A Diverse Array of Cancer-Associated MTOR Mutations Are Hyperactivating and Can Predict Rapamycin Sensitivity

In mammals, the PI3K–AKT–mTOR pathway regulates cell size, mRNA translation, autophagy, and many metabolic processes, including lipid synthesis (1). A variety of upstream regulators control the activity of the serine/threonine protein kinase mTOR, and many of these are deregulated in cancer, resulting in pathway hyperactivation. Such activation occurs most commonly through loss-of-function mutations in tumor suppressors, such as PTEN, tuberous sclerosis 1/2 (TSC1/2), neurofibromin 1/2 (NF1/2), or oncogenic mutations in KRAS, PIK3CA, or AKT (2). However, few cancer-associated mutations have been functionally characterized in MTOR itself, with only two reports thus far describing such mutations. In the first report (3), the authors tested six cancer-associated MTOR mutations and observed that two (S2215Y from a colorectal sample and R2505P from a kidney sample) conferred mTOR complex 1 (mTORC1) activation. In the second report (4), the authors identified an mTORC1-activating MTOR mutation (L2431P) that was present in a portion but not the entirety of a primary kidney tumor. Although these initial reports establish that activating MTOR mutations do arise in cancer, they were based on limited sample sets that do not reflect the diverse subtypes of cancer. More recently, cancer genome sequencing projects, such as The Cancer Genome Atlas (TCGA) and the Cancer Cell Line Encyclopedia (CCLE), have identified a vast number of somatic mutations in thousands of tumors from more than 40 cancer subtypes (5–7).

Using publicly available databases of cancer genome sequence data, we cataloged all mutations in mTOR pathway components. We annotated more than 400 samples with missense mutations in the MTOR gene from dozens of cancer subtypes, most of which lie within six clusters in the part of the gene that encodes the C-terminal portion of mTOR. Furthermore, through functional analyses, we identify 33 novel mTOR pathway–activating mutations, some of which affect the capacity of mTOR to interact with its partner proteins. None of the activating mutations affect the sensitivity of mTORC1 activity in cells to mTOR inhibitors, but a subset confers mTORC1 signaling resistance to nutrient deprivation. Importantly, cancer cells that naturally express a subset of the mutations are hypersensitive to the mTOR inhibitor rapamycin. These findings may have translational relevance, as rapamycin analogs (rapalogs) are clinically approved for the treatment of cancer, and several ATP-competitive mTOR kinase inhibitors are in development (8). As tumor biopsies are increasingly subjected to whole-exome sequencing, the hyperactivating MTOR mutations we characterize may serve as biomarkers in predicting cancer response to mTOR-targeting drugs.

To generate a comprehensive catalog of all cancer-associated missense mutations in canonical genes of the mTOR pathway, we analyzed partial genome sequencing data from the TCGA, CCLE, International Cancer Genome Consortium (ICGC), and Catalogue of Somatic Mutations in Cancer (COSMIC) databases (5–7, 9, 10). This analysis revealed that almost every gene in the mTOR pathway harbored somatic point mutations (Supplementary Table S1). To enrich for mutations that are more likely to affect pathway function, only mutations that alter the same codon more than once were counted. When normalized for gene length, this analysis revealed that MTOR contained the highest percentage of recurring mutations, and thus we focused most of our attention on these mutations (Supplementary Fig. S1). Collectively, through data mining and a literature search, we curated more than 400 samples with nonsynonymous MTOR point mutations (Fig. 1B; Supplementary Table S2; refs. 11–17). Although the majority of these mutations are represented by only one sample in our database, approximately 40% are recurrent, most of which have not been previously described. In addition, the mutations cluster in six distinct regions of mTOR centered on highly recurrent mutations that alter amino acids C1483, E1799, T1977, S2215, L2427, and R2505 (Fig. 1C). The presence of these clusters suggests that there is a selective advantage to acquiring mutations in certain regions of the mTOR protein. Furthermore, these mutations are present in multiple cancer subtypes, with the highest number in colorectal, endometrial, and lung cancers, although these cancer subtypes are also considered to have the highest mutation rates (Fig. 1D and E; ref. 18). Interestingly, a random mutagenesis screen in the Schizosaccharomyces pombe mTOR homolog (tor2) identified activating mutations in homologous amino acid positions to many of those we find in MTOR to be recurrently mutated in cancer (Supplementary Table S3; ref. 19).

