Regulation of PTEN translation by PI3K signaling maintains pathway homeostasis

Summary

The PI3K pathway regulates cell metabolism, proliferation and migration and its dysregulation is common in cancer. We now show that both physiologic and oncogenic activation of PI3K signaling increase the expression of its negative regulator PTEN. This limits the duration of the signal and output of the pathway. Physiologic and pharmacologic inhibition of the pathway reduces PTEN and contributes to the rebound in pathway activity in tumors treated with PI3K inhibitors and limits their efficacy. Regulation of PTEN is due to mTOR/4E-BP1 dependent control of its translation and is lost when 4E-BP1 is deleted. Translational regulation of PTEN is therefore a major homeostatic regulator of physiologic PI3K signaling and plays a role in reducing the pathway activation by oncogenic PIK3CA mutants and the antitumor activity of PI3K pathway inhibitors. However, pathway output is hyperactivated in tumor cells with coexistent PI3K mutation and loss of PTEN function.

eTOC blurb

Mukherjee et al show that PI3K pathway homeostasis is mediated by its control of translation of PTEN, its negative regulator. Perturbations of the pathway are buffered by changes in PTEN expression, thus limiting the effects of physiologic and oncogenic pathway activation and reducing the therapeutic effects of pathway inhibitors.

Graphical Abstract

Introduction

The PI3K pathway is activated by insulin and other growth factors and regulates metabolism, proliferation, and migration(Auger et al., 1989; Cain and Ridley, 2009; Ruderman et al., 1990; Ward and Thompson, 2012). Receptor activation of Class 1 PI3-kinases catalyzes phosphorylation of their substrate PI-(4,5)diphosphate (PIP2) and the formation PI-(3,4,5)triphosphate (PIP3)(Alessi et al., 1996; Auger et al., 1989; Metz and Houghton, 2011; Ruderman et al., 1990). PIP3 binds to pleckstrin homology domains in proteins and causes them to compartmentalize to various membranes where they become activated. These proteins include the PDK1 and AKT kinases, which cause activation of mTOR-dependent signaling, protein translation, and ribosomal biogenesis (Alessi et al., 1997; Dibble and Cantley, 2015; Ma and Blenis, 2009; Manning and Toker, 2017). The pathway is negatively regulated by PTEN, a phosphatase that dephosphorylates PIP3 on the 3-position and other phosphatases as well as by AMP kinase and the TSC complex which negatively regulates mTOR (Garami et al., 2003; Stambolic et al., 1998). The PI3K signal is also limited by AKT and mTOR-dependent feedback inhibition of receptor expression and signaling (Chandarlapaty et al., 2011; Hsu et al., 2011; O’Reilly et al., 2006; Yu et al., 2011).

Genetic lesions that dysregulate PI3K signaling, most commonly PI3K mutation, HER2 amplification, and events that inactivate PTEN are prevalent in cancer (Millis et al., 2016; Vivanco and Sawyers, 2002; Zhao and Vogt, 2008). With the exception of HER2 amplification, they usually coexist with other drivers (RAF, RAS mutation, activated ER and AR signaling) that are, by themselves, sufficient to drive tumor growth (TCGA PanCancer Atlas dataset, cBioPortal). The therapeutic effects of inhibitors of components of the PI3K pathway (PI3K, AKT, mTOR) tend to be modest (Fruman and Rommel, 2014; Lee et al., 2015). This has been attributed to toxicity and relief of PI3K pathway-dependent feedback inhibition of upstream signaling (Chandarlapaty et al., 2011; Rozengurt et al., 2014; Schwartz et al., 2015; Serra et al., 2011)

Here we report a new mechanism for regulating PI3K signaling. We find that translation of the PTEN protein is driven by mTOR phosphorylation of 4E-BP1. Physiologic or oncogenic activation of PI3K/AKT/mTOR signaling leads to increased PTEN expression that limits the duration of ligand-induced signaling and the pathway output driven by PI3K mutants. By contrast, the effects of physiologic or pharmacologic inhibition of the pathway are limited by decreasing expression of PTEN. Regulation of PTEN expression by changes in PI3K signaling buffers the effects of these perturbations and comprises a powerful mechanism for homeostatic regulation of the pathway. It provides an explanation for the co-occurrence of PTEN loss and PI3K mutation in some cancers. Finally, the decrease in PTEN expression that occurs in tumors treated with PI3K inhibitors plays a major role in mediating the rebound in pathway activity that limits their efficacy.

Results

PTEN expression is regulated by the PI3K pathway

Activation of the PI3K pathway induces feedback inhibition of receptor expression and signaling. Feedback is exaggerated in tumor cells in which the pathway is hyperactivated and feedback is relieved when the pathway is inhibited pharmacologically. This reactivates signaling and attenuates the therapeutic effects of pathway inhibition (Chandarlapaty et al., 2011; Rodrik-Outmezguine et al., 2011; Schwartz et al., 2015). Combined inhibition of PI3K-alpha (with BYL-719) and PI3K-beta (with AZD8186) in BT474, a breast cancer cell line with HER2 amplification and mutant PI3KCA, caused an initial reduction of AKT phosphorylation and mTOR activity (S6K, S6 and 4E-BP1 phosphorylation) followed by a rebound, accompanied by induction of expression of HER3 (Fig 1A) and other RTKs (Chandarlapaty et al., 2011). AKT phosphorylation began to rebound 2 hours after drug addition. To our surprise, PTEN expression declined as well, 2 to 4 hours following inhibition and fell 50% by 20 hours, and to 30% of initial levels by 48 hours (Fig 1A, 1B). PI3K-beta inhibition had little effect in these cells and the kinetics of PTEN reduction with combined inhibition of PI3K-alpha and - beta approximated those caused by PI3K-alpha inhibition alone (Fig S1A).

Figure 1. PTEN expression is regulated by the PI3K pathway.

A-B. A. BT474 cells were treated with BYL-719 (1uM) and AZD8185 (250nM) for the indicated times and probed for PTEN expression and AKT pathway activation by immunoblotting B. PTEN and pAKT T308 from 1A and two other independent experiments (n=3) were quantified and normalized to actin expression; the mean and S.E.M. values were plotted as a function of time C. BT474 cells were starved for serum or amino acids D-E. D. MCF7 and CHO cells were stimulated with Insulin, Heregulin and EGF (100ng/ml each) E. The fold-increases in PTEN expression and pAKT T308 from their basal values after ligand stimulation of MCF7 was quantified and the means and S.E.M.s were plotted (n=3) F. Immunoblots for PTEN and phosphorylated downstream targets of the PI3K pathway in MCF10A, MCF10A-PIK3CA-H1047R, and MCF10A-HER2. PTEN and pAKT T308 levels were quantified and normalized as above and represented as fold-increase over MCF10A G. MCF7 parental (MCF7 with mutant PIK3CA) and MCF7 PIK3CA-WT cells were analyzed for PI3K signaling by immunoblotting. Data was quantified as in 1F and represented as fold-reduction compared to parental. Note; Lanes for targets in 1F and 1G were run on the same gel and extra lanes cropped out, indicated by space.

We asked whether PI3K inhibition affected PTEN expression in other cancer and non-cancer cell lines. These include cancer cells with activated PI3K mutants treated with inhibitors of PI3K-alpha and PI3K-beta in breast cancers-T47D (PIK3CA H1047R), MCF7 (PIK3CA E545K), BT20 (H1047R), endometrial cancer-MFE280 (H1047Y), colorectal cancer-NCIH508 (E545K). A diffuse large B-cell lymphoma cell line (SUDHL10) with WT PI3K was treated with an inhibitor of PI3K-delta (CAL101), the dominant PI3K in these tumors (Fig S1B–C). PI3K inhibition reduced PTEN expression in the above models and in untransformed cell lines (mouse embryonic fibroblasts-3T3L1, breast epithelial line -MCF10A, kidney epithelial line-HEK293T) within 2 to 8 hours (Fig S1B–D). In all of these models, rebounds in AKT phosphorylation occurred between 1 and 4 hours after PI3K inhibition. These data suggest that downregulation of PTEN expression after PI3K inhibition is common and occurs in untransformed and tumor cells.

Serum and growth factors such as insulin activate PI3K signaling whereas glucose and amino acids are required for mTOR activation (Sancak et al., 2008; Ward and Thompson, 2012). Upon serum or amino acid starvation of MCF7 or BT474 cells, mTORC1 kinase activity (S6K and 4E-BP phosphorylation) declined in 4 to 8 hours with a concomitant decline in PTEN expression (Fig 1C and Fig S1E). In amino acid deprived cells, these phenomena were associated with a marked increase in AKT T308 and S473 phosphorylation. In serum-starved cells, AKT phosphorylation initially decreased and subsequently rebounded (Fig 1C and Fig S1E). Thus, both physiologic and pharmacologic inhibition of PI3K signaling caused decreased expression of the PTEN protein that was temporally associated with increased AKT phosphorylation.

