PIK3CA C2 Domain Deletions Hyperactivate Phosphoinositide 3-kinase (PI3K), Generate Oncogene Dependence, and Are Exquisitely Sensitive to PI3Kα Inhibitors

PIK3CA is the most mutated gene in breast cancer with an overall rate of approximately 40%. More than 80% of these mutations are located in “hotspots” within exons 9 and 20 in the helical and kinase domain, respectively (1, 2). The remainder of PIK3CA mutations are widely distributed over its entire coding sequence (3). PIK3CA encodes the p110α catalytic subunit, which dimerizes with the p85α regulatory subunit and induces the formation of the second messenger phosphatidylinositol (3,4,5)-trisphosphate (PIP3; ref. 4). Hotspot mutations in PIK3CA have been shown to partially destabilize the association of p110α with p85α and, thus, relieve the enzymatic activity of p110α from p85-mediated inhibition (5–7). This results in increased residence time of p110α at the plasma membrane, enhanced formation of PIP3, and subsequent hyperactivation of AKT/mTOR signaling, resulting in growth factor–independent growth, cellular transformation, invasion, and metastases (8–10).

Early studies showed that mutations in the C2 domain make up less than 5% of PIK3CA mutations and also enhance kinase activity (3). However, more recent next-generation sequencing studies suggest that C2 domain mutations make up greater than 10% of PIK3CA mutations (Supplementary Table S1). Rudd and colleagues first reported PIK3CA^delP447_L455 in ovarian cancer and observed that it was associated with higher levels of phosphorylated AKT than the more common C2 domain missense point mutation E453K (11). Although deletions occur in only 3% of PIK3CA mutations, a significant portion of these deletions occur in a small range of critical residues within the C2 domain. This is equivalent to approximately 1,100 new breast cancers per year. Herein, we report that deletions in the C2 domain of PIK3CA result in disruption of p85α binding, increased p110α activity, and cellular transformation. As a result, cells harboring these alterations are highly vulnerable to treatment with PI3K inhibitors. These data have important implications for the selection of patients for trials with therapies targeted to the PI3K/AKT pathway.

MCF10A breast epithelial cells (ATCC CRL-10317; purchased in 2012) were from ATCC. Cell lines were authenticated by ATCC prior to purchase by the short tandem repeat (STR) method. Isogenically modified MCF10A cells containing PIK3CA E545K or H1047R were developed previously (12). The Gateway cloning system (Thermo Fisher Scientific) was used to generate pLX302-PIK3CA plasmids encoding PIK3CA^WT and the two C2 deletions, PIK3CA^P447_L455del and PIK3CA^H450_P458del. Transfected MCF10A cells were maintained in MCF10A complete media (DMEM/F12 supplemented with 5% horse serum, 20 ng/mL EGF, 10 μg/mL insulin, 0.5 μg/mL hydrocortisone, 0.1 μg/mL cholera toxin, and 1× penicillin/streptomycin). For experiments, 2% charcoal/dextran-stripped serum (CSS) serum was used, and EGF and insulin were removed as indicated. Experiments were carried out on cells with passage numbers below 20.

The pDONR-223 vector encoding PIK3CA^WT was subjected to site-directed mutagenesis (Genewiz) to generate PIK3CA^P447_L455del and PIK3CA^H450_P458del. The PIK3CA^WT and mutant plasmids were recombined into the lentiviral expression vector pLX-302 containing a C-terminal V5 epitope tag. Lentiviral supernatant was produced in early-passage 293FT cells by transfection with psPAX2 and pMD2.G packaging plasmids along with the generated pLX302-PIK3CA plasmids. Viral supernatant was filtered and applied to MCF10A cells in the presence of polybrene. After 48 hours, cells were selected with 1.5 μg/mL puromycin and tested for plasmid expression via immunoblot analysis with a V5 antibody. Stable cell lines were maintained in media containing 1 μg/mL puromycin.

The frequency of PIK3CA C2 deletions and mutations was determined using cBioPortal (ref. 13; including METABRIC and Genie), COSMIC (14), and databases from Foundation Medicine and Guardant Health.

