modeled with the cSH2 domain of 2Y3A. (C) Structural information for hotspot oncogenic mutations in p110α using the crystal structure of the H1047R p110α mutant (PDB ID 3HIZ) and the structure of nonmutated p110α in complex with the nSH2 and iSH2 domains of p85 (PDB ID 3HHM).

CREDIT: J. E. BURKE AND O. VADAS/MEDICAL RESEARCH COUNCIL, LABORATORY OF MOLECULAR BIOLOGY, UK

Although the ability of soluble pY peptides to mimic activation of PI3Ks by receptor tyrosine kinases is well known, the mechanism of this activation is only beginning to become apparent. It is now clear that soluble phosphopeptide binding activates PI3Ks by two distinct mechanisms: (i) It causes a substantial increase in phosphotransferase activity, even in the absence of lipid substrate, and (ii) it greatly increases the inherent affinity of the p110-p85 heterodimer for phospholipid membranes. This implies that SH2 domains of p85 inhibit membrane binding in the basal state (22, 23). A study using deuterium-exchange mass spectroscopy has shed light on the structural determinants in PI3Ks involved in membrane binding (22, 23). In the kinase domain, the membrane-interacting region includes the activation (substrate-binding) loop and the regulatory arch that cradles this loop.

PI3K activity can also be modulated by phosphorylation of the p85 regulatory subunit by the Src family tyrosine kinases Abl and Lck, by cyclic adenosine monophosphate (cAMP)–dependent protein kinase (PKA), by the p110 catalytic subunit itself, and by protein kinase C (PKC) family members (2831). Phosphorylation of Tyr688 in the cSH2 domain of p85α by Lck increases enzyme activity (32). Dephosphorylation of Tyr688 by the protein tyrosine phosphatase SH2 domain-containing phosphatase–1 (SHP-1) decreases PI3K activity (33). However, the mechanism of the activation by phosphorylation of Tyr688 remains to be determined. Furthermore, because other tyrosine phosphatases have been reported to increase PI3K activity, the role of phosphorylation of Tyr688 in the regulation of endogenous PI3K is unclear (34). Consequently, the phosphorylation state of Tyr688 is not a reliable indicator of PI3K activity. Phosphorylation of Ser608 in p85α by p110 and phosphorylation of Ser652 by PKC family members both result in decreased PI3K activity, although by different mechanisms (31, 35). Even when present in the same cells, the class IA isoforms are not uniformly stimulated by receptor tyrosine kinases. Despite the ability of phosphopeptides to stimulate class IA isotypes in vitro, most cellular data indicate that p110β is fairly unresponsive to stimulation by receptor tyrosine kinases and is predominately stimulated downstream of GPCRs (5). However, when p110β does not have to compete with p110α and p110δ, it does mediate tyrosine kinase signaling (36). This puzzling behavior might be due to the three brakes imposed on p110β activity by the nSH2, iSH2, and cSH2 domains of p85. In contrast, p110α may be more readily activated because it is inhibited only by the nSH2 and iSH2 domains. Furthermore, phosphorylated tyrosines motifs on receptor tyrosine kinases and adaptor proteins should bind with higher affinity to the cSH2 domain from a p110α-p85 complex, because they do not have to disrupt an inhibitory cSH2-p110 contact (22). It may also be that p110β-p85 is sequestered by interactions with other partners that limit its responsiveness to receptor tyrosine kinases.

Activation by Gβγ heterodimers

Among class I PI3Ks, p110γ and p110β are the only isotypes that are activated by Gβγ heterodimers (Gβγ) downstream of GPCRs (35, 3739). Whereas Gβγ and pY phosphopeptides synergistically activate p110β in vitro, p110γ is activated by Gβγ and not pY phosphopeptides (3, 4). p110β can be activated by both receptor tyrosine kinases and Gβγ heterodimers. These properties may explain why this isotype is preferentially engaged in neutrophils activated by low concentrations of immunoglobulin G–containing complexes, which stimulate Fcγ receptors (which are receptor tyrosine kinases) as well as the GPCR BLT1 (leukotriene B4 receptor 1) through an autocrine-paracrine pathway dependent on the leukotriene B (40). The p110β and p110γ catalytic subunits alone can be stimulated by Gβγ; however, full activation of p110γ requires association with its p101 regulatory subunit, which enables recruitment of p110γ to plasma membrane and shifts substrate specificity toward PIP2 (4, 39, 41, 42). Activation by and interaction with Gβγ is more pronounced for the p110γ-p101 complex compared with the p110γ-p84 complex or with p110γ alone (41, 43).

