Activating mTOR Mutations Are Detrimental in Nutrient-Poor Conditions

Cellular metabolic reprogramming to promote growth and proliferation is one of the hallmarks of cancer. Normal cells, after being stimulated by growth factors, initiate a receptor tyrosine kinase (RTK)–PI3K–Akt-pathway that leads to the activation of the kinase mTOR. mTOR is the catalytic subunit of both the mTOR complex 1 (mTORC1) and the mTOR complex 2 (mTORC2; refs. 1, 2). mTORC2 contributes to growth factor signaling and the regulation of cytoskeletal rearrangements, whereas mTORC1 promotes cell growth by increasing nutrient uptake, de novo lipid, nucleotide, and protein synthesis, and regulating cell-cycle progression (3–5). mTORC1 also inhibits catabolic processes such as autophagy and lysosomal biogenesis (Fig. 1A; refs. 2, 6–8). In the absence of growth signals, mTORC1 remains inactive as a result of tuberous sclerosis complex (TSC) inhibition of Rheb, the GTPase that activates mTOR kinase. Growth factor signaling through Akt inhibits the TSC complex and promotes mTORC1 activation. In addition to regulation by growth factor signaling, mTORC1 activity is also modulated by a nutrient-sensing regulatory arm (9).

Although some oncogenic drivers can indirectly activate mTOR, the canonical oncogenic pathway regulating mTOR is the RTK–PI3K–Akt–mTOR axis (10, 11). This pathway is negatively regulated by two major tumor suppressors, the TSC1/TSC2 complex and PTEN. Mutations in each of these upstream components are thought to have the ultimate effect of activating mTOR to promote uncontrolled growth. Supporting a role of mTOR in oncogenesis, gain-of-function mutations in mTOR have been isolated from tumor cells and demonstrated to exhibit constitutive kinase activity (12–14). Yet clinically, mTOR inhibitors have shown limited activity (15). In this study, we show that although mutations in RTKs and PI3K are significantly enriched in multiple tumor types, mutations of mTOR are significantly enriched only in renal clear cell carcinoma (RCCC). Examining the properties of cancer cells in which activating mTOR mutants are ectopically introduced, we consistently found that such mutations were detrimental to cell growth whenever either amino acids, glucose, or oxygen became limiting. This suggests that the ability to modulate mTORC1 in response to variations of nutrient levels in the tumor microenvironment was maintained in most human tumors. Consistent with this, using analysis of The Cancer Genome Atlas (TCGA) pan-cancer database we found that mutations in the components of the nutrient-sensing arm of mTORC1 regulation are exceedingly rare, despite the fact that such mutations lead to constitutive activation of mTORC1.

The following reagents antibodies were from: Cell Signaling Technology #9234 phospho-T389 S6K1 RRID:AB_2269803, #2708 S6K1 RRID:AB_390722, #2920 Akt RRID:AB_1147620, #2983 mTOR RRID:AB_2105622, #4060 phospho‐S473 Akt RRID:AB_2315049, #2855 phospho-T37/46 4E‐BP1 RRID:AB_560835, #6888 phospho-S757 Ulk1 RRID:AB_10829226, #8054 Ulk1 RRID:AB_11178668. Secondary antibodies were from: GE Healthcare (HRP‐linked anti‐rabbit RRID:AB_772206, HRP‐linked anti‐mouse RRID:AB_772209). Inhibitors were from: Tocris Bioscience (PD98059, Torin 1). Amino acid starvation media were from Athena (DMEM/F12 0423) or prepared by the Memorial Sloan Kettering Cancer Center media core facility. Gibco Human Plasma-Like Medium (HPLM) was from Thermo Fisher Scientific, A4899101. Dialyzed FBS was from Gemini Biosystems and BSA (A1470) was from Sigma.