We asked whether induction of PI3K signaling increased PTEN expression as well (Fig 1D–E, S1F). MCF7 cells or immortalized hamster ovarian epithelial cells (CHO) were serum-starved for 24 hours and then treated with a combination of Heregulin, insulin and EGF. PTEN expression increased 2–4 hours after pathway stimulation, reaching a maximum of 3.2-fold induction in MCF7 cells and a 4-fold induction in CHO cells. Ligand stimulation caused a rapid induction of AKT T308 and S6K phosphorylation followed by a decline that correlated temporally with induction of PTEN expression (Fig 1D–E, S1F).

Oncogenic Activation of PI3K Signaling Increases PTEN Expression

We asked whether chronic activation of PI3K signaling by oncogenic lesions also increases PTEN expression. In MCF10A cells, expression of an activating mutant of PI3K (H1047R) or overexpression of HER2 increased phosphorylation of AKT, PRAS40 and mTOR substrates (S6K and 4E-BP) and concomitantly increased PTEN expression (Fig 1F). PTEN increased 3-fold in the PIK3CA mutant cells and 8-fold in those with HER2 overexpression, whereas AKT phosphorylation increased 10-fold in the former and 20-fold in the latter (Fig 1F). These data suggest that levels of PTEN expression can serve as a measure of PI3K pathway output. Consistent with this idea, replacement of the PIK3CA E545K mutation in MCF7 with WT PIK3CA resulted in an approximately 2-fold reduction in PTEN expression and a 3.5-fold reduction in phosphorylated AKT (Fig 1G). Induction of PTEN in the HER2 overexpressing cells was PI3K dependent as it was suppressed by PI3K inhibitors (Fig S2A).

PIP3-phosphatase activity of PTEN immunoprecipitates from whole cell lysates was assessed (see Methods). Activity declined 4 hours after PI3K inhibition in BT474 cells and fell 30% in 24 hours (Fig S2B). Ligand stimulation of MCF7 cells resulted in concomitant increases in PTEN lipid phosphatase activity and PTEN expression (Fig S2C). Moreover, PIP3-phosphatase activity was elevated in both MCF10A PIK3CA H1047R (30% over control MCF10A) and in HER2 overexpressing MCF10A (100% over control) while PTEN protein levels increased 3-fold and 8-fold respectively (Fig S2D–E). Thus, cellular PTEN-associated PIP3 phosphatase activity was directly correlated with changes in PTEN expression.

Regulation of PTEN expression by PI3K is mediated by mTOR

In BT474, selective inhibitors of the kinase activity of each of three nodes of the pathway (Lapatinib-HER2, MK2206-AKT1,2, AZD8055-mTOR), reduced PTEN expression with similar kinetics (Fig 2A) and was accompanied in each case with a rebound in AKT phosphorylation. Inhibition of mTOR kinase with AZD8055 in MEFs (fibroblasts) or YUM3.3 (a BRAF V600E mutant, PTEN WT murine melanocyte) also reduced PTEN expression (Fig S2G–H).

Figure 2. Regulation of PTEN expression by PI3K is mediated by mTOR.

A-B. A. BT474 cells were treated with HER2 inhibitor Lapatinib (2uM), AKT inhibitor MK-2206 (2uM) or mTOR kinase inhibitor AZD8055 (500nM) and changes in the indicated targets were assessed by immunoblotting B.BT474 cells treated with Rapamycin (50nM) C. PI3K/AKT/TSC regulation of mTOR D.LAM TSC2 null cells treated with PI3K inhibitors ((BYL-719 (1uM) + AZD8186 (250nM)) or with mTOR kinase inhibitor AZD8055 (500nM) (PTEN; LI; Long Isoform, SI; Short Isoform) E.TSC2−/− and control MEFS were analyzed for PI3K signaling by immunoblot.

Since inhibitors of PI3K, AKT or mTOR all downregulated PTEN expression, we asked whether inhibition of mTOR, the most downstream component, is sufficient. We inhibited mTORC1 with Rapamycin. Rapamycin uncouples mTORC1 from AKT by inhibiting the mTORC1 complex while relieving upstream feedback and inducing AKT phosphorylation (O’Reilly et al., 2006). In BT474, 50nM Rapamycin inhibited the phosphorylation of mTORC1 targets S6K and 4E-BP and induced AKT phosphorylation within 1 hour of drug administration. This was associated with a reduction in PTEN expression 4 hours later (Fig 2B). These data suggest that the reduction of PTEN expression by PI3K inhibitors is mediated by mTORC1 inhibition.

To test this hypothesis, we utilized TSC2-null lymphangioleimyomatosis (LAM) cells (Steagall et al., 2018). AKT activates mTOR by phosphorylating and inhibiting its negative regulator, the TSC1/2 complex (Inoki et al., 2002) (Fig 2C). LAM cells lack TSC2 inhibitory function, so that, in these cells, mTOR activation is PI3K independent (Darling et al., 2010; Huang and Manning, 2009). In these cells, PI3K inhibition inhibited phosphorylation of AKT and its substrate PRAS40, but not that of mTOR substrates S6K and 4E-BP and PTEN levels remained roughly constant (Fig 2D). In contrast, the mTOR kinase inhibitor AZD8055 inhibited the phosphorylation of mTOR substrates, caused a marked decrease in PTEN expression and relieved mTOR-dependent feedback inhibition of AKT. Moreover, stimulation of LAM cells with growth factors induced AKT phosphorylation, but not phosphorylation of mTOR substrates or expression of PTEN (Fig S2I). Thus, PI3K signaling affects PTEN expression by regulating mTOR (Fig 2C). In support of this conclusion, in TSC2 null MEFs, phosphorylation of mTOR substrates and the expression of PTEN are increased compared to that observed in TSC2 WT controls (Fig 2E).

mTOR controls PTEN expression by regulating its 4E-BP1-dependent translation.

PTEN mRNA levels did not change until 48 hours of PI3K inhibition at which point it increased by 2.5-fold (Fig S3A). mTOR is a regulator of cap-dependent translation (Mamane et al., 2006), so we employed L-azidohomoalanine (AHA), an azide-containing methionine analog, to determine whether it regulates PTEN expression in this way (Tom Dieck et al., 2012). BT474 cells were starved for methionine in the absence or presence of PI3K inhibitors (BYL-719 and AZD8186), or the mTOR kinase inhibitor AZD8055 or the translation inhibitor cycloheximide for 30 minutes and were then pulsed with AHA for 2 hours (see methods). AHA-labeling of cyclin D1, a protein known to be regulated by mTOR-driven cap-dependent translation, (Cowling, 2010) and of PTEN were significantly reduced by PI3K or mTOR inhibitors (Fig 3A). Although translation of both cyclin D1 and PTEN was markedly inhibited by all three treatments, total cyclin D1 declined as well but total PTEN did not. This result is consistent with the short half-life of the former and the longer half-life of the latter. In contrast, the translation of HER2 protein was induced by PI3K inhibitors, consistent with the known induction of both HER2 mRNA and protein expression by AKT pathway inhibitors (Chandarlapaty et al., 2011; Muranen et al., 2012). Thus, PI3K signaling affects PTEN expression by regulating the mTOR-dependent translation of PTEN.

Figure 3. mTOR controls PTEN expression by regulating its 4E-BP1-dependent translation.

A. BT474 cells were starved of methionine and treated with vehicle or combination of PI3K inhibitors BYL-719 (1uM) and AZD8186 (250nM) or mTOR inhibitor AZD8055 (500nM) or Cycloheximide (50ug/ml) for 30 min followed by an AHA pulse (50uM) for two hours. Strep-avidin beads were used to pull down labeled translated proteins from cell lysates and the indicated proteins were analyzed by immunoblotting (see methods). B-C. B. Methionine starved BT474 cells were first pulsed with ³⁵S-methionine for 30 min and then chased with cold complete media in the absence or presence of PI3K inhibitors for the indicated time intervals (see methods). PTEN protein was immunoprecipitated and analyzed by autoradiography. PTEN protein was quantified, normalized to t=0 and the mean and S.E.M. were plotted as a function of time (n=3) C. Autoradiography analysis of ³⁵S-methionine pulse-chase of PTEN from B. D. BT474 cells were treated with BYL-719 (1uM) and AZD8186 (250nM) and analyzed for changes in 4E-BP and eIF4E expression and phosphorylation E. BT474 with a doxycycline-inducible 4E-BP-4A were treated over time with 100ng/ml of doxycycline (Note; the arrow indicates the pAKT T308 band) F. BT474 cells were transduced with Cas9 and sgRNAs targeting GFP or 4E-BP1 and treated with BYL-719 (1um) and AZD8186 (250nM) G-I. G. Illustration of the dual luciferase reporter system. H. BT474 WT and 4E-BP1KO cells expressing pGL3 (Vec) or pGL3-PTEN 5’UTR were treated with AZD8055 (500nm) (AZ) or DMSO (C) for 24h and relative luciferase activity (RLU) was determined. The mean and S.E.M are represented as bar graphs (n=3). I. Decrease in RLU in AZD8055 treated cells compared to control is represented as % inhibition of translation. Note; Lanes for targets in 3E were run on the same gel and extra lanes cropped out, indicated by space.