Activation of PI3Ks by the Ras superfamily of G proteins

Class I PI3Ks can be activated by small G proteins of the Ras superfamily with differences in the requirement for particular Ras members or in the degree of activation among PI3K isotypes. This is not surprising given that the p110 isoforms show considerable sequence divergence in the RBD regions responsible for Ras binding (44). In contrast to p110δ, which is activated only downstream of R-Ras and TC21, a broader range of Ras family GTPases (proteins that can hydrolyze GTP) can activate p110α and p110γ (45). Activation of p110β by Ras family GTPases has not been firmly established, with some experimental setups supporting activation (46, 47) and some not (45). On the other hand, Rab5 interacts with p110β but not p110α (1, 48). Mutations in the RBD that prevent binding of p110 to Ras show that Ras is required for the signaling and transforming ability of p110α (49), p110β, and p110γ (12, 46, 50). PI3K regulatory subunits appear to modulate responsiveness to Ras by controlling sequential activation of the catalytic subunits by receptor tyrosine kinases and Ras (51). The p110γ complex containing the p84 regulatory subunit depends to a greater extent on activation and membrane translocation by Ras compared with the complex containing p101 (41).

Structural Basis for Increased Activity by Cancer-Related Mutants of PI3Ks

The PI3K–phosphatase and tensin homolog deleted on chromosome 10 (PTEN) pathway is one of the most frequently mutated pathways in various human cancers. Class I catalytic isoforms can promote cell proliferation and survival (36) and cause cell transformation (46) when overexpressed. Among the class I PI3K catalytic subunits, p110α is the only one that contains somatic oncogenic mutations that confer gain of function (13, 52). Over 3000 independent somatic mutations have been identified in tumor samples for the gene encoding p110α (PIK3CA, http://www.sanger.ac.uk/genetics/CGP/cosmic) and detected in different cancer types, including breast cancer, endometrial cancers, and glioblastomas (53). Many somatic mutations have also been identified in the class IA regulatory subunit p85α (52, 54, 55). Mutations in the p85 regulatory subunit activate all class IA isoforms in vitro and promote growth factor–independent cell survival and anchorage-independent cell growth (54).

The majority of cancer-linked mutations in p110α cluster around two locations, one in the helical domain (Glu542 and Glu545) and the other in the C-terminal end of the kinase domain (His1047) (Figs. 3A and 4). Biochemical and structural evidence indicates that the helical-domain mutations interfere with the inhibitory binding of the nSH2 domain of p85α (26, 27) because these residues form an electrostatic interaction with the nSH2 domain. Helical domain mutants are no longer sensitive to activation by pY phosphopeptides (56) but still require Ras to induce oncogenic transformation (57). A mutation in the nSH2 domain of p85α engineered to break this interaction (Lys379→Glu379; K379E) also stimulates activation of lipid kinase activity and is oncogenic (27, 55). The structure of the p110α His1047→Arg1047 (H1047R) mutant shows a conformational shift of two loops (kα4-kα5 and kα11-kα12) that are proposed to interact with the membrane surface, thereby implying that activation by this mutation occurs through increased association with membranes (26). This mechanism is consistent with the observation that cell transformation by the H1047R mutant does not require the presence of Ras, which is proposed to activate PI3Ks by increasing lipid binding (57).