The prevalence of the recurrent mutations, their clustering, and correlation with activating tor2 mutations strongly suggested that the MTOR mutations affect mTOR pathway activity. To test this possibility, wild-type (WT) or 10 different mTOR mutants (L1460P, C1483F, E1799K, F1888L, T1977R, V2006I, S2215Y, I2500F, R2505P, and D2512H) were expressed in cells, and phosphorylation of the mTORC1/2 substrates S6K1, 4EBP1, or AKT1 was examined (Fig. 2A–C). All the mTOR mutants conferred varying degrees of pathway activation, and, interestingly, a few displayed some substrate preference (L1460P, C1483F, S2215Y, and R2505P toward S6K1/4EBP1 and V2006I toward AKT1), implying that these mutations have greater effects on mTORC1 or mTORC2. Thus, although three pathway-activating MTOR mutations have previously been identified (3, 4), our initial analyses uncovered an additional eight recurrent MTOR mutations that activate mTORC1 and/or mTORC2.

It is well established that two activating mutations (E545K and H1047R) within the phosphoinositide 3-kinase, catalytic subunit α (PI3K-p110α), encoded by the PIK3CA gene, can lead to mTOR pathway activation (20). To determine the degree of mTOR pathway activation conferred by mTOR mutation, S6K1 phosphorylation from cells expressing WT or S2215Y mTOR was compared with cells expressing WT or mutant PI3K-p110α (Supplementary Fig. S2A and S2B). This experiment revealed that although expression of only mutant mTOR conferred mTORC1 activation, expression of either WT or mutant PI3K-p110α induced mTORC1 activity.

We next examined in more detail the nature of the activating MTOR mutations, and observed that although some codons are mutated to only one other residue (e.g., E1799K), others are mutated to several different amino acids (e.g., S2215F/P/T/Y). This is reminiscent of the diverse amino acids to which the G12 residue of KRAS is mutated in cancers (21). We expressed 20 such MTOR mutations and found that many (C1483R/W/Y, F1888I, T1977K, S2215F/P, and I2500M), but not all (A41P/S/T, F1888V, T1977S, V2006L, S2215T, R2505Q/*, and D2512G/Y), led to increased S6K1 phosphorylation (Supplementary Figs. S3 and S5 and data not shown). That most of these activating mutations are found only once in our dataset implies that other nonrecurrent mutations may also activate mTORC1 signaling. To identify other such mutations, we turned our attention to the mutational clusters identified in Fig. 1 and observed that although most are represented by several tissue types, those in the C1483 cluster are particularly prevalent in kidney cancer (Fig. 1E and Supplementary Fig. S3D). Interestingly, two recent cancer genome sequencing efforts reported that kidney cancers have an increased percentage of mutations in mTOR pathway genes, but the functional consequences of these mutations were not evaluated (13, 22). Expression of the 10 kidney cancer-associated MTOR mutations from the C1483 cluster revealed that in addition to the previously evaluated L1460P and C1483F/Y mutants (Fig. 1B and Supplementary Fig. S2), three nonrecurrent mutants (L1433S, A1459P, and E1519T) also induced mTORC1 pathway activity (Fig. 2D). Similarly, expression of eight nonrecurrent mutations from the S2215 cluster (N2206S, L2209V, A2210P, L2216P, R2217W, L2220F, Q2223K, and A2226S) activated mTORC1 signaling as measured by S6K1 phosphorylation, with the L2209V mutant conferring higher activation than even S2215Y (Supplementary Fig. S3E). Finally, other than the S2215Y and R2505P mutations, the only other cancer-associated MTOR mutation previously characterized is L2431P from a patient with kidney cancer (4). Expression of mTOR containing this nonrecurrent mutation caused mTORC1 activation that was much weaker than that caused by expression of the S2215Y mutant (Supplementary Fig. S3F). Collectively, these results suggest that although most nonrecurrent MTOR mutations found in cancer are part of the mutational “noise,” there exists a diverse set of MTOR mutations that can drive mTORC1 pathway activity. Likely, as more cancer genomes are sequenced, the activating nonrecurrent mutations we identified will prove to be recurrent.