Structural modeling of the PIK3CA^P447_L455del was performed using Rosetta. The experimental crystal structure complex of p110α and p85α (PDBID: 3HIZ) was used to examine the binding interface of wild-type proteins and deletion mutants. The N-terminal expression tag was removed prior to modeling and all mutated residues were reverted to wild-type. Finally, the structure was refined with Rosetta FastRelax protocol while constraining backbone atoms to their crystallographic coordinates. To model deletion mutants, residues corresponding to 447–455 (447del) and 450–458 (450del) were removed; resulting chain breaks were fused with cyclic coordinate descent (CCD) followed by kinematic refinement in Rosetta. All conditions were refined once more with constraints to starting coordinates with 3,000 trajectories, and a final ensemble of 1,000 models was selected on the basis of overall pose energy. To quantitate changes in interface energy between the three conditions, the dG_separated (change in energy between isolated and chains in complex) was calculated with the Interface Analyzer application; the top 10 models per condition by dG_separated were compared. To identify specific interactions lost in the deletions, Residue Energy Breakdown was run for all models. Residue pairs contributing to at least 0.1 Rosetta Energy Units (REU) across the interface in the wild-type, and not in a deletion mutant, were considered as potentially lost interactions.

Cells were washed with PBS and lysed on ice in NP-40 lysis buffer plus protease and phosphatase inhibitors. Protein concentrations were measured using BCA protein assay reagent (Pierce). Lysates were subjected to SDS-PAGE and transferred to nitrocellulose membranes (Invitrogen) as described previously (15). Immunoreactive bands were detected by enhanced chemiluminescence following incubation with horseradish peroxidase–conjugated secondary antibodies. Primary antibodies included V5 Tag (CS13202), p110α (CS4249), p85 (CS4257), p-p70S6K (CS9205), p90RSK (CS9333), PathScan Multiplex Western Cocktail I (for p-AKT S473, p-p90RSK, p-ERK, and p-S6), and GAPDH. The Invitrogen NuPage system was used. Nitrocellulose membranes were cut horizontally to probe with multiple antibodies.

For dose–response curves to generate an IC₅₀, cells were seeded in 96-well plates at a density of 200 cells per well in growth factor–depleted media. The next day, media were replaced with media containing 1 μmol/L alpelisib ± EGF and/or insulin as indicated. Media and growth factors were changed every 3 days. Cells were stained using crystal violet, resuspended in 1% SDS, and imaged using a GloMax plate reader. Additional growth assays were carried out in 12-well dishes at a seeding density of 2,000 cells/well. Growth assays were quantified using a LI-COR Odyssey Infrared plate reader. For three-dimensional (3D) growth assays, cells were seeded on growth factor–reduced Matrigel (BD Biosciences) in 8-well chamber slides as described previously (11).

All experiments were performed using three technical replicates and at least two independent times. P values were calculated using GraphPad Prism (version 6.0) by ANOVA followed by Tukey multiple comparisons test.

Cell lysates were harvested using ice-cold lysis buffer (1% Triton X-100, 10% glycerol, 100 mmol/L NaCL, 50 mmol/L HEPES, 100 mmol/L Na₃VO₄, 10 mmol/L NaF, 1 Roche Minitab) and rotated at 4°C for 1 hour. Lysates were then clarified by spinning at 10,000 × g at 4°C for 15 minutes. Protein concentrations were measured using BCA standard curves (Pierce). One microgram anti V5-Tag antibody (Cell Signaling Technology, 13202) was added to 1 mg of clarified cell lysate and rotated at 4°C overnight. Fifty microliters of Protein G agarose beads (Life Technologies 10003D) was added to the lysates and rotated at 4°C for 4 hours. Tubes were placed on magnets, washed three times with lysis buffer, and eluted in 20 μL of LDS sample buffer. Lysates were subjected to immunoblot analysis with a p85 antibody (Cell Signaling Technology, 4257). Unbound lysate was analyzed via Western blot for loading controls. Immunoblots were analyzed using ImageJ software.