Along with the hotspot mutations, there are also numerous sporadic mutations located in every domain of p110α except the RBD (Fig. 4). In the ABD domain, the two most frequently found mutations [Arg38→His38 (R38H) and Arg88→Gln88 (R88Q)] are located at the interface of the ABD domain and N-terminal lobe of the lipid kinase domain (24). Mutations have also been discovered in the C2 domain, with the most frequent mutations occurring at Asn345, Cys420, and Glu453. Asn345 is located in CBR1 (loop β1-β2) of the p110 C2 domain, at the interface with the iSH2 domain of p85α (Fig. 4) (24). Cancer-linked mutations in p85α also map to this interface [Asp560→Tyr560 (D560Y), Asn564→Asp564 (N564D), or Asn564→Lys564 (N564K)], and mutations that affect the C2-iSH2 interface promote lipid kinase activity as well as increase cell transformation (58). The Cys420→Arg420 (C420R) mutation in p110α is located in the CBR3 (β5-β6) loop of the C2 domain, close to the interface with the iSH2 domain, and the mutation of Glu453 to Ala, Lys, or Gln in the β7-β8 loop at the interface with both the nSH2 and iSH2 domains also increases lipid kinase activity and cellular transformation (59, 60).

The C-Terminal PI3K Regulatory Arch

The structures of class I PI3Ks show that the three C-terminal helices of the p110 kinase domain—kα10, kα11, and kα12—form an important structural element for the regulation of p110 activity (Fig. 3) (22, 61). This regulatory arch encircles the catalytic and activation loops that are the key elements implicated in catalysis. The C-terminal helix (kα12) of the regulatory arch takes on various conformations in the structures that have been reported (Fig. 3B), and it is critical for membrane binding and consequently kinase activity on lipid membranes (16, 22). This helix also has a second role in maintaining the enzyme in a closed, inactive conformation in the absence of lipid membranes. In this closed conformation, substrate entry into the active site appears to be restricted. The two helices in the regulatory arch preceding kα12 also appear to have important roles, because the nSH2 and cSH2 domains and several external regulatory proteins that modify the activity of PI3Ks interact with sides of this arch. The nSH2 domain contacts kα10, whereas the cSH2 domain inhibits p110 activity by positioning kα12 (22, 26). When in complex with the iSH2 and cSH2 domains of p85β, the p110β C-terminal helix (kα12) folds over the catalytic loop, presumably holding it in an inactive conformation. Thus, inhibition by the cSH2 domain might operate by clamping kα12 in a closed conformation, preventing access of substrate to the catalytic site. Many of the mutated residues in the p110α kinase domain identified in tumor samples are located along the regulatory arch region (Fig. 3A). Moreover, the Vps15 regulator binds Vps34 in this C-terminal region (62). Changes in the interaction network of the arch may result in allosteric regulation of enzyme activity, influencing several possible steps in the reaction, such as lipid binding, phosphoryl transfer, or product release during catalysis.

On the basis of the 6.6-Å DNA–protein kinase crystal structure, the overall fold of the kinase domain in PI3K-related kinases (PIKKs) is proposed to be similar to that of the kinase domain of PI3Ks (63). Both sequence and structural similarities suggest the possible presence of the regulatory arch in PIKKs (61, 64). As with the PI3Ks, this C-terminal region in the PIKK target of rapamycin (TOR) is important for membrane interaction (65). Thus far, three-dimensional structures of several PIKK family members derived from EM reconstruction seem to support the importance of the arch in transmitting the regulatory influences of interacting protein partners. For example, the electron microscopic structure of the budding yeast TOR1-KOG1 complex shows that the C terminus of the kinase domain and the FATC domain of TOR1 may contact the regulatory subunit KOG1 (an ortholog of raptor in mammals) (66). Similarly, the C terminus of mammalian TOR (mTOR) interacts with raptor in the dimeric structure of mTOR complex I (mTORC1) (67).

Catalytic Mechanism: PI3Ks Compared to Protein Kinases

PI3Ks and protein kinases share a common bilobal organization of their kinase domains with the ATP bound between the lobes. Structural work on protein kinases have contributed to understanding what constitutes the catalytically competent “active” conformation of a prototypical protein kinase as well as a range of intermediate or “inactive” conformations that they can assume. Various regulatory partners modulate kinase activity by interacting with the N lobe of the kinase domain (68). In contrast, the PI3Ks show a greater degree of conservation, and the structures show relatively little variation in the N lobe. This lobe is largely inaccessible to regulatory partners because it is tightly packed against the helical domain and the ABD. The C lobe of PI3Ks accommodates the lipid substrate, interacts with membranes and regulatory subunits, and is more structurally divergent from the C lobe of protein kinases.

