A Tumor Suppressor Complex with GAP Activity for the Rag GTPases That Signal Amino Acid Sufficiency to mTORC1

See allHide authors and affiliations

Science  31 May 2013:
Vol. 340, Issue 6136, pp. 1100-1106
DOI: 10.1126/science.1232044

Limiting mTORC1

The mTORC1 protein kinase complex has important functions linking metabolism to cell growth and its functions are disrupted in common diseases, including cancer and diabetes. Bar-Peled et al. (p. 1100; see the Perspective by Shaw) discovered regulatory components that help turn down signaling by mTORC1 when cells are deprived of amino acids. Two complexes of proteins, GATOR1 and GATOR2, have opposite effects on activity and cellular localization of mTORC1. Components of the GATOR1 complex negatively regulate mTORC1 and appear to function as tumor supressors. Cancers with loss of GATOR1 function may be particularly amenable to therapeutic strategies that limit activity of mTORC1.


The mTOR complex 1 (mTORC1) pathway promotes cell growth in response to many cues, including amino acids, which act through the Rag guanosine triphosphatases (GTPases) to promote mTORC1 translocation to the lysosomal surface, its site of activation. Although progress has been made in identifying positive regulators of the Rags, it is unknown if negative factors also exist. Here, we identify GATOR as a complex that interacts with the Rags and is composed of two subcomplexes we call GATOR1 and -2. Inhibition of GATOR1 subunits (DEPDC5, Nprl2, and Nprl3) makes mTORC1 signaling resistant to amino acid deprivation. In contrast, inhibition of GATOR2 subunits (Mios, WDR24, WDR59, Seh1L, and Sec13) suppresses mTORC1 signaling, and epistasis analysis shows that GATOR2 negatively regulates DEPDC5. GATOR1 has GTPase-activating protein (GAP) activity for RagA and RagB, and its components are mutated in human cancer. In cancer cells with inactivating mutations in GATOR1, mTORC1 is hyperactive and insensitive to amino acid starvation, and such cells are hypersensitive to rapamycin, an mTORC1 inhibitor. Thus, we identify a key negative regulator of the Rag GTPases and reveal that, like other mTORC1 regulators, Rag function can be deregulated in cancer.

The mTOR complex 1 (mTORC1) kinase is a master regulator of growth, and its deregulation is common in human disease, including cancer and diabetes (1). In response to a diverse set of environmental inputs, including amino acid levels, mTORC1 regulates many anabolic and catabolic processes, such as protein synthesis and autophagy (1, 2). The sensing of amino acids by mTORC1 initiates from within the lysosomal lumen (3) and requires a signaling machine associated with the lysosomal membrane that consists of the Rag guanosine triphosphatases (GTPases) (4, 5), the Ragulator complex (6, 7), and the vacuolar adenosine triphosphatase (V-ATPase) (3). The Rag GTPases exist as obligate heterodimers of RagA or RagB, which are highly homologous, with either RagC or RagD, which are also very similar to each other (4, 5, 8). Through a poorly understood mechanism requiring the V-ATPase, luminal amino acids activate the guanine nucleotide exchange factor (GEF) activity of Ragulator toward RagA and RagB that, when guanosine triphosphate (GTP)–loaded, recruits mTORC1 to the lysosomal surface (7). There, mTORC1 interacts with its activator Rheb, which is regulated by many upstream signals, including growth factors (1). Upon amino acid withdrawal, RagA and RagB become guanosine diphosphate (GDP)–bound (4), and mTORC1 leaves the lysosomal surface, and that leads to its inhibition. The negative regulators that inactivate the Rag GTPases are unknown.