We asked whether PI3K signaling regulates PTEN stability as well. Pulse labeling of PTEN with ³⁵S-methionine for 30 minutes followed by a chase with cold methionine showed that the half-life of methionine-labeled PTEN is approximately 20 hours. PI3K inhibitors did not alter this half-life, suggesting that they do not induce PTEN degradation (Fig 3B–C). Furthermore, in BT474 cells treated with either RapaLink-1 (Rodrik-Outmezguine et al., 2016), a more potent mTOR kinase inhibitor or cycloheximide, PTEN expression declined with the same kinetics observed in BT474 treated with PI3K inhibitors (Fig S3B–D) and with the kinetics of decrease of pulse-labeled PTEN during the chase. Thus, the regulation of PTEN expression by PI3K signaling is predominantly due to regulation of its translation by mTOR.

The major effects of mTOR on translation are mediated by its phosphorylation of eIF4E binding proteins (4E-BPs). Dephosphorylated 4E-BPs bind to eIF4E/5’capped-mRNA and, by doing so, interfere with the assembly of the pre-initiation complex (Gingras et al., 1999; Sonenberg and Hinnebusch, 2009). Phosphorylation of 4E-BPs causes their release from eIF4E, allows the initiation complex to form and protein translation ensues. Inhibition of PI3K signaling in BT474 caused a rapid decrease in 4E-BP1 phosphorylation, reaching a minimum in two hours (Fig 3D and Fig S3E). This was associated with subsequent reduction in PTEN expression beginning 2–4 hours after PI3K inhibition. Dephosphorylation of 4E-BP1 causes it to bind to eIF4E and prevents phosphorylation of eIF4E on S209 (Furic et al., 2010; Muller et al., 2013). The phosphorylation of eIF4E on S209 declined after PI3K inhibition suggesting its increased binding to 4E-BP1; expression of eIF4E and 4E-BP1 did not change (Fig 3D, FigS3F).

4E-BP-4A is a 4E-BP1 in which its phosphorylation sites were replaced with alanines (T37, T46, S65 and T70). This mutant is unphosphorylatable and binds constitutively to eIF4E causing inhibition of cap-dependent translation (Cai et al., 2014; She et al., 2010). Stable expression of a doxycycline-inducible 4E-BP1–4A vector in BT474 resulted in its expression 1 hour after doxycycline induction (lower dark band) (Fig 3E). Phosphorylation of eIF4E S209 began to decline between 1 and 4 hours after induction concomitantly with a decline in PTEN levels and an induction of AKT phosphorylation. Thus, dephosphorylated 4E-BP1 is sufficient to inhibit PTEN expression. Levels of AKT phosphorylation decreased from their maxima between 8 and 16 hours after doxycycline treatment, possibly due to inhibition of upstream signaling by activated AKT (Chandarlapaty et al., 2011), but remained significantly elevated 24 hours after doxycycline induction. Despite induction of AKT phosphorylation, PTEN levels remained low 24 hours after induction of 4E-BP-4A. This is consistent with the inability of mTOR activation to overcome inhibition of PTEN translation by the unphosphorylatable 4E-BP1.

These data suggest that mTOR regulates PTEN translation by regulating 4E-BP phosphorylation. To confirm this, we deleted 4E-BP1 from MCF7 and BT474 with CRISPR-Cas9 technology. In contrast to parental cells, PI3K pathway inhibitors did not inhibit PTEN expression in cells without 4E-BP1(4E-BP1KO) (Fig 3F and Fig S3G–I). The initial decline in phosphorylation of downstream pathway components (pAKT S473, pPRAS40, pS6K) was similar in both the parental and 4E-BP1 depleted BT474 (Fig 3F). However, rebound of pAKT and pPRAS40 was both delayed and reduced in the cells without 4E-BP1. In MCF7 4E-BP1KO cells, PI3K inhibitors rapidly inhibited PI3K signaling but PTEN was unaffected and there was almost no rebound in AKT phosphorylation (Fig S3G). PTEN reduction was completely abrogated in the 4E-BP1KO cells treated with the mTOR kinase inhibitor AZD8055 and the rebound in AKT phosphorylation on T308 was significantly reduced (Fig S3H, I).

4E-BP1 regulates translation of capped mRNAs by binding to their 5’UTRs. The 5’UTR of PTEN mRNA is an approximately 1000bp sequence upstream of the start codon (Han et al., 2003). We generated a construct in which the PTEN 5’UTR was upstream of a Firefly luciferase reporter (GenBank Accession Number U47298) (Fig 3G). The internal control for the experiment was a Renilla luciferase reporter (GenBank Accession Number AF025844), the translation of which was driven by a cap-independent IRES element. When these reporters were transiently expressed in BT474 cells, inhibition of mTOR kinase for 24 hours caused a 50% reduction in relative luciferase activity (Fig 3H, 3I, S4A). In contrast, mTOR inhibition had no effect on relative luciferase activity when the reporters were expressed in BT474 4E-BP1KO cells (Fig 3H, Fig 3I, S4A). Overall, these results provide multiple lines of evidence that PTEN is controlled by PI3K-mTOR regulation of its 4E-BP1-sensitive cap-dependent translation.

A computational model predicts the PTEN-dependent feedback regulation of the PI3K network

Our data reveals a previously unrecognized mechanism of feedback regulation of the PI3K pathway. We tested whether our experimental results could be predicted by simple mathematical models employing a set of previously known characteristics of PI3K signaling as well as its regulation of PTEN expression. We examined the relationship between AKT S473 phosphorylation and PTEN expression in BT474 and MCF7 after PI3K pathway inhibition with drugs (Fig 1A, S1B) or activation with growth factors (Fig 1D, S4B). When fold-changes in PTEN and pAKT 24 and 48 hours (long-term) after inhibition or stimulation were plotted, there was a positive linear relationship between pAKT and PTEN, with slopes of 0.55 and 0.5 (shown as the shaded regions) for BT474 and MCF7 cells respectively (Fig 4A–B). In contrast, at earlier time points of 1, 4 and 8 hours, values of pAKT and PTEN obtained experimentally diverged after stimulation or inhibition (Fig 4A and Fig 4B). After pathway inhibition, pAKT decreased rapidly and then rebounded as PTEN levels decreased. Over this period, pAKT and PTEN expression were linearly anti-correlated (Fig 4A–B). This transient anti-correlation is consistent with the different kinetics of change of pAKT and PTEN expression. The negative correlation of the short-term values and the positive correlation of the long-term values of PTEN and pAKT can be explained by negative feedback regulation.

Figure 4. A computational model predicts the PTEN-dependent feedback regulation of the PI3K network.

A-B. BT474 and MCF7 cells were inhibited with PI3K inhibitors BYL719, AZD8186 (Fig 1A, S1B) or stimulated with a combination of EGF, Heregulin and Insulin-100ng/ml each (Fig 1D, S4B) for 1, 4, 8, 24 and 48 hours. PTEN and pAKT S473 values were normalized to actin expression and plotted along the X and Y axes respectively and connected by dotted lines. Slopes of straight lines passing through the 24 and 48 hours PTEN-pAKT values were also computed C. Schematic of the PI3K/AKT/mTOR computational model D. The model in C was trained (parameters of the model evaluated) on the experimental values as a function of time of pAKT S473, PTEN and p4E-BP T37/46 in BT474 treated with PI3K inhibitors (n=3, mean and S.E.M plotted, for representative experiment see Fig S3E). The experimental data (symbols with standard error of mean) and final model fit (line joining the symbols) are shown E. Model predictions (solid lines) and experimental data (symbols) of the effects of PI3K inhibition on PTEN and pAKT S473 in BT474 4E-BP1KO cells are shown. 4E-BP1WT experimental data and model fit are shown for reference F. Experimental data from BT474, BT20 and SUDHL10 cell lines (symbols) and the model predictions for BT20 and SUDHL10 (solid lines) of the effects of PI3K inhibitors on PTEN and pAKT S473 are shown.