The N lobe of PI3Ks and protein kinases are generally organized around a conserved five-stranded β sheet and the helix kα3, an important regulatory element of protein kinases (also known as helix C) (nomenclature for the protein kinase is derived from PKA) (69). The P loop, which interacts with the phosphates of the bound ATP, is conserved in both kinase families, although conserved Gly residues in protein kinases are absent in PI3Ks (Fig. 5A). Lys72 in PKA, which is at the center of the N-lobe β sheet, coordinates α and β phosphates of ATP and makes an ion pair with Glu91 in helix α3, thereby stabilizing the helix α3 in an active conformation. PI3Ks possess an equivalent Lys (Lys802 in human p110α) and an Asp (Asp810) equivalent to the Glu91 of PKA. However, in PI3K structures, these residues are too distant to form an ion pair, even when ATP is bound. Although the lysine equivalent in PI3Ks to Lys72 in PKA is important for catalysis (this residue is covalently modified by wortmannin in PI3Ks, thereby inactivating these enzymes), it is not clear that the salt link is a signature of the active conformation of the PI3Ks.

Fig. 5

Structural comparison of PI3Ks and protein kinases. (A) Structure alignment of the N and C lobes of protein kinase A (PKA) and PI3K (p110β) with ATP shown in pink. The major corresponding features of each lobe are identically colored in both structures. The residue numbers shown for PI3K are from the mouse p110β (PDB ID 2Y3A). The His residues from the conserved DRH motif of three different PI3K structures (p110β, p110γ, and Vps34) are shown on the PI3K C lobe. (B) Structure alignment of the kinase domains of PKA and p110β with the catalytic spines and regulatory spines shown in yellow and red dots, respectively. Residue numbering for PI3K is shown for mouse p110β and for human p110α in parentheses. His915 and Trp1053 from p110β are also represented in dots, using the same colors as in (A).

CREDIT: O. VADAS/MEDICAL RESEARCH COUNCIL, LABORATORY OF MOLECULAR BIOLOGY, UK

The C lobe is built mainly of α helices and contains the lipid substrate-binding site and part of the ATP-binding site. The ATP-binding region and two loops essential for catalysis—namely, the catalytic and activation loops—are present in the PI3Ks and the protein kinases, but the rest of the C lobe differs. The catalytic loop, the activation loop, and the helix kα12 all show conformational diversity among PI3Ks, and features of each of these three elements can be classified as active or inactive conformations. The Asp-Arg-His (DRH) motif in the catalytic loop, which is equivalent to the Tyr-Arg-Asp (YRD) or His-Arg-Asp (HRD) motif in protein kinases, is strictly conserved in PI3Ks, and the different conformations of the catalytic loop captured in the crystal structures of p110α, β, γ, and δ are thought to represent inactive states of this loop, with the Asp, the His, or both residues turned away from the catalytic center (Fig. 5A) (16, 70). In Vps34, both residues are oriented toward the catalytic center, and this is likely to represent the active conformation of this loop (16, 70). Similar to the catalytic loop, conformational changes are likely to occur in the activation loop during catalysis, as observed in protein kinases (71). The highly conserved Asp-Phe-Gly (DFG) motif at the beginning of the activation loop often changes from “out” to “in” conformation in protein kinases when the enzyme goes through different steps of the catalytic cycle (68). Some conformational changes in this motif are also observed in PI3Ks, because the Asp residue in the DFG motif adopts more than one conformation. However, there is no evidence for a conformational change of the Phe residue in the DFG motif in PI3Ks. In contrast to protein kinases, PI3Ks are not regulated by phosphorylation of residues in the activation loop.