We suspected that important regulators of the Rags might have escaped prior identification because their interactions with the Rags are too weak to persist under standard purification conditions. Thus, to preserve unstable protein complexes (9), we treated human embryonic kidney–293T (HEK-293T) cells expressing FLAG-tagged RagB with a chemical cross-linker and identified via mass spectrometry proteins that coimmunoprecipitate with FLAG-RagB. This analysis revealed the presence in the immunoprecipitates of known Rag-interacting proteins, as well as Mios, a 100-kD WD40-repeat protein not previously studied (fig. S1A). Consistent with this finding, endogenous RagA and RagC coimmunoprecipitated with recombinant Mios expressed in HEK-293T cells and isolated under similar purification conditions (Fig. 1A). Suppression of Mios, by RNA interference (RNAi) in human cells, strongly inhibited the amino acid–induced activation of mTORC1, as detected by the phosphorylation state of its substrate S6K1 (Fig. 1B and fig. S2B). Moreover, in Drosophila S2 cells, double-stranded RNAs (dsRNAs) targeting Mio (10), the fly ortholog of Mios, ablated dTORC1 signaling and also reduced cell size (Fig. 1, C and D). Thus, in human and fly cells, Mios is necessary for amino acid signaling to TORC1.

Fig. 1 GATOR is a Rag-interacting complex, whose Mios component is necessary for the activation of mTORC1 by amino acids.

(A) Mios interacts with endogenous RagA and RagC. HEK-293T cells were transfected with the indicated cDNAs in expression vectors. Cells were treated with a cell-permeable chemical cross-linker, lysates were prepared and subjected to FLAG immunoprecipitation (IP) followed by immunoblotting for the indicated proteins. (B) Mios is necessary for the activation of the mTORC1 pathway by amino acids. HEK-293T cells expressing short hairpin RNAs (shRNAs) targeting GFP or Mios were starved of amino acids for 50 min or starved and then restimulated with amino acids for 10 min. Cell lysates were analyzed for the phosphorylation state of S6K1. (C) S2 cells treated with dsRNAs targeting Mio or GFP were starved of amino acids for 90 min or starved and restimulated with amino acids for 30 min. The indicated proteins were detected by immunoblotting. (D) Cell size histogram of S2 cells after dsRNA-mediated depletion of Mio. (E and F) GATOR is an octomeric complex defined by two distinct subcomplexes and interacts with the Rag GTPases. HEK-293T cells were transfected and processed as in (A) with the exclusion of the cross-linking reagent, and cell lysates and FLAG immunoprecipitates were subjected to immunoblotting. (G) HEK-293T cells stably expressing FLAG-tagged DEPDC5 or WDR24 were lysed, and cell lysates and FLAG immunoprecipitates were analyzed by immunoblotting for endogenous RagA, RagC, Mios, and Nprl3. (H) Schematic summarizing GATOR-Rag interactions. GATOR2 (Mios, Seh1L, WDR24, WDR59, and Sec13) interacts with GATOR1 (DEPDC5, Nprl2, and Nprl3), which likely then binds the Rags.

In vitro, we failed to detect a strong interaction between purified Mios and the Rag heterodimers, which suggested that, within cells, other components exist that are needed for complex formation. Indeed, in FLAG-Mios immunoprecipitates prepared from HEK-293T cells, we detected seven additional proteins (WDR24, WDR59, Seh1L, Sec13, DEPDC5, Nprl2, and Nprl3) by mass spectrometry. The proteins varied in abundance, however, with much greater amounts of WDR24, WDR59, Seh1L, and Sec13 coimmunoprecipitating with Mios than DEPDC5, Nprl2, and Nprl3 (fig. S1B). In contrast, in FLAG-DEPDC5 immunoprecipitates, Nprl2 and Nprl3 were more abundant than Mios, WDR24, WDR59, Seh1L, and Sec13, and experiments with FLAG-Nprl2 gave analogous results (fig. S1B). These findings suggest that two subcomplexes exist, one consisting of Mios, WDR24, WDR59, Seh1L, and Sec13, and the other of DEPDC5, Nprl2, and Nprl3. To test this notion, we coexpressed FLAG-WDR24 or FLAG-Nprl2 together with hemagglutinin (HA)–tagged versions of the other seven proteins. As expected, DEPDC5 and Nprl3 coimmunoprecipitated with Nprl2 much more strongly than with WDR24, whereas the opposite was true for Mios, WDR59, Seh1L, and Sec13 (Fig. 1E). For reasons described later, we call the eight-protein complex GATOR for GTPase-activating protein (GAP) activity toward Rags and the two subcomplexes GATOR1 (DEPDC5, Nprl2, and Nprl3) and GATOR2 (Mios, WDR24, WDR59, Seh1L, and Sec13) (Fig. 1H).

When the eight proteins were coexpressed with RagB and RagC, GATOR interacted strongly with the Rag heterodimer (Fig. 1F) and, like the Rags and Ragulator (6, 7), its DEPDC5 component localized to the lysosomal surface (fig. S1E). Experiments in which single GATOR proteins were omitted revealed complex relations between the components but suggested that GATOR1 mediates the GATOR-Rag interaction (fig. S1C). Consistent with this conclusion, when stably expressed in HEK-293T cells, FLAG-DEPDC5 coimmunoprecipitated much more endogenous RagA and RagC than FLAG-WDR24, as detected by immunoblotting (Fig. 1G) and mass spectrometric analysis (fig. S1D). Amino acid starvation increased the amount of RagA and RagC that coimmunoprecipitated with DEPDC5, which suggested a regulatory role for GATOR1 (fig. S1F).

The finding that GATOR components interact with the Rag GTPases was intriguing because their likely budding yeast orthologs (IML1, NPR2, and NPR3) positively regulate autophagasome formation, a TORC1-dependent process (11), and, at least in certain yeast strains, also inhibit TORC1 signaling upon nitrogen starvation (1214). Recently, the likely yeast orthologs of GATOR2 (Sea2, Sea3, Sea4, Seh1L, and Sec13) were shown to interact with IML1, NPR2, and NPR3 to form a complex that has been called SEA (15). However, unlike GATOR, the SEA complex does not appear to consist of two distinct subcomplexes, as its components are found in stoichiometric amounts.

We used RNAi in HEK-293T and Drosophila S2 cells to examine the function of each GATOR component in amino acid sensing by mTORC1 and dTORC1, respectively. We excluded Sec13 from further analysis, as it functions in several protein complexes (16), and so its inhibition might have effects that are difficult to interpret. Consistent with Mios being required for amino acids to activate mTORC1 (Fig. 1B), depletion of other GATOR2 components or their Drosophila orthologs strongly blunted amino acid–induced activation of mTORC1 and dTORC1, respectively (Fig. 2, A, C, and D; and fig. S2, A to D). It is interesting that in contrast, loss of GATOR1 proteins had the opposite effect and prevented the inactivation of mTORC1 and dTORC1 normally caused by amino acid deprivation (Fig. 2, B and E, and fig.S2, A and D). Consistent with the opposite roles of GATOR1 and GATOR2 on dTORC1 signaling, dsRNAs targeting dSeh1L or dDEPDC5 decreased and increased, respectively, S2 cell size (Fig. 2F). To clarify the relation between GATOR1 and GATOR2, we used RNAi to inhibit dDEPDC5 at the same time as Mio or dSeh1L in S2 cells. In the background of GATOR1 inhibition, loss of GATOR2 had no effect on dTORC1 activity, which indicated that GATOR2 functions upstream of GATOR1 (Fig. 2, G and H). Thus, GATOR2 is an inhibitor of an inhibitor (GATOR1) of the amino acid–sensing branch of the TORC1 pathway.

Fig. 2 The GATOR complex is required for the regulation of the TORC1 pathway by amino acids.

(A) shRNA-mediated depletion of the GATOR2 components Seh1L, WDR24, or WDR59 in HEK-293T cells inhibits amino acid–induced S6K1 phosphorylation. (B) In HEK-293Ts expressing shRNAs targeting the GATOR1 components DEPDC5, Nprl2, and Nprl3, S6K1 phosphorylation is insensitive to amino acid withdrawal. In (A) and (B), cells were starved of amino acids for 50 min or starved and restimulated with amino acids for 10 min. Cell lysates were immunoblotted for the phosphorylation state of S6K1. dsRNA-mediated depletion in S2 cells of (C) dSeh1L; (D) dWDR59 and dWDR24; and (E) dDEPDC5, dNprl2, and dNprl3. S2 cells were treated with the indicated dsRNAs and were starved of amino acids for 90 min or starved and restimulated with amino acids for 30 min. Immunoblotting was used to detect the phosphorylation state of dS6K. (F) S2 cell sizes after dsRNA-mediated depletion of dSeh1L and dDEPDC5. (G) GATOR2 functions upstream of GATOR1. S2 cells were treated with the indicated combinations of dsRNAs and then starved and restimulated with amino acids and analyzed as in (C) to (E). (H) Schematic depicting the relationship between GATOR1 and GATOR2 in their regulation of mTORC1.

A key step in the amino acid–induced activation of mTORC1 is its recruitment to the lysosomal surface, an event that requires known positive components of the amino acid–sensing pathway, like Ragulator (7) and the V-ATPase (3). Consistent with a positive role for GATOR2, in HEK-293T cells expressing shRNAs targeting Mios (Fig. 1B) or Seh1L (Fig. 2A), mTOR did not translocate to LAMP2-positive lysosomal membranes upon amino acid stimulation (Fig. 3A). In contrast, in cells expressing an shRNA targeting DEPDC5 (fig. S2A), mTOR localized constitutively to the lysosomal surface, regardless of amino acid availability (Fig. 3B). Moreover, just overexpression of DEPDC5 was sufficient to block the amino acid–induced translocation of mTOR to the lysosomal surface (Fig. 3C). Unlike Ragulator, which tethers the Rags to the lysosomal surface (6, 7), GATOR2 is not needed for the proper Rag localization (fig. S3A). Thus, GATOR1 and GATOR2 have opposite effects on the activity and subcellular localization of mTORC1.

Fig. 3 GATOR regulates mTORC1 localization to the lysosomal surface and functions upstream of the Rag GTPases.

(A) RNAi-mediated depletion of the GATOR2 components Mios and Seh1L prevents amino acid–induced mTOR lysosomal translocation. HEK-293T cells expressing the indicated shRNAs were starved or starved and restimulated with amino acids for the specified times before coimmunostaining for mTOR (red) and Lamp2 (green). (B) Reduced expression of DEPDC5 in HEK-293T cells results in constitutive mTOR localization to the lysosomal surface. HEK-293T cells treated with the indicated lentiviral shRNAs were processed as described in (A). (C) Images of HEK-293T cells stably expressing FLAG-DEPDC5 starved of, or starved and restimulated with, amino acids. Cells were processed as described in (A). In all images, insets show selected fields that were magnified five times and their overlays. Scale bar, 10 μM. (D) GATOR1 functions upstream of the nucleotide binding state of the Rags. HEK-293T cells transfected with the indicated cDNAs in expression vectors were starved of amino acids for 50 min or starved and restimulated with amino acids for 10 min. The indicated proteins were detected by immunoblotting.

Consistent with the finding that GATOR1 inhibited the mTORC1 pathway, concomitant overexpression of its three components blocked the amino acid–induced activation of mTORC1 (Fig. 3D) to a similar extent as RagBT54N-RagCQ120L, a Rag heterodimer that is dominant negative because the RagBT54N mutant cannot bind GTP (7). In contrast, expression of the dominant active RagBQ99L-RagCS75N heterodimer blocked not only amino acid deprivation but also GATOR1 overexpression from inhibiting mTORC1 signaling. Because RagBQ99L is constitutively bound to GTP (17) and RagCS75N cannot bind GTP (7), this result suggests that GATOR1 functions upstream of the regulation of the nucleotide binding state of the Rags.

To test the possibility that GATOR1 is a GEF or a GAP for the Rags, we prepared Rag heterodimers consisting of a wild-type Rag and a RagX mutant (see methods) (7). The RagX mutants are selective for xanthosine rather than guanine nucleotides, which allowed us to prepare heterodimers in vitro in which the wild-type Rag is loaded with radiolabeled GTP or GDP while the RagX partner is bound to XDP or XTP (7). In vitro, purified GATOR1 (fig. S4E) did not stimulate the dissociation of GDP from RagB or RagC when each was bound to its appropriate RagX partner (fig. S4, A and B); this ruled out its function as a GEF. In contrast, GATOR1 strongly increased, in a time- and dose-dependent manner, GTP hydrolysis by RagA or RagB within RagCX-containing heterodimers, irrespective of which nucleotide RagCX was loaded with (Fig. 4, A, B, and E; and fig. S4, C and D). GATOR1 also slightly boosted GTP hydrolysis by RagC within a RagBX-RagC heterodimer (Fig. 4C) but had no effect on the GTPase activity of Rap2A (Fig. 4D). Leucyl–transfer RNA synthetase (LRS), a putative GAP for RagD (18), did not alter the basal GTP hydrolysis by RagA, RagB, or RagC (Fig. 4, A, B, and C). Consistent with the binding preference of many GAPs for the GTP-loaded state of target GTPases, in vitro GATOR1 preferentially interacted with the RagBQ99L-containing heterodimer (Fig. 4F). Thus, the GATOR1 complex has GAP activity for RagA and RagB, which provides a mechanism for its inhibitory role in mTORC1 signaling.

Fig. 4 GATOR1 is a GTPase-activating protein complex for RagA and RagB.

(A to D) GATOR1 stimulates GTP hydrolysis by RagA and RagB. RagA-RagCX, RagB-RagCX, RagC-RagBX, or the control GTPase Rap2A were loaded with [α-32P]GTP and incubated with GATOR1 (20 pmol) or the control LRS (20 pmol). GTP hydrolysis was determined by thin-layer chromatography (see methods). Each value represents the normalized mean ± SD (n = 3). (E) GATOR1 increases GTP hydrolysis by RagB in a time-dependent manner. RagB-RagCX was loaded with [γ-32P]GTP and incubated with GATOR1 or a control, and hydrolysis was determined by phosphate capture (see methods) Each value represents the normalized mean ± SD (n = 3). (F) GATOR1 preferentially interacts with the dominant active Rag heterodimer. In vitro binding assay in which FLAG-GATOR1 was incubated with immobilized HA–glutathione S-transferase (HA-GST)–tagged RagBT54N-RagCQ120L (dominant negative), RagBQ99L-RagCS75N (dominant active) or Rap2A. HA-GST precipitates were analyzed by immunoblotting for the levels of FLAG-GATOR1.

Because the pathways that convey upstream signals to mTORC1 are frequently deregulated by mutations in cancer [reviewed in (19)], we thought it possible that GATOR1 components might be mutated in human tumors. Indeed, previous studies identified in lung and breast cancers deletions of a 630–kilobase (kb) region of 3p21.3 that includes NPRL2 (20, 21), and one study reported two cases of glioblastoma with deletions in a three-gene region of 22q12.2 that contains DEPDC5 (22). Moreover, in cancer cells with 3p21.3 deletions, expression of Nprl2 inhibited their capacity to grow as tumor xenografts, which identified it as a tumor suppressor (21, 23). Our analyses of publically available data from the Cancer Genome Atlas identified a subset of glioblastomas and ovarian cancers with nonsense or frameshift mutations or truncating deletions in DEPDC5 or NPRL2. In most of these tumors, DEPDC5 or NPRL2 also underwent a loss of heterozygosity (LOH) event, which indicated that the tumors were unlikely to retain a functional copy of the gene products (Fig. 5, A, B, and C; and fig. S5A). In addition, in both tumor types, focal homozygous or hemizygous deletions, as well as missense mutations accompanied by LOH, were also detected in DEPDC5 and NPRL2 (Fig. 5A). NPRL3 is located too proximal to the 16p telomere to adequately access copy number alterations in it using high-density microarray analysis. In aggregate, inactivating mutations in GATOR1 components are present in low single-digit percentages of glioblastomas and ovarian cancers, frequencies that may change upon better assessment of NPRL3.

Fig. 5 GATOR1 components are mutated in cancer and GATOR1-null cancer cells are hypersensitive to the mTORC1 inhibitor rapamycin.

(A) Table summarizing genomic alterations in DEPDC5 and NPRL2 and their frequencies in glioblastoma and ovarian cancer. The ratios of nonsense and frameshift mutations to missense mutations in DEPDC5 (P = 0.00015) and NPRL2 (P = 0.00342) in glioblastoma differ significantly from the ratio of all nonsense and frameshift mutations to missense mutations in glioblastoma genomes as determined by a Fisher's Exact test. (B and C) Mutations and deletions identified in DEPDC5 in glioblastomas and ovarian cancers. (D) In GATOR1-null cancer cells the mTORC1 pathway is resistant to amino acid starvation. Cells were starved of amino acids for 50 min or starved or restimulated with amino acids for 10 min. Cell lysates were analyzed by immunoblotting for levels of the indicated proteins. (E) Cancer cells were starved or starved and restimulated with amino acids (a.a.) for the specified times before coimmunostaining for mTOR (red) and Lamp2 (green). In all images, insets show selected fields that were magnified five times and their overlays. Scale bar, 10 μM. (F) Reintroduction of Nprl2 into the SW780 cell line (NPRL2−/−) restores amino acid–dependent regulation of mTORC1. Cells stably expressing a control protein or Nprl2 were treated and analyzed as in (D). (G) GATOR1-null cancer cells are hypersensitive to rapamycin. Rapamycin IC50 values for indicated cancer cell lines. Values are presented as means ± SD (n = 3). (H) Model for the role of the GATOR complex in the amino acid–sensing branch of the mTORC1 pathway. GATOR2 is a negative regulator of GATOR1, which inhibits the mTORC1 pathway by functioning as a GAP for RagA.

In order to study the effects of GATOR1 loss on cancer cells, we used the Cosmic and CCLE resources (see methods) to identify human cancer cell lines with homozygous deletions in DEPDC5, NPRL2, or NPRL3, which we confirmed via immunoblotting or polymerase chain reaction (PCR) of genomic DNA (fig. S5, B, C, and F). In seven such lines, but not in Jurkat or HeLa cells, mTORC1 signaling was hyperactive and completely insensitive to amino acid deprivation and V-ATPase inhibition, irrespective of which GATOR1 component was lacking (Fig. 5D and fig. S5, D, E, and I). Furthermore, in GATOR1-null cells mTOR localized to the lysosomal surface even in the absence of amino acids (Fig. 5E). When DEPDC5 and Nprl2 were reintroduced into cancer cell lines lacking them, the mTORC1 pathway regained sensitivity to amino acid regulation (Fig. 5F and fig. S5, G, and H), which indicated that it is indeed the loss of GATOR1 proteins that is driving aberrant mTORC1 signaling in these cells.

The proliferation of the GATOR1-null cancer cells was very sensitive to the mTORC1 inhibitor rapamycin, with median inhibitory concentration (IC50) values in the 0.1 to 0.4 nM range (Fig. 5G). These values are many orders of magnitude less than for cell lines that are not considered rapamycin-sensitive, like HeLa and HT29 cells, and at the low end of cancer cell lines, like PC3 and Jurkat cells, which have lost PTEN function (2426), an established negative regulator of the mTORC1 pathway. In addition, the forced expression of DEPDC5 in the MRKNU1 (DEPDC5−/−) cell line led to a marked reduction in its proliferation (fig. S5J).

In conclusion, we identify the octomeric GATOR complex as a critical regulator of the pathway that signals amino acid sufficiency to mTORC1 (Fig. 5G). The GATOR1 subcomplex has GAP activity for RagA and RagB and its loss makes mTORC1 signaling insensitive to amino acid deprivation. Inactivating mutations in GATOR1 are present in cancer and may help identify tumors likely to respond to clinically approved pharmacological inhibitors of mTORC1.

Supplementary Materials

Materials and Methods

Figs. S1 to S5

References (2731)

References and Notes

  1. Acknowledgments: We thank all members of the Sabatini lab for helpful suggestions, E. Spooner for the mass spectrometric analysis of samples, and N. Kory for technical assistance. This work was supported by grants from the NIH (CA103866 and AI47389) and Department of Defense (W81XWH-07-0448) to D.M.S. and the National Cancer Institute (NIH) (U24CA143867) to M.M. and awards from the David H. Koch Graduate Fellowship Fund to L.B.-P.; the NSF Graduate Research Fellowship Program to L.C.; the Harvard-MIT Health, Sciences, and Technology IDEA2 program to W.W.C.; and the American Cancer Society to B.C.G. D.M.S. is an investigator of the Howard Hughes Medical Institute.
View Abstract

Navigate This Article