Our simple mathematical model connects pAKT and PTEN with a feedback loop and described in two equations shows that the apparent steady state values of pAKT and PTEN can be positively correlated with each other (Fig S4C, Methods S1). The same model predicts transient negative linear correlation between pAKT and PTEN expression before they reach their steady state values. The model also provides an explanation for the similarity of the slopes of pAKT-PTEN values (long-term) in two different cell lines (see Methods S1). It accurately simulates the effects of amino acid starvation or mTORC1 inhibition on PTEN and pAKT if the kinetic constant relating PTEN levels to pAKT is decreased (see Methods S1) (Fig 1C, Fig 2B). Thus, this simple model reproduces responses of the system to negative or positive perturbations of the PI3K-AKT-PTEN pathway.

The model does not, however, capture the complexity of the real network and thus does not fully predict the non-steady state signaling response. We therefore developed a more complex model incorporating a more detailed description of activation of PI3K/mTOR signaling (Fig 4C and Methods S2) including receptor activation, the lipid phosphatase activity of PTEN, AKT kinase activation and AKT regulation of RTKs via FOXO transcription factors. It includes a quantitative description, based on our data, of mTOR/4E-BP regulation of PTEN expression. The model was trained on experimental data sets that describe the dynamic changes in expression and phosphorylation of components of the pathway in BT474 cells after PI3K inhibition (Fig 4D and Methods S2). Model parameters were determined by an optimization algorithm that varied the values until the model output matched experimental data as shown in Fig 4D (also see Methods S2).

To test the predictability of the model, we removed the effect of 4E-BP1 on PTEN expression and compared the results of the revised model with the data obtained when 4E-BP1 KO cells were treated with PI3K inhibitor (Fig 4E). There was close agreement between the experimental results and the predictions of the model for the kinetics of changes in PTEN expression and AKT S473 phosphorylation. These results support the importance of mTOR regulation of PTEN expression for controlling the dynamics of pathway inhibition. We then tested the ability of the model to reproduce the data obtained when PI3K was inhibited in different cell lines (Fig 4F, Fig S1B–C). First, we considered the BT20 cell line. We found that the limited response of PTEN expression to PI3K inhibition in the BT20 cell line was reproduced by the model when the experimental values for total 4E-BP1 expression and the ratio of p4E-BP T37/46 to total 4E-BP1 in BT20 were substituted for those obtained in BT474 (Fig 4F, see Methods S2 for relative quantifications). The model prediction of the pAKT S473 response to PI3K inhibition in BT20 also matched the experimental data (Fig 4F). These data suggest that levels of 4E-BP1 and the ratio of phosphorylated to total 4E-BP1 may be important determinants of PTEN dynamics and pAKT kinetics in response to PI3K inhibition.

In the SUDHL10 B cell Lymphoma cell line, the decrease in PTEN was more profound and occurred 2-fold faster than in BT474 cells. It was sufficient to substitute this increased rate of PTEN loss (see Methods S2) in the model to reproduce the dynamics of PTEN and pAKT changes after PI3K inhibition in this cell (Fig 4F). Finally, the computational model trained for BT474 cells was able to account for the levels of pAKT and PTEN expression in nine of ten breast cancer cell lines with PI3K mutation, HER2 amplification or both (Methods S2). Therefore, these integrative experiment-modeling results support the importance of PI3K regulation of PTEN feedback loop in determining the static and dynamic states of the PI3K pathway and its response to inhibition.

4E-BP1-dependent regulation of PTEN expression limits the effects of perturbations of the pathway

Ligand stimulation of cells results in pathway activation and an induction of PTEN expression that correlates temporally with the decline in phosphorylated AKT that occurs after its initial induction (Fig 1D). We asked whether PTEN induction, like its downregulation, is 4E-BP1 dependent and whether its induction limits the duration of ligand-stimulated signal.

We used BT474 and 3T3L1 (mouse adipocytes). Versions of each cell line in which 4E-BP1 was knocked out were compared to parental cells. BT474 cells were stimulated with insulin, Heregulin and EGF after serum starvation overnight. In BT474, growth factors rapidly induced pAKT T308 (Fig 5A). Induction reached a peak after 4 hours of treatment, markedly declined by 8 hours and fell to undetectable levels between 8 and 24 hours after treatment. pS6K levels remained elevated longer, peaking at 8 hours, and fell almost to the baseline by 24 hours. p4E-BP T37/46 levels remained elevated longer and were still elevated at 24 hours (Fig 5A). These changes were accompanied by an induction of PTEN expression within 4 hours of ligand stimulation that increased approximately 2.5-fold 24 hours later. By contrast the PTEN levels in the BT474 4E-BP1KO ligand stimulated cells remained roughly constant and increased slightly 24 hours later (Fig 5A–B). Thus, both reduction and induction of PTEN expression by PI3K perturbations are 4E-BP1 dependent. Peak phosphorylation of AKT T308 after ligand stimulation was unaffected by depletion of 4E-BP1, but the subsequent decline in phosphorylation was blunted, with pAKT levels still elevated 8 and 24 hours after stimulation and similar to that seen one hour after stimulation (Fig 5A–B). Levels of pS6K were still significantly elevated after 24 hours of stimulation, compared to controls. Thus, duration of signaling activation in parental cells is inversely correlated with induction of PTEN expression. In cells in which 4E-BP1 is knocked out, the duration of ligand induced PI3K signaling is increased and PTEN levels remain constant. Similar results were obtained in 3T3L1 adipocytes treated with 1ug/ml of insulin (Fig S5A). These data support the conclusion that activation of 4E-BP-dependent translation by growth factor stimulation limits the duration of the signal. We believe it is likely that induction of PTEN plays role in this process.

Figure 5. 4E-BP1-dependent regulation of PTEN expression limits the effects of perturbations of the pathway.

A-B. A. BT474 and BT474 lacking 4E-BP1 were treated with 100ng/ml each of Insulin, Heregulin and EGF probed for the indicated proteins by immunoblotting. B. PTEN and pAKT T308 values from A. were quantified, normalized and plotting as a function of time C. BT474 and BT474 lacking 4E-BP1 were treated with increasing doses of BYL-719 and 250nM of AZD8186 for 24h. pAKT T308 values normalized to t=0 was plotted. D-E. D. BT474 4E-BP1WT and KO xenografts were treated with BYL-719 (30mg/kgs) and AZD8186 (70mg/kgs) and the indicated targets evaluated by immunoblotting (n=2) E. BT474 4E-BP1WT and KO xenografts were treated with BYL-719 30mg/kgs daily and AZD8186 70mg/kgs B.I.D. Tumor volumes were measured and mean and S.EM. values represented (n=5). p value was computed using Student’s t-test. Inset; Percent tumor volume of PI3K inhibitor treated 4E-BP1WT and KO xenografts normalized to vehicle treated animals F-G. F. BT474 cells were treated with BYL-719 (1uM) and AZD8186 (250nM) for 2 weeks G. BT474 xenografts (n=5) were treated with BYL-719 30mg/kgs daily and AZD8186 70mg/kgs B.I.D. for 3 weeks (the arrow indicates the pAKT T308 band) Note; Lanes for targets in 5A and 5G were run on the same gel and extra lanes cropped out, indicated by space.

Previous studies show that the therapeutic effects of PI3K inhibitors are limited by pathway reactivation. Our data suggests that 4E-BP1 dependent suppression of PTEN translation plays a role in this process (Fig 3F, S3G, Fig 4E–F). We tested whether 4E-BP1 loss enhances the efficacy of PI3K inhibitors. BT474 4E-BP1 WT and KO cells were treated with the indicated doses of BYL-719 and a fixed dose of AZD8186 (250nM). Cells lacking 4E-BP1 were more sensitive to inhibition of PI3K than those with 4E-BP1, as shown by the 3-fold shift in the IC50 (450nm to 150nM) for inhibition of pAKT T308 after 24 hours of treatment (Fig 5C, Fig S5B). These results were recapitulated in vivo in BT474 4E-BP1WT and KO xenografts. PTEN levels were downregulated within 8 hours of PI3K inhibition in the 4E-BP1WT tumors (Fig 5D) but did not decline in the 4E-BP1KO tumors. Phosphorylation of AKT T308 decreased markedly in both models 8 hours after drug administration. In the 4E-BP1WT xenografts, it rebounded and was supra-induced by 24 hours whereas, in 4E-BP1KO tumors, pAKT T308 did not rebound and remained significantly inhibited. BT474 4E-BP1KO xenografts grew faster than the WT xenografts and were also inhibited more potently by PI3K inhibitors (Fig 5E). In the parental BT474, tumor growth was slowed by PI3K inhibitors, but not arrested, and 28 days after treatment, tumors had increased in size to 50% that of the size attained by the control (inset). In contrast, after 28 days of treatment, 4E-BP1KO BT474 tumor volume remained 16% of that of the DMSO control (inset). The growth of the 4E-BP1KO BT474 tumors was completely arrested by PI3K inhibitors compared to a 600% increase in the control. Thus, inhibition of 4E-BP1-dependent translation by PI3K inhibitors reduces PTEN expression and the antitumor effects of the inhibitors.

We asked whether PTEN downregulation persisted with prolonged PI3K inhibition. Two weeks of administration of PI3K-alpha and PI3K-beta inhibitors to BT474 cells in tissue culture or three weeks of daily administration of drugs in vivo to BT474 xenografts in mice resulted in a new steady state characterized by reduced 4E-BP1 phosphorylation and PTEN expression, and increased AKT T308 and S473 phosphorylation (Fig 5F, 5G). Thus, it is possible that prolonged pharmacologic inhibition of PI3K will reduce effective anti-tumor activity after the initial exposure to the drug (Will et al., 2014).

Inhibition of PTEN translation by PI3K inhibitors causes reactivation of the pathway

4E-BP1-dependent inhibition of translation reduces PTEN expression in cells treated with PI3K inhibitors. Moreover, knocking out 4E-BP1 markedly reduces the rebound in PI3K signaling in cells treated with these drugs and improves their ability to inhibit tumor growth (Fig 3F, 5D–E). We asked whether these phenomena are connected, that is, to what degree is the 4E-BP-dependent reduction in the efficacy of PI3K inhibitors due to reduction in PTEN expression.

We approached this issue by generating a vector that drives PTEN expression in an mTOR -independent manner. We asked whether the translation of PTEN mRNA lacking the 5’UTR, would be insensitive to PI3K pathway inhibition and, if so, whether expression of this mRNA would affect the rebound in AKT phosphorylation that follows PI3K inhibition. Doxycycline-inducible vectors were generated encoding PTEN mRNA with either the full length 5’UTR (5’UTR-PTEN) or just the coding sequence without the 5’UTR (CDS-PTEN). MCF7 cells with either construct were treated with 100ng/ml of doxycycline (Fig S6A). Induction of 5’UTR-PTEN and CDS-PTEN reached significant levels 24 hours after drug treatment and remained upregulated for up to 72 hours. In both models, induction downregulated PI3K signaling (Fig S6A).

Cells were treated with doxycycline for 24 hours (at which point, only partial suppression of the PI3K pathway components was observed) and then treated with PI3K-alpha and - beta inhibitors. In 5’UTR-PTEN transfected cells, PTEN levels began to decrease 4 to 8 hours after PI3K inhibition and steadily thereafter (Fig 6A). Pathway inhibition occurred rapidly after drug addition and rebounded 4 hours later. By contrast, in cells transfected with CDS-PTEN, PTEN levels did not significantly change during the 48 hours of PI3K inhibition and pathway rebound was both delayed and markedly reduced in amplitude (Fig 6A). AKT phosphorylation did not rebound. Thus, 4E-BP control of PTEN translation requires the PTEN 5’UTR. Moreover, the rebound in PI3K signaling that occurs after its initial inhibition is mediated, to a great extent, by 4E-BP-dependent reduction of PTEN. We tested this conclusion in another model. The PIK3CA H1047R mutant (Fig 1F) was transduced into MCF10A cells to generate a PI3K-mutant PTEN-wild type mammary cell line (MCF10A PIK3CAmut WT-PTEN). A PTEN knockout of this cell was generated with CRISPR methodology. We transiently expressed a pCMV driven CDS PTEN vector in this cell. We compared the effects of PI3K inhibition on the PIK3CAmut WT-PTEN cells, with those elicited in the PI3KCAmut CDS-PTEN cells. In cells expressing WT-PTEN, PTEN levels decreased within 4 hours of PI3K inhibition and remained significantly depressed for 24 hours, whereas PTEN was unaffected in treated cells expressing CDS-PTEN (Fig 6B–C). In WT-PTEN cells, PI3K/AKT signaling was inhibited 4 hours after PI3K inhibition and rebound of pAKT S473 began 8 hours after PI3K inhibition and of pAKT T308, p-PRAS40, p4E-BP T37/47 approximately 16 hours later. The levels of phosphorylation of downstream components of the PI3K pathway were similar in the untreated WT-PTEN and CDS-PTEN cells. However, after 4 hours of PI3K inhibitor treatment there was a greater reduction in PI3K signaling in the CDS-PTEN cells. This is consistent with inhibition of PTEN expression in the WT-PTEN cells but not in CDS-PTEN cells (Fig 6B–C). In the CDS-PTEN cells, initial inhibition of phosphorylation of pathway components was stable and followed by minimal rebound in AKT phosphorylation, significantly less than that observed in WT-PTEN cells. These data confirm that the 4E-BP dependent rebound in PI3K signaling in tumors treated with PI3K inhibitors is, in large part, due to inhibition of PTEN translation.

Figure 6. Inhibition of PTEN translation by PI3K inhibitors causes reactivation of the pathway.

A. 5’UTR-PTEN or CDS-PTEN were induced with 100ng/ml of doxycycline in MCF7 cells for 24 hours followed by treatment with BYL-719 (1uM) and AZD8186 (250nM) B-C. B. MCF10A-PIK3CAmut-CDS-PTEN or MCF10A-PIK3CAmut-WT-PTEN cells were treated with the PI3K inhibitors. C. Quantification of PTEN and pAKT S473 normalized to t=0 from B. D. MCF7 5’UTR-PTEN or CDS-PTEN cells were treated with 100ng/ml doxycycline for 24 hours and then with the indicated doses of BYL-719 and 250nM of AZD8186 for 2 days. Cell growth was assayed by Alamar blue and normalized to vehicle treated controls (n=8, mean and S.E.M. plotted).

We asked whether the decrease in PTEN translation reduces the sensitivity of tumor cells to PI3K inhibition. Expression of either the 5’UTR-PTEN or CDS-PTEN was induced in MCF7 cells for 24 hours (Fig S6A). The cells were then treated with a range of concentrations of BYL-719 together with a fixed dose of AZD8186 (250nM) in the continued presence of doxycycline. In CDS-PTEN cells, loss of downregulation of PTEN by PI3K inhibition increased their sensitivity to the drug as compared to the 5’UTR-PTEN cells (IC50 100nM vs. 1000nM respectively) and was also associated with decreased rebound in PI3K signaling (Fig 6D). Thus, inhibition of PTEN translation by PI3K inhibitors significantly reduces the sensitivity of tumor cells to inhibition of PI3K.

Coexistent PI3K mutation and PTEN inactivation in tumors disables feedback and increases pathway output

The duration and output of ligand stimulation of PI3K signaling is reduced by induction of PTEN expression and by AKT- and mTOR-dependent feedback inhibition of upstream signaling. In tumors, PI3K mutants induce PTEN expression and cause the feedback inhibition of receptor expression and signaling (Fig 1F). In tumors with PTEN loss, the resulting activation of AKT and mTOR results in feedback inhibition of RTKs and hormone receptor signaling (Carver et al., 2011).

Our model suggests that PI3K pathway output is substantially limited by negative feedback in tumors with mutant PI3K or PTEN inactivation. This may explain why these lesions coexist with primary drivers in many tumors. Coexistent PIK3CA mutations and PTEN inactivation would limit negative feedback and might significantly elevate pathway output. These lesions coexist in small subsets of many tumor types, but most are common in endometrial cancer, occurring in 60–70% of cases (cBioPortal-Pan Cancer Atlas TCGA dataset). In these tumors, PTEN is inactivated by truncating or frameshift or inactivating mutations or by deletions, and the great majority of these also harbor mutations in the catalytic subunit PIK3CA or the regulatory subunit PIK3R1 of PI3K or both (cBioPortal, and Figure 7A). Eight endometrial cancer cell lines were studied, two with activating catalytic domain mutations of PIK3CA and WT PTEN, two with WT PIK3CA and frame-shift mutations in PTEN and four with coexistent PIK3CA and PTEN lesions (Fig 7B, Fig S7A). The PTEN mutant lines had almost undetectable levels of PTEN protein except for the cell line (MFE296) that harbored the hotspot R130* PTEN mutation, which impairs phosphatase function (Papa et al., 2014). Despite the small size of this cohort, we did observe potentially important associations. The two cell lines with PI3K mutation and wild type PTEN had markedly lower levels of phosphorylation of AKT, PRAS40, and mTORC1 substrates than the others. The two PTEN mutant/PIK3CA wild type cell lines and one cell line with coexistent lesions had higher levels of AKT phosphorylation than the others. The most notable findings were the significantly higher levels of PRAS40, S6K and 4E-BP phosphorylation in the cell lines with coexistent mutations (Fig 7B). The data suggests that coexistent lesions significantly enhance mTORC1 signaling in these tumors.

Figure 7. Coexistent PI3K mutation and PTEN inactivation in tumors disables feedback and increases pathway output.

A-B. A. Prevalence of PTEN, PIK3CA and PIK3R1 mutations in endometrioid endometrial carcinoma (EEC) (TCGA PanCancer Atlas; cBioPortal) B. EEC cell lines with PIK3CA or PTEN mutation or both were analyzed for PI3K pathway activation C-F. C. PI3K pathway targets were assessed in isogenic MCF10A models-parental, PIK3CA H1047R or PTEN null or PIK3CA H1047R/PTEN D. The cell lines in A. were assayed for migration as assessed by impedance changes (see methods) E. BT474 parental and BT474 PTEN knockout cells were assayed for the indicated targets F. BT474 parental and BT474 PTEN knockout cells grown in soft agar were stained in crystal violet (inset 20X magnification) Note; Lanes for targets in 7B were run on the same gel and extra lanes cropped out, indicated by space. In 7C and 7E S.E is Short Exposure and L.E. Long Exposure.

Interpretation of these data must be tempered by the small number of models and the genetic complexity of these tumors. We therefore examined an isogenic model of single or coexistent PTEN loss and/or PIK3CA mutation on signaling in MCF10A breast epithelial cells (Fig 7C). The PIK3CA H1047R mutant markedly induced phosphorylation of AKT and mTOR substrates, and induced PTEN expression in MCF10A. Knocking out PTEN increased AKT phosphorylation to roughly the same degree as PI3K mutants did but increased the phosphorylation of mTORC1 substrates to a significantly greater degree. In cells with coexistent PIK3CA mutation and PTEN loss, phosphorylation of AKT was greatly induced, and phosphorylation of mTORC1 substrates S6K and 4E-BP T37/46 much greater than in cells with either lesion alone. Neither MCF10A nor any of the cells with mutants formed colonies in soft agar, nor was the proliferation of any of the mutant cells in 2D different from that of MCF10A cells (data not shown). Cell migration was enhanced in the PIK3CA mutant MCF10A cells as compared to the parental cell while the migration of the cells with PTEN loss was suppressed (Fig 7D). A marked increase in migration was observed in the double mutants, increasing by almost 7-fold compared to the parental cells and 2-fold compared to the PIK3CA mutants alone. Thus, the combination of PTEN loss and PIK3CA mutation can serve to increase the functional output of the pathway.

Knocking out PTEN in the HER2 amplified/PIK3CA mutant BT474 cells also increased PI3K and mTORC1 signaling, as assessed by increased phosphorylation of AKT and 4E-BP (Fig 7E). In contrast to MCF10A, in BT474, PTEN knockout resulted in a significant increase in colony formation in a soft agar assay but did not accelerate migration or proliferation (Fig 7F). The variations in these findings suggest that tumor lineage and the particular oncogenic mutants in the tumor may be determinants of the effects of coexistent mutations on output. For instance, in MCF10A, HER2 overexpression is a much greater inducer of PTEN than a PIK3CA mutant is (Fig1F). These data suggest, in three different models, coexistent PI3K mutation and PTEN loss result in hyperactivation of mTORC1 signaling and support a number of different phenotypes associated with transformation, depending on the system.

Discussion

We show here that PI3K signaling regulates PTEN expression by controlling its mTOR dependent translation. This buffers the response of the pathway to diverse environmental, genetic and pharmacological perturbations. Activation of the pathway by insulin and other growth factors is accompanied by an induction of PTEN expression that limits the duration of the signal and prevents overactivation. PI3K mutation induces PTEN expression in tumors, thus, limiting its output. Physiologic inhibition of mTOR due to nutrient or growth factor deprivation results in a reduction of PTEN expression that moderates the decline in PI3K/AKT signaling and could support cell survival and response to refeeding.

PTEN has been shown to be regulated by both single and polycistronic miRNA clusters including the mir21, mir22, mir29, mir17–92 cluster that bind to the 3’UTR region of PTEN mRNA and reduce its expression (Bermudez Brito et al., 2015; Dou et al., 2015; Tumaneng et al., 2012). The majority of this work suggests that this reduces the half-life of the PTEN mRNA, although some studies also invoke inhibition of translation of the PTEN mRNA. In no case has microRNA-dependent physiologic regulation of PTEN expression been demonstrated.

EGF-dependent induction of E3 ubiquitin ligases like MKRN1 and stabilization of p85 regulatory subunit of PI3K have been shown to affect PTEN stability and activity (Chagpar et al., 2010; Cheung et al., 2011; Taniguchi et al., 2006). Deubiquitinases such as USP11 and USP13 have also been shown to enhance the stability of PTEN thereby affecting its expression and PI3K signaling (Zhang et al., 2013; Park et al., 2019). There is thus a body of work that reports post-transcriptional modifications of PTEN expression that affect its stability but there is no evidence yet that these processes are regulated or that they affect PTEN translation. Regulation of PTEN by mTOR is mediated by the mTOR/4E-BP1 dependent regulation of PTEN translation. The cells we studied express both 4E-BP1 and 4E-BP2, yet 4E-BP1 expression was necessary and sufficient for the regulation of PTEN. Deletion of 4E-BP1 had no compensatory effect on 4E-BP2 expression (data not shown). The computational model we developed supports the role of 4E-BP in controlling PTEN dynamics and demonstrates that relative levels of phosphorylated 4E-BP protein is correlated with the differences in PTEN kinetics on PI3K inhibition in different cell lines.

Activation of PI3K signaling is attenuated by a complex set of AKT and mTOR - dependent mechanisms that cause feedback inhibition of receptor expression and signaling (Chandarlapaty et al., 2011; O’Reilly et al., 2006; Yu et al., 2011). Our results here demonstrate that ‘upstream feedback’ is accompanied by mTOR-dependent regulation of PTEN expression-a ‘downstream feedback’. These mechanisms cooperate to maintain homeostasis in multiple settings and may shed light on the biology of oncogenic activation of the PI3K pathway in tumors and on why PI3K pathway inhibitors have had limited success in treating patients.

PI3K inhibitors have only achieved modest benefit in PIK3CA mutant tumors. Inhibitors of PI3K, AKT or mTOR, cause relief of upstream feedback and a reactivation of receptor signaling that attenuates the antitumor activity of these drugs (Chakrabarty et al., 2012; Chandarlapaty et al., 2011; Rodrik-Outmezguine et al., 2011). Inhibition of the PI3K pathway also relieves ‘downstream’ feedback by inhibiting PTEN expression. Together, these findings show that pathway rebound depends on both activation of upstream signaling and inhibition of PTEN expression. Reduction of PTEN lipid phosphatase activity is predicted to increase the half-life of PIP3. We show that knocking out 4E-BP1 or expression of a PTEN construct that is insensitive to mTOR inhibition substantially reduces pathway rebound and increases drug efficacy. Enhanced effectiveness of PI3K inhibition in the setting of a marked decline in the expression of PTEN will require very potent inhibition of PI3K or combined inhibition of two components of the pathway e.g. mTOR in combination with PI3K or AKT (Rodrik-Outmezguine et al., 2011).

Our preliminary data in three systems showed a marked increase in phosphorylation of mTORC1 substrates (4E-BP and S6K) in cells with coexistent PI3K activation and loss of PTEN function compared to cells with PI3K mutation or PTEN inactivation alone. mTORC1 activation induces the cap-dependent translation of key growth regulatory proteins and, we now know, that of the PTEN phosphatase. We hypothesize that its induction of PTEN serves to limit mTORC1 activation. Coexistent activation of PI3K and PTEN may be selected, in part, to deregulate mTORC1 output. Our results are consistent with those obtained by others in genetically engineered mouse models in which PIK3CA mutants have been shown to cooperate with PTEN loss to accelerate disease progression (Kinross et al., 2012; Pearson et al., 2018; Schade et al., 2009) and in breast cancer patients in whom PTEN loss has been shown to be responsible for acquired resistance to PI3K inhibition (Juric et al., 2015). PI3K mutations co-occur with PTEN in approximately 70% of endometrial cancers (PanCancer Atlas Dataset, TCGA cBioPortal), and suggests that PI3K signaling may drive these tumors. Why this combination of mutations is so prevalent in endometrial cancers is unknown and answering these questions will require development of faithful in vivo and cell culture models of these tumors and of regimens that effectively inhibit PI3K signaling.

Limitations of the Study

Our study has limitations. We showed that PI3K/mTOR signaling regulates PTEN translation and thus buffers perturbations of the pathway, but we have not integrated the effects of feedback inhibition of receptor signaling and expression along with the PTEN effects. These analyses require further experimental and modeling studies. Furthermore, current assays of PI3K and mTOR output are limited and more quantitative measures must be developed. Second, we demonstrate the role of PTEN regulation in mediating adaptive resistance to PI3K inhibitors, but other consequences were addressed more superficially and its roles in regulating normal metabolism and metabolic disease were not investigated. Finally, we show that coexistent PI3K mutation and PTEN inactivation elevate mTORC1 activity by escaping feedback induction of PTEN. Why this occurs in 70% of endometrial cancers, what roles it plays in in tumor evolution and how to effectively inhibit the pathway in these tumors are unknown.

STAR Methods

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Neal Rosen (rosenn@mskcc.org).

Materials Availability

Plasmids and Cell lines generated in this study as available on request

Data and Code Availability

Matlab Code for the computational mathematical model is available upon request.

EXPERIMENTAL MODELS AND SUBJECT DETAILS

293T packaging cell line was from Clontech. All other cell lines were from the American Type Culture Collection. BT-474, LAM TSC2−/−, MDA-MB-361, MCF7, CHO, HCC1954, HCC202, SKBR3, AU565, HCC1419, BT483, HCC1187, HCC1806, MDA-MB-231, MFE280, HEC1A, SNGII, HEC265, HEC6, HEC59, MFE296, HEC251, NCIH508 were grown in DMEM-F12 medium; T-47D, SU-DHL-10, were grown in RPMI1640 medium; BT-20 was grown in MEM medium; 293T, 3T3L1, MEF, MEF TSC2−/− were grown in DMEM medium. MCF10A (Sigma Aldrich), MCF10A HER2, MCF10A PIK3CA H1047R/+ (Horizon), MCF10A PTEN−/−, MCF10A PIK3CA H1047R/+ PTEN−/− (Sigma Aldrich) grown in MEGM bullet kit from Lonza. All other media were supplemented with 100μg/mL penicillin, 100mg/mL streptomycin, 4mM glutamine, and 10% fetal bovine syndrome. All cells were maintained at 37°C in 5% CO₂. MCF7 PIK3CAwt cells were a gift from Josh Lauring at Johns Hopkins. MEF TSC2−/− was generously provided by John Blenis at Weill Cornell. C57BL/B6 mice were obtained from Jackson’s laboratory.

METHOD DETAIL

Immunoblotting

Cells in culture were washed thrice in cold PBS and lysed with Cell Lysis Buffer (Cell Signaling #9803) supplemented with Halt protease and phosphatase inhibitors (Pierce Chemical). Lysates were briefly sonicated before centrifugation at 16,000×g for 15 minutes at 4°C. The supernatant was collected, and protein concentration was determined using the BCA kit (Pierce) per manufacturer’s instructions. Protein samples were diluted in 4X LDS sample Buffer with 10X Sample Reducing Agent (both from Invitrogen).

Xenograft tumors were homogenized in SDS lysis buffer (50mM Tris-HCL pH 7.4, 10% Glycerol, 2% SDS) and boiled at 95°C for five minutes. Lysates were then briefly sonicated, boiled again for 5 minutes, before clearing by centrifugation at 14,000rpm for 10 minutes at room temperature. The supernatant was collected and protein concentration was determined using the BCA kit (Pierce) per manufacturer’s instructions. Protein samples were diluted in SDS sample buffer (final concentration: 62.5mM TrisHCL pH 6.8, 2%SDS, 10% Glycerol, 15.5mg/mL DTT, 0.02mg/mL Bromophenol blue).

25–50μg of protein was loaded onto each lane of a 4–12% BisTris mini gel or midi gel (Invitrogen) for immunoblotting. Transfer was onto nitrocellulose membranes (0.2μm, GE Health Care) before blocking for 1h at room temperature and incubating with primary antibodies overnight at 4°C. Membranes were incubated with secondary rabbit antibody (Sigma) or secondary mouse antibody (GE Health Care) for 1h at room temperature. Blots were developed in Perkin-Elmer’s Western Lightning ECL or Millipore’s Immobilon HRP reagents per manufacturer’s instructions.

Viable Cell Counting

Cells were plated at 5000 cells per well in a 96 welled plate and grown in 8 replicates per condition, then treated with drug the following day. At indicated times, each plate was treated with 20X Alamar Blue (Thermo Fisher Scientific), incubated for 4 hours at 37 degree Celsius and the fluorescence measured.

Transfections

pGL3 promoter vector and pGL3-PTEN 5’UTR were transiently transfected into BT474 cells plated at 2 million cells per well of a 6 welled plate, using the transient transfection protocol from Thermo Fisher Scientific with Lipofectamine 2000 at a ratio of 3ul of Lipofectamine/ug of plasmid DNA.

4E-BP4A was cloned into the EcoR1/BamH1 site in the retroviral vector TTIGFP-MLUEX obtained from Scott Lowe (Zuber et al., 2011). Retroviral constructs were all transfected into amphotropic 293T packaging cells using Lipofectamine 2000 (Invitrogen). After 48h, virus-containing medium was filtered with a 0.45μm PVDF syringe filter (Millipore), collected, and used to infect target cells in the presence of 4μg/mL polybrene (Millipore). The stable transfectants were obtained by selection with 2μg/mL puromycin.

4E-BP1Knockout cell lines:

4E-BP1knockout was done in BT474, MCF7 cell lines using the following guide RNAs, followed by selection of clonal sub populations.

sgGFP-F: CACCGGGGCGAGGAGCTGTTCACCG

sgGFP-R: AAACCGGTGAACAGCTCCTCGCCCC

sgRNA guide 3-F: CACCGGGAAATTCCTGATGGAGTGT

sgRNA guide 3-R: AAACACACTCCATCAGGAATTTCCC

The guides were cloned into V2 lentiviral crispr and https://www.addgene.org/52961/

Guide 3 was the best and has been used for the experiments.

4E-BP1–4A:

Doxycycline inducible 4E-BP1(T37A/T46A/S65A/T70A) BT474 were constructed in two steps using a retroviral hygromycin selectable vector containing rtTA3, followed by infection with the following lentiviral construct https://www.addgene.org/38240/

mRNA extraction and RTPCR:

mRNA extraction was done using the RNeasy mini kit from Qiagen, cDNA was made using the SuperScript III First Strand Synthesis from Thermo Fisher Scientific. Taqman reactions were done using Applied Biosystems Taqman probes and ABI 7500 real-time quantitative PCR system. For data analysis, raw counts were normalized to housekeeping gene average for the same time point and condition (ΔCt). Counts are reported as fold change relative to the untreated control (2^–ΔΔCt). The Taqman probes were PTEN Hs02621230_s1 and GAPDH Hs03929097_g1.

AHA labeling and pull down:

BT474 cells were starved of Methionine for 30 minutes with simultaneous treatment with vehicle or combination of PI3K inhibitors BYL-719 (1uM) and AZD8186 (250nM) or mTOR inhibitor AZD8055 (500nM) or Cycloheximede (50ug/ml) and then pulsed with AHA (50uM) for 2 hours. Lysates were conjugated with biotin-alkyne via click chemistry and AHA-biotin-alkyne labeled proteins were pulled down with streptavidin beads and the indicated proteins analyzed by immunoblotting.

Met S35 labelled Pulse-Chase:

BT474 were plated at 5 million cells per 10cm dish and starved of Methionine and Cysteine for 45 min, pulsed with radiolabelled MetS35 at 0.2mCi/ml for 30 min, washed and chased with regular medium, with or without PI3K inhibitors for different time intervals. PTEN protein was immunoprecipitated and analyzed via autoradiography.

Luciferase Assay:

870bp of PTEN 5’UTR from −114 to −983 was synthesized by Genewiz and cloned into HindIII/NcoI site of the pGL3 promoter vector. Cells were transfected with either pGL3 promoter vector or pGL3-PTEN 5’UTR and co- transfected with Renilla luciferase plasmid as internal control. Firefly and Renilla fluorescence was measured and the final values are expressed as normalized Firefly to Renilla units of fluorescence.

Cell Migration assay:

Cell Migration was assayed using the xCELLigence® RTCA DP (Acea Biosciences, Inc.,) instrument. Cells were plated on the upper chamber of the microporous membrane and the cells passed through the microporous membrane and deposit onto gold impedance electrodes. The impedance signal is measured as a readout of cell migration.

PTEN in vitro activity assay:

BT474, MCF7, MCF10A, MCF10A PIK3CA H1047R and MCF10A HER2 cells were lysed and PTEN protein immunoprecipitated using the PTEN-antibody-sepharose beads conjugates (CST #4326) or the rabbit mAB IgG isotype control beads by overnight IP from whole cell lysates. PTEN activity was analysed using the PTEN activity ELISA kit from Echelon Biosciences (#K-4700). PTEN-sepharose bead conjugates were incubated with 16uM of PIP3 for 4 hours at room temperature and pmols of PIP2 from the reaction products was measured by ELISA and interpolated from the PIP2 standard curve. The assay conditions were in the linear range and was confirmed by running a standard curve of increasing concentrations of recombinant PTEN (#E-3000) for the given substrate concentration, at 2 hours and 4hours at room temperature.

QUANTIFICATION AND STATISTICAL ANALYSIS

Pairwise t test

Pairwise t test was used to test statistical significance in all cases and was implemented in MATLAB as ttest2.

Standard Error of the Means

Standard error of the means (SEM) was used to depict the variance of the data in the figures as Mean+/−SEM.

The statistical analysis used has been specified in the relevant Figure Legends.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit pAKT Ser473, clone D9E	Cell Signaling Tech.	Cat# 4060, RRID:AB_2315049
Rabbit pAKT T308	Cell Signaling Tech.	Cat #2965
Rabbit P-HER2 (Y1221/1222)	Cell Signaling Tech.	Cat# 2249, RRID:AB_2099241
Rabbit P-FoxO1 (T24)/FoxO3a (T32)/FoxO4 (T28)	Cell Signaling Tech.	Cat# 2599, RRID:AB_2106814
Rabbit P-S6 (S235/236)	Cell Signaling Tech.	Cat# 4858, NA
Rabbit p-S6K (T389)	Cell Signaling Tech.	Cat# 9234, RRID:AB_2269803
Rabbit phospho-4E-BP1 (S65)	Cell Signaling Tech.	Cat# 9451
Rabbit 4E-BP1	Cell Signaling Tech.	Cat# 9452
Rabbit Actin	Cell Signaling Tech.	Cat# 4970
Rabbit p-eIF4E S209	Cell SignalingTech.	Cat# 9742, RRID:AB_823488
Rabbit P-PRAS40	Cell Signaling Tech.	Cat# 2997, RRID:AB_2258110
Rabbit PTEN-antibody-sepharose beads conjugates	Cell Signaling Tech.	Cat# 4326, RRID:AB_10691828
Rabbit HER3	Cell Signaling Tech.	Cat# 4754, RRID:AB_10691324
Rabbit HER2	Millipore Sigma	Cat# 06–562
Rabbit CyclinD1	ThermoFisher	Cat# PA516777
Rabbit PTEN	Cell Signaling Tech.	Cat# 9559, RRID:AB_390810
Bacterial and Virus Strains
Biological Samples
Chemicals, Peptides, and Recombinant Proteins
Rapamycin	Calbiochem	# CAS 53123–88-9
BYL-719	Novartis	NA
MK2206	Merck Research Labs	NA
Lapatinib	Selleck Chem	#S2111
AZD8186	AstraZeneca Pharma	NA
AZD8055	AstraZeneca Pharma	NA
RapaLink-1	Revolution Medicines	NA
Silvestrol	MedChemExpress	Cat# HY-13251
Cycloheximide	Sigma-Aldrich	Cat# 66–81-9
Cal-101	Selleck Chem	Cat# S2226
Insulin, Human Recombinant	Sigma-Aldrich	CAS# 11061–68-0
Heregulin beta-1 human	Sigma-Aldrich	Cat# SRP3055
EGF, h(EGF)	Sigma-Aldrich	Cat# 11376454001
Doxycycline hydrochloride	Sigma-Aldrich	Cat# D3447
EasyTag™ L-[35S]-Methionine, 500μCi(18.5MBq), Stabilized Aqueous Solution	Perkin-Elmer	NEG709A500UC
Alamar Blue	Thermo-Fisher	DAL1100
Critical Commercial Assays
TaqMan™ Fast Advanced Master Mix	ThermoFisher Scientific	Cat #4444556
Click-IT™ AHA (L-Azidohomoalanine)	ThermoFisher Scientific	Cat #C10102
PTEN Activity ELISA Kit	Echelon Biosciences	Cat #K4700
Deposited Data
Experimental Models: Cell Lines
BT474	ATCC	Cat# HTB-20, RRID:CVCL_0179
LAM TSC2−/−	MSKCC	NA
MDA-MB-231	ATCC	Cat# HTB-26, RRID:CVCL_0062
MCF7	ATCC	Cat# HTB-22, RRID:CVCL_0031
HCC1954	ATCC	Cat# CRL-2338, RRID:CVCL_1259
HCC202	ATCC	Cat# CRL-2316, RRID:CVCL_2062
SKBR3	ATCC	Cat # HTB-30, RRID:CVCL_0033
AU565	ATCC	Cat# CRL-2351, RRID:CVCL_1074
HCC1419	ATCC	Cat# CRL-2326, RRID:CVCL_1251
BT483	ATCC	Cat# HTB-121, RRID:CVCL_2319)
HCC1187	ATCC	Cat# CRL-2323, RRID:CVCL_1248
HCC1806	ATCC	Cat# CRL-2335, RRID:CVCL_1258
MFE280	MSKCC	RRID:CVCL_1405
MDA-MB-361	ATCC	Cat# HTB-27, RRID:CVCL_0620
HEC1A	ATCC	Cat# HTB-112, RRID:CVCL_0293
SNGII	MSKCC	RRID:CVCL_3170
HEC265	MSKCC	RRID:CVCL_2928
HEC6	MSKCC	RRID:CVCL_2931
HEC59	MSKCC	RRID:CVCL_2930
MFE296	MSKCC	RRID:CVCL_1406
HEC251	MSKCC	RRID:CVCL_2927
NCI-H508	ATCC	Cat# CCL-253, RRID:CVCL_1564
T-47D	ATCC	Cat# HTB-133, RRID:CVCL_0553
SU-DHL-10	ATCC	Cat# CRL-2963, RRID:CVCL_1889
BT-20	ATCC	Cat# CRL-7912, RRID:CVCL_0178
293T	ATCC	Cat# CRL-11268
3T3L1	ATCC	Cat# CL-173, RRID:CVCL_0123
MEF	MSKCC	NA
MEF TSC2−/−	John Blenis lab	NA
MCF10A	ATCC	Cat# CRL-10317, RRID:CVCL_0598
MCF10A HER2	MSKCC	NA
MCF10A PIK3CA H1047R/+	Cellosaurus	RRID:CVCL_LD55
MCF10A PTEN−/−	Cellosaurus	RRID:CVCL_RR05
MCF10A PIK3CA H1047R/+ PTEN−/−	Millipore Sigma	Cat# CLLS1199–1SET, NA
MCF7 PIK3CAwt	Josh Lauring, Johns Hopkins University.	NA
CHO	ATCC	Cat# CRL-12023, RRID:CVCL_JL49
Experimental Models: Organisms/Strains
C57BL/B6	The Jackson Laboratory	Cat#000664
Oligonucleotides
The Taqman probes for PTEN Hs02621230_s1	ThermoFisher Scientific	Cat# 4331182
The Taqman probes for GAPDH Hs03929097_g1	ThermoFisher Scientific	Cat# 4331182
sgGFP-F: CACCGGGGCGAGGAGCTGTTCACCG	This paper	NA
sgGFP-R: AAACCGGTGAACAGCTCCTCGCCCC	This paper	NA
4E-BP1 sgRNA guide 3-F: CACCGGGAAATTCCTGATGGAGTGT	This paper	NA
sgRNA guide 3-R: AAACACACTCCATCAGGAATTTCCC	This paper	NA
Recombinant DNA
pGL3-Basic Vector	Promega	(GenBank Accession Number U47298) Cat No #E1751
pRL-CMV Vector	Promega	(GenBank Accession Number AF025843), Cat No-E2231
pRL-CMV-IRES Vector	This paper	NA
pCW57.1–4EBP1_4xAla	Addgene	Cat# 38240
LentiCRISPRv2	Addgene	Cat#52961
TTIGFP-MLUEX	Lab of Scott Lowe	NA
4E-BP4A-TTIGFP-MLUEX	This paper	NA
Software and Algorithms
Other
xCELLigence® RTCA DP	Agilent	NA

Highlights.

• PI3K regulates PTEN translation via mTOR, thus buffering changes in PI3K signaling

• Physiologic or oncogenic PI3K activation increase PTEN and limits pathway output

• In tumors, inhibitors of PI3K signaling reduce PTEN, thus limiting their efficacy

• In tumors, coexistent PTEN loss and PI3K mutation hyperactivates pathway output