Communicating substrate binding to catalytic activity

Two spatially connected hydrophobic networks are conserved in the catalytic domains of all active protein kinases (72). These two networks, named regulatory and catalytic spines (R and C spines respectively), connect the N and the C lobes together, providing a scaffold for the catalytic elements and positioning ATP and the peptide substrate. The adenine ring of ATP forms part of the C spine. These R and C spines are connected through the helix αF, which completes the spines, and create an arch that coordinates catalysis (Fig. 5B). The catalytic loop also stretches between the two spines, and the conformation of the activation loop has an important impact on formation of the spines. We wondered whether equivalent hydrophobic spines exist in PI3Ks.

The active conformations of protein kinases have an assembled R spine, but in inactive kinases the R spine becomes disassembled. Superimposition of PI3K structures on the active conformation of PKA shows that residues at the center of the R spine are conserved in all PI3Ks (Fig. 5B). PI3Ks have a continuous R spine, although all of the structures have at least some features characteristic of inactive conformations.

The C spine of protein kinases is also conserved in PI3Ks, except for residues in helix αF of protein kinases. PI3Ks have no clear equivalent element to the helix αF in protein kinases. Instead, PI3Ks have the regulatory arch, which may play a role in coordinating the binding of lipid substrate by the activation loop with formation of the active catalytic center. The kα12 helix, which is part of the regulatory arch, reaches over the activation loop and catalytic loop, blocking them in an inactive conformation. In the structure of p110β, the His residue of the DRH motif in the catalytic loop extends the C spine, forming a hydrophobic contact with a conserved Trp residue (p110β-Trp1053) in helix kα12 that caps the C spine (Fig. 5B). In the presence of lipid membranes, which bind to both the activation loop and helix kα12, there must be a conformational change to accommodate lipid substrate and reorganize the catalytic center to its active state. Ordered activation loops have been observed for p110β, p110γ, and Vps34, but none of these bind to lipid, and it is likely that lipid substrate binding plays a substantial role in the conformation of the activation loop. One possibility is that the hydrophobic spines coordinate membrane binding with changes in the ATP-binding pocket.

Regulatory Associations of PI3Ks

Activation of PI3K signaling by the influenza virus protein NS1

Influenza viruses cause severe contagious respiratory diseases that have led to extensive morbidity and mortality worldwide. Influenza viruses activate the PI3K-Akt signaling pathway to support efficient replication (73). The viral nonstructural protein 1 (NS1) is a nonessential virulence factor in influenza A (74) that activates the PI3K pathway in cells through interaction of its effector domain with the p85β regulatory subunit (75). Insight into the PI3K activation mechanism was revealed by the crystal structure of a NS1 effector domain with the iSH2 domain of p85β (76). A structural model suggests that NS1 interacts with the iSH2 domain in a manner that would prevent the nSH2 domain from making its inhibitory contacts with the catalytic subunit. The model also suggests that NS1 contacts the activation loop of p110, an interaction that has been proposed to directly activate p110. However, it still remains to be shown that NS1 influences PI3K activity in vitro.

Convergence in cAMP and PI3K signaling

The p110γ-p84 complex regulates cardiac contractility. This is achieved by p110γ acting as an A-kinase anchoring protein (AKAP) for PKA (10). A loop in the ABD of p110γ (residues 126 to 150) that is not conserved in other class I PI3Ks binds to the RIIα subunit of PKA, and the p84 regulatory subunit interacts with and recruits phosphodiesterase (PDE) 3B (10, 77). In this complex, PKA has dual functions: It phosphorylates and decreases activity of p110γ, and it also phosphorylates and increases activity of PDE3B. These phosphorylation events reduce the abundance of two second messengers, because activated PDE3B enhances cAMP degradation and inhibition of p110γ decreases PIP3 abundance, thereby reducing the extent of β-adrenergic receptor (βAR) internalization and degradation. PKA decreases activity of cardiac p110γ by phosphorylating Thr1024 in the loop between kα8 and kα9 (10), which is close to the regulatory arch (Fig. 5A). Additional structural work would shed light on the mechanism linking PKA phosphorylation and p110γ activity. In heart failure, the lipid kinase activity of p110γ is increased, thereby decreasing cell-surface βAR density (10).

