The PI3K (phosphoinositide 3-kinase) p110α isoform is activated by oncogenic mutations in many cancers. This has stimulated intense interest in identifying inhibitors of the PI3K pathway as well as p110α-selective inhibitors, and understanding the mechanisms underlying activation by the oncogenic mutations. In the present article, we review recent progress in the structure and function of the p110α enzyme and two of its most common oncogenic mutations, the development of isoform-selective inhibitors, and p110α pharmacology.
The class I PI3Ks (phosphoinositide 3-kinases) are a family of closely related enzymes that are among the first level of cell signalling molecules regulated by tyrosine kinases, by small G-proteins Rac and Ras and by the βγ subunits of heterotrimeric G-proteins . They are encoded by four genes (PIK3CA, PIK3CB, PIK3CG and PIK3CD), each of which codes for a protein of approximately 110 kDa (p110α, p110β, p110γ and p110δ respectively). The kinase domain of these enzymes is closely related to class II and class III PI3Ks and to the phosphoinositide 3-kinase-related kinases, including mTOR (mammalian target of rapamycin), DNA-PK (DNA-dependent protein kinase), ATM (ataxia telangiectasia mutated) and ATR (ataxia telangiectasia- and Rad3-related) .
Regulation of class I PI3Ks by tyrosine kinases occurs because three of the enzymes (p110α, p110β and p110δ) are tightly associated with a regulatory subunit encoded by the PI3KR1, PI3KR2 and PI3KR3 genes. These contain two SH2 (Src homology 2) domains that allow recruitment to tyrosine-phosphorylated proteins and explains how the PI3K isoforms are activated during normal stimulation of RTKs (receptor tyrosine kinases) (e.g. in insulin receptor signalling ) and also how they can be continually activated by oncogenic forms of RTKs [1,3]. Regulation via Gβγ subunits is by direct interaction with the p110β and p110γ catalytic subunits. All four isoforms contain a Ras-binding domain that allows interactions with a range of small GTPases . These mechanisms result in recruitment to the plasma membrane where the substrate is located.
Class I PI3Ks phosphorylate the 3-position of the inositol ring of the ubiquitous membrane phospholipid PtdIns(4,5)P2 producing PtdIns(3,4,5)P3 . PtdIns(4,5)P2 is a relatively abundant membrane phospholipid, but PtdIns(3,4,5)P3 is found at much lower levels. PtdIns(3,4,5)P3 levels rise rapidly after stimulation of many RTKs and G-protein-coupled receptors and fall rapidly when the stimulus is removed. In most cells, the decrease is due to lipid phosphatases including INPP4b (inositol polyphosphate-4-phosphatase), SHIP (SH2-domain-containing inositol phosphatase) 1 and 2, and PTEN (phosphatase and tensin homologue deleted on chromosome 10) [3,6,7]. Class I PI3Ks can also act as protein kinases, although very little is known about the cellular role of this capability [8,9]. PtdIns(3,4,5)P3 recruits a limited subset of PH (pleckstrin homology)-domain-containing proteins, and a consequence is transient co-localization of signalling proteins. The most widely studied such event is co-localization of PDK1 (phosphoinositide-dependent kinase 1) with Akt/PKB (protein kinase B), which results in phosphorylation of Akt/PKB on Thr308 and the activation of a wide range of downstream signalling events involved in regulating cell growth, cell division and cell metabolism [3,7].
Hyperactivation of this pathway is observed in many cancers; it can arise from oncogenic mutations in growth factor receptors and also explains, in part, the powerful oncogenic effects of Ras mutations. Another activating mechanism in tumours is loss of the phosphatases that dephosphorylate the lipids. Both PTEN and INPP4b are commonly bi-allelically mutated or deleted in tumours . Overexpression of various class Ia PI3K isoforms has also been reported in many tumours . Oncogenic mutations in the p110α enzyme are also common in tumours , and an oncogenic p110β mutation was reported recently . The mutations in PIK3CA cluster in two hotspots , with the most common mutations including the helical domain E545K and kinase domain H1047R substitutions. Both cause constitutive activation of the kinase, but result in different kinetic properties . They activate p110α signalling through the PI3K–Akt/PKB–mTORC (mTOR complex) pathway, and are oncogenic in vivo . Mutations are also found in the PI3KR1 gene encoding the p85α regulatory subunit, and these also appear to act through p110α . Contributions of PI3Ks to tumour development do not have to be intrinsic to the tumour cells themselves. Some of the responses of stromal cells that support tumour growth are dependent on different PI3K isoforms including p110α  and p110δ . These findings have led to intense interest in developing inhibitors that selectively block individual isoforms while retaining therapeutic efficacy in the hope of gaining a greater therapeutic window. In the present article, we review the biology behind efforts to develop selective inhibitors of p110α.