Approximately 10⁷ cells were resuspended in 500 μL of PBS with Roche Complete protease inhibitor tab and subjected to three freeze/thaw cycles. Whole-cell lysates were aliquoted into PCR strips and exposed to a thermal gradient ranging from 44.5°C to 67°C. Tubes were spun down at 20,000 × g for 20 minutes, and supernatants were collected followed by separation by SDS-PAGE.

A 63-year-old postmenopausal woman with advanced estrogen receptor (ER)⁺ breast cancer resistant to endocrine therapy exhibited an excellent and sustained clinical response (11 months) to the PI3Kα inhibitor alpelisib (BYL719) in combination with the aromatase inhibitor letrozole (16). Targeted capture next-generation sequencing (NGS) of DNA from a liver metastasis identified a P447_L455 deletion in the C2 domain of PIK3CA with an allele frequency of 11% (Fig. 1A). PIK3CA C2 domain mutations have been previously under-reported, and deletions are frequently not reported by tumor and plasma cell–free DNA NGS panels. Interrogation of cBioPortal, Metabric, Genie, Foundation Medicine, and Guardant Health suggested that C2 domain mutations make up approximately 11% of all PIK3CA mutations, which is much higher than previously thought (Supplementary Table S1–S4; ref. 3). Unlike PIK3CA missense mutations, which are spread across the entire gene, PIK3CA deletions cluster in two specific regions, the p85α-binding domain and the C2 domain (Fig. 1B). Both of these regions are responsible for the contacts of p110α with its regulatory subunit, p85α.

To understand how this deletion would affect the binding to p85α, we performed computational structural modeling of the dimeric complex with PIK3CA^P447_L455 using Rosetta, a computerized algorithm to model protein mutations and their interactions (17). We calculated the predicted change in free energy. The model was constructed from the X-ray crystal structure of the p110α/p85α complex (18). The deletion removes nine residues of p110α that form a loop in direct contact with p85α, resulting in the loss of six specific strong interactions between these two molecules (Fig. 1C). Computational structural analysis determined four residues within the C2 domain deletion (Pro449, His450, Glu453, and Leu455) that significantly affected the interaction with p85α (Table 1). These predicted interactions include hydrogen bonds and salt bridges that we hypothesize disrupt the binding of p110α to the iSH2 domain of p85α, thus altering the regulatory effect of p85α on p110α.

To determine whether the C2 domain deletion, PIK3CA^P447_L455del, affects the activity of p110α through its interaction with p85α, we stably transduced MCF10A breast epithelial cells with lentiviral vectors encoding PIK3CA wild-type (WT), PIK3CA^P447_L455del (447del), and an additional deletion found in cbioportal, PIK3CA^H450_P458del (450del) (Fig. 2A). Through sequencing, PCR and immunoblot analysis, we were able to confirm the 27 base-pair cDNA and 9 amino acid shift, respectively, due to each deletion (Fig. 2B). To quantify the structural consequences of the P447_L455 deletion on binding to p85α, the average computed binding energy for the best 10 out of 3,000 computational models was calculated for the deletion mutant, and compared with wild-type p110α. The deletion of the four previously mentioned residues led to the loss of seven interactions as indicated by an average change of greater than 0.1 Rosetta Energy Units (REU; Fig. 2C). Three of these interactions (His450-Asp464, His450-Tyr467, and Glu453-Lys567) had an average change greater than 1.0 REU with strong interaction energy terms indicating the loss of electrostatic forces and hydrogen bonds (His450-Asp464 and Glu453-Lys567). Both PIK3CA^P447_L455del and PIK3CA^H450_P458del showed a statistically significant decrease in binding energy to p85α (Fig. 2D). When compared with PIK3CA^WT, PIK3CA^P447_L455del showed a modest decrease in protein stability, while PIK3CA^P450_458del exhibited a statistically significant decrease (Fig. 2E).