Another link between PIP3 and cAMP signaling pathways involves the tethering of EPAC1 (exchange protein activated by cAMP-1) and p110γ by PDE3B to p84 (78). The interaction of EPAC1 with PDE3B indirectly activates PI3Kγ and promotes angiogenesis. These two studies confirm the importance of the p110γ-p84 complex in cardiac functions, establishing p110γ as a potential target for the treatment of heart failure. They also support the distinct regulatory functions p84 and p101 on p110γ activity.

A direct interaction between PI3K and βAR kinase (βARK) affects the surface density of βAR. Desensitization of the βAR by internalization is controlled through phosphorylation by βARK, a process that requires PIP3 (79). Furthermore, direct contact between the helical domain of p110γ and βARK mediates βAR sequestration (80).

PKA also regulates class IA PI3Ks signaling in cells through phosphorylation of the p85α regulatory subunit at Ser83 (30). Two proteins interact with Ser83-phosphorylated p85α, estrogen receptor α (30) and 14-3-3ζ (81), leading to increased PI3K activity, possibly by increasing membrane recruitment.

Regulation of DNA replication and repair by p110β/p85β

Some class I PI3Ks also have nuclear functions: p110β controls DNA replication and double-strand-break DNA repair by both kinase-dependent and -independent mechanisms (82, 83). In contrast to p110α, which is found exclusively in the cytosol, a large portion of p110β is found in the nucleus, where it can activate nuclear Akt (84). Association of p110β with p85β but not with p85α results in nuclear localization of the p110β-p85β complex, and a potential nuclear localization sequence is located in the CBR3 loop of the C2 domain of p110β that controls nuclear localization of the complex (84). A nuclear export sequence in p85β regulates distribution of the p110β-p85β complex between the nucleus and the cytosol. In the nucleus, p110β regulates binding of the DNA damage sensor protein Nbs1 to sites of DNA damage (83). These studies have revealed that cell survival is controlled not only by cytoplasmic but also nuclear PI3Ks. How p110β is stimulated in the nucleus still remains ambiguous, because none of the known PI3K activators is present in this subcellular compartment.

PI3K-PTEN crosstalk

PTEN is a tumor suppressor protein frequently inactivated or deficient in human cancers. This lipid phosphatase dephosphorylates the D3 position of PIP3, thus antagonizing the action of class I PI3Ks. Tumorigenesis in PTEN-deficient prostate cancer is selectively driven by p110β. Accordingly, inactivation of p110β by conditional knockout in a PTEN-deficient mouse model prevents tumorigenesis, and short hairpin RNA (shRNA)–mediated depletion of p110β inhibits PI3K signaling and growth in PTEN-deficient cell lines (38, 85). Given that p110β activity contributes to tumorigenesis, p110β selective inhibitors may be useful anticancer agents tailored to PTEN-deficient cancers.

Earlier work with a conditional knockout of p85α had shown that this protein can regulate PTEN activity in vivo (86). p85α binds directly to unphosphorylated PTEN through the SH3 and BH domains in the N terminus of p85, leading to stimulation of PTEN activity (87, 88). These results highlight a dual function of p85 regulatory subunits in regulating formation of PIP3 through interaction with p110 and at the same time increasing dephosphorylation of PIP3 to PIP2 by increasing the phosphatase activity of PTEN.

The Development of PI3K-Selective Inhibitors

The importance of increased PI3K activity in cancer, thrombosis, and inflammatory diseases has led to intense efforts to develop selective and potent PI3K inhibitors (89, 90). This is a difficult undertaking because the residues in PI3K that contact ATP are conserved among the four classes of PI3Ks (Fig. 6A). Accordingly, selectivity needs to be tailored by exploiting the chemical diversity and pockets surrounding the ATP-binding site. Crystal structures of PI3Ks have revealed six regions within the ATP-binding pocket: hydrophobic regions I (the affinity pocket) and II, the hinge region, the P loop, the start of the activation loop (DFG motif), and the specificity pocket (91) (Fig. 6B). Designing compounds that explore or create these pockets have improved potency and selectivity of PI3K inhibitors (6, 92, 93). With the determination of crystal structures of all class I isoforms as well as that of Vps34, development of isotype-specific inhibitors is entering a new era. Several highly selective PI3K and dual inhibitors of PI3K and mTOR have been developed and are currently in clinical trials for the treatment of cancers and inflammatory diseases (53, 94).