K-RasG12D and wild-type (WT) mouse embryonic fibroblasts (MEF) were generated in the Thompson laboratory, as described previously (6). PTEN KO MEFs and MEFs expressing myr–Akt were generated in the Thompson laboratory, as described previously (16). Caki-1 cells were kindly provided by J. Hsieh. KPC (KrasLSL-G12D Trp53LSL-R272H Pdx1-Cre) murine pancreatic ductal adenocarcinoma cell lines were kindly provided by Scott Lowe. Generation of cell lines expressing tetracycline-inducible WT or mutant mTOR cells were made as described, with plasmids for TRE-Flag tagged WT or mutant mTOR kindly provided by J. Hsieh (Washington University School of Medicine in St. Louis, St. Louis, MO; ref. 17). Expression of mTOR WT and Mutants was induced by adding 1μg/mL doxycycline (Sigma) to culture medium.

gRNAs targeting DEPDC5 were cloned into LentiCRISPR v2. Plasmids were cotransfected with lentivirus packaging plasmids into HEK293T cells using Lipofectamine 2000 Transfection Reagent (Life Technologies), fresh media were added after 16 h, and viral supernatants were collected 2 days after infection. Target cells were infected by addition of viral supernatant and 8 μg/mL polybrene. Twenty-four hours after infection, cells were selected with appropriate antibiotics.

All cell lines were verified to be Mycoplasma-negative by MycoAlert Mycoplasma Detection Kit (LT07-318, Lonza). Cell lines were tested biannually. All cells used for experiments were maintained for less than 20 passages.

Cell culture experiments were performed at 37°C and 5% CO₂ in DMEM or DMEM/F12 with 10% dialyzed FBS (molecular weight cutoff value of 10,000), 100 U/mL penicillin, 100 μg/mL streptomycin, 17.5 mmol/L glucose, and 4 mmol/L glutamine with 1 μg/mL doxycycline to induce mTOR WT or Mutant expression, unless otherwise specified. Glucose and glutamine consumption and lactate secretion were measured as previously described (18).

For proliferation assays, MEFs were plated in complete medium, 5 hours later or overnight briefly rinsed, and then cultured in starvation medium as indicated. For 5% amino acid (AA) experiments, AAs were supplemented to total 5% of DMEM/F12 amino acid solution. For HPLM experiments, cells were grown in HPLM medium supplemented with 10% dialyzed FBS. For hypoxia experiments, cells were cultured in a hypoxia chamber at 1% or 0.5% oxygen for the indicated length before counting. Cell numbers at the start of treatment (d0) and 1–4 days later were counted in triplicates using the Multisizer 3 Coulter Counter (Beckman) and normalized to the cell number at d0.

TCGA pan-cancer studies and MSK IMPACT data were accessed using cBioPortal version 3.6 (RRID:SCR_014555; refs. 19, 20). Putative driver mutations were identified by filtering variant annotation counts identified by OncoKB driver annotation (RRID:SCR_014782), Hotspots, cBioPortal > = 10, and COSMIC ≥ 10 (the number of occurrences of mutations at the same amino acid position present in the COSMIC database; RRID:SCR_002260; refs. 21). Significance of mutual exclusion and co-occurrence of alterations of MTOR, PTEN, and PIK3CA in RCCC and endometrial cancer were also obtained from cBioPortal. Graphs were created using GraphPad Prism software (RRID:SCR_002798). In vitro experimental data are representative of three independent experiments, and each independent experiment is done with samples in triplicate. Statistical significance is based on one-way ANOVA. P values are as indicated: ns, not significant; *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001. R package maftools = v2.6.0 was used to format TCGA MC3 public access MAF files ahead of analysis with MutSigCV = v1.41 using default full-coverage and covariate files (22). Statistical significance reported on TCGA MC3 data is based on Bonferroni correction for the number of cancer types considered (n = 33). Code used to generate figures, and reproducibility details are available at https://github.com/morrislab/mtor-rcc/.

Protein extracts were prepared by using 1 × RIPA buffer (20–188, Millipore) supplemented with protease (1860932, Thermo Fisher Scientific) and phosphatase inhibitors (78428, Thermo Fisher Scientific). Equal amounts of total protein were separated on NuPAGE Bis-Tris, transferred to nitrocellulose membranes and subjected to Western blotting with indicated primary antibodies.