Thus, we stably transduced MCF10A cells with V5-tagged PIK3CA^WT, PIK3CA^P447_L455del, and PIK3CA^P450_458del. We confirmed equal expression of the V5-tagged constructs by immunoblot analysis (Supplementary Fig. S1). Cells transfected with PIK3CA^H450_P458del, or “450del,” exhibited slightly lower levels of the mutant protein when compared with PIK3CA^P447_L455del. We speculate this may be due to the decrease in protein stability suggested by the previously discussed computational calculations. To support the predicted decrease in contacts between each PIK3CA^P447_L455del and PIK3CA^H450_P458del with p85α, we coprecipitated p85α with V5 antibodies from lysates of stably transduced MCF10A cells. Both MCF10A cells expressing the C2 domain deletions showed a reduction in p85α binding when compared with PIK3CA^WT (Fig. 2F). Quantification of this reduction is show in Fig. 2G. To confirm the reduction in p85α binding, a thermal shift assay was performed in which cell lysates were subjected to a gradient of temperatures ranging from 44.5°C to 67°C. As the temperature increases, the p110α–p85α complex becomes less stable, denatures, and degrades. Both C2 deletions degraded at a lower temperature, suggesting that the p110α-p85 complex is less stable in intact cells compared with WT PIK3CA (Supplementary Fig. S4).

MCF10A cells require EGF and insulin to propagate. Previous studies demonstrated that stable transduction with an activating PIK3CA mutation results in EGF- and/or insulin-independent growth. The ability of MCF10A cells to proliferate in the absence of growth factors has been used as an experimental surrogate of cellular transformation (12). In monolayer growth (2D) assays, MCF10A cells transfected with the PIK3CA C2 domain mutations exhibited a growth advantage when compared with parental MCF10A cells and to a lesser extent with MCF10A/PIK3CA^WT cells (Fig. 3B–E). PIK3CA is an oncogene; therefore, overexpression of the N-terminal tagged wild-type protein is also expected to induce EGF-independent growth. However, upon removal of both EGF and insulin, only cells expressing the C2 domain mutations were able to grow (Fig. 3D and E). Cells with each of the C2 deletions exhibited similar growth phenotypes when compared with isogenic cells expressing PIK3CA E545K, and PIK3CA H1047R. In the absence of both EGF and insulin, cells expressing H1047R exhibited decreased growth when compared with cells with either a PIK3CA C2 deletion or E545K. Previous studies have shown that E545K weakens the p85α–p110α interaction, thus removing inhibition of p110α and allowing it to remain active in the absence of both ligands (19). In contrast, cells with PIK3CA H1047R remain dependent on insulin for its activation. Similar to cells expressing either E545K and H1047R hotspot mutations (9), MCF10A cells expressing PIK3CA^P447_L455del or PIK3CA^H450_P458del showed invasive branching and irregular morphology, consistent with a transformed phenotype. Upon the removal of EGF, only cells harboring the C2 deletions were capable of forming acini (Fig. 3F). We next examined activation of the PI3K/AKT/TOR pathway in cells expressing the PIK3CA C2 deletions (20, 21). Cells expressing PIK3CA^P447_L455del or PIK3CA^H450_P458del showed elevated levels of phosphorylated AKT at both the S473 and T308 sites in the absence of EGF and/or insulin (Fig. 3G). In addition, they showed elevated levels of p70S6K and p90RSK and slightly elevated levels of pERK. Collectively, these results suggest that PIK3CA C2 domain deletions hyperactivate the PI3K/AKT/TOR pathway and induce cell transformation.