We also observed that other mTOR pathway genes, such as RPTOR, RICTOR, and RHEB, harbored recurrent mutations (Supplementary Table S1). Expression of RHEB Y35N/C/H and, to a lesser degree, E139K increased phosphorylation of endogenous S6K1 more than that of WT RHEB (Fig. 2E), whereas recurrent mutations in RPTOR (R788C) or RICTOR (S1101L) failed to induce any mTORC1 or mTORC2 activity (data not shown). In support of the relevance of these mutations to cancer, RHEB was recently highlighted as a novel cancer gene due to the presence of the Y35N mutation (23). Thus, cancer-associated mutations in either MTOR or RHEB can induce mTORC1 activity; whether this is true of other mTOR pathway genes remains to be determined.

One potential mechanism through which the mutations in MTOR (Fig. 2) might increase S6K1/AKT1 phosphorylation is by altering mTORC1 or mTORC2 assembly. Alternatively, or in addition, the MTOR mutations may diminish the binding of mTOR to its endogenous inhibitor DEPTOR, which binds to the FAT domain of mTOR (24). To test these possibilities, the interactions of exogenously expressed WT or mutant mTORs with endogenous RAPTOR, RICTOR, and DEPTOR were measured. All mutant mTOR proteins bound equally well to RAPTOR and RICTOR, suggesting that the mutations do not affect formation of either mTOR complex (Fig. 3A). In contrast, DEPTOR binding to the immunoprecipitated mTOR proteins was reduced in cells expressing the mTOR mutants as compared with WT mTOR (Fig. 3A). Given that increased mTOR pathway activity correlates with a reduction in the mTOR–DEPTOR interaction (24), it is likely that the observed decrease in DEPTOR binding reflects the increased mTOR pathway activity caused by the MTOR mutations. However, two mTORC1-activating mutations that are in the FAT domain of mTOR, L1460P and C1483Y, bound much less DEPTOR than the other mutants tested (Fig. 3A), suggesting that these mutations directly perturb the DEPTOR binding site on mTOR. Consistent with this, examination of the mutations from the C1483 cluster described in Fig. 2D revealed that two other highly activating mutations in the FAT domain (A1459P and C1483Y) also strongly reduced the mTOR–DEPTOR interaction (Supplementary Fig. S4A). Although these results suggest a possible mechanism of action for a subset of activating mTOR mutations, they do not preclude other mechanisms by which MTOR mutations may affect mTORC1 signaling, such as by increasing the activity of the mTOR kinase domain.

We next considered the possibility that the mTORC1-activating MTOR mutants have altered sensitivity to established mTOR inhibitors. Engineered mutations in the FRB domain of mTOR, which binds the FKBP12–rapamycin complex, confer marked rapamycin resistance to mTOR (25). To test whether MTOR mutation alters pathway inhibition by rapamycin or Torin1 (an investigational ATP-competitive mTOR inhibitor), we used HEK-293 cells with an integrated flippase recombination site (TREX cells) to stably express WT or several different mTOR mutants (A1459P, C1483Y, E1799K, F1888I, L2209V, S2215Y, L2431P, I2500F, and R2505P), as the MTOR cDNA is too large to be packaged and expressed via a lentiviral delivery system (Supplementary Fig. S4B). Baseline phosphorylation of endogenous S6K1 was higher in TREX cells expressing the A1459P, L2209V, S2215Y, I2500F, and R2505P mTOR mutants than in parental or WT-expressing cells (Fig. 3B and Supplementary Fig. S4C and S4D), consistent with the results obtained upon transient expression of these mutants (Fig. 2A and D and Supplementary Fig. S3D). Importantly, mTORC1 activity in none of the TREX cells was resistant to rapamycin, Torin1, MLN0128 (an mTOR ATP-competitive inhibitor), and GDC0980 (a dual PI3K/mTOR ATP-competitive inhibitor; Fig. 3B; Supplementary Fig. S4C and S4D). Thus, on a molecular level, although cells with MTOR mutations may display high levels of mTORC1/2 activity, this activity is still sensitive to pharmacologic mTOR inhibition.