Fig. 6

Structural insights into PI3K inhibitor selectivity. (A) Surface representation of the ATP-binding pocket in p110γ with residues colored according to their degree of conservation among the four class I PI3K members. (B) Close-up view of the ATP-binding pocket in p110δ in complex with the p110δ-specific inhibitor SW13. The main regions that define distinct structural parts of this pocket are highlighted. The specificity pocket is formed between Trp760 and Met752 (p110δ numbering) by an outward movement of Met752 to accommodate the inhibitor. (C) Same view as in (B), with an overlay of the inhibitor GDC-0941 crystallized in complex with p110β (PDB 2Y3A), p110δ (PDB 2WXP), and p110γ (PDB 3DBS).

CREDIT: A. BERNDT/MEDICAL RESEARCH COUNCIL, LABORATORY OF MOLECULAR BIOLOGY, UK

Thienopyrimidine GDC-0941 (Pictrelisib, Roche), a multitargeted class I PI3K inhibitor, was developed by optimization of the pyridofuranpyrimidine PI-103 (Yamanouchi) and is currently in phase II clinical trials for the treatment of advanced solid tumors (93). GDC-0941 exhibits good pharmacokinetics, is orally bioavailable, and shows a good in vivo tolerability (93). Crystal structures of three class I PI3K isoforms (PI3Kβ, γ, or δ) bound to GDC-0941 are available (Fig. 6C). In these structures, the mostly flat compound binds in the same manner with its morpholino oxygen, contacting the hinge and the indazole group, which projects deep in the affinity pocket. The terminal sulfonylpiperazine group slightly projects from the ATP-binding pocket toward the solvent, where it hydrogen-bonds with the amide nitrogen of a P-loop residue. This sulfonylpiperazine group shows different orientations depending on the isoform, which might explain its slight selectivity for α and δ over β and γ within the class I PI3K family (22, 92, 93).

Calistoga Pharmaceuticals disclosed the structure of the PI3Kδ inhibitor CAL-101 [acquired by Gilead (Foster City, California) in 2011 and renamed GS-1101], which is currently in phase II clinical trials for the treatment of leukemias, lymphomas, and melanomas (95). Chemically, CAL-101 is related to the quinazolinone purine compounds PIK-39 and IC87114 (ICOS) (96), and co-crystal structures of PIK-39 with p110γ (6) and PIK-39 or IC87114 with p110δ (92) have demonstrated a flexibility-based mechanism for how these compounds achieve selectivity. They cause a conformational rearrangement of a conserved methionine residue (Met804 in p110γ and Met752 in p110δ) that induces the creation of a hydrophobic pocket (specificity pocket) in the ATP-binding site between this residue and a conserved tryptophan residue (Trp812 in p110γ). These and other p110δ-selective compounds, such as SW13, adopt a propeller-shaped conformation in the active site, and their quinazolinone moiety fits snugly in this newly formed pocket (Fig. 6B). The active site of p110δ allows the opening of this new pocket with only a local change of the P-loop conformation because of a higher conformational flexibility, whereas in p110γ this P-loop movement leads to a global movement of the N lobe relative to the C lobe.

Isotype-specific PI3K inhibitors offer a promising direction for clinical application. However, they have also emerged as routine tools for studying the roles of specific PI3Ks in cellular processes, providing a valuable extension to gene targeting approaches. One current challenge is the lack of specific, potent, commercially available inhibitors for some isotypes, such as p110α, Vps34, and class II PI3Ks.