TCGA MC3 data are publicly available via the NIH Genomic Data Commons (https://gdc.cancer.gov/about-data/publications/mc3-2017). All PCAWG analysis results were downloaded from the ICGC data portal (https://dcc.icgc.org/releases/PCAWG/drivers/p-values).

Using the TCGA Pan-Cancer Atlas and Memorial Sloan Kettering IMPACT tumor sequencing data, the prevalence of mTOR alterations was examined. Despite mTOR's central role in directing cellular growth (Fig. 1A), mTOR alterations are rare; occurring in only 4% of tumors sequenced in the TCGA Pan-Cancer Atlas dataset (Fig. 1B), and 3% of tumors sequenced within the MSK IMPACT dataset (Supplementary Fig. S1A; refs. 19, 20, 23). Many of these identified alterations are unlikely to be oncogenic (24, 25). When more stringent criteria were used to identify putative driver alterations, such as OncoKB annotation and hotspot analysis, potential activating alterations in mTOR only made up 1.1% of all samples within the TCGA pan-cancer atlas (Fig. 1B), and 0.8% in IMPACT (Supplementary Fig. S1A).

We then looked at the frequency of activating alterations in upstream effectors within this pathway, including RTKs, PI3K, and Akt; or deletions in PTEN, the main suppressor of pathway activity (Fig. 1C; Supplementary Fig. S1B). RTKs were found to be altered in >40% of all tumor samples within the TCGA pan-cancer dataset; with nearly 75% of these being identified as potential oncogenic drivers (Fig. 1C). Thus, oncogenic alterations in RTKs occur nearly 30-fold more often than potentially activating mTOR alterations. The 16 most commonly altered RTKs were grouped in the above analysis as they are cell-type–specific and we analyzed a pan-cancer dataset. Recognizing that a comparison of multiple genes with one gene is imperfect, we performed additional comparison of alteration frequencies within the Breast Invasive Carcinoma TCGA tumor sequencing data. ERBB2 as a representative RTK was altered in 14% of samples, with nearly all of these alterations acting as putative drivers. However, only 0.4% of samples contained potentially activating mutations in mTOR (Supplementary Fig. S1C). Other upstream activators of the mTOR pathway are also frequently altered in cancers. PIK3CA, PTEN, and Akt are altered in 17%, 12%, and 8% of tumor samples, respectively; and the vast majority of these mutations have been classified by OncoKB and hotspot analysis as putative drivers. Thus, potentially activating alterations of mTOR are rare compared with oncogenic alterations within upstream effectors of the RTK/PI3K/Akt/mTOR pathway.

Although the putative driver filters within cBioportal are an expedient way to differentiate passenger from driver alterations noted on sequencing datasets, it introduces bias as less-well studied genes may have true driver mutations that are not identified and included within the putative driver annotation. To further address the issue of passenger versus driver mutations, we performed a MutSigCV analysis on mTOR mutations as well as PIK3CA or PTEN, across the TCGA pan-cancer studies. MutSigCV tests whether mutations within genes occur at a rate higher than expected by chance, by taking into account covariates such as background mutation rate, gene expression level, replication timing, gene size, and chromatin state—all of which affect the expected mutational rate within a gene (24).

Using MutSigCV analysis on whole-exome sequencing TCGA Pan-cancer Atlas data, we found that mTOR mutations were significantly enriched only in tumors reported in the Kidney RCCC TCGA cancer study (Fig. 1D). A similar analysis on whole-genome sequenced cancers also revealed significant enrichment in mTOR mutations exclusively in RCCC (Supplementary Fig. S1D; refs. 26, 27). In contrast, PIK3CA and PTEN mutations were significantly enriched in many cancer types (Fig. 1D; Supplementary Fig. S1D). To further validate our MutSigCV analysis, we examined the mutational significance of other common oncogenes and tumor suppressors, and found that these were significantly mutated within cancer types where they are known to act as drivers. As a control we analyzed titin, which is commonly mutated in cancers due to its size, which was not significantly enriched in any cancer type (Supplementary Fig. S1E).