Enzyme structure and function
X-ray crystal structures of wild-type p110α  and its H1047R oncogenic mutant have been determined , the former with small-molecule inhibitors PIK108  and BYL719  bound in the ATP-binding site, the latter with wortmannin bound . These structures illustrate the three-dimensional arrangement of the domains that make up the p110 catalytic unit including the ABD (adaptor-binding domain), RBD (Ras-binding domain), C2 domain, helical domain and the kinase domain. They also show where the N-terminal SH2 (nSH2) and coiled-coil inter-SH2 (iSH2) domains of p85α bind to the p110α catalytic unit. The iSH2 domain interacts with the ABD and C2 domains, whereas the nSH2 domain interacts with the C2 and helical domains. The phosphotyrosine-binding site of the nSH2 domain forms an interface with the p110α helical domain  and inhibits kinase activity .
The p110α–p85α structures provide insight into the spatial arrangement of cancer-associated mutations within the two proteins, and in some cases their mechanistic effects. The E545K mutation is located at the interface between the helical and nSH2 domains , whereas the H1047R mutation is found in the C-terminal tail of the kinase domain in a helical region described as the regulatory arch [23,24]. Structural  and biochemical  data indicate that the E545K substitution disrupts the helical and nSH2 domain interaction, relieving the inhibitory effect and rendering the p110α enzyme unresponsive to further activation by phosphotyrosine-containing peptides; it is, however, still responsive to activation by Ras. By contrast, the activated state of the H1047R kinase domain mutation retains responsiveness to phosphotyrosine peptides , and is not activated further by Ras . This suggested that the different mutations mimic activated states of the enzyme [13,27].
Detailed insight into the conformational changes during activation for the wild-type and oncogenic p110α–p85α complexes upon phosphotyrosine peptide binding and membrane binding was provided recently by deuterium exchange experiments . Burke et al.  proposed that the wild-type enzyme undergoes four distinct events in moving from its inactive cytosolic form to the active membrane-bound state. Although the order is unknown, these include breaking the nSH2–helical domain interface, disrupting the iSH2–C2 interface, movement of the ABD relative to the kinase domain and interaction of the kinase domain with lipid. Mutations activate the enzyme by either mimicking or enhancing different steps. The H1047R mutation affects the lipid-interaction surface within the C-lobe of the kinase domain, and the E545K helical domain mutant affected the p85α nSH2 interface as well as the iSH2–C2 interaction and ABD–kinase domain orientation.
The effect of these mutations on lipid binding was investigated recently . Hon et al.  showed that primary lipid-binding sites in the wild-type enzyme include hydrophobic residues in the C-terminal tail as well as basic residues in the activation loop, with lipid binding involving both an electrostatic and a hydrophobic component. A strong correlation between lipid kinase activity and lipid binding was noted for both wild-type and oncogenic mutant enzymes. Releasing the p85α nSH2 domain by phosphotyrosine peptide binding or E545K mutation increased membrane binding, with basic residues in the activation loop contributing to the electrostatic component, whereas the C-terminal tail has a more complex contribution. Upon phosphotyrosine peptide binding, the H1047R mutant had increased hydrophobic and electrostatic interactions . On the basis of a superimposition of existing p110α/p85α protein structures, Hon et al.  also proposed that release of the nSH2 domain constraint promotes a concerted conformational change that propagates from the helical domain and may effect changes in the conformation of the activation loop and C-terminal region rendering them lipid-binding-competent.