Conclusions and Perspectives

In addition to the well-established regulators of PI3Ks—such as receptor tyrosine kinases and their adaptors, GPCRs and Ras GTPases—other regulators such as PKA and the viral protein NS1 are emerging that display isoform-selective regulatory mechanisms. Relocalization to membranes is one means of regulating activity of PI3Ks, and the importance of factors that dictate recruitment to membrane are widely appreciated. PI3Ks can assume active and inactive conformations, which differ from those observed for protein kinases. We know that the inactive state is not the same for all isotypes—for example, the p110α isoform lacks the inhibitory constraint imposed by the cSH2 domain. However, quantitative dose-response analyses of activity, localization, and membrane residence time for each isotype in cells stimulated with various ligands are missing. The PI3Ks are slow enzymes that are only transiently in contact with membranes, where residence time can crucially affect activity. Thus, many regulatory influences on PI3Ks can be viewed as shifting the likelihood of the various states of the enzymes (Fig. 7). Recruitment of PI3Ks to membranes could involve (i) translocation of a static state of a PI3K, (ii) uncovering of a membrane-binding interface that is hidden in the inhibited (basal) state, or (iii) conformational changes that remodel the enzyme’s interface (Fig. 7). Repositionings of the membrane-binding C-terminal tail caused by dislodging the cSH2 domain from p110β or p110δ with phosphopeptides or by the cancer hotspot mutation H1047R in p110α are examples of conformational changes that affect membrane binding (23). Similarly, truncation of the ABD releases free, active p110α and increases PI3K signaling activity (97). This suggests that the p85 regulatory subunit is not required for membrane binding, but rather that it has an important role in occluding an intrinsic membrane-interacting property of the catalytic subunit. However, the observations that phosphopeptides cause an increase in enzyme activity in the absence of membranes indicate that activation is not a simple process of translocation to membranes (22).

Fig. 7

Model of class IA PI3K activation by receptor tyrosine kinases, phosphotyrosine (pY) adaptor molecules, Gβγ heterodimers, and Ras. Activation by membrane recruitment and conformational changes are shown. Phosphopeptide binding or mutation releases the SH2-mediated brakes, whereas the iSH2 brake may be partially released by membrane binding or mutation. Each individual activator only partially stimulates PI3K (a few partially activated states are illustrated), and full activation requires contributions from more than one stimulus. Ras and Gβγ both participate in PI3K membrane recruitment, but the binding site for Gβγ on p110β has not been characterized, so this interaction is represented with dotted lines.

CREDIT: Y. HAMMOND/SCIENCE SIGNALING

Specific mutations in the RBDs of the PI3Ks were important tools for dissecting the contribution of Ras to PI3K signaling (46, 49, 50). Similar approaches could be taken to understand the quantitative contributions of each of the isoforms to a given cell stimulus. Early in vitro proof-of-principle experiments along these lines have been undertaken for brakes imposed by the nSH2 (27), iSH2 (25), and cSH2 (22) domains. Isotype-specific inhibitors and gene-targeting approaches are other alternatives to understanding the function of each PI3K isoforms in cells, and some compounds show promise in the treatment of human cancers.

Although several p110s share the same regulatory partners, the response to upstream inputs differs among the isoforms. These differences could reflect either the selective ability of the catalytic subunits to interact with different partners (for example, the specificity of p110β and p110γ for Gβγ) or the intrinsic enzymatic properties of the catalytic subunits based on sequence differences in the activation loop, C2 domain loops, and C-terminal tail. These distinctions could affect catalysis once the enzymes are present at the membrane. The selective inhibitory functions of p85-like regulatory subunits and their potency to stimulate PTEN activity add a further level of complexity to the already diverse regulatory interactions that can affect PI3K signaling.

Altogether, each PI3K isoform has a distinct order of priorities in its list of activation mechanisms, resulting in graded signal outputs depending on the upstream stimulus.Class IB p110γ interferes with the cAMP signaling pathway to regulate cardiac functions, and similar crosstalk between the PI3K pathway and other signaling pathways may emerge in the future.

References and Notes

  1. Acknowledgments: We are grateful to members of R.W.’s group for fruitful discussions. Funding: O.V. was supported by a Swiss National Science foundation fellowship (grant number PA00P3_134202) and a European Molecular Biology Organization (EMBO) fellowship (ALTF 690-2010), J.E.B. by an EMBO fellowship (ALTF 268-2009), and X.Z. by a Medical Research Council–LMB Cambridge Scholarship and Cambridge Overseas Trust. Work in the Williams lab was funded by the Medical Research Council.
View Abstract

Stay Connected to Science Signaling


Editor's Blog

Navigate This Article