The kinase activity of mTOR is tightly regulated by nutrient availability; when cells are starved of nutrients (specifically glucose and amino acids), they downregulate mTORC1 signaling, whereas retreatment of cells with these nutrients restores mTORC1-mediated signaling (26). Therefore, we sought to determine whether any of the activating mTOR mutants affect the sensitivity of mTORC1 signaling to amino acid or glucose withdrawal. Indeed, S6K1 phosphorylation was highly resistant to deprivation of either nutrient in cells expressing several mutants (A1459P, S2215Y, I2500F, and R2505P; Fig. 3C; Supplementary Fig. S4E and S4F). This observation may be relevant in the context of developing tumors, which are often glucose starved (27, 28), where inappropriate mTORC1 activity may provide a growth advantage.

To determine whether cancer cells with MTOR mutations are dependent on mTORC1 activity, we evaluated the impact of rapamycin on the proliferation of a set of cancer cell lines harboring WT or mutant mTOR (Fig. 4A). We identified four cell lines with mTORC1-activating MTOR mutations (C1483Y in MOLT16, E1799K in HEC59 and SNU349, and S2215Y in JHUEM7) and two with MTOR mutations that have no apparent impact on mTORC1 signaling (V2006L in TUHR10TKB and R2152C in A375; Supplementary Fig. S5A and S5B). HeLa cells were used as a negative control cell line, as they are mildly rapamycin sensitive and are WT for mTOR (7, 29). MCF7 and SW780 cells, which are also WT for mTOR, served as positive controls for rapamycin sensitivity because the MCF7 cells harbor an activating PI3K-p110α mutation (E545K), whereas the SW780 cells lack NPRL2, which encodes a component of the GATOR1-negative regulator of the amino acid–sensing pathway (30, 31). Importantly, only the cell lines with activating MTOR mutations, the PI3K-p110α mutant, or the NPRL2-null SW780 cells were hypersensitive to rapamycin treatment (Fig. 4B). Consistent with these and published results, the growth of HeLa xenografts was partially rapamycin sensitive, whereas rapamycin completely halted the growth of HEC59 xenografts, which carry the hyperactivating mTOR E1799K mutation (Fig. 4C; refs. 32, 33). These results suggest that the presence of hyperactivating MTOR mutations in cancer cells may serve as biomarkers to identify tumors that are likely to respond to mTOR inhibitors.

Early cancer genome sequencing projects led to the identification of three mutations in MTOR that were shown to activate mTORC1 signaling (3, 4). Here, we have identified an additional 33 previously unknown activating MTOR mutations (some are recurrent, others are not; Supplementary Table S4). Mechanistically, we find that the activating mutations diminish the mTOR–DEPTOR interaction and, interestingly, when mapped onto the recently solved crystal structure of mTOR, the mutations cluster in several distinct locations within the protein itself, suggesting that other activating mechanisms may exist (Supplementary Fig. S6; ref. 34). Although the mutations do not prevent pharmacologic mTOR inhibition, a subset protects against the effects of nutrient deprivation on mTORC1 signaling. Finally, we demonstrate that cell lines with activating MTOR mutations are particularly sensitive to rapamycin in cell culture and in xenografts, likely due to mTOR pathway dependency.

