A new functional assay reveals that membrane binding is critical for overactivation of the phosphoinositide 3-kinase H1047R mutant

Abstract

PIK3CA encodes the catalytic subunit of class I PI3Kα (p110α), a key enzyme in receptor-mediated signaling that phosphorylates phosphatidylinositol-4,5-bisphosphate to the 3,4,5-triphosphate lipid. This gene is frequently mutated in cancers, with H1047R and E545K being the most prevalent mutations. However, the mechanisms of mutants’ overactivation remain incompletely understood. Here, we report the development of a PI3Kα activity assay using reconstituted liposomes, which we employed to address the role of lipid membranes on mutant overactivation. The assay was validated by assessing typical enzymatic features and by confirming the IC₅₀ of well-known inhibitors and the activation by a phosphopeptide mimicking receptor-mediated stimulation. Additionally, we examined the impact of lipid membranes on mutant overactivation by comparing WT and mutant activities with soluble or liposomal PIP₂. Interestingly, we discover that the membrane form of the substrate is crucial for the catalytic overactivation of H1047R, whereas E545K overactivation occurs independently of membranes. Consistently, molecular dynamics simulations of ΔABD p110α on a model membrane, performed for WT and H1047R, revealed structural and dynamic changes induced by H1047R, including stabilization of the C-terminal tail on the membrane and altered dynamics of membrane-binding loop 2 residues, suggesting enhanced membrane binding. Intriguingly, in agreement with the above findings, surface plasmon resonance assays reveal higher rate of association of H1047R-PI3Kα with the membrane. Altogether, our data suggest that the overactivation of the H1047R mutant is due to increased rate of membrane binding, providing novel mechanistic insights into how this hot spot mutation leads to catalytic overactivation and contributes to oncogenesis.

PI3Ks are a family of lipid kinases that phosphorylate the hydroxyl group at position 3 of the inositol ring of phosphatidylinositols (PtdIns), in response to extracellular stimuli. These enzymes are regulated by a complex network of interactions (1, 2, 3, 4, 5, 6, 7), controlling tightly the phosphorylated enzymatic products, which in turn are responsible for diverse cellular functions (5, 8, 9, 10, 11). Class I PI3Ks, the most widely implicated in cancer (12, 13), phosphorylate, mainly, phosphatidylinositol-4,5-biphospate (PIP₂) to generate phosphatidylinositol-3,4,5-triphosphate (PIP₃) (4), a reaction that takes place mostly at the plasma membrane (14, 15). PIP₃ acts as a second lipid messenger that recruits effector proteins at the membrane, thus regulating signaling cascades that control cellular processes such as cell proliferation, differentiation, cell motility, and apoptosis (16, 17, 18).

PI3Kα, the prototype of the class I family, consists of p110α catalytic and p85α regulatory subunit, which exerts an inhibitory effect on p110α. PI3Kα is recruited to activated RTKs, resulting in a conformational change on p85α that relieves the inhibitory action on the catalytic subunit p110α, and thereby activating PI3Kα (19, 20, 21, 22, 23). The p110α catalytic subunit of PI3Kα is encoded by PIK3CA, one of the most frequently mutated genes in human cancers (24, 25, 26). Although mutations occur throughout PIK3CA, three specific hotspot mutations constitute more than 70% of cancer mutations (24, 26). These include the E542K and E545K mutations, located in the helical domain encoded by exon 9, and the H1047R mutation in the kinase domain encoded by exon 20 (24, 26, 27). It has been shown that these activating mutations, which result in higher production of PIP₃, lead to increased downstream phosphorylation of PDK1 and Akt, thereby promoting carcinogenesis and metastasis (24, 28, 29, 30, 31, 32). Thus, mutated PI3Kα is considered a significant target for cancer therapies. Yet, in order to develop mutant-specific therapies, it is essential to understand how these somatic mutations activate the oncogenic potential of p110α. The hot spot E545K and H1047R mutations operate by two different and independent mechanisms (33, 34, 35). E545K mutation releases the inhibitory role of p85α regulatory subunit, thereby resulting in the constitutive activation of p110α (36, 37, 38). On the other hand, interestingly, the H1047R mutant, similar to the WT enzyme, is regulated by p85α, while it is also believed that the mutation causes catalytic overactivation by enhancing the interaction with the plasma membrane (35, 39, 40, 41). However, the precise relationship between the conformational changes caused by hotspot mutations and membrane association, and how they influence the catalytic activity of p110α, has yet to be fully elucidated. To understand the mechanism of activation of PI3Kα mutants, a prerequisite for the discovery of mutant-specific inhibitors, it is critical to develop enzymatic assays assessing quantitatively and faithfully the activity of PI3Kα.

Early assays for PI3Kα activity primarily relied on chromatographic separation techniques using thin layer chromatography. Many well-known pan-PI3K inhibitors, such as Wortmannin and LY294002, were identified by this approach. Although thin layer chromatography and other radioactivity-based assays, such as capture on nitrocellulose membrane, provide high sensitivity and precise quantification of the product, they come with notable limitations, including the use of hazardous reagents, labor-intensive and time-consuming procedures, and restricted throughput capacity (42). Later studies developed more elaborated and high-end technology methods that are based on fluorescence polarization, homogeneous time-resolved fluorescence, FRET, and luminescence. Although these approaches are sensitive and prone to high-throughput screenings, they come with increased reagent costs and the need for specialized and costly equipment (42, 43, 44, 45).

To overcome the limitations of previous methods, and to additionally include the membrane environment of PI3Kα, we sought to develop a new cell-free assay that assesses quantitatively the enzymatic activity of PI3Kα, using as substrate PIP₂ incorporated into membranes, thus faithfully recapitulating the properties of the membrane-bound enzyme. Our assay is simple, rapid, inexpensive, robust, and applicable to high-throughput screening, using in-house prepared components, chemically defined reagents and standard laboratory equipment. Using this method, as well as molecular dynamics (MD) simulations and surface plasmon resonance (SPR) experiments, we provide data suggesting that the mechanism of overactivation of the H1047R mutant arises from its enhanced membrane-binding rate, underlining the critical role of the enzyme–membrane interface in the oncogenic potential of PI3Kα. In turn, these findings highlight the need for using a membrane-based assay when evaluating/developing inhibitors of PI3Kα Η1047R mutants instead of a soluble PIP₂ assay, as the protein–membrane interface plays a crucial role in the design of mutant-specific inhibitors.

Results

Principle of the nοvel PI3Kα activity assay

To develop a membrane-based, cell-free assay of PI3Kα activity assay, we took advantage of the property of the pleckstrin homology (PH) domain of GRP1 (general receptor for phosphoinositides) to bind the PIP₃ product of the PI3Kα reaction with very high selectivity (46, 47, 48). Thus, we used the PH segment of GRP1 fused with glutathione-S-transferase (GST) (Lindsay et al., 2006), in order to immobilize this tagged reagent on the walls of 96-well plates coated with glutathione (Fig. 1, A and B), to capture the PIP₃ product on the plate. Subsequent steps of the assay aim at the quantification of the amount of bound PIP₃, thus resulting in assessment of the PI3Kα activity. More specifically, the main steps of the assay, in the absence (negative control) or the presence of PI3Kα, are listed in Figure 1A, while Figure 1B shows schematically the principle of the assay and the interactions that take place. Briefly, in the first step of the assay, GST-GRP1 (PH segment) was bound to glutathione-coated 96-well plates (step 1 in Fig. 1, A and B). In step 2 (consisting of three substeps), the PI3Kα reaction is set up in separate tubes (step 2a) and run for 10 min. After termination of the reactions, biotin-PIP₃ (b-PIP₃) is added to the tubes (step 2b) and the mixture is incubated with the plates (step 2c). In this last substep, the PIP₃ that is generated by the action of PI3Kα competes with the exogenously added b-PIP₃ for binding to the immobilized GST-GRP1 (competition-based assay). Next, horseradish peroxidase–conjugated streptavidin (streptavidin-HRP) is added (step 3) to assess the amount of b-PIP₃ bound to GST-GRP1 (streptavidin interacts strongly with the biotin moiety of b-PIP₃). For spectrophotometric quantification of HRP, the chromogenic substrate o-phenylenediamine (OPD) is then added (step 4), and the chromophore that is generated is measured at 492 nm (step 5). Accordingly, the higher the PI3Kα activity, the greater the amount of enzymatically produced PIP₃, the lower the amount of the b-PIP₃ competitor and its corresponding streptavidin-HRP reporter that is finally bound to the plate, resulting in a lower value of the absorbance at 492 nm. Thus, when candidate inhibitors are tested in this assay, the PIP₃ levels are reduced proportionally to the effectiveness of the inhibitor, which results in a higher binding of b-PIP₃ and corresponding streptavidin-HRP to the wells of the plate, leading to higher absorbance readings at 492 nm. Consequently, the stronger the inhibitor, the higher the absorbance.

Figure 1.

Principle of the PI3Kα activity assay and titration of its basic components GST-GRP1 and b-PIP₃.A, table listing the key steps (1–5) of the assay. B, schematic illustration of the assay’s principle. The number of each individual step of the assay in (B) corresponds to the same number of the step shown in (A) (numbers are shown in red fonts, both in (A) and (B). In step 1, GST-GRP1 is immobilized to the walls of the plate. In step 2, in the absence of PI3Kα (negative control), the plates are incubated with b-PIP₃ alone. In the presence of PI3Kα, the enzymatic reactions are set up in separate tubes (step 2a), b-PIP₃ is added to the tubes post termination of the PI3Kα reaction (step 2b) and the mixture is added to the wells (step 2c), where PIP₃ (generated by PI3Kα) competes with b-PIP₃ (added exogenously) for binding to GST-GRP1. Next, streptavidin-HRP is added (step 3) to assess the amount of bound b-PIP₃, followed by the addition of OPD (step 4), the specific substrate of HRP, used to generate a chromophore measured at 492 nm (step 5). C, the optimal amount of GST-GRP1 in the assay was determined by testing increasing concentrations of this component, followed by addition of 0.9 ng b-PIP₃. Two replicates from n = 5. D, the optimal amount of b-PIP₃ was assessed by adding varying quantities of b-PIP₃ to glutathione-coated wells, preincubated with 4 μg GST-GRP1. Two replicates from n = 5. The individual data points displayed above represent the means of replicates in each experiment. The statistical significance of difference between groups was examined by one-way ANOVA, followed by Tukey’s multiple comparisons test. ns, no significance, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. Data were shown as mean ± SD. GRP, general receptor for phosphoinositide; GST, glutathione-S-transferase; PIP₃, phosphatidylinositol-3,4,5-triphosphate; b-PIP₃, biotin-phosphatidylinositol-3,4,5-triphosphate; HRP, horseradish peroxidase.

Optimization of the PI3Kα activity assay

To set up our PI3Kα activity assay, we titrated all reagents and optimized the conditions, aiming at low noise, maximum dynamic range and specificity. First, recombinant GST-GRP1 was produced in bacteria and affinity-purified (Fig. S1), with a yield of 2 mg/lt of bacterial culture, and then attached to the surface of glutathione-coated multiwell plates. The optimum amount of GST-GRP1 for reliable absorbance reading was determined by adding increasing amounts of GST-GRP1 to the plates, followed by sequential incubation with b-PIP₃, streptavidin-HRP and measurement of HRP activity at 492 nm. As shown in Figure 1C, 4 μg of GST-GRP1 is required to reach sufficiently high absorbance response, within a practically linear range of the absorbance scale, while a higher amount of protein (8 μg) does not substantially increase the signal. Thus, for further experiments, 4 μg of GST-GRP1 was added to each well.

Tο determine the optimum amount of required b-PIP₃, increasing amounts of this reagent were added to the wells, with already bound GST-GRP1. As shown in Figure 1D, increasing levels of b-PIP₃ resulted in steadily higher absorbance values and the quantity of 0.9 ng was selected for subsequent experiments, as it resulted in sufficiently high absorbance value (∼1.5) (Fig. 1D), which is below the maximum level that can be monitored with the ELISA reader used in this study. We chose not to use higher levels of b-PIP3 for two reasons. First, there is no statistical significance between 0.9 and 2.7 ng of b-PIP₃. Second, since the assay relies on the competition between enzymatically produced PIP₃ and exogenous b-PIP₃, higher levels of b-PIP₃ would necessitate larger quantities of PI3Kα and PIP₂, thereby increasing the assay's cost without providing a significant improvement to the method.

We next proceeded with the optimization of the PI3Kα substrate, PIP₂, which is commonly used in a water-soluble form (sPIP₂) with short hydrocarbon side chains diC8PtdIns(4,5)P₂ (44, 49). This form is convenient because of its solubility in aqueous buffers. We thought, however, that it would be advantageous to simulate the physiological microenvironment of the enzymatic reaction at cellular membranes, and sought to employ as substrate the lipid form of PIP₂ (lPIP₂) reconstituted in membrane liposomes (henceforth called lPIP₂ liposomes). Thus, we isolated total cellular lipids from HCT116 cells, a colon cancer cell line that carries the most common H1047R hotspot mutation of PI3Kα (50). The use of liposomes derived from these cells has been shown to be essential for accurately assessing the overactivation of the H1047R mutant relative to WT PI3Kα (39), a key aspect for identifying mutant-specific inhibitors. In addition, the use of this cell line provides a cost-effective lipid source, eliminating the need to purchase multiple individual lipids. To this lipid preparation we added lPIP₂, to produce reconstituted liposomes (lPIP₂ liposomes), in the form of small unilamellar vesicles, at a final lPIP₂ to lipids ratio 1:10 or 1:2. First, it was critical to assess whether the substrate interferes with the assay, because lPIP₂ exhibits weak binding to the PH domain of GRP1 (51, 52), despite the preference (∼60-fold higher affinity) of GRP1 for PIP₃ (46, 47, 52). Indeed, as shown in Figure 2A, increasing amounts of lPIP₂ liposomes caused substantial dose-dependent inhibition of the signal. To prevent this interference of lPIP₂ liposomes due to nonspecific binding, we added detergents to the mixture, prior to incubation with the plates. Among the three tested detergents, the polar detergent sodium deoxycholate was not sufficiently effective, while the nonpolar Triton X-100 and Tween 20 prevented effectively the inhibitory effect of lPIP₂ liposomes (Fig. 2B) in a dose-dependent fashion (Fig. 2C). Based on these data, in all subsequent PI3Kα assays, we added 1% Tween 20, but after termination of the PI3Kα reaction, so that it could not perturb the membrane based PI3Kα enzymatic reaction. The mixture was then added to the GST-GRP1–containing plate.

Figure 2.

The substrate lPIP₂ liposomes interferes with b-PIP₃ for binding to GRP1.A, various amounts of lPIP₂ liposomes along with 0.9 ng b-PIP₃ were added to GST-GRP1 prebound glutathione coated wells, as schematically presented in (D). Note a dose-dependent reduction in the signal. Two replicates from n = 6. B, three detergents, 0.1% X-Triton, or 1% Tween-20, or 0.167% sodium deoxycholate, were used to prevent the interference of lPIP₂ to the assay. Each detergent was added to the lPIP₂/b-PIP₃ mixture followed by incubation with the GST-GRP1 containing wells. The assay then proceeded with addition of HRP and OPD, as outlined in (D). Note that of the three detergents tested, the polar sodium deoxycholate proved ineffective, while the non-polar detergents X-Triton and Tween-20 effectively overcame the inhibitory effect of lPIP₂. Two replicates from n = 4. C, different concentrations of Tween-20 were tested to identify the optimal amount needed to overcome the inhibitory effect of the substrate lPIP₂. Two replicates from n = 4. D, schematic diagram of the experimental set up used in (A). The individual data points displayed above represent the means of replicates in each experiment. The statistical significance of difference between groups was examined by one-way ANOVA followed by Tukey’s multiple comparisons test. ns: no significance, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. Data were shown as mean ± SD. lPIP₂, lipid form of phosphatidylinositol-4,5-biphospate; OPD, o-phenylenediamine; GRP, general receptor for phosphoinositide; GST, glutathione-S-transferase; HRP, horseradish peroxidase; b-PIP₃, biotin-phosphatidylinositol-3,4,5-triphosphate.

Subsequently, toward optimization of the ratio between lPIP₂ and total lipids in the liposomal preparation, we tested the ratios 1:2 and 1:10, while using the same final concentration of lPIP₂ (7 μΜ) in the reaction mixture, so that the two conditions could be directly compared. We observed that at the low ratio of lPIP₂ (1:10), the activity of PI3Kα was hardly detectable, while it was significantly higher at the 1:2 ratio (Fig. 3A). For comparative purposes, we also assessed the PI3Kα activity using the soluble form of PIP₂ (sPIP₂). Interestingly, we found that the activity of PI3Kα was significantly lower when sPIP₂ was used instead of lPIP₂ liposomes, even when the concentration of sPIP₂ was 14-fold higher. Thus, the water-soluble form of PIP₂ is significantly poorer as a substrate, in comparison to lipid PIP₂ prepared in the form of liposomes, indicating that the interaction of PI3Kα with the membrane is critical for optimal enzymatic activity, consistently with previous studies (53). Therefore, we used the lPIP₂/total lipid ratio of 1:2 in subsequent experiments. Subsequently we determined the optimal final concentration of lPIP₂ in the reaction, using increasing concentrations of lPIP₂. As shown in Figure 3B, at low concentrations of lPIP₂ (1.75, 3.5, up to 7 μM), there was progressively a positive dose-dependent effect of the substrate, while at concentrations >7 μM the effect was less prominent and approached a plateau (7 μΜ and 14 μΜ do not exhibit a statistically significant difference). Thus, in subsequent experiments, we used the amount of 7 μM of substrate, which provides sufficient detection signal.

Figure 3.

Establishment and titration of the optimal form of the substrate PIP₂.A, the optimal form of substrate was determined by comparing two different ratios of lPIP₂ to total lipids, 1:10 and 1:2, both at a final lPIP₂ concentration of 7 μM. A water-soluble PIP₂ was also tested, at a concentration of 100 μM. Note that the most advantageous substrate preparation is a 1:2 mixture of lPIP₂ to total lipids. Two replicates from n = 10. B, different concentrations of lPIP₂ (1:2 mixture of lPIP₂ to total lipids) were tested. Note the dose-dependent effect. Two replicates from n = 4. C, schematic diagram of the experimental set up used in (A) and (B). The individual data points displayed above represent the means of replicates in each experiment. The statistical significance of difference between groups was examined by one-way ANOVA, followed by Tukey’s multiple comparisons test. ns: no significance, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. Data were shown as mean ± SD. Specific activity values for PI3Kα, calculated as pmol PIP₃ produced per minute per nanogram of enzyme, are indicated in red fonts in the figure. lPIP₂, lipid form of PIP₂; PIP₂, phosphatidylinositol-4,5-biphospate; PIP₃, phosphatidylinositol-3,4,5-triphosphate.

Finally, for quantitative assessment of relative PI3Kα activity, we constructed a standard curve of “absorbance versus PIP₃” (Fig. 4A), to be used for the quantitative estimation of the product PIP₃ and the calculation of PI3Kα activity in all experiments. This curve has a hyperbolic profile, typical for a competition-based assay. In the current example, the antagonist PIP₃ displaces competitively the agonist b-PIP₃ from GST-GRP1 (Fig. 4A) in a dose-dependent manner. To further evaluate the performance characteristics of the assay, we assessed its reproducibility, inter- and intra-experimental, as well as its sensitivity. As shown in Figure 4, A and B, the dataset values reveal very low CV across the range of tested concentrations of PIP₃, while the sensitivity analysis produced a lower limit of detection of 1 ng PIP₃. Together, these data demonstrate that the assay is both reproducible and sensitive, supporting its suitability for quantitative determination of PI3Kα lipid-kinase activity.

Figure 4.

Assessment of reproducibility and sensitivity of the assay.A, to assess the reproducibility and sensitivity of the assay, we generated a standard curve (Abs versus PIP₃) from four independent experiments, with three technical replicates per PIP₃ concentration. Percent of CV = SD/mean × 100 is shown for each sample, reflecting interassay reproducibility. To evaluate the sensitivity of the assay, we calculated the Abs of the lower detectable amount of PIP₃ based on the formula (Abs = mean Abs_blank − 3SD Abs_blank), which results in Abs = 1.73, which, based on the standard curve, corresponds to 1 ng PIP₃ (lower limit of detection). Error bars represent the SD across experiments. This plot served as the standard curve for quantifying PIP₃ produced in PI3Kα activity assays. Data were analyzed by nonlinear regression analysis (Gaussian). B, to assess intra-assay reproducibility, a standard curve was generated from a single experiment with 8 technical replicates per PIP₃ concentration. Percent of CV = SD/mean × 100 is shown for each sample, reflecting intra-assay reproducibility. Error bars indicate the SD of the replicates. The plot shows a representative experiment from three independent experiments. C, schematic diagram of the experimental set up used in (A) and (B), illustrating the competition between pure PIP₃ and b-PIP₃ for binding to GST-GRP1. PIP₃, phosphatidylinositol-3,4,5-triphosphate; b-PIP₃, biotin-PIP₃; GRP, general receptor for phosphoinositide; GST, glutathione-S-transferase.

Validation of the assay specificity

To validate the specificity of the assay, we first confirmed that it complies with the typical characteristics of enzymatic reactions, that is, it exhibits enzyme dose dependence (Fig. 5A), as well as time and temperature dependence (Fig. 5B). Moreover, we tested the efficacy of well-known PI3Kinase inhibitors, that is, Wortmannin and LY-294002 (54, 55), on the activity of the enzyme and found that the IC₅₀ values were 2.3 nM and 2.8 μM, respectively (Fig. 5, C and D), in agreement with previously reported data (56).

Figure 5.

Validation of the specificity of the assay.A, to assess the dose dependence of the catalytic reaction (on the amount of the enzyme), PI3Κα reaction mixtures were set up by incubating varying amounts of PI3Kα with 7 μΜ lPIP₂, at 25 °C for 16 min. The mixture was then added to 96-well plates along with b-PIP₃ and processed as described in Experimental procedures section. Two replicates from n = 3. B, to test the time-dependence of the enzymatic reaction, the reaction mixture was prepared as described in (A), with the exception that the enzyme concentration was kept constant at 4 ng, while the incubation time varied from 4 to 60 min. Temperature dependence was also examined by evaluating the reaction progress at both 25 °C and 0 °C. Two replicates from n = 3. C and D, determination of the IC₅₀ values of Wortmannin (C), or LY294002 (D), well-established PI3Kα inhibitors. The steps of the reactions are shown schematically in (E). Three replicates from n = 2 in (C), or n = 3 in (D). The individual data points displayed above represent the means of replicates in each experiment. E, schematic diagram illustrating the steps involved in the PI3Kα enzymatic reaction in the presence of an inhibitor. The potential inhibitor is first incubated with PI3Κα for 10 min at 25 °C. The reaction is then initiated by adding lPIP₂ and the mixture is incubated for an additional 10 min at 25 °C, before the reaction is terminated. The process continues as described in Experimental procedures section. Data in (A) and (B) were analyzed by nonlinear regression analysis (first order polynomial-straight line). IC₅₀ values in (C) and (D) were calculated by nonlinear regression analysis (binding-competitive). Specific activity values for PI3Kα, calculated as pmol PIP₃ produced per minute per nanogram of enzyme, are indicated in red fonts in the figure. lPIP₂, lipid form of phosphatidylinositol-4,5-biphospatePIP₃, phosphatidylinositol-3,4,5-triphosphate; b-PIP₃, biotin-PIP₃.

Finally, we evaluated whether the conditions of the assay recapitulate the physiological receptor-mediated activation of the enzyme. In the PI3Kα complex, the regulatory p85α subunit serves an inhibitory role by binding to the catalytic p110α subunit. Phosphorylated receptors interact with the nSH2 domain of the p85α regulatory subunit, and abrogate its inhibitory binding to the catalytic subunit p110α, thereby leading to catalytic activation (21). This activation mechanism applies to the WT PI3Kα (57) and also to the H1047R mutant, but not to the E545K form, as the later mutation prevents binding of the SH2 domain of p85α to p110α, thereby resulting in constitutive activation of the catalytic subunit (34, 36, 49). In order to test whether the conditions of our assay recapitulate the above regulatory mechanism for the different PI3Kα forms (WT and mutants), we tested the effect of a doubly phosphorylated platelet derived growth factor receptor (PDGFR)-derived peptide on the enzymatic activity of PI3Kα WT and on the mutants H1047R and E545K. As shown in Figure 6, B–D, increasing concentrations of the PDGFR phosphopeptide led to a higher enzymatic activity of the WT and of the H1047R mutant of PI3Kα, achieving more than 2.5-fold activation, while there was no effect on the E545K mutant, exactly as expected (34, 36, 49). Considering altogether the above described tests, we conclude that the assay that we have developed faithfully recapitulates the native enzymatic reaction of PI3Kα on cellular membranes.

Figure 6.

The assay recapitulates the physiological receptor-mediated activation of PI3Kα.A, schematic diagram illustrating the steps involved in the PI3Kα enzymatic reaction in the presence of a doubly phosphorylated platelet derived growth factor receptor-derived peptide ESDGGY∗MDMSKDESIDY∗YPMLDMKGDIKYADIE (the asterisk indicates the phosphorylated residues). Different quantities of this p-peptide were preincubated for 10 min at 25 °C with the different forms of PI3Kα; WT in (B), mutant H1047R in (C) and mutant E545K in (D) before assembling the PI3Kα reactions, and proceeded as described in Experimental procedures section. Three replicates from n = 4 in (B), n = 6 in (C) and n = 4 in (D). The individual data points displayed above represent the means of replicates in each experiment. The statistical significance of difference between groups was examined by one-way ANOVA, followed by Tukey’s multiple comparisons test. ns: no significance, ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. Data were shown as mean ± SD. Specific activity values for PI3Kα, calculated as pmol PIP₃ produced per minute per nanogram of enzyme, are indicated in red fonts in the figure. PIP₃, phosphatidylinositol-3,4,5-triphosphate.

Role of membranes in the overactivation of the H1047R mutant form of PI3Kα

Interestingly, our membrane-based assay proved to be useful for gaining insights into the mechanism responsible for mutant PI3Kα overactivation. Although it is unquestionable that the oncogenic potential of the hotspot mutations is owed to their ability to enhance the catalytic activity of PI3Kα (58, 59, 60), the contribution of lipid membranes to this overactivation is not completely understood. Thus, to address this question, we took advantage of our membrane-based assay in order to examine whether the membrane per se is responsible for the overactivation of PI3Kα mutants. For this purpose, we compared the enzymatic activity between WT and mutant PI3Kα, either H1047R or E545K, using membranous substrate (lPIP2 liposomes) versus water-soluble PIP₂ (sPIP₂), a comparison which, to the best of our knowledge, is reported for the first time. We observed that, when lPIP₂ liposomes were used as substrate, both mutant forms of PI3Kα were more active than the WT enzyme (Fig. 7A). On the other hand, with sPIP₂ as substrate, only the E545K form of PI3Kα was overactivated in comparison with the WT (Fig. 7B), whereas, quite surprisingly, H1047R was less active than the WT form (equal nanogram of enzyme was compared). These data suggest that the lipid bilayer plays a critical role in the overactivation of the H1047R mutant, but is not involved in the hyperactivation of E545K.

Figure 7.

Role of membranes in the overactivation of the mutants H1047R and E545K. We compare the PI3Kα enzymatic activity of H1047R, E545K, and WT, using as substrate 7 μM lPIP₂ (A), or 100 μM sPIP₂ (B). The activities of the mutants were normalized to WT, which was set to value “1.” Note that E545K shows higher activity than WT with either soluble or membranous PIP₂, whereas H1047R exhibits increased activity only when membranes are used. Two replicates from n = 6 in (A) and n = 7 in (B). The individual data points displayed above represent the means of replicates in each experiment. The statistical significance of difference between groups in graphs (A) and (B) was examined by one-way ANOVA, followed by Tukey’s multiple comparisons test. ns: no significance, ∗∗p < 0.01, ∗∗∗p < 0.001, and ∗∗∗∗p < 0.0001. Data were shown as mean ± SD. Specific activity values for PI3Kα, calculated as pmol PIP₃ produced per minute per nanogram of enzyme, are indicated in red fonts in the figure. lPIP₂, lipid form of phosphatidylinositol-4,5-biphospate; SPR, surface plasmon resonance; sPIP₂, soluble form of PIP₂; PIP₃, phosphatidylinositol-3,4,5-triphosphate.

MD simulations of PI3Kα/HRAS on model membrane reveal membrane-based structural differences between WT and H1047R

To further explore the mechanism of how the H1047R mutation interacts with the membrane, we employed MD simulation to interpret experimental results on complex membrane systems and to gain insight into the relevant interactions at the atomic level, in sync with experimental studies. We performed three independent unbiased MD simulations of 600 ns each, on a model membrane, for both the WT and H1047R mutant active-like ΔABD p110α systems, in complex with HRAS, a critical PI3Kα interactor with known structure (61) (Fig. 8A). The simulations were considered converged based on RMSD and principal component analysis calculations (Figs. S2 and S3, and ref. (62)), while the final 400 ns of each trajectory were used for trajectory analysis. Before analyzing the simulations, we validated our MD data with available hydrogen-deuterium (H/D) exchange rates for the ΔABD p110α WT in solution and in the presence of the membrane (63, 64, 65, 66, 67, 68), by comparing the differences between them, using the Python package HDXer (63, 64, 65, 66, 67, 69). The mean deviation between the experimental and calculated H/D exchange rates was 16.7% (SD 13.4%) for the ΔABD p110α system in solution and 20.3% (SD 7.9%) for the membrane-bound ΔABD p110α WT system. Our MD simulations’ findings align with the hydrogen–deuterium exchange mass spectrometry experimental data (68) (Figs. S4 and S5) and support a model in which ΔABD p110α interacts with the membrane at a surface region comprising C2 domain (region 1), kα1 helix of the helical domain (region 2), kα3 helix of the N-lobe of the kinase domain (region 3), and the activation loop (region 4) (Fig. 8, A and B; negative values in the plots of Fig. 8B signify membrane interaction). On the other hand, other regions of PI3Kα show no change upon membrane binding, for example, C2β1 of the C2 domain and residues of the Ras binding domain (Fig. 8B).

Figure 8.

Results from molecular modeling of the active-like ΔABD p110α WT and H1047R mutant in complex with HRAS, on a model membrane.A, the model depicts the ΔABD p110α in complex with HRAS, highlighting their domains and annotating the functionally important residues of PI3Kα. B, the plots 1 to 6 show H/D exchange rate differences between ΔABD p110α WT in solution and membrane-bound ΔABD p110α WT for residues (1) 343 to 354 (C2 domain), (2) 713 to 734 (membrane binding loop 1, kα1), (3) 799 to 811 (Ν-lobe of kinase domain, kα3), (4) 930 to 961 (activation loop), (5) 279 to 287 (C2 domain, C2β1), and (6) 335 to 342 (Ras binding domain). Negative values signify membrane interaction. Residues colored green show no change upon membrane binding, while blue residues show reduced H/D exchange rates in the membrane-bound state. ΔABD p110α interacts with the membrane via the C2 domain, kα1 helix of the helical domain, N-lobe regions, and the activation loop. C, H/D exchange rate differences between WT and H1047R mutant on the membrane. In H1047R, C-terminal tail residues are stabilized and protected on the model membrane, while residues of the membrane binding loop 2 show increased exchange rates compared to the WT. RMSF of Cα atoms in the C-terminal tail (D) and the membrane binding loop 2 (E) of ΔABD p110α WT and H1047R mutant in solution and on the membrane. H/D, hydrogen-deuterium; RMSF, root-mean-square-fluctuation

Following validation of our model against the hydrogen–deuterium exchange mass spectrometry data for the ΔABD p110α WT system both in solution and on the membrane, we advanced to predict the H/D exchange rates for the membrane-bound ΔABD p110α H1047R system. This allowed us to explore the differences in interactions with the membrane between WT and H1047R mutant. Notably, among all membrane-interacting domains, the C-terminal tail of the H1047R mutant exhibits greater protection on the membrane, as reflected by a reduction in H/D exchange rates of up to 27.8% compared to the membrane-bound WT (Fig. 8C, plot at the bottom left and Figs. S5 and S6). This observation highlights the influence of the H1047R mutation on membrane binding, demonstrating an enhanced interaction between the C-terminal tail of H1047R and the membrane, consistently with previous studies (35, 40, 68). A slight increase in H/D exchange rates (up to 7.9%) of residues in the membrane-binding loop 2 of membrane-bound H1047R, as compared to WT (Fig. 8C plot at the bottom right and Figs. S5 and S6), could provide intramolecular flexibility in mutant PI3Kα, potentially contributing to enhanced membrane-binding capability of its C-terminal tail. These findings are further supported from our previous results from the dynamical network analysis of p110α trajectories for the WT and H1047R mutant in solution, which demonstrated allosteric communication between the two regions and stronger residue-residue coupling induced by the H1047R mutation (70).

Additional analysis of these two regions by root-mean-square-fluctuation (RMSF) of the MD trajectories for the membrane-bound WT and H1047R systems (41), versus the corresponding PI3Kα systems in solution, further highlighted differences in residue mobility between the WT and H1047R mutant (Fig. 8, D and E). In membrane-binding loop 2, residues 863 to 867 show increased flexibility in the H1047R mutant in solution, compared to the WT, while residues 868 to 873 are more stabilized in the mutant (Fig. 8D, left plot). Conversely, in the membrane-bound systems, the H1047R mutant exhibits consistently greater flexibility across residues 863 to 873 than the WT (Fig. 8D, right plot), aligning with the observed differences in H/D exchange rates between the two systems (Fig. 8C). Strikingly, in the C-terminal tail, although both H1047R and WT systems in solution display similar flexibility (Fig. 8E, left plot), the H1047R mutant on the membrane shows significantly enhanced stability of residues 1059 to 1068 (Fig. 8E, right plot), while, in sharp contrast, the membrane rather destabilizes the C terminus in the WT system (Fig. 8E). This RMSF data further supports the influence of the H1047R mutation on the dynamics of the membrane-binding loop 2 and the C-terminal tail, which could contribute to enhanced membrane-binding of the mutant.

SPR experiments show that H1047R mutation causes increased rate of association with the membrane

To validate experimentally the above MD findings, we conducted SPR experiments to compare H1047R with WT PI3Kα, for binding to HCT116-derived liposomes in the presence or absence of lPIP₂. To ensure a robust comparison, we performed experiments using the same biosensor chip system, capable of monitoring of up to 36 interactions. Quantification of response unit levels (RU) at the end of SPR injections (association phase) showed a 2- to 4-fold higher levels of association for the mutant PI3Kα, than equimolar amounts of WT PI3Kα (Fig. 9A). Further kinetic analysis of PI3Kα binding to lPIP₂-loaded liposomes revealed that the H1047R mutant exhibits a 2-fold higher association rate constant (K_on) than the WT (Fig. 9, Table 1). Surprisingly, the dissociation rate constant (K_off) for the mutant was also higher (9.4 × 10⁻⁴ s⁻¹) than the WT (3.5 × 10⁻⁴ s⁻¹), resulting in an overall similar equilibrium dissociation constant (K_D) between the WT and the H1047R mutant (Fig. 9, Table 1). To assess the implications of these data on the catalytic output of the enzyme, we sought to corelate the “catalytic efficiency” (K_cat/K_m), a fundamental parameter of enzymatic performance, with the binding and dissociation parameters of the enzyme, using the following equation:

$$\frac{K_{\text{cat}}}{K_{\mathrm{m}}}=\frac{K_{\text{cat}}}{\left(\frac{K_{\text{off}}+K_{\text{cat}}}{K_{\text{on}}}\right)}$$ (1)

Figure 9.

SPR assays of H1047R and WT PI3Kα binding to the membrane.A, quantification of association RUs (reporting points) show that H1047R exhibits on average a 2- to 4-fold higher binding to liposomes in comparison to WT enzyme, in a lPIP₂-dependent manner. B, representative sensograms from SPR experiments demonstrating the binding capacity of H1047R mutant, compared to WT, for lPIP₂ liposomes versus plain liposomes (without lPIP₂). Data were obtained under the same experimental conditions and reagents, on the same biosensor chip. lPIP₂, lipid form of phosphatidylinositol-4,5-biphospate; SPR, surface plasmon resonance.

Table 1.

Kinetic parameters of the interaction between PI3K and lPIP₂-loaded liposomes

Protein	Kon ± SEM (105 M−1 s−1)	Koff ± SEM (10−4 s−1)	Rmax ± SEM (RU)	KD ± SEM (10−9 M)
PI3Kα WT	1.13 ± 0.05a	3.6 ± 0.1	144 ± 21.4	3.19 ± 0.15
PI3Kα H1047R	2.44 ± 0.46a	9.41 ± 0.01	296 ± 38.0	3.99 ± 0.75

Surface plasmon resonance (SPR) analysis was used to determine the binding kinetics of WT PI3Kα and the oncogenic H1047R mutant to lPIP₂ liposomes. The association rate constant K_on, the dissociation rate constant K_off, and equilibrium dissociation constant K_D were calculated as described in the experimental procedures. Statistics are derived from kinetic SPR fits from three independent experiments for the WT, two independent experiments for the H1047R mutant and five technical replicates per experiment.

lPIP₂, lipid form of phosphatidylinositol-4,5-biphospate.

^aThe higher MU K_on value compared to WT K_on was statistically significant according to ANOVA analysis (p value < 0.0103).

To simplify the above equation, we calculated the K_cat value of PI3Kα WT and H1047R. Using the specific activities of PI3K forms found by our assay (0.3 ± 0.02 for WT and 0.51 ± 0.05 for H1047R, in pmol PIP₃/min/ng enzyme), the K_cat values were calculated to be 1 ± 0.03 s⁻¹ for WT and 1.7 ± 0.09 s⁻¹ for H1047R, in line with the values reported previously {Carson, 2008 #80;Sun, 2020 #156;Gong, 2023 #166}. Interestingly, these K_cat values are four orders of magnitude higher than the dissociation rate constants (K_off) determined by our SPR assays (Table 1). This indicates that, once the enzyme–substrate complex is formed, the conversion of PIP₂ to PIP₃, governed by K_cat, proceeds much faster than the dissociation of the complex into free enzyme and substrate (dictated by K_off). Consequently, since K_cat is much higher than K_off, the above Equation 1 is approximated as shown below in Equation 2.

$$\frac{K_{\text{cat}}}{K_{\mathrm{m}}}=\frac{K_{cat}}{\left(\frac{K_{\text{off}}+K_{\text{cat}}}{K_{\text{on}}}\right)}\approx \frac{K_{\text{cat}}}{\left(\frac{K_{\text{cat}}}{K_{\text{on}}}\right)}=K_{\text{on}}$$ (2)

Thus, the “catalytic efficiency” of the enzyme (K_cat/K_m) is determined by K_on alone (the rate of binding between enzyme and substrate), while K_off has no significant impact on enzyme’s catalytic yield. As a result, the >2-fold higher Kon of H1047R mutant, in comparison to WT, determines the higher catalytic efficiency of the mutant.

Altogether, our activity assays, MD simulations, and SPR membrane-binding experiments indicate that the H1047R mutation enhances the rate of membrane association of PI3Kα, compared to the WT, thereby resulting in an overall increased catalytic activity.

Discussion

The PIK3CA hotspot mutations E545K and H1047R are among the most commonly found in human cancers (24). Given the role of these mutations in the overactivation and the oncogenic potential of PI3Kα (58, 59, 60), they constitute therapeutic targets of high interest. This objective has led to the development of various PI3Kα activity assay technologies, using radioactive, fluorescence, luminescence, and absorbance detection techniques (43, 44, 45). However, these methods have a number of limitations. Radioactivity-based assays, for example, despite their high sensitivity, are hazardous and involve multiple steps, in addition to frequent high-noise problems. On the other hand, fluorescence and luminescence assays face challenges such as high reagent costs and the need for specialized equipment (42, 45). To overcome such shortcomings, we developed a novel cell-free assay, based on the GRP1 competition principle (44). Οur assay was developed entirely de novo, using membrane-incorporated PIP₂, thus recapitulating the conditions for membrane-based activation of oncogenic PI3Kα. The protocol is established on an ELISA-based principle, is developed entirely with individually obtained and chemically defined reagents, as well as in-house prepared components (home-made GST-GRP1 and liposomal substrate), is simple, rapid, specific, and applicable to high-throughput screening. Moreover, it has low background, high dynamic range and sensitivity, and requires only standard laboratory equipment.

The development of this membrane-based PI3Kα assay was challenging due to the necessity of using lipid PIP₂ reconstituted in membranes, as a substrate, in order to simulate the physiological environment of the PI3Kα reaction. We fulfilled this requirement by utilizing lPIP₂ liposomes prepared by using lipids from cancer cells. Interestingly, lPIP₂ liposomes proved to be a much better substrate than soluble sPIP₂ (for both WT and H1047R mutant forms), thus confirming that the use membranes is essential for ensuring optimal output of the assay.

Within this optimized system, our assay produced PI3Kα activity values that depend on the dose of PI3K, the time and temperature. The enzyme was fully inhibited by known inhibitors, with IC₅₀ values consistent to previous reports. WT and H1047R forms show a modest activation of PI3Kinase by the phosphopeptide, similarly to previous reports (19, 36, 49), unlike other studies that observe higher levels of activation (34, 71). These differences, in levels of activation by the phosphopeptide, more likely arise from variations in experimental conditions, assay formats, or other methodological parameters unique to each study.

To understand how the PI3Kα hotspot mutations contribute to carcinogenesis, it is essential to elucidate the mechanisms by which these mutations trigger catalytic overactivation. Previous studies suggested that the H1047R and E545K mutants exhibit enhanced enzymatic activity, compared to the WT protein, through distinct mechanisms (33, 35). The E545 K mutation abolishes the inhibitory effect of the p85α regulatory subunit, leading to constitutive activation of p110α (36, 37, 38). In contrast, the H1047R mutant remains regulated by p85α like the WT but is thought to enhance catalytic activity by strengthening plasma membrane interactions (35, 39, 40, 41). Yet, a direct comparison of the effect of the membranous form of the substrate, versus soluble substrate, on the activities of WT and H1047R and E545K mutants, has not been addressed. In this regard, taking advantage of the membrane-based assay developed here, we show for the first time that E545K possesses higher activity than WT regardless of the nature of the substrate (lPIP₂ in liposomes versus soluble PIP₂), whereas, importantly, H1047R exhibits enhanced activity, compared to WT, only when lPIP₂ liposomes are utilized (Fig. 7, A and B). These data imply that the lipid nature and membrane environment of the substrate play a crucial role in the overactivation mechanism of the H1047R mutant. Our findings are consistent with the notion that the H1047R mutation alters the structural conformation of the membrane binding domain of PI3Kα, thus, increasing its affinity for the membranes (34, 35, 39, 40). On the basis of related data (68), it was recently proposed that a reorientation of the WIF motif toward the membrane surface enhances membrane binding of PI3Kα and prompts the activation loop to adopt a more competent conformation.

To further investigate the structural and dynamical changes of the active-like state of the WT and H1047R bound to a membrane, we employed, for the first time, all-atom MD simulations of the active-like state of ΔABD p110α in complex with HRAS on a PIP2/1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine/1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine bilayer. Interestingly, this analysis revealed how the H1047R mutation and the lipid bilayer contribute to structural and dynamical changes compared to the WT protein and provide atomistic information for the membrane-binding mechanism of mutant PI3Kα. After confirming the validity of our model, we compared the predicted H/D exchange rates of the ΔABD p110α H1047R mutant and the WT membrane-bound system. Interestingly, this analysis reveals that residues in the C-terminal tail are more stable in the H1047R mutant, in contrast to the increased flexibility observed in its membrane-binding loop 2, as compared to WT (Fig. 8C). RMSF analysis of the membrane-bound systems, along with comparisons to the respective ΔABD p110α systems in solution, further corroborated these observations (Fig. 8, D and E). These findings suggest that the H1047R mutation enhances interactions at the C-terminal tail while modulating the dynamics of residues near membrane-binding loop 2, thereby contributing to the improved membrane-binding ability of the mutant. Our SPR experiments, comparing H1047R PI3Kα to the WT in binding to membranous lPIP₂ (Fig. 9), supported these results, revealing a faster rate of association of the mutant with the membrane, relatively to the WT. Further analysis of the SPR data allowed us to calculate the association and dissociation rate constants (K_on and K_off), for the first time. Interestingly, taking into account these values, we conclude that the catalytic efficiency (K_cat/K_m) of PI3Kα is primarily determined by K_on (see Equation 2 in “Results”), highlighting the critical role of the association rate (2-fold higher for the mutant) in determining the enzymatic performance, thereby providing a deeper understanding of the mechanism of mutant overactivation.

Overall, our experimental data and MD simulations advance our understanding on the membrane binding mechanism of mutant PI3Kα and highlight the structural and dynamical changes induced by the H1047R mutation and the lipid bilayer, in line with previous studies (34, 35, 39, 40, 68). By shedding light into the membrane interactions triggered by the oncogenic mutation H1047R, the present study offers potential insights that may be exploited for the development of new cancer therapeutic drug discoveries. Our MD methodology capturing the active membrane-bound conformation of PI3Kα is crucial, as it provides a rational framework for structure-based drug design targeting of the oncogenic H1047R mutation.

Given the ongoing need for the development of new targeted small molecule anticancer drugs (11, 12, 72, 73, 74, 75, 76, 77, 78), the PI3Kα activity assay developed here offers opportunities for high-throughput approaches targeting the membrane-based overactivation of PIK3CA-H1047R, one of the most frequent mutations in cancer. This approach, assisted by MD simulations of drug-PI3Kα binding screenings, could lead to the development of mutant-specific inhibitors with minimal impact on WT PI3Kα, thus, avoiding potential side effects. Our assay can also provide the basis for a more profound understanding of the mechanism(s) behind hotspot mutation-mediated overactivation, by reconstituting into membranes the complete signaling machinery, including ligand–receptor complexes, Ras, and trimeric G proteins, followed by assessment of their role on membrane-based PI3Kα activation.

Experimental procedures

Reagents

Glutathione-coated plates as well as streptavidin poly-HRP were from PIERCE/Thermo Fisher Scientific. ATP, dimethyl sulfoxide, bovine serum albumin (BSA), OPD, Wortmannin, and LY294002 were from Sigma-Aldrich/Merck, while isopropyl β-D-1-thiogalactopyranoside was from Thermo Fisher Scientific. Proteins PI3Kα WT, PI3Kα H1047R, and PI3Kα E545K were from Millipore/Merck. Lipid PIP₂ [L-α-phosphatidylinositol-4,5 bisphosphate (Brain, Porcine) (ammonium salt)] and soluble PIP₂ [1,2-dioctanoyl-sn-glycero-3-phospho-(1′-myo-inositol-4′,5-bisphosphate) (ammonium salt)] were from Avanti Polar Lipids. Lipid PIP₃ [PtdIns-(3,4,5)-P₃ (1-stearoyl, 2-arachidonoyl) (sodium salt), soluble PIP₃ (PtdIns-(3,4,5)-P3 (1,2-dioctanoyl) (sodium salt), and biotinylated PIP₃ [PtdIns-(3,4,5)-P₃ biotin (sodium salt) were from Cayman Chemicals. Doubly phosphorylated PDGFR peptide was from Cambridge Peptides Ltd. Protease inhibitors PMSF, Pepstatin, and Aprotinin were from Roche/Merck, while Leupeptin was from Sigma-Aldrich/Merck. From Sigma-Aldrich/Merck were also DNase and RNase.

Preparation of liposomes

For the preparation of PIP₂-containing liposomes (lPIP₂ liposomes), we used the HCT116 cell line, a colorectal cancer line in which the PIK3CA gene carries the H1047R mutation in exon 20. From a large culture of these cells (45 plates, 10 cm diameter each), we isolated 14 mg of mixed lipids reconstituted in 39 ml CHCl₃:EtOH:H₂O (20:9:1) (79). In an aliquot of this mixture, 560 μl (containing 200 μg lipids), we added 200 μg of long side chains PIP₂ (lPIP₂, isolated from brain, Avanti), already reconstituted in 40 μl CHCl₃:EtOH:H₂O (20:9:1). This mixed lipid preparation was dried completely under N₂ stream, followed by speed vacuum for 1 h. Then, 0.2 ml of an aqueous solution (20 mM Tris pH = 7.4, 100 mM KCl, 1 mM EDTA pH = 8, and 1 mM EGTA) was added and the mixed lipid preparation and incubated at room temperature (RT) for 1 h, with occasional vortexing (every 10 min). Subsequently, the sample was subjected to 5 freeze/thaw cycles (liquid N₂/water bath at RT), followed by sonication in a water bath (VWR Ultrasonic Cleaner USC-TH), for 30 min. Finally, the preparation was subjected to extrusion, using the Avanti mini-Extruder, according to manufacturer’s instructions, with a 100 nm membrane. This procedure is known to result in a homogeneous mixture of small unilamellar vesicles with diameter below 100 nm (80). The final preparation contained 1 μg/μl lPIP₂ and 1 μg/μl mixed lipids. This final lipid composition (1:2 PIP₂:total lipids), rich in PIP₂, provided optimal detection of the PI3K activity under our conditions (Fig. 3A), consistently with previous reports (39, 43). Multiple independent lipid preparations, obtained from distinct HCT116 cell cultures, gave consistent specific activity values for WT and mutant PI3K, suggesting lack of prep-to-prep variability. Finally, regarding the possible concern that endogenous PIP₃ in these cells could interfere with the assay, we reasoned that, since this lipid represents <0.05 mol% of phosphoinositides (81, 82), which in turn constitute <1% of total cellular lipids (83), such interference is unlikely to occur.

Expression and purification of GST-GRP1 domain

A GST-GRP1 cDNA (kindly provided by Nicholas Leslie, University of Dundee) was used to transform the Escherichia coli strain BL21 (DE3). A 4-L culture of the transformed bacteria was grown at 37 °C until reaching an absorbance (A) of 0.4 to 0.6, at 600 nm. IPTG was added to the bacterial culture to a final concentration of 200 μM, to induce protein expression of GST-GRP1, followed by overnight incubation at 22 °C. The bacterial cells were harvested by centrifugation, resuspended in 50 mM Tris–HCl pH = 7.4, 100 mM NaCl, 1 mM PMSF, 1 μΜ Leupeptin, 1 μg/ml Aprotinin, 0,7 μg/ml Pepstatin, 5 μg/ml DNase, 5 μg/ml RNase and subjected to lysis using French Press, according the manufacturer’s instructions. The lysate was centrifuged at 100.000g, to pellet cell debris and insoluble material, and was then further processed to purify GST-GRP1. For that purpose, glutathione-Sepharose 4B resin (Amersham Pharmacia) was used according to the manufacturer’s standard protocols. The final preparation was 1.3 ml containing 6.2 mg/ml GST-GRP1, while the purity of the protein was determined by SDS-PAGE (Fig. S1).

PI3Kα activity assay

For the activity assay of PI3Kα, 96-well plates coated with glutathione were incubated with 0.2 ml blocking buffer (50 mM Tris pH = 7.4, 150 mM NaCl, 0.05% Tween and 0.2% gelatin) for 1 h at RT, under shaking. Then, 4 μg purified GST-GRP1 in 0.2 ml blocking buffer was added to each well and incubated for 1 h at RT, under shaking. Afterward, the wells were washed 5 times with blocking buffer, for 2 min per washing step. In parallel, PI3Kα reactions were set up in micro-tubes by the addition of 5 μl reaction buffer (250 mM Tris pH = 7.4, 10 mM MgCl₂), 2.5 μl ATP (1 mM in water), 2 μl BSA (10 μg/μl in water), 1 to 5 μl PI3Kα (1 ng/μl in 25 mM Hepes pH = 7.4, 10 mM MgCl₂, and 1 μg/μl BSA), 1 μl of the potential inhibitor (dissolved in dimethyl sulfoxide) at a range of final concentration from 1 to 100 μM, and water, to makeup a final volume of 20 μl. This mixture was incubated for 10 min at 25 °C (preincubation step), to allow binding of the potential inhibitor to the enzyme. Following this preincubation step, the reaction was initiated by the addition of 5 μl lPIP₂ liposomes (substrate, prepared as described above), containing 40 ng/μl lPIP₂ and 40 μg/μl mixed lipids. The reaction was allowed to proceed for 5 to 16 min at 25 °C, while it was terminated by the addition of 75 μl stop solution, containing 5 mM EDTA, 1% BSA, 1% Tween and 0.9 ng biotinylated-diC8PtdIns(3,4,5)P₃ (b-PIP₃), pH = 8, in Tris buffer saline (50 mM Tris pH = 7.4, 150 mM NaCl). The resulting mixture was then transferred from the microtubes to the wells of the 96-well plate, where GST-GRP1 was already bound (as described above), and was further incubated for 1 h at RT, under shaking. During this step, the product of the PI3Kα reaction, PIP₃, competes with b-PIP₃ for binding to GST-GRP1, that is, the higher the PI3Kα activity, the higher PIP₃ is produced the lower b-PIP₃ binds to GST-GRP1. The wells were then washed 5 times with blocking buffer, 2 min each, and were incubated with 0.1 μg/ml streptavidin-HRP conjugate, in 1% BSA in PBS, for 40 min at RT, under shaking. Finally, the wells were washed 5 times, 2 min each, and were incubated with the HRP substrate (200 μl, 0.04% OPD, 0.01% H₂O₂ in phosphate/citrate buffer, pH = 5.0), for 1 to 5 min, until the development of a yellow color. In order to terminate the HRP reaction and stabilize the chromophore product, a solution of 50 μl 1 M H₂SO₄ was added and the absorbance was measured at 492 nm using an ELISA Reader (TECAN, Infinite F50 Plus).

In order to assess the levels of PI3Kα activity, we calculated the amount of produced PIP₃ in each tested reaction by employing the standard curve shown in Figure 4. The measured A_492nm signal was converted into the corresponding PIP₃ quantity, which was then used to determine the final PI3Kα activity. The specific activity of PI3Ka was calculated as pmol PIP₃/min/ng enzyme. Notably, the values obtained by our assay (0.17 ± 0.02 pmol PIP₃/min/ng enzyme) were similar with those reported by the supplier using the homogeneous time-resolved fluorescence approach (0.18–0.31 pmol PIP₃/min/ng enzyme, depending on the batch), when using the same substrate conditions (soluble PIP₂), confirming the validity of our assay.

Model construction of the ΔABD p110α systems in complex with HRAS on the model membrane by MD simulations

The lipid bilayer was designed in a rectangular box with a composition consisting of 2.5 mol% PIP2 mono-protonated on the 4-phosphate (SAPI24), 2.5 mol% PIP₂ mono-protonated on the 5-phosphate (SAPI25), 30.0 mol% 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine, and 65.0 mol% 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine, in accordance with the lipid composition reported in (68), starting from different lipid arrangements. The bilayer was generated using the CHARMM-GUI Membrane Builder. The constructed bilayer was 160 × 160 Ų with 440 lipids per leaflet (see the SI for the membrane setup). The membrane was then solvated with TIP3P transferable intermolecular potential with 3 points) (TIP3P) water molecules and neutralized with Na⁺ and Cl^- ions to a salt concentration of 100 mM. The system underwent a 400 ns equilibration process according to the CHARMM-GUI protocol, which included an energy minimization step followed by a series of restrained MD simulations to stabilize the lipid environment (see the SI for the exact protocol). Following this, the membrane-binding regions of the kinase were determined (35, 68, 71) and the model was placed on the model membrane. In the generated model, the ATP and Mg²⁺ ions were appropriately placed in the ATP pocket using the 1E8X structure (56) and the PIP₂ molecule was built within the binding pocket using diC4-PIP₂ as template (84). Then, the model of the active-like state of PI3Kα in complex with HRAS was built using the 1HE8 structure (61). For the HRAS model, the HVR of HRAS was built based on a model of KRAS-4B and palmitoylated C181 and C184 (85). The last step was to place GTP and change rotamers of HRAS and ΔABD p110α to reflect the interactions between the active state of PI3Kα and HRAS. The resulting ΔABD p110α WT and H1047R complexes with HRAS on the lipid bilayer were finally solvated with TIP3P water molecules and neutralized with Na⁺ and Cl^- ions, achieving a physiological salt concentration of 150 mM.

Setup of the unbiased MD simulations

The unbiased MD trajectories for the systems were generated with ACEMD3 package (86). Protein atoms were modeled using the CHARMM36m (87) all-atom force field for protein atoms, while CHARMM36 was applied to all-atom force field for the rest of the system (lipids, solvent, ions). The TIP3P potential was used for modeling water molecules. Each system was subjected to an initial 10,000 steps of energy minimization using the steepest descent algorithm of ACEMD3 (86). Then, the systems were equilibrated using ACEMD3 first in the NPT ensemble for 30 ns with the Langevin leap-frog integrator, a Monte Carlo isotropic barostat (88) to keep the pressure constant at 1 bar and the Langevin thermostat (89) with damping constant of 0.1 ps⁻¹ to keep the temperature at 310 K with restraints on the deprotonated PIP₂ molecule in the active site (5 kcal/mol/Ų) and on the Cα atoms (1 kcal/mol/Ų). Over time, the force constants were reduced linearly, reaching 0 kcal/mol/Ų at 20 ns. The second equilibration phase was in the NPT ensemble for 40 ns with the Langevin leap-frog integrator, a Monte Carlo isotropic barostat (88), keeping the pressure constant at 1 bar and the Langevin thermostat (89) with damping constant of 0.1 ps⁻¹, to keep the temperature at 310 K without restraints. All hydrogen-involving bonds were constrained using the M-SHAKE algorithm (90) and the hydrogen mass was repartitioned to 4.0 au to allow the use of a 4 fs timestep (91). The cut-off distance for van der Waals and electrostatic interactions was set at 12 Å, with a switching function applied beyond 10 Å. Long-range electrostatic interactions were applied using the particle-mesh Ewald summation method (86, 92). After equilibration, the production runs were performed under constant pressure, temperature, and number of particles, with the same algorithms and parameters used in the second equilibration phase described above, for three independent replicas, of 600 ns each, starting from different conformations.

SPR experiments

For the SPR experiments we used the ProteOn XPR36 biosensor (Bio-Rad). In this setting, ProteOn uses a unique 6 × 6 chip two-dimensional array, creating 36 interaction spots, allowing to run experiments in a grid format. The LCP memLayer kit (Bio-Rad) was used for tethering liposomes onto the chip surface in two consecutive layers (one on top of the other), reaching approximately 4500 RU. The concentration of the injected liposomal sample was 1 mg/ml. Blank control channels, within the same chip, which underwent the same treatment as the liposome-loaded channels but without liposomes, served as reference. WT and H1047R PI3Kα, prepared by diluting in SPR running buffer immediately before the experiment, were injected over the chip using the horizontal (A) channels, at concentrations ranging from 1.25 to 80 nM. To maintain experimental consistency and reliable SPR data comparison, both WT and H1047R were injected simultaneously in different horizontal channels. Background PI3Kα binding and bulk effects were referenced using blank channels. The experiments were repeated at least three times, with fresh liposome loadings. For SPR kinetic analysis, the Langmuir 1:1 binding model was used to fit data from WT PI3Kα, while a heterogeneous ligand model was used for the mutant due to its biphasic-binding profile. The SPR running buffer used contained 10 mM NaPi (pH 7.4), 150 mM NaCl, and 0.1 mg/ml BSA.

Statistics

Data plotting including the determination of IC₅₀ and statistical analysis were performed in GraphPad Prism (https://www.graphpad.com) using nonlinear regression analysis, where indicated. The statistical significance of difference between groups was examined by one-way ANOVA followed by Tukey’s multiple comparisons test for Figures 1, C and D, 2, A–C, 3, A and B, 5, B–D, 6, A, and B. The values reported in the figures represent mean ± SD calculated from at least two replicates for each experimental setting. Data in Figure 4A were analyzed by nonlinear regression analysis (Gaussian) and in Figure 5, A and B by nonlinear regression analysis (first order polynomial-straight line). IC₅₀ values in Figure 5, C and D were calculated by nonlinear regression analysis (binding competitive). SPR data analysis was performed using ProteOn Manager.

Data availability

All data are contained within the article. The input files for the simulations and the scripts used for data analysis are available in the Zenodo repository: https://doi.org/10.5281/zenodo.14359253.

Supporting information

This article contains supporting information.

Conflict of interest

The authors declare that they have no conflicts of interest with the contents of this article.

Reviewed by members of the JBC Editorial Board. Edited by Philip A. Cole