The excellent clinical response to treatment with a PI3Kα inhibitor that was observed in the patient with breast cancer harboring PIK3CA^P447_L455del (Fig. 1A) suggested such tumor was at least in part dependent on the oncogenic role of the C2 domain deletion mutant. Thus, we next examined the effect of alpelisib over a dose range on MCF10A cells expressing PIK3CA^P447_L455del and PIK3CA^H450_P458del. MCF10A cell viability over a dose range of alpelisib showed that cells with the PIK3CA C2 deletions were more susceptible to the PI3K inhibitor when compared with MCF10A/PIK3CA^WT control cells in the absence of EGF ± insulin (Fig. 4A and B). These studies were conducted in the two conditions in which the C2 deletions showed the largest growth advantage (absence of EGF and absence of both EGF and insulin). Parental MCF10A cells were not included as they did not grow in the absence of EGF. Both C2 domain deletions exhibited complete growth inhibition in the presence of 1 μmol/L alpelisib (Fig. 4C; Supplementary Fig. S5). Similarly, growth of cells expressing PIK3CA hotspot mutations was markedly inhibited upon treatment with 1 μmol/L alpelisib. In 3D Matrigel containing EGF and insulin, treatment with 1 μmol/L alpelisib reduced the invasiveness of acini expressing the PIK3CA C2 deletions (Fig. 4D, first row). In the absence of EGF and/or insulin, the addition of alpelisib completely ablated acinar formation induced by the PIK3CA C2 deletions (Fig. 4D, fourth row). A decrease in p110α activity in the C2 deletions in the presence of alpelisib was confirmed by immunoblot analyses. Almost a complete loss of AKT phosphorylation in the presence of both EGF and insulin was observed for all four cell lines (Fig. 4E).

Following an excellent clinical response to a PI3Kα inhibitor in a patient with breast cancer harboring a PIK3CA delP447_L455 deletion, we report herein the function of this mutant and its implications for other alterations in the C2 domain of PIK3CA. Although previous studies suggested C2 domain mutations comprise less than 5% of PIK3CA mutations (3), the databases we report here suggest that these mutations make up over 10% of all PIK3CA mutations. The region around the deletion observed in the patient's cancer involves residues that are within a significant portion of deletions reported in PIK3CA (Supplementary Fig. S2). Scanning various databases, we noticed that deletions in PIK3CA cluster in two regions, the p85 binding domain and C2 domain, both involved in binding to the p85α regulatory subunit (Fig. 1B). This nonrandom distribution suggests there is a selective advantage to the deletion of these two regions in PIK3CA. Our interrogation of the p110α–p85α interface revealed that the residues deleted in the patient's tumor strongly interacted with residues in the iSH2 domain of p85α, specifically clustering around p85α residues Y452-E462 and N564-M582 (Supplementary Fig. S3). When analyzing these sites of interaction, the corresponding residues binding to the amino acids within the p110α C2 domain deletion are the most frequently mutated or deleted sites (amino acids Y452-E469 and N564-M582) in p85α, and are mutually exclusive with the C2 domain deletions (Supplementary Fig. S3; ref. 22).

On the basis of these data, we hypothesized that deletions in the PIK3CA C2 domain are gain-of-function by reducing contacts with p85 and, thus, relieving p85α-mediated inhibition. Other studies have suggested that mutations in the PIK3CA C2 domain changed the binding affinity of p110α with the lipid membrane (23, 24). However, more recent structural studies of the commonly mutated C2 residue, Asn345, suggest mutations in this region would alter the interaction of p110α with p85α (6). Through computational analysis, this study also noted that Glu453, a residue with prominent interactions within the deleted region, was located at the interface between PIK3CA C2 and p85 iSH2; however, the density for the side chains was not defined enough at the time to identify direct interactions. Furthermore, residues 581–593 of p85α (those found at the C2 deletion interface) have been proposed to constrain the location of the iSH2 domain of p85α and deletion of these residues removes these orientation constraints (25). This further suggested that mutations and, more specifically, deletions in the C2 domain directly disrupt the ability of p85α to regulate p110α. Indeed, computational structural modeling revealed that the deletion of select residues within the PIK3CA C2 domain, specifically 449–455, caused a significant decrease in the ability of p85α to bind p110α (Fig. 2D; Table 1). Consistent with a reduction in p110α–p85α binding, V5 pulldowns showed an observable decrease in the coprecipitated p85α in the PIK3CA C2 deletion mutants (Fig. 2F and G). In addition, both C2 deletions exhibited a decrease in their thermal stability of the p110α–p85 complex in intact cells (Supplementary Fig. S4). These results support a reduction in the contacts between p110α and p85α but not a complete dissociation of the dimer, as it is well established that p110α is unlikely to survive as a monomer in vivo (26).