The diversity of activating MTOR mutations we characterized is reminiscent of the various activating mutations in the prototypical lipid kinase PIK3CA, which encodes for the PI3K catalytic subunit p110α (35). However, unlike the activating PIK3CA mutations, which confer equal activation of PI3K-mediated signaling pathways, the diverse MTOR mutations can differentially activate mTORC1 or mTORC2, leading to either S6K1/4EBP1 or AKT1 phosphorylation, respectively. This finding is particularly relevant in a clinical setting, as patients with mTORC1-activating mutations may respond well to U.S. Food and Drug Administration (FDA)–approved rapalogs, whereas patients with mTORC2-activating mutations might best be treated with ATP-competitive mTOR inhibitors, although such inhibitors are still in clinical trials. This distinction is especially relevant for patients with mTORC2-activating mutations, because rapalogs can in some cases upregulate mTORC2 signaling through modulation of an mTORC1-dependent feedback loop (36).

Rapalog-based therapies are currently approved or are in clinical trials for several cancer subtypes, and some patients have had dramatic and durable responses to these therapies (37). Given that some cells with activating MTOR mutations are particularly rapamycin sensitive (Fig. 4), it is possible that some of the rapalog-responsive patients had tumors with similar mutations. Indeed, it was recently found that the tumor of a patient with bladder cancer with two activating MTOR mutations displayed exquisite sensitivity to rapalog-based therapy (38). Many more cancer-associated MTOR mutations will likely be identified in future studies, but functional evaluation of these mutations will be required to determine their impact on the mTOR pathway. As tumor sequencing becomes more common, functional studies such as ours may aid in arriving at a more personalized treatment regimen that eventually promotes patient survival.

All mutations in genes of the mTOR pathway were acquired from the cBIO, COSMIC, and ICGC databases or through a literature search with a data-freeze date of September 1, 2013. To identify regions of mTOR significantly enriched for somatic point mutations, we determined the probability of seeing at least the observed number of mutations in a 50–amino acid sliding window with the null hypothesis being that all 463 MTOR mutations randomly occurring across the 2,549 amino acid protein. Specifically, we calculated a threshold for significance of at least 15 mutations per 50 amino acids by the formula

which equals 0.043 for k = at least 15 mutations per 50 amino acids.

Reagents were obtained from the following sources: antibodies to hemagglutinin (clone C29F4), S6K1 (clone 49D7), phospho–T389-S6K1 (clone 108D2), and phospho–S473-AKT1 (clone D9E) from Cell Signaling Technology; antibody to glyceraldehyde-3-phosphate dehydrogenase (GAPDH; clone GT239) from Genetex; antibodies to DEPTOR and RAPTOR, X-ray film, and polyvinylidenedifluoride (PVDF) membrane from Millipore; antibody to RICTOR from Bethyl; antibody to the FLAG epitope (clone M2), FLAG M2-coupled agarose beads, glucose and amino acid mixtures from Sigma; horseradish peroxidase–labeled anti-mouse and anti-rabbit secondary antibodies from Santa Cruz Biotechnology; FuGENE HD and CellTiter-Glo from Promega; X-tremeGENE 9 and Complete Protease Mixture from Roche Applied Science; QuikChange II XL from Stratagene; enhanced chemiluminescence reagent and glutathione-coupled agarose beads from Thermo; 8-week-old nude mice from Taconic; rapamycin from LC Laboratories; and MLN0128 and GDC0980 from Selleckchem. Torin1 was synthesized in the Gray laboratory (39). cDNAs for mTOR and PI3K-p110α were from Addgene; expression vectors for HA-GST-S6K1, HA-GST-Akt1, and HA-GST-RHEB1 were described previously (24, 40, 41); and Flag-14-3-3β was a generous gift from the Yaffe laboratory. The primers used to generate the mTOR and Rheb1 mutations were from Integrated DNA Technologies and are listed in Supplementary Table S5. The HEK-293T-TREX cell system, 4% to 12% Bis–Tris SDS-PAGE gels, and 20× MOPS buffers were from Invitrogen.