We also examined clinically annotated mTOR driver mutations and found that RCCC and endometrial cancer had the highest proportion of clinically annotated putative driver mTOR mutations across MSK IMPACT and the TCGA studies (Fig. 1E; Supplementary Fig. S1F). The relatively high proportion of annotated mTOR mutations within endometrial cancer was surprising, as mTOR mutations were not found to be enriched in endometrial cancers in our MutSigCV analysis. Additional analysis of the endometrial cancer TCGA dataset showed that mTOR mutations co-occur with PIK3CA or PTEN mutations in 90% of tumor samples. This was in contrast with RCCC, where only 5% of mTOR mutations had co-occurring PIK3CA or PTEN mutations (Supplementary Tables S1 and S2 and Supplementary Fig. S1G). Further analysis and study will be necessary to investigate why mTOR activating mutations in endometrial cancer are enriched in tumors with upstream PIK3CA or PTEN mutations.

mTOR point mutations have been reported and validated as activating, with hotspot mutations mostly occurring in mTOR's FAT, FRB, or kinase domains (Fig. 2A; refs. 12, 13). The hotspot mutations reported to date in mTOR have been associated with increased tumor cell growth when assayed in vitro (12). Despite their potential to occur, it is surprising that activating mTOR mutations appear to be significantly enriched only in RCCC. Recently, it was found that WT mTOR is a negative regulator of exogenous protein utilization initiated by RTK/Ras/PI3K stimulation of macropinocytosis (6). Macropinocytosis has been shown to be an important pathway to support tumor cell growth in vivo in KRas-driven pancreatic adenocarcinoma, where essential nutrients like amino acids become limiting, and inhibition of mTOR is beneficial for tumor growth in these conditions (28, 29). To test whether activating mutations in mTOR can suppress growth dependent on exogenous protein utilization, we expressed several recurring gain-of-function mTOR mutants in KRas-transformed cells, known to be able to use exogenous proteins for growth when essential amino acids are limiting.

Three previously validated activating mTOR mutants were selected representing the most common naturally occurring variants within different functional domains of mTOR: C1483F in the F1 cluster of the FAT domain, T1977K in the F3 cluster of the FAT domain, and S2215F in the K1 cluster of the kinase domain (Fig. 2A; refs. 12, 17). As expected, both KRas-transformed cells expressing empty vector (control) and those expressing mTOR WT retained the ability to sense leucine deprivation and inactivate mTORC1, as shown by decreased phosphorylation of mTOR target S6K1 (pS6K1; Fig. 2B). Cells expressing mutant mTOR had higher levels of pS6K1 at baseline in complete media, and the cells maintained a higher level of pS6K1 under leucine starvation compared with mTOR WT or control cells. Consistent with previous studies, these mTOR mutants had a growth advantage over control KRas-transformed cells or cells expressing WT mTOR in complete medium (Fig. 2C). However, KRas-transformed cells expressing activating mTOR mutants were unable to use exogenously supplied BSA as an amino acid source for growth in the setting of leucine deprivation—whereas control and mTOR WT Ras-transformed cells were able to grow under these same conditions (Fig. 2D). Moreover, the mTOR mutants were detrimental in conditions of leucine deprivation without supplementation of exogenous BSA, leading to more cell death whereas Control and mTOR WT Ras-transformed cells remained viable (Fig. 2E).