Our studies also suggest that deletions in the PIK3CA C2 domain are activating, generate dependence on PI3K, and are exquisitely sensitive to PI3Kα inhibitors currently in clinical development. A reduction in p85α-mediated repression results in upregulation of the catalytic activity of p110α with subsequent enhanced activation of AKT/TOR signaling and ligand-independent growth. It was previously shown that p85α can be a critical modulator of insulin sensitivity and that changes in the stoichiometric balance of p85α and p110α can directly affect PI3K-dependent signaling and response to growth factor stimuli (27, 28). It is tempting to speculate that the PIK3CA C2 deletions may be causal to the observed insulin-independent growth of transfected MCF10A cells (Fig. 3D and E). Furthermore, the C2 deletions exhibit a comparable phenotype with the PIK3CA helical hotspot mutation, E545K, which has been shown to weaken the p85-p110α and, as a result, is less dependent on insulin-induced activation (19). In contrast, H1047R does not weaken the p85–p110α interface and, therefore, remains dependent on insulin for its activation (Fig. 3D and E).

Currently, several targeted capture NGS panels do not systematically examine deletions in PIK3CA in tumor DNA extracted from tissue biopsies and/or plasma. Data shown here suggest that C2 domain mutations and deletions are more prominent than previously determined (Supplementary Table S1 and S2; ref. 3). One implication of these findings is that PIK3CA C2 deletions may make up a portion of patients with cancers previously categorized as PIK3CA wild-type. Therefore, we propose that, in addition to hotspot PIK3CA mutations, both deletions and mutations in the C2 domain of PIK3CA should be included in comprehensive tumor gene panels so that patients with PI3K-dependent cancers and, thus, potential sensitivity to PI3K inhibitors can be identified.

J. He is an employee of Foundation Medicine. R.B. Lanman is an employee of and holds ownership interest (including patents) in Guardant Health, Inc. V. Miller is an employee of and holds ownership interest (including patents) in Foundation Medicine. L.C. Cantley reports receiving commercial research grants from, is a consultant/advisory board member for, and holds ownership interest (including patents) in Petra Pharmaceuticals. C.L. Arteaga is a member of an advisory board for Novartis and a member of the scientific advisory board for the Komen Foundation. No potential conflicts of interest were disclosed by the other authors.

Conception and design: S. Croessmann, J.M. Balko, L.C. Cantley, J. Meiler, C.L. Arteaga

Development of methodology: S. Croessmann, J.M. Balko, R.B. Lanman, J. Meiler

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Croessmann, K.-M. Lee, R.J. Nagy, D.A. Riddle, I.A. Mayer, R.B. Lanman, V. Miller, C.L. Arteaga

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Croessmann, J.H. Sheehan, G.R. Sliwoski, J. He, J.M. Balko, J. Meiler, C.L. Arteaga

Writing, review, and/or revision of the manuscript: S. Croessmann, J.H. Sheehan, R.J. Nagy, D.A. Riddle, I.A. Mayer, J.M. Balko, R.B. Lanman, V. Miller, L.C. Cantley, J. Meiler, C.L. Arteaga

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Croessmann, R.J. Nagy

Study supervision: S. Croessmann, I.A. Mayer, J. Meiler, C.L. Arteaga

This study was funded by NIH Breast SPORE grantP50 CA098131, Vanderbilt-Ingram Cancer Center Support grantP30 CA68485, Susan G. Komen for the Cure Foundation grantSAC100013 (to C.L. Artega), grants from the Breast Cancer Research Foundation (to C.L. Artega and L.C. Cantley), R01 GM041890 (to L.C. Cantley), R35 CA197588 (to L.C. Cantley), and U54 CA210184. J.M. Balko is supported by NIH/NCI 4R00 CA181491 and Susan G. Komen Career Catalyst Research awardCCR 299052.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.