MCF7, HeLa, HEK-293T, HCT116, A375, and SW780 cells were from the American Type Culture Collection (ATCC); JHUEM7, MOLT16, NCIH446, SNU349, and TUHR10TKB cells were from the Broad Institute CCLE; HEK-293E cells were a generous gift from the Blenis Laboratory; and HEC59 cells were a generous gift from the Lippard Laboratory. The cell lines have not been reauthenticated, although those from the ATCC and the Broad Institute (Cambridge, MA) were used within 6 months of resuscitation. The HEK-293T, HEK-293E, HCT116, A375, HEC59, and TUHR10TKB cells were cultured in Dulbecco's Modified Eagle Medium (DMEM); JHUEM7 and MOLT16 cells were cultured in RPMI; and the SW780 cells were cultured in Iscove's Modified Dulbecco's Medium (IMDM). All media were prepared with 10% heat-inactivated FBS and 1% penicillin/streptomycin. All cell lines were maintained at 37°C, 5% CO₂.

All MTOR and RHEB point mutations were generated in the parental vectors using site-directed mutagenesis with the QuikChange II XL Kit according to the manufacturer's protocol. Primer sequences are listed in Supplementary Table S5. Point mutations were verified using Sanger sequencing as conducted by Eton Biosciences. The HA-GST-4EBP1 reporter vector was generated by synthesizing a gBlock from IDT encoding for 4EBP1 and cloning it into the pRK5-HA-GST vector using SalI and NotI restriction digestion followed by ligation. WT and mutant MTOR cDNAs were subcloned directly from the pcDNA3 vector into the pcDNA5/FRT/TO vector using single-site NotI restriction digestion followed by ligation. Flag–mTOR-expressing TREX cells were generated according to the manufacturer's protocol.

For S6K1 and AKT1 cotransfection assays, 3 × 10⁵ HEK-293T or HEK-293E cells per well were seeded into 6-well tissue culture plates and transfected 24 hours later with 2 μg of mTOR-encoding plasmid and 1 ng of S6K1- or AKT1-encoding plasmid using X-tremeGENE 9 or FuGENE HD, followed by whole-cell lysis 48 hours after transfection in NP-40 lysis buffer (50 mmol/L Tris–HCl at pH 7.5, 150 mmol/L NaCl, 1% NP-40 substitute, 1 mmol/L EDTA at pH 8.0, 50 mmol/L NaF, 10 mmol/L Na-pyrophosphate, 15 mmol/L Na₃VO₄, 100 mmol/L β-glycerophosphate, and 1 tablet of protease inhibitor per 50 mL of buffer). Whole-cell lysates (200 to 400 μg) were mixed with 5× SDS sample buffer to a final concentration of 1 to 2 μg/μL, boiled for 5 minutes, and then used directly for immunoblotting or frozen at −20°C. For glutathione pull-down assays, 2 × 10⁶ HEK-293T cells were seeded onto 10-cm tissue culture plates, transfected 24 hours later with 10 μg of mTOR-encoding plasmid and 2 ng of S6K1-encoding plasmid using X-tremeGENE 9, and lysed 48 to 72 hours after transfection in NP-40 lysis buffer. The resultant whole-cell lysates (2 to 3 mg) were incubated with 30 μL of prewashed glutathione-coupled agarose beads in a 1-mL volume, which were rotated at 4°C for 2 hours, spun at 2,500 relative centrifugal force for 1 minute, and washed in 500 μL lysis buffer three times, and the beads directly boiled in 50 μL of 2× SDS sample buffer. For mTOR immunoprecipitations, the cells were treated as with the glutathione pull-down, except they were lysed/washed in CHAPS buffer (50 mmol/L HEPES at pH 7.4, 150 mmol/L NaCl, 0.4% CHAPS, 50 mmol/L NaF, 10 mmol/L Na-pyrophosphate, 100 mmol/L β-glycerophosphate, and 1 tablet of protease inhibitor per 50 mL of buffer) and incubated with FLAG-M2–coupled agarose beads.