We then wanted to determine whether the growth detriment seen in Ras-transformed cells with activating mTOR mutants was dependent on Ras activation, or whether it was generalizable to cells that do not robustly activate macropinocytosis and exogenous protein utilization under periods of amino acid starvation. Thus, we did similar experiments in Ras-WT MEFs expressing empty vector (control), WT mTOR, or mutant mTOR. As expected, the mTOR mutants showed a growth advantage when nutrients were replete (Fig. 3A). Unlike the KRasG12D cells, these Ras WT cells were unable to significantly grow in the setting of leucine starvation with BSA supplementation. However, the control and mTOR WT cells remained viable under conditions of leucine starvation (with minimal growth with supplemental BSA), whereas the mTOR mutants had increased cell death under leucine starvation with or without BSA (Fig. 3B and C). This survival-defect was dependent on mTOR activity—as mTOR inhibition with Torin rescued the death of leucine-starved mTOR-mutant cells (Fig. 3D). Similar to leucine starvation, decreased survival was shown with arginine deprivation, where the mTOR mutants uniformly had more cell death whereas the control and mTOR WT cells maintained viability during this same period (Fig. 3E). Glutamine starvation, which is not sensed by mTOR via the canonical RagGTPase signaling pathway, was equally deleterious to cells regardless of whether they expressed activating mTOR mutations (Supplementary Fig. S2A).

In addition to starving cells of individual amino acids known to be sensed by mTOR, we also tested whether generally depleting all amino acids would similarly be detrimental in growth or survival of these mTOR mutants. Thus, we grew the mTOR mutant cells in medium containing only 5% of the normal amino acid composition. Similar to experiments with leucine or arginine deprivation, the mTOR mutants showed a significant disadvantage in growth in 5% amino acid media as compared with the control or mTOR WT-transfected cells (Fig. 3F). These findings are in contrast with expression of active Akt (myr-Akt) or deletion of PTEN in cells growing in 5% AA. Neither of these alterations were deleterious to cell growth or survival when grown in 5% amino acid media, with or without the supplementation of BSA (Supplementary Fig. S2B and S2C). We then examined the effect of activating mTOR mutations grown in human plasma-like medium (HPLM), which is medium synthesized to mimic the nutrient conditions found in human plasma (30). Cells expressing activating mTOR mutations did not have any growth advantage or disadvantage relative to control cells or those expressing WT mTOR grown in HPLM over three days (Supplementary Fig. S2D).

As expected, the mTOR mutants had increased mTORC1 activity under complete conditions, as seen by increased levels of pS6K1. pS6K1 was retained under amino acid starvation only in cells expressing mTOR mutants, and was maintained at 24 hours (Fig. 3G; Supplementary Fig. S2E). The mTOR mutants also increased phosphorylation of 4EBP1 (p4EBP1), another mTORC1 target, in complete conditions. Although p4EBP1 levels remained higher in cells expressing activating mTOR mutants compared with WT mTOR or control cells under amino acid starvation, this change was small, and did not persist at 24hours of amino acid starvation (Fig. 3G; Supplementary Fig. S2E). We did not see significant increases in Ulk1 phosphorylation (pUlk1) in cells expressing activating mTOR mutants in either complete or amino acid deficient conditions. We also did not see increased mTORC2 activity, as seen by phosphorylation of Akt at serine 473 (pAkt S473). In fact, cells expressing either WT mTOR or the activating mTOR mutants had decreased pAkt S473 compared with control cells expressing an empty vector.

Glucose is critical to cellular growth and survival—and has been shown to be depleted within tumors (29, 31). mTOR senses glucose directly via its metabolite DHAP, as well as indirectly via AMPK (32). Cells with loss of TSC are “addicted” to glucose, and have more cell death upon glucose withdrawal (33). We hypothesized that glucose limitation, similar to amino acid limitation, would be detrimental to cells with hyperactive mTOR mutations. In agreement with this hypothesis, we found that cells expressing hyperactive mTOR mutants had significantly more cell death in conditions of glucose starvation as compared with control cells or cells expressing mTOR WT (Fig. 4A and B). We examined glucose utilization of these cells and found that the mTOR mutants utilized significantly more glucose and secreted more of this as lactate, although the glucose consumption/lactate secretion ratio remained unchanged (Fig. 4C and D). Given the increased utilization of glucose by the mTOR mutant cells, we examined cell growth in various glucose-limiting conditions by doing a glucose titration. We found a detrimental effect of mTOR mutants compared with Control or WT mTOR on growth and survival in the low glucose conditions of 1 or 0.1 mmol/L as well as with complete glucose starvation (Supplementary Fig. S3A–S3D).

In addition to sensing metabolites such as amino acids and glucose, mTORC1 senses cellular oxygen availability, both as a result of changes to energy charge through AMPK, as well as sensing oxygen directly via REDD1 (34). Previous work showed that TSC null cells had increased growth when grown under hypoxic conditions with 1% oxygen (34). In agreement with these studies, we found that cells expressing activating mTOR mutants had a growth advantage under hypoxic conditions with 1% O₂ (Supplementary Fig. S3e). However, this was not true when oxygen levels were lowered to 0.5% O₂, a level potentially seen in a tumor environment that not only stabilizes HIF, but also interferes with respiration (35). Cells expressing activating mTOR mutants grew slower in 0.5% O₂ compared with control cell or cells expressing WT mTOR (Fig. 4E). Cells expressing mTOR-activating mutants also had more cell death when cultured in low glucose at 0.5% O₂ (Fig. 4F) and low amino acids at 0.5% O₂ (Supplementary Fig. S3F).

Despite the survival disadvantage conferred by activating mTOR mutations under nutrient-limited conditions, these mTOR mutations are significantly enriched in RCCC. We hypothesize that the enrichment of mTOR mutants in RCCC may occur because these cancers are hypervascular and thus rarely deprived of nutrients. Supporting this hypothesis, we found that sporadic mTOR alterations significantly co-occur with VHL alterations when looking across all TCGA cancer studies (P < 0.001; Supplementary Table S3). However, an alternative possibility is that renal cell cancers are uniquely able to tolerate mTOR hyperactivation during periods of nutrient limitation due to some cell-type–specific behavior. To further evaluate this possibility, we expressed mTOR WT and hyperactive mutants in Caki-1 cells, an RCCC cell line (Fig. 5A; Supplementary Fig. S4A). Similar to experiments in KRas mutant or WT MEFs, these mTOR mutants retained mTORC1 signaling despite amino acid deprivation (Fig. 5A). Caki-1 cells expressing mutant or WT mTOR had similar growth properties (Fig. 5B), yet a clear detriment with increased cell death was seen in the mTOR mutant-expressing cells starved of arginine or leucine (Fig. 5C; Supplementary Fig. S4B). We also tested the effect of glucose limitation and hypoxia, and similar to our result in MEFs, the Caki-1 mTOR mutant cells had a growth disadvantage under low-glucose conditions (Fig. 5D), as well as in hypoxic conditions with 0.5% O₂ alone (Fig. 5E), or in combination with low amino acids or low-glucose conditions (Supplementary Fig. S4C and S4D).

In addition to examining the effect of hyperactivating mTOR mutations within a hypervascular cancer such as RCCC, we also examined the effects of these mutations within hypovascular cancers such as pancreatic adenocarcinoma. We expressed WT and hyperactive mTOR mutants in KPC cells, a murine pancreatic cancer cell line derived from Kras and TP53-mutated pancreatic cells (Supplementary Fig. S4E). Consistent with all other cell lines examined, we found increased cell death upon arginine starvation in KPC cells expressing mTOR mutants compared with those expressing WT mTOR (Fig. 5F and G).

mTORC1 activity is regulated by two regulatory arms: a growth factor sensing arm, and a nutrient sensing arm. Rheb, the final effector of the growth factor sensing arm, can only activate mTORC1 if it is appropriately anchored at the lysosome by the Rag GTPases. The functions of the Rag GTPases are regulated by the GATOR1 complex, a component of the nutrient-sensing arm regulating mTOR (Fig. 6A; ref. 2). The localization of mTORC1 at the lysosome is disrupted if specific amino acid and other nutrient conditions are not met. Thus, a lack of these signals leads to incomplete activation of mTORC1 even in the presence of growth factors (9). Given the detrimental effects of hyperactive mTOR in the setting of amino acid or glucose limitation, we hypothesized that the ability of mTOR to sense nutrient availability—and inactivate in the setting of nutrient limitation—might be critical for cellular survival and continued growth during periods of nutrient stress.

To test this hypothesis, we examined the prevalence of mutations within the nutrient-sensing arm of the mTORC1 pathway. Only rare alterations within each gene examined in this pathway were found in the TCGA pan-cancer dataset. Furthermore, when filtered for oncogenic alterations using OncoKB annotation and hotspot analysis, alterations within nutrient-sensing genes were exceedingly rare (Fig. 6B). To avoid bias arising from relying on putative driver annotation among these less-well studied genes, we performed a MutSigCV analysis of all GATOR1 and GATOR2 complex components across the TCGA pan-cancer studies. Of this set of 8 genes, none were significantly enriched in any cancer type (Fig. 6C).

Given the rarity of driver mutations within the nutrient-sensing arm of mTORC1, we hypothesized that alterations in this pathway that would activate mTORC1 might behave similarly to activating mTOR mutations. To test this, we made DEPDC5 knockout (KO) MEFs, as DEPDC5 is an essential component of GATOR1. As expected, the KOs retained S6K1 phosphorylation under leucine deprivation compared with control cells (Fig. 6D,i). The DEPDC5 KOs also grew better under nutrient replete conditions; however, they exhibited increased cell death under leucine starvation (Fig. 6D ii and iii). Thus, inactivating nutrient-sensing and thereby inappropriately retaining mTORC1 activity is detrimental in amino acid–deficient conditions.

In this study, we analyzed existing tumor databases to determine the frequency with which activating mutations in mTOR are observed relative to other components in the RTK/PI3K/Akt/mTOR signaling pathway. Surprisingly, we found that mTOR mutations are only significantly enriched in RCCC. This result was unexpected given mTOR's central role in promoting cell growth. Although the present findings lend support to previous work suggesting that mTOR driver mutations undergo positive selection in RCCC (25), the lack of a significant number of mTOR mutations across the rest of the cancer landscape was unexpected. However, the present findings are consistent with clinical data showing poor overall efficacy of mTOR inhibitors as cancer therapy, with the notable exception of RCCC (15). In contrast, mutations in upstream activators of the mTOR pathway, including RTKs and PI3K, were found to be significantly enriched in most cancer types. These results confirm the importance of the RTK/PI3K/Akt/mTOR pathway to oncogenesis. However, the data suggest that RTKs or PI3K are the central genes within this pathway that contribute to oncogenesis.

One potential reason for the apparent advantage of RTK or PI3K driver mutants over mTOR driver mutants is that PI3K activates multiple simultaneous pathways in addition to mTOR that contribute to oncogenesis, including Akt-mediated inhibition of FoxO and GSK3 (36). In addition, we and others have shown that PI3K activates alternative nutrient scavenging via macropinocytosis (16). When nutrients are abundant, PI3K-mediated activation of mTORC1 inhibits the catabolism of these scavenged proteins in the lysosome. However, in the setting of nutrient stress, mTORC1 is inhibited and catabolism of PI3K-mediated uptake of exogenous proteins or necrotic cells provides an additional nutrient-source.

Another potential reason for the rarity of activating mTOR mutations is that mTOR activation negatively feeds back to decrease signaling through PI3K (37). The impact of this negative regulation is highlighted by work investigating the different malignant potential of tumors arising in mice with deletion of PTEN versus TSC. Both of these alterations led to increased mTOR activity (38). However, PTEN-deleted tumors maintained high levels of PI3K-induced PIP3 as well as activated Akt and mTOR. In contrast, TSC null cells had high mTOR activity, but mTOR-mediated negative feedback ultimately led to decreased PI3K and Akt activity. Concomitant PTEN deletion with TSC deletion overcame the negative regulation of PI3K/Akt and allowed TSC-deleted tumors to become more malignant similar to PTEN-deleted tumors (38). We found that a similar situation may be occurring in endometrial cancer where PTEN and mTOR are mutated in the same tumor.

A growing body of evidence suggests that mTORC1 needs to be inactivated for cells to survive periods of nutