Oncogenic PIK3CA corrupts growth factor signaling specificity

Abstract

Technical limitations have prevented understanding of how growth factor signals are encoded in distinct activity patterns of the phosphoinositide 3-kinase (PI3K)/AKT pathway, and how this is altered by oncogenic pathway mutations. We introduce a kinetic, single-cell framework for precise calculations of PI3K-specific information transfer for different growth factors. This features live-cell imaging of PI3K/AKT activity reporters and multiplexed CyTOF measurements of PI3K/AKT and RAS/ERK signaling markers over time. Using this framework, we found that the PIK3CA^H1047R oncogene was not a simple, constitutive activator of the pathway as often presented. Dose-dependent expression of PIK3CA^H1047R in human cervical cancer and induced pluripotent stem cells corrupted the fidelity of growth factor-induced information transfer, with preferential amplification of epidermal growth factor receptor (EGFR) signaling responses compared to insulin-like growth factor 1 (IGF1) and insulin receptor signaling. PIK3CA^H1047R did not only shift these responses to a higher mean but also enhanced signaling heterogeneity. We conclude that oncogenic PIK3CA^H1047R corrupts information transfer in a growth factor-dependent manner and suggest new opportunities for tuning of receptor-specific PI3K pathway outputs for therapeutic benefit.

Keywords: Growth Factor Specificity; Information Transfer; PI3K Signaling Dynamics; Single-cell Biology.

Figure 1. IGF1 and EGF induce stereotypical PIP₃/PI(3,4)P₂ signaling dynamics.

Dynamic TIRF microscopy measurements of IGF1- and EGF-induced PIP₃/PI(3,4)P₂ levels in live HeLa, MEF, or A549 cells expressing WT or loss-of-function PIK3CA as indicated. The cells were serum-starved for 3 h prior to stimulation with either saturating doses (100 nM) of IGF1 or EGF and treatment with the PI3Kα inhibitor BYL719 (500 nM). The traces represent the mean PH_AKT2 reporter fold change relative to baseline (the median signal of the first four time points). The shading signifies bootstrapped 95% confidence intervals of the mean. The number (n) of single cells for each genotype is indicated on the plots. For wild-type (WT) PIK3CA HeLa cells, two independent CRISPR/Cas9 clones were used, with and without silent mutations. The 3xFS HeLa cells originate from a single CRISPR/Cas9 clone, engineered with a frameshift (FS) mutation in all three PIK3CA alleles (see also Appendix Fig. S1). The MEFs were from polyclonal cultures established from mice with WT or CRE-deleted (KO) PIK3CA, followed by immortalization in vitro (Foukas et al, 2010). The A549 cells were from a single CRISPR/Cas9 clone per genotype, with knock-out (KO) of PIK3CA caused by a frameshift mutation in exon 3 (Gong et al, 2023). HeLa datasets for IGF1 and EGF are from six and seven independent experiments, respectively. MEF and A549 IGF1 and EGF data are from 3 independent experiments each. Non-PI3Kα activity refers to the PI3K activity that remains following pharmacological inhibition of PI3Kα.

Figure 2. Oncogenic PIK3CA^H1047R in HeLa cells reduces the information capacity in the PIP₃/PI(3,4)P₂ dynamics for IGF1 but not EGF.

(A) TIRF microscopy measurements of IGF1- and EGF-induced PIP₃/PI(3,4)P₂ kinetics in live HeLa cells with endogenous, dose-controlled expression of PIK3CA^H1047R (see also Appendix Fig. S1). The cells were serum-starved for 3 h prior to stimulation with the indicated growth factors and treatment with the PI3Kα inhibitor BYL719 (500 nM). Measurements were obtained every 70 s for a total of 60 min. The traces represent the mean PH_AKT2 reporter fold change relative to the median signal for the 40–60 min time window, used here to capture the baseline signaling elevation in PIK3CA^H1047R mutant cells. The shaded areas represent bootstrapped 95% confidence intervals of the mean. The shading signifies the 95% confidence intervals of the mean. The number (n) of single cells for each genotype is indicated on the plots. For WT HeLa cells, two independent CRISPR/Cas9 clones were used, with and without silent mutations. The data are from two independent WT, two independent 1xH1047R, and three independent 2xH1047R CRISPR/Cas9 clones. The data are from the following number (n) of independent experiments: n = 6 for 100 nM IGF1; n = 7 for 100 nM EGF; n = 2 for 10 nM IGF1 and 10 nM EGF; n = 3 for 1 nM IGF1; n = 4 for 1 nM EGF. (B) Median information capacity in bits (log2) for IGF1 and EGF calculated from the trajectory responses from all independent experiments and individual cells specified in (A). Capacity is a measure of the maximum amount of information that flows from the pathway input to its output. The theoretical maximum for three inputs (doses) is 1.5 bits if all the information is captured by the PIP₃/PI(3,4)P₂ dynamics. The height of each bar specifies the median capacity value, with error bars corresponding to the interquartile range (the distance between the first and the third quartiles). These were estimated following 40 bootstrap repetitions using 80% of the initial observations (default diagnostic settings in the SLEMI package (Jetka et al, 2019)). (C) Median information capacity in bits (log2) calculated from snapshot measurements at the indicated time points from the datasets in (A).