Conformational differences between the wild-type and oncogenic mutant proteins also provide opportunity for altered interactions with other proteins as reported recently for the E545K mutant and the IRS1 (insulin receptor substrate 1) protein . Here, the oncogenic mutation facilitated a p110α–IRS1 interaction independent of p85 binding to IRS1 phosphotyrosine amino acids and may contribute to the oncogenic effect of the E545K mutation.
Since PI3K isoforms play such an important role in signalling pathways, it is not surprising that drugs targeting all isoforms have significant side effects [30,31]. This has led to efforts to develop isoform-selective inhibitors in the hope that these will have greater therapeutic index.
Few p110α-specific compounds have been reported to date, with the aminothiazole-based molecules developed by Novartis, including A66  and the closely related NVP-BYL719 , being the most widely studied selective compounds [15,33,34]. The latter compound entered clinical trials in 2010 [21,34]. In the absence of protein structure information for these compounds bound to p110α, molecular docking and analogue synthesis was used to characterize features of A66 that control its isoform-specificity. Selectivity and potency for p110α and its H1047R and E545K oncogenic forms was associated with the inhibitors’ carboxamide group and a predicted hydrogen bond with the p110α-specific amino acid Gln859 [21,33]. Site-directed mutagenesis illustrated the role of Gln859 as the main ATP-binding site residue influencing A66 activity , and a crystal structure of BYL719 bound to the p110α enzyme confirmed the predicted electrostatic interaction . Identifying interactions that influence the p110α-selectivity of PIK75  and related compounds  has also been the subject of medicinal chemistry , molecular modelling [39–41] and biochemical studies [35,42] and is likely to involve the p110α-specific amino acid Ser773, although other residues may play a role . Nacht et al.  reported the first p110α-specific irreversible inhibitor on the basis of the chemical scaffold of GDC-0941 that covalently modifies p110α-specific Cys862.
In contrast with the electrostatic interaction underlying the p110α-selectivity of A66 and BYL719, p110δ-specific inhibitors including IC87114 and PIK-39 achieve isoform-specificity by steric effects. These ligands have a ‘propeller’ shape in the p110δ ATP-binding site  and require formation of an allosteric pocket termed the specificity pocket. This pocket is created by Met752 side chain rearrangement in the ATP-binding site , and is also formed in p110γ with PIK-39 bound  and in p110α with PIK-108 bound  by rearrangement of analogous methionine amino acids. Differences in flexibility of this region is the likely controller of selectivity as indicated by MD simulations  and site-directed mutagenesis studies . Not all p110δ-specific inhibitors use the specificity pocket. The inhibitor AS15 was shown to make p110δ-specific interactions in a site adjacent to Met752 with its side chain in a more conventional conformation .
The discovery of p110α-selective inhibitors improves the pharmacological tool set that can be used to investigate specific PI3K isoform signalling in cell or animal systems. We used p110α-selective inhibitors A66 and PIK75 to probe the role of p110α in PI3K signalling through the PI3K–Akt/PKB–mTORC pathway in a panel of 12 cancer cell lines of different tissue origin including low-passage-number melanoma cell lines . These studies found that, upon insulin stimulation after serum starvation, the cell lines could be classified as sensitive or resistant to drug treatment on the basis of Akt/PKB phosphorylation, a proximal marker of pathway activation. Sensitive cells harboured the p110α H1047R mutation, whereas resistant cell lines harboured the E545K mutation, or were PTEN-null, implicating mutation-dependent differences in p110α signalling from the insulin growth factor receptor. Combinations of p110α-, p110β- and p110δ-specific inhibitors were able to reduce Akt/PKB phosphorylation in PIK75-resistant cells, pointing to residual phosphorylation by isoforms other than the p110α, and functional redundancy reported previously for insulin signalling . Sensitivity may in part relate to the total levels of H1047R oncogenic protein expressed and its levels relative to other p110 isotypes. Drug sensitivity or resistance was retained by tumour xenografts in mice, and growth of tumours from A66-sensitive cell lines was restricted by drug treatment, and corresponded to decreased activity of the PI3K–Akt/PKB–mTORC pathway, as indicated by decreased phosphorylation of Akt/PKB and pS6 (p70 S6 kinase), markers of proximal and distal pathway activation respectively .