Ten to 50 μg of whole-cell lysates, 10 to 30 μL of glutathione pull-down samples, or 10 to 30 μL of Flag-M2 immunoprecipitates were loaded into lanes of 4% to 12% Bis–Tris SDS-PAGE gels, run at 120 V for 2 hours in 1× MOPS buffer, and then transferred to 0.45-μm PVDF membrane at 60 V for 2 hours in 1× transfer buffer (100 mmol/L CAPS, 123 mmol/L NaOH, and 10% ethanol). The membranes were then immunoblotted for the indicated proteins. All primary antibodies were diluted at 1:1,000 in 5% bovine serum albumin w/v TBS+Tween 20 (TBST), with the exception of GAPDH and FLAG antibodies, which were diluted at 1:2,000. All secondary antibodies were diluted at 1:5,000 in 5% milk w/v TBST. Densitometry was performed using ImageJ, and a Student t test was used to determine the statistical significance.

For the mTOR inhibitor experiments, 3 × 10⁵ TREX cells per well were seeded into 6-well plates, grown for 1 day, treated with 100 nmol/L rapamycin, 250 nmol/L Torin1, 250 nmol/L MLN0128, 250 nmol/L GDC0980, or dimethyl sulfoxide (DMSO) for 60 minutes. The cells were then lysed as above. For the nutrient-deprivation experiments, 3 × 10⁵ TREX cells per well were seeded into 6-well plates, grown for 1 day, rinsed briefly in 1× amino acid- or glucose-free RPMI, incubated in 1× amino acid- or glucose-free RPMI for 60 minutes, and then stimulated with free amino acids or glucose to the levels in complete RPMI for 15 minutes.

Six replicates each of 800 to 1,200 cells were seeded into 96-well plates, treated with eight doses of rapamycin 24 hours later, and assayed for ATP content using CellTiter-Glo 96 hours following rapamycin treatment according to the manufacturer's protocol.

A total of 3 × 10⁶ cells per injection site were implanted subcutaneously into the right and left flanks of nude mice. Once tumors were palpable in all animals (>50 mm³ volume by caliper measurements), mice were assigned randomly into rapamycin-treated or rapamycin-untreated groups, and caliper measurements were taken every 3 to 4 days until tumor burden approached the limits set by institutional guidelines. Tumor volume was assessed according to the formula (0.5)(W)(W)(L), and a Student t test was used to determine the P values. Rapamycin (4 mg/kg) or vehicle was delivered by daily intraperitoneal injection, 100 mL per injection. All experiments involving mice were carried out with approval from the Committee for Animal Care at MIT (Cambridge, MA) and under supervision of the Department of Comparative Medicine at MIT.

No potential conflicts of interest were disclosed.

Conception and design: B.C. Grabiner, V. Nardi, D.M. Sabatini

Development of methodology: B.C. Grabiner, R. Possemato, D.M. Sabatini

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): B.C. Grabiner, V. Nardi, K. Birsoy, R. Possemato, S. Sinha, A. Jordan, A.H. Beck

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.C. Grabiner, R. Possemato, K. Shen, S. Sinha, A.H. Beck

Writing, review, and/or revision of the manuscript: B.C. Grabiner, V. Nardi, K. Birsoy, R. Possemato, S. Sinha, A.H. Beck, D.M. Sabatini

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): B.C. Grabiner, A. Jordan, D.M. Sabatini

Study supervision: B.C. Grabiner, D.M. Sabatini

The authors thank N. Wagle and L. Garraway for providing whole-exome sequencing data from a renal cell carcinoma patient who harbored an MTOR missense mutation.

This research is supported in part by a postdoctoral fellowship (#118855-PF-10-069-01-TBG) from The American Cancer Society (to B.C. Grabiner), The Leukemia and Lymphoma Society and The Jane Coffin Childs Fund (to K. Birsoy), NIH K99 CA168940 (to R. Possemato), and grants from the Starr Cancer Consortium, the David H. Koch Institute for Integrative Cancer Research at MIT, The Alexander and Margaret Stewart Trust Fund, and the NIH (CA103866, CA129105, and AI07389) to D.M. Sabatini. D.M. Sabatini is an investigator of the Howard Hughes Medical Institute.