Figure 3. Oncogenic PIK3CA^H1047R blurs the dynamic encoding of ligand identity in HeLa cells.

(A) Live-cell fluorescence-based measurements of a miniFOXO-based AKT kinase translocation reporter (KTR) (Gross et al, 2019), stably expressed in HeLa clones with the indicated PIK3CA genotypes. The total duration of the time course was 300 min, with measurements obtained every 6 min. For each time point, the traces correspond to the mean proportion of cytoplasmic KTR signal, with shaded areas representing bootstrapped 95% confidence intervals of the mean (note that these may be too small to be seen on the figure). Y axis represents an approximation of AKT activity: C, cytoplasmic; N, nuclear. Single-cell numbers (n) are shown in the plots. For each condition, WT PIK3CA-expressing cells were also seeded at low density to confirm intra-experimental consistency irrespective of cell crowding. The data are representative of a minimum of two independent experiments per condition, performed in two independent CRISPR/Cas9 clones per genotype. Plots from all independent experiments are shown in Fig. EV2 and include control experiments with the 3xFS PIK3CA LOF mutant line. (B) Mutual information (MI) in bits (log2) for IGF1 versus each one of the indicated growth factors (EGF, epigen, insulin), calculated using the corresponding KTR trajectory responses (A) prior to inhibitor addition. MI values from individual experimental replicates are indicated as dots overlaid on barplots which correspond to the respective mean of each set of measurements. Because IGF1 gave highly robust KTR dynamics, associated with relatively low single-cell noise as reflected in consistently high MI values, it was chosen as the control stimulus in all experimental replicates. (C) A graphic summarizing the biochemical signal blurring caused by oncogenic PIK3CA^H1047R.

Figure 4. PIK3CA^H1047R amplifies EGF-dependent signaling in a time- and allele dose-dependent manner.

(A) Overview of the multiplexed mass cytometry (CyTOF) workflow for profiling of single-cell signaling markers in scaffold-free spheroid models. Following fixation and thiol-reactive organoid barcoding in situ (TOBis) (Qin et al, ; Sufi et al, 2021), up to 126 conditions are combined into a single sample for non-enzymatic single dissociation which ensures preservation of antibody epitopes, including post-translational modifications (PTMs). Subsequent staining with experimentally validated, metal-conjugated antibodies captures information about cell cycle state (such as cycling, non-cycling, apoptotic) and signaling state. (B) Mass cytometry (CyTOF) data from cycling, non-apoptotic HeLa spheroid cells with endogenous expression of WT PIK3CA or one (1xH1047R) or two (2xH1047R) copies of the oncogenic PIK3CA^H1047R. The spheroids were serum-starved for 4 h prior to stimulation with 100 nM EGF or IGF1, with and without the PI3Kα inhibitor BYL719 (alpelisib; 500 nM) as a control for signal specificity. Note that BYL719 was added at the same time as the growth factor, not as pre-treatment. The stippled line indicates the position of the peak in WT spheroids treated with vehicle (H₂O). The gray shading highlights the response region not shown by WT PIK3CA-expressing cells in the absence of stimulation. (C) Earth mover’s distance (EMD)-PHATE embedding of the signaling trajectories observed in the indicated HeLa cell genotypes. Single-cell distributions for the following signaling markers were used for EMD-PHATE processing (see also Fig. EV4): pAKT Ser⁴⁷³, pERK1/2 Thr²⁰²/Tyr²⁰⁴; Thr¹⁸⁵/Tyr¹⁸⁷, pNDRG1 Thr³⁴⁶, pS6 Ser^240/244, pSMAD2/3 Ser^465/467; Ser^423/425.

Figure 5. PIK3CA^H1047R amplifies an EGF-driven transcriptional signature and increases phenotypic diversity in an allele dose-dependent manner.

(A) Bulk transcriptional profiling of EGF-dependent immediate early and delayed early gene expression in HeLa spheroids with endogenous expression of either WT PIK3CA or one or two copies of PIK3CA^H1047R (1-2xH1047R). Expression values are relative and represented as log2 fold-changes, normalized internally to each genotype’s control (H₂O) response after 30 min of stimulation. All data were further normalized to the expression values of TBP (housekeeping gene). The data are representative of two independent experiments (indicated with solid and stippled lines) with one CRISPR-derived clone per genotype. IGF1 and EGF were used at 100 nM, either alone or in combination as indicated. Note the log2 scale of the y axis. (B) Representative fluorescence images of HeLa cells with the indicated genotypes during normal maintenance culture. The cells express a nuclear mCherry marker and were stained with CellMaskBlue and Phalloidin to demarcate their cytoplasm and actin cytoskeleton, respectively. The cells are representative of images from three independent wells per clone and one CRISPR/Cas9 clone per genotype (see also Appendix Fig. S2A for brightfield images of independent HeLa clones for each genotype). The scale bar in (B) corresponds to 100 µm. (C, D) The cytoplasmic images from all replicates were used for deep learning-based segmentation with Cellpose (Stringer et al, 2020). Cell shape solidity (C) and area (D) were quantified for n > 900 single cells per genotype. The P values in (C, D) were calculated according to a one-way ANOVA with Tukey’s Honest Significant Difference to correct for multiple comparisons. The exact P value in (C) is 0.00066. The P value in (D) is <0.0000001. The boxplots display five summary statistics. The lower and upper hinges correspond to the first and third quartiles (the 25th and 75th percentiles). The upper whisker extends from the upper hinge to the largest value no further than 1.5 * IQR from the hinge (where IQR is the interquartile range, or distance between the first and third quartiles). The lower whisker extends from the lower hinge to the smallest value at most 1.5 * IQR of the hinge. Data beyond the end of the whiskers are individually plotted outliers.

Figure 6. Corrupted signal transfer and EGF response amplification is conserved in homozygous PIK3CA^H1047R iPSCs.

(A) Mass cytometry data from cycling (pRB⁺) iPSC spheroid cells with WT PIK3CA or heterozygous or homozygous PIK3CA^H1047R expression. The spheroids were serum-starved for 2 h prior to stimulation with 100 nM EGF or IGF1, with and without the PI3Kα inhibitor BYL719 (alpelisib; 250 nM) as a control for signal specificity. Note that BYL719 was added at the same time as the growth factor, not as pre-treatment. The stippled line indicates the position of the peak in WT spheroids treated with vehicle (H₂O). The gray shading highlights the response region not shown by WT PIK3CA-expressing cells in the absence of stimulation. (B) Thresholding of the data in (A) to quantify the percentage of cells within each condition with a pAKT or pERK signal above the corresponding 95th percentile of vehicle (H₂O)-treated WT iPSCs at 30 min. The stippled lines indicate the maximum fraction of cells within this threshold for each genotype prior to growth factor stimulation. The data are from a confirmatory screen with one iPSC clone per genotype and n > 260 single cells per condition. (C) Graphical summary of the key observations of the impact of PIK3CA^H1047R expression in HeLa and iPSC cells. Created with BioRender (https://BioRender.com/v54q309).

Figure EV1. Systematic benchmarking of pleckstrin homology (PH) domain-based class I PI3K biosensors.

(A) Schematic of the optimized live-cell imaging setup to ensure that multiple comparisons could be performed in the same microenvironment, aided by fluidics for minimal physical perturbation during compound additions. To bring down the baseline of PI3K signaling, serum was removed from the cells 3 h prior to imaging start. D1, D2, D3 refer to day 1, day 2 and day 3 of the experimental workflow. (B) Schematic of the different wild-type and mutant PH domain constructs used for benchmarking, all cloned into the same plasmid backbone for consistent comparisons. The portion of the PIP₃-binding region in the PH domain of GRP1, which has often been used for live-cell detection of PIP₃, has an identical sequence to that in the ARF GEF ARNO. We therefore chose to include the latter in our comparisons given the tandem-dimer, modified version of this PH domain as biosensor for PIP₃ (Goulden et al, 2019).The shown alignments cover the conserved β1 strand, variable loop 1, and β2 strand of the PH domain fold. Of the four PH domains, only PH-AKT2 is capable of binding both PIP₃ and PI(3,4)P₂. The remaining PH domains only bind PIP₃ (Posor Y et al, 2022). The alanine (A) mutation in the phosphoinositide (PI) signature motif renders the mCherry-tagged mutant PH domain versions unable to bind phosphoinositides. (C) Quantification of total internal reflection fluorescence (TIRF) microscopy experiments comparing the response rate and dynamic range of individual PH domain-based PI3K reporters in response to pharmacological PI3Kα activation in HeLa cells. To correct for non-specific increases in biosensor signal at the plasma membrane, the intensity of each GFP-tagged wild-type PH domain was normalized to that of its mCherry-tagged mutant version. Experimental replicates and single-cell numbers are indicated. Two different configurations were tested for the BTK-derived PH domain: with and without the adjacent Tec homology (TH) domain. Only one experiment was performed with PH-BTK without TH because most of the cells failed to tolerate its expression. (D) TIRF microscopy of the PH-AKT2-derived biosensor in HeLa cells stimulated with 5 µM 1938, then treated with the PI3Kα inhibitor BYL719 (500 nM). Two independent experiments are superimposed to illustrate the expected inter-experimental variability. (E) Evaluation of the performance of the PH-TH version of BTK with N-terminal or C-terminal fluorescent protein fusion, with simultaneous removal of the nuclear export sequence, as in the original plasmid DNA used for subcloning of this reporter. All plots in (C, D, E) represent mean normalized reporter signal relative to time 0, with shading corresponding to +/- 1 standard deviation (SD).

Figure EV2. FOXO-based AKT kinase translocation reporter (KTR) setup and full set of experimental outputs.

(A) Schematic of the reporter, which was expressed stably in cells using transposon-based technology, and its mechanism of action. (B) Overview of the computational image and KTR data analysis pipeline which has been deposited on the accompanying OSF project site (10.17605/OSF.IO/4F69N). (C, D) The data in (C, D) are from all independent experiments performed across different genotypes, HeLa clones, cell densities, KTR reporter doses, operators and experimental sites for a robust evaluation of reproducibility. Experiments with high and low KTR transposon dose are shown in (C) and (D), respectively. For each time point, the traces correspond to the mean proportion of cytoplasmic KTR signal, with shaded areas representing bootstrapped 95% confidence intervals of the mean (note that these may be too small to be seen on the figure). Although we observed operator-dependent differences in EGF-induced signaling dynamics in WT PIK3CA cells, the overall pattern relative to IGF1, including the blurring of the response in mutant cells, remained consistent. This technical variability in EGF responses in WT cells is likely due to their sensitivity to the pressure/rate of delivery of the stimulus through the manual fluidics system (see https://doi.org/10.17504/protocols.io.261gedjkjv47/v1).

Figure EV3. Representative CyTOF data demonstrating that growth factor-induced signaling responses are observed only in cycling and non-apoptotic HeLa spheroid cells.

The plots on the left-hand side show the single-cell signal for cascade 3 (CASP3) cleaved at Asp¹⁷⁵ in the different pRB gates (pRB-negative^- or pRB-positive⁺ at Ser^807/811). The plots on the right show the corresponding pS6 Ser^240/244 signal in each gate. The overall experimental setup is as shown in Fig. 4. The shown data are from n = 1 clone per genotype but are representative of four independent experiments across two independent clones per genotype.

Figure EV4. Additional CyTOF data and independent experimental replicate with independent clones.

(A) The pNDRG1 and pSMAD2/3 signaling responses captured as part of the dataset shown in Fig. 4. The phosphorylation of NDRG1 on Thr³⁴⁶ is a marker of mTORC2 activation (García-Martínez and Alessi, 2008). The phosphorylation of SMAD2/3 (Ser^465/423; Ser^467/425) is a marker for activated TGFβ signaling which is associated with PIK3CA^H1047R phenotypes in human iPSCs (Madsen et al, 2021). (B) CyTOF data from an independent repeat of the experiment in Fig. 4, using independent CRISPR/Cas9-engineered, 3D-cultured HeLa clones, including the PIK3CA loss-of-function 3xFS clone as an additional control. The spheroids were serum-starved for 4 h prior to the indicated perturbations. The signaling data are from cycling, non-apoptotic cells. The stippled line indicates the position of the peak in WT spheroids treated with vehicle (H₂O). The gray shading highlights the response region not shown by WT PIK3CA-expressing cells in the absence of stimulation.

Figure EV5. Dose- and time-dependent IGF1 and EGF single-cell signaling responses in HeLa spheroid cells with WT or PIK3CA^H1047R (1–2 copies) expression.

(A, B) The plots in (A, B) are from two independent CyTOF datasets using independent CRISPR/Cas9-engineered, 3D-cultured HeLa clones stimulated with 1, 10 or 100 nM of IGF1 or EGF as a function of time. The spheroids were serum-starved for 4 h prior to the indicated perturbations. The signaling data are from cycling, non-apoptotic cells. The stippled line indicates the position of the peak in WT spheroids treated with vehicle (H₂O). The gray shading highlights the response region not shown by WT PIK3CA-expressing cells in the absence of stimulation. (C) Graphical summary of the key observations in the datasets in (A, B). A positive response is indicated with (+), the size of which indicates the response magnitude.