Resistance to PI3K inhibitors can arise from activation of downstream components of the PI3K signalling pathway which can partially compensate for the loss of PI3K activity. Elkabets et al.  found that across a panel of 20 breast cancer cell lines harbouring a range of p110α mutations, including the H1047R and E545K, some were sensitive and others resistant to p110α-specific inhibition by BYL719 with respect to proliferation. Resistance both in vitro and in vivo was related to persistent mTORC1 signalling indicated by phosphorylation of the distal PI3K pathway marker pS6, whereas phosphorylation of Akt/PKB, remained sensitive to p110α blockade. Notably, resistance to p110α inhibition by BYL719 and non-selective PI3K inhibition by GDC-0941 could be acquired in vitro by reactivation of mTORC1 signalling and increased S6 phosphorylation after long-term drug exposure. In tumour biopsies from patients that responded to BYL719 treatment, strong suppression of S6 phosphorylation was observed, whereas tumours that did not respond showed only weak suppression. Moreover, in patients with a BYL719-sensitive tumour that progressed, S6 phosphorylation was increased. Resistance could be suppressed in cell lines and xenografts by combining BYL719 with the mTORC1 inhibitor RAD001 .
p110α inhibition in cancer
Many compounds under clinical development are dual class I PI3K and mTOR inhibitors (reviewed recently in [10,48]), although the p110α-specific BYL719 was reported to show promising clinical activity . Recent analysis of clinical trials of mTORC1 and PI3K inhibitors has shown an association between the H1047R mutation and a positive response, and is in line with our pre-clinical findings, although it was noted that specifically designed clinical studies are needed to explore this observation .
It also seems likely that p110α inhibition will have a range of side effects. Animal studies have shown that A66 induces impairment of in vivo insulin action indicating that p110α is the most important form of PI3K in the pathways acutely regulating glucose metabolism [30,31]. Furthermore, detrimental effects on bone structure after 1 month of dosing mice with A66 suggested that osteoporosis might be a side effect of treatment with p110α inhibitors. These effects may limit the therapeutic window for a standard p110α-selective inhibitor, and taken together with the possibility of mutation-specific responses indicate that new strategies that target cancer-specific p110α signalling or that increase relative drug concentration selectively in the tumour might need to be employed.
Future directions for drug discovery
Since oncogenic mutations mimic or enhance different activation steps of the p110α enzyme, exploiting the conformational differences may lead to novel inhibitors with some selectivity for the oncogenic mutant forms, and these might have a broader therapeutic window. Indeed, very recent reports indicate that this may be possible . Also, the recent report of a p110α-selective irreversible inhibitor indicates the potential to develop PI3K inhibitors that might achieve better therapeutic responses by more persistent knockdown of tumour signalling.
Finally, another area of interest in selectively targeting oncogenic PI3K signalling might be in disrupting the interactions between p110 catalytic subunits and the small GTPases Rac or Ras, or Gβγ subunits, as emerging evidence indicates that isoform-selective responses could be generated . It was shown recently that p110β/Gβγ signalling could be blocked in cells using a peptide targeting the interaction .
In conclusion, the overall aim of the next wave of drug discovery in this area will be to explore these new possibilities to determine whether such second-generation PI3K inhibitors might be more therapeutically advantageous than the first-generation ones.
Signalling 2013: from Structure to Function: A Biochemical Society Focused Meeting held at the University of York, U.K., 17–19 June 2013. Organized and Edited by Nicholas Brindle (University of Leicester, U.K.), Sandip Patel (University College London, U.K.) and Stephen Yarwood (University of Glasgow, U.K.)
insulin receptor substrate 1
mammalian target of rapamycin
protein kinase B
p70 S6 kinase
phosphatase and tensin homologue deleted on chromosome 10
receptor tyrosine kinase
Src homology 2
We thank Roger Williams and John Burke for comments on the paper before submission.
This work was supported by the Health Research Council of New Zealand [programme grant number 13/763] along with the Maurice Wilkins Centre for Biodiscovery.