mTORC1 activity repression by late endosomal phosphatidylinositol 3,4-bisphosphate

See allHide authors and affiliations

Science  02 Jun 2017:
Vol. 356, Issue 6341, pp. 968-972
DOI: 10.1126/science.aaf8310

Local specificity of growth signals

The mechanistic target of rapamycin complex 1 (mTORC1) regulates cell growth in response to nutrients. Marat et al. found that the lipid phosphatidylinositol 3,4-bisphosphate [PI(3,4)P2], which, when generated at the cell surface, is linked to stimulation of mTORC1 and promotion of cell growth, does the opposite when synthesized at the late endosome or lysosome. In cells deprived of nutrients, PI(3,4)P2 was produced at lysosomes, where it recruited an inhibitor of mTORC1. The results elucidate the complex regulation of mTORC1, which is altered in human diseases such as cancer and neurodegeneration.

Science, this issue p. 968


Nutrient sensing by mechanistic target of rapamycin complex 1 (mTORC1) on lysosomes and late endosomes (LyLEs) regulates cell growth. Many factors stimulate mTORC1 activity, including the production of phosphatidylinositol 3,4,5-trisphosphate [PI(3,4,5)P3] by class I phosphatidylinositol 3-kinases (PI3Ks) at the plasma membrane. We investigated mechanisms that repress mTORC1 under conditions of growth factor deprivation. We identified phosphatidylinositol 3,4-bisphosphate [PI(3,4)P2], synthesized by class II PI3K β (PI3KC2β) at LyLEs, as a negative regulator of mTORC1, whereas loss of PI3KC2β hyperactivated mTORC1. Growth factor deprivation induced the association of PI3KC2β with the Raptor subunit of mTORC1. Local PI(3,4)P2 synthesis triggered repression of mTORC1 activity through association of Raptor with inhibitory 14-3-3 proteins. These results unravel an unexpected function for local PI(3,4)P2 production in shutting off mTORC1.

Nutrient sensing and signaling by mechanistic target of rapamycin complex 1 (mTORC1) on lysosomes or late endosomes (LyLEs) integrate internal and external cues to regulate cell metabolism, growth, proliferation, and survival. Dysregulation of mTORC1 activity is implicated in diseases (1, 2). Nutrients and mitogens activate mTORC1 through class I phosphatidylinositol 3-kinases (PI3K)–mediated synthesis of phosphatidylinositol 3,4,5-trisphosphate [PI(3,4,5)P3] (3, 4). PI(3,4,5)P3 stimulates the proto-oncogenic kinase AKT, which facilitates the activation of mTORC1 (5, 6). How mTORC1 signaling is repressed under conditions of starvation and how this may be locally controlled by membrane lipids—in particular, phosphoinositides (PIs) (7, 8)—is not clear.

Class II phosphatidylinositol 3-kinases (PI3KC2) in addition to phosphatidylinositol 3-phosphate PI(3)P (9, 10) can synthesize phosphatidylinositol 3,4-bisphosphate [PI(3,4)P2] (11). As this lipid, when present at the plasma membrane, can activate AKT (12), we hypothesized that PI3KC2 may stimulate mTORC1 signaling. We tested this by probing mTORC1 activity using specific antibodies against S6 kinase (S6K) phosphorylated at Thr389 (pS6K) or phosphorylated 4E-binding protein 1 (4E-BP1) (p-4E-BP1) in human epithelial carcinoma HeLa cells depleted of class II PI3K 2α or 2β (PI3KC2α or PI3KC2β). Depletion of PI3KC2β but not of PI3KC2α (Fig. 1A and fig. S1A) resulted in increased phosphorylation of S6K and 4E-BP1 (fig. S1C), indicative of increased, rather than repressed, mTORC1 activity. Inhibition of mTOR kinase completely repressed phosphorylation of S6K and 4E-BP1 in PI3KC2β-depleted cells (fig. S1B). PI3KC2β depletion also increased mTORC1 activity in noncancer cell lines, such as African Green Monkey Cos7 fibroblasts cells or human embryonic kidney 293T (HEK293T) cells (fig. S1D). PI3KC2β depletion (knockdown) did not affect phosphorylation of AKT1 or AKT2 (fig. S1, E and G), indicating that PI3KC2β repression of mTORC1 activity is not via AKT. We analyzed the effect of PI3KC2β loss under conditions of starvation and nutrient replenishment. Depletion of PI3KC2β facilitated recovery of mTORC1 activity after starvation and replenishment of amino acids irrespective of the presence or absence of serum growth factors and, thus, independent of AKT activity in various cell types (Fig. 1, B and C, and figs. S1, F and G, and S2, A to D). Elevated mTORC1 activity in the absence of PI3KC2β was further confirmed in genome-engineered (fig. S3) PI3KC2β knockout (KO) HEK293T cells (Fig. 1D), in mouse embryonic fibroblasts (MEF) and adult (MAF) fibroblasts isolated from mice lacking PI3KC2β (13) (fig. S1, H and I). Reexpression of PI3KC2β in KO cells rescued this phenotype (see below).

Fig. 1 PI3KC2β represses mTORC1.

(A) Steady-state phospho-Thr389–p70 S6K/total p70 S6K (p70S6K) ratio in HeLa cells depleted of PI3KC2β by two small interfering RNAs (siRNAs). Means ± SEM; n = three independent experiments; one-way analysis of variance (ANOVA) plus Dunnett’s post test; **P < 0.01, ***P < 0.001. (B) Response to recovery from starvation in HeLa cells depleted of PI3KC2β (siPI3KC2β). Phospho-Thr389–p70S6K/total p70S6K ratio after readdition of Dulbecco’s minimum essential medium (DMEM) with or without fetal bovine serum (FBS). Means ± SEM; n = three independent experiments; unpaired two-tailed t test, *P < 0.05 (DMEM + FBS, P = 0.0294; DMEM – FBS, P = 0.0128). (C) Phospho-Thr308 and phospho-Ser473–AKT to total AKT ratios in PI3KC2β-depleted HeLa cells starved then fed with FBS-containing medium. Means ± SEM of two (Thr308) or three (Ser473) independent experiments; unpaired two-tailed t test (Thr308, P = 0.3924; Ser473, P = 0.6309). (D) Steady-state phospho-Thr389–p70S6K/total p70S6K ratio and amounts of phospho-Ser473 AKT in HEK293T wild-type (WT) or PI3KC2β-knockout cells (KO). Means ± SEM; n = three independent experiments; one-sample two-tailed t test with hypothetical mean of 1, *P < 0.05 (P = 0.0223). (E) FACS analysis of the size of HEK293T WT versus HEK293T PI3KC2β KO cells. Means of WT: 68,939 versus KO: 92,515 arbitrary units (a.u.) (n = 50969 cells per condition). (F) Dispersion of LAMP1-positive LyLEs in PI3KC2β-depleted HeLa cells. Blue, 4′,6-diamidino-2-phenylindole (DAPI)–stained nuclei; white, outline of cell bodies. Scale bar, 20 μm. (G) Mean distance of the majority (75%) of LAMP1 intensity from the center of the nucleus (scrambled, n = 144 cells; siPI3KC2β, n = 143 cells). Means ± SEM; n = three independent experiments; unpaired two-tailed t test, **P < 0.01 (P = 0.0031).

mTORC1 signaling controls cell size (1, 2). Consistent with elevated mTORC1 activity, PI3KC2β KO cells were larger than wild-type (WT) HEK293T cells when measured by fluorescence-activated cell sorting (FACS) (Fig. 1E), and PI3KC2β-depleted HeLa cells displayed enlarged footprints (fig. S4A), in line with (14). Increased mTORC1 activity also represses protein degradation through the autophagy and lysosomal pathway (15). The amounts of p62, an autophagy substrate and cargo receptor for ubiquitylated proteins, were increased in cells depleted of PI3KC2β (fig. S4, B to D), whereas depletion of PI3KC2α was without effect. Consistently, in PI3KC2β-depleted cells, p62 failed to undergo autophagy-mediated degradation in response to starvation (fig. S4, E and F), and the autophagy marker microtubule-associated protein 1A/1B–light chain 3 isoform II (LC3-II) was elevated (fig. S4G). These data identify PI3KC2β as a repressor of mTORC1 signaling under conditions of low growth factors when the canonical class I PI3K-AKT pathway is inactive.

High mTORC1 activity correlates with the peripheral dispersion of LyLEs (16, 17). In agreement with increased mTORC1 activity, loss of PI3KC2β caused the peripheral redistribution of lysosome-associated membrane protein 1 (LAMP1) (Fig. 1, F and G), mTOR (fig. S5B), and the late endosomal marker CD63 (fig. S5A). The appearance of organelles such as early endosomes or the Golgi complex remained unaltered (fig. S5C). Loss of PI3KC2β did not affect levels of LAMP2A, Vps41, HOPS, or Rab7, proteins located at LyLEs (fig. S4B). Thus, PI3KC2β may repress mTORC1 through an association with LyLEs. As antibodies to detect endogenous PI3KC2β were not available, we genome-engineered HEK293T knock-in cells expressing enhanced green fluorescent protein (eGFP)–tagged PI3KC2β from its endogenous locus (fig. S3). Although eGFP-PI3KC2β had a largely cytoplasmic localization that was lost upon prepermeabilization before fixation in cells grown in the presence of serum, in cells deprived of serum, endogenous eGFP-PI3KC2β was recruited to punctate structures (fig. S5E) identified as CD63-containing LyLEs (Fig. 2, A and B, and fig. S5F). Exogenously expressed PI3KC2β also localized to LyLEs (fig. S6B) but not early endosomes or endocytic structures (fig. S5D). To identify proteins that associate with glutathione S-transferase–tagged PI3KC2β and may aid in its recruitment, we used affinity chromatography paired with mass spectrometry and immunoblotting. This analysis revealed that PI3KC2β associates with the mTORC1 subunit Raptor, as well as with mTOR itself, but not with Rictor, a component of mTORC2 (fig. S6A). Biochemical mapping analyses revealed two Raptor-binding sites within the amino-terminal domain of PI3KC2β (fig. S7) that bound to the WD40 domain of Raptor (fig. S6D). Both proteins colocalized at LyLEs when expressed in cells (fig. S6, B and C). Furthermore, complex formation between endogenous eGFP-PI3KC2β and Raptor was facilitated under conditions of serum deprivation (Fig. 2, C and D). Consistent with their physical association, endogenous eGFP-PI3KC2β cοlocalized with mTOR on CD63-containing LyLEs in serum-deprived cells (Fig. 2A and fig. S6G). Under such conditions, inactive mTORC1 is retained at lysosomes because Rag guanosine triphosphatases remain active in the presence of amino acids (18). Recruitment of PI3KC2β to LyLEs was reduced in serum-deprived cells that were also depleted of Raptor (fig. S6, E and F). In contrast, cells cultured without amino acids or serum growth factors—conditions that lead to dissociation of mTORC1 from the lysosome (18)—showed little complex formation between PI3KC2β and Raptor-mTORC1 (fig. S6H). These data show that recruitment of PI3KC2β to LyLEs is induced by its association with Raptor when cells are deprived of growth factors and that negative regulation of mTORC1 signaling by PI3KC2β is not caused by removal of the complex from the lysosomal surface (18, 19).

Fig. 2 Growth factor deprivation–induced PI(3,4)P2 synthesis by PI3KC2β at LyLEs.

(A) Confocal images of serum-starved HEK293T cells expressing endogenous eGFP-PI3KC2β stained for endogenous mTOR, CD63, and GFP. Scale bar, 5 μm. (B) Percentage of cells with prominent GFP-PI3KC2β punctae in the fed (324 cells) or serum-starved (263 cells) states. Mean ± SEM; n = three independent experiments; unpaired two-tailed t test ***P < 0.0001. (C) HEK293T cells expressing endogenous eGFP-PI3KC2β or wild-type (WT) untagged HEK293T were starved of serum (+) or put in fresh medium with FBS (–). Cell lysates were subjected to GFP nanobody immunoprecipitation (IP αGFP) and analyzed by immunoblotting against the indicated proteins. SM, 2% of cell lysate used for IP. (D) Ratio of Raptor bound to endogenous GFP-PI3KC2β in HEK293T cells deprived of serum or fed with FBS. Means ± SEM; n = five independent experiments, one-sample two-tailed t test with hypothetical mean of 1, *P < 0.05 (P = 0.0363). (E) Kinase activity of purified PI3KC2β (5 ng/μl) using PI, PI(4)P, or PI(4,5)P2 as substrates assayed by ADP-Glo. (F) Normalized perinuclear PI(3,4)P2 levels of fed or starved HeLa cells determined by confocal imaging. Scatter plots of the intensity of each cell normalized to the mean of the fed condition in each experiment. n = three experiments; one-sample two-tailed t test, *P < 0.05 (P = 0.0155). (G) Normalized perinuclear PI(3,4)P2 levels of starved HeLa cells depleted of PI3KC2β similar to data in (F). n = three experiments; one-sample two-tailed t test, *P < 0.05 (P = 0.0173).

To mechanistically understand the function of PI3KC2β in repressing mTORC1 signaling in serum-deprived cells, we analyzed its lipid kinase activity. Like PI3KC2α, which can produce PI(3,4)P2 (11), PI3KC2β within its substrate-binding loop contains basic residues that coordinate the 4-phosphate group (fig. S8A), thereby producing PI(3,4)P2 from phosphatidylinositol 4-phosphate [PI(4)P] (Fig. 2E). Recombinant, enzymatically active PI3KC2β purified from insect cells almost exclusively synthesized PI(3,4)P2 from PI(4)P, a lipid enriched at LyLEs (20, 21), the Golgi complex, and the plasma membrane. hVps34 analyzed in parallel as a control almost exclusively synthesized PI(3)P from PI (fig. S8C). Minor activity of PI3KC2β toward phosphatidylinositol to produce PI(3)P, as well as PI(4,5)P2 to produce PI(3,4,5)P3, was detectable (Fig. 2E). A similar preference for PI(3,4)P2 production was observed for immunoprecipitated hemagglutinin epitope (HA)– or endogenous GFP-tagged PI3KC2β from HEK293T cells (fig. S8, B and D). As Raptor partially colocalized with PI(4)P, a substrate for PI3KC2β, at LyLEs but not with PI(3)P (fig. S9D), we tested whether PI3KC2β repressed mTORC1 activity under conditions of serum starvation by locally producing PI(3,4)P2 at LyLEs. PI(3,4)P2 was measured with specific antibodies that enabled the semiquantitative detection of this rare PI lipid (11). Although PI(3,4)P2 was barely detectable at Raptor-containing LyLE in fed HeLa cells, its abundance increased significantly in growth factor–deprived cells (Fig. 2F and fig. S8, E and G). Depletion of PI3KC2β completely prevented the serum starvation–induced synthesis of PI(3,4)P2 at LyLEs (Fig. 2G and fig. S8, F and H). Similar observations were made in starved Cos7 cells (fig. S9C). Loss of PI3KC2β did not affect the amount of PI(3)P at endosomes (fig. S9, A and B). Note that enzymatic PI(3,4)P2-synthesizing activity of PI3KC2β was required to restore increased mTORC1 signaling in PI3KC2β KO cells (Fig. 3A) and to restore the peripheral dispersion of LyLEs induced by depletion of the endogenous enzyme (Fig. 3B and fig. S10, D and E), whereas it was dispensable for PI3KC2β recruitment to LyLEs or for its binding to Raptor (fig. S10, C and F). Moreover, overexpression of PI3KC2β WT, but not catalytically inactive (KI) enzyme (fig. S10A), partially repressed mTORC1 signaling in WT cells (fig. S10B).

Fig. 3 PI(3,4)P2 represses mTORC1 signaling via 14-3-3.

(A) Immunoblot of steady-state phospho-Thr389–p70S6K and total S6 kinase levels in WT versus PI3KC2β KO HEK293T cells and KO cells expressing HA-PI3KC2β WT or kinase-inactive (KI) PI3KC2β. (B) Rescue of LAMP1-positive LyLE dispersion in PI3KC2β-depleted HeLa cells by WT but not KI PI3KC2β as described in the legend to Fig. 1G. Mean distance of the majority (75%) of LAMP1 intensity from the center of the nucleus (scrambled, n = 150 cells; siPI3KC2β, n = 168 cells; siPI3KC2β + GFP-PI3KC2β WT, n = 97 cells; siPI3KC2β + KI GFP-PI3KC2β, n = 100 cells). Means ± SEM; n = three independent experiments; one-way ANOVA followed by Tukey’s post test; ***P < 0.001. (C) Exogenously added PI(3,4)P2 represses mTORC1 signaling in HEK293T cells evidenced by reduced ratio of phospho-Thr389–p70S6K/total p70S6K. Means ± SEM; n = three independent experiments; one-sample two-tailed t test with a hypothetical mean of 1, *P < 0.05 (P = 0.0107). (D) Ratio of phospho-Thr389–p70S6K/total p70S6K in control (scrambled siRNA) or PI3KC2β-depleted HeLa cells, treated with dimethyl sulfoxide (DMSO) or 50 μM PI(3,4)P2. Means ± SEM; n = two independent experiments. (E) Coprecipitation of HA–14-3-3γ with FLAG-tagged Raptor in stably transfected control (scrambled) or PI3KC2β-depleted HEK293T cells. SM, 5% starting material. (F) Quantification of data in (E), means ± SEM; n = three independent experiments; one-sample two-tailed t test, *P < 0.05 (P = 0.0186). (G) Confocal images of control or PI3KC2β-depleted HeLa cells (outlined in white) treated with solvent (DMSO) or PI(3,4)P2 and stained for 14-3-3 (red) and CD63 (green) as marker for LyLEs. Scale bar, 10 μm. Scatter plot, quantification of 14-3-3 and CD63-positive puncta per cell (15 images per condition). (H) Model of 14-3-3–mediated repression of mTORC1 signaling by PI3KC2β-dependent local PI(3,4)P2 synthesis at LyLEs upon serum starvation.

Thus, repression of mTORC1 signaling in growth factor–deprived cells appears to require local production of PI(3,4)P2 by PI3KC2β at LyLEs. To acutely manipulate cellular PI content, we applied cell-permeable acetoxy-methylester–protected PI derivatives (22) that, after uptake, are hydrolyzed by intracellular esterases, resulting in the release of the native PI lipid. Acute increases in PI(3,4)P2 concentration reduced mTORC1 activity in control cells (Fig. 3C and fig. S10G) and restored normal mTORC1 activity in cells depleted of endogenous PI3KC2β (Fig. 3D and fig. S10G). In contrast, exogenous supply of PI(3)P neither repressed mTORC1 signaling in control cells nor restored increased mTORC1 activity in cells depleted of PI3KC2β (fig. S10G). Addition of PI(3,4)P2 also fully restored dispersion of LyLEs in PI3KC2β-depleted cells (fig. S10H). Thus, PI(3,4)P2 synthesized in growth factor–deprived cells by PI3KC2β at LyLEs appears to repress mTORC1 signaling.

PI(3,4)P2 synthesized at LyLEs might repress mTORC1 activity by recruiting a specific lipid-binding effector protein that regulates mTORC1 activity. Quantitative proteomic analyses have identified 14-3-3 proteins as specific PI(3,4)P2-binding proteins (23), which we confirmed (fig. S11B). Association of 14-3-3 with a phosphopeptide in Raptor underlies repression of mTORC1 activity in response to energy deprivation (24). We therefore hypothesized that repression of mTORC1 activity by PI(3,4)P2 might involve complex formation of 14-3-3 with Raptor. (23, 24). Indeed, 14-3-3γ associated with Raptor in control cells but much less well in cells lacking PI3KC2β (Fig. 3, E and F). Loss of PI3KC2β did not affect Raptor phosphorylation (fig. S11A). Impaired complex formation between 14-3-3 and Raptor in PI3KC2β-depleted cells was associated with loss of 14-3-3 from its normal localization at LyLEs (Fig. 3G and fig. S11C). Exogenous supply of PI(3,4)P2 restored 14-3-3 recruitment to LyLEs in PI3KC2β-depleted cells (Fig. 3G), indicating that PI(3,4)P2 is required for complex formation of 14-3-3 with Raptor to repress mTORC1 activity.

Our results identify a function for PI3KC2β-mediated local synthesis of PI(3,4)P2 [a PI lipid hitherto associated with growth promotion through activation of AKT at the cell surface (3, 4)] in the shutdown of mTORC1 signaling at LyLEs in response to growth factor deprivation—that is, under conditions in which class I PI3K and AKT signaling are inactive. Thus, the same lipid [e.g., PI(3,4)P2] can exhibit different cell physiological functions depending on its site of synthesis and the physiological status of the cell. In this model, local PI(3,4)P2 synthesis by PI3KC2β at LyLEs triggers recruitment of inhibitory 14-3-3 proteins to locally repress mTORC1 signaling (Fig. 3H) and to reposition lysosomes at the cell center [e.g., via Arl8-based mechanisms (16)] when growth factor signaling ceases. How PI3KC2β senses this change remains to be determined. Inhibition of AMPK activity was insufficient to disrupt PI3KC2β-mTORC1 complex formation (fig. S12). This suggests that additional kinase or phosphatase pathways are involved in regulating PI3KC2β function.

Given the multiple links mTORC1 signaling and its repressor AMPK have with diseases including diabetes, obesity, and cancer (25), we suggest that pharmacological targeting of PI3KC2β-mediated PI(3,4)P2 synthesis may open new avenues for the treatment of these diseases.


Materials and Methods

Figs. S1 to S12

References (2630)


  1. Acknowledgements: We thank D. M. Sabatini and E. Y. Skolnik for reagents; E. Krause and colleagues for proteomic analyses; L. von Oertzen, D. Loewe, and S. Zillmann for technical assistance; and M. Krauß for critical reading and discussions. This work was supported by Alexander von Humboldt Foundation and Banting postdoctoral fellowships (to A.L.M.) and grants from the German Research Foundation (SFB740/C08 to V.H. and TRR186/A08 to V.H. and C.S.). G.D.N. received funding from the Telethon Foundation (GGP13002), the Ministero della Salute (GR-2011-02346974), Fondazione Cariplo (2016-0852), and an Aspire Cardiovascular grant (2016-WI218287). M.F. acknowledges support by the Avner Pancreatic Cancer Foundation and Diabetes Australia, and the infrastructure and staff support provided by Curtin Health Innovation Research Institute, School of Biomedical Sciences and Faculty of Health Sciences, Curtin University.
View Abstract

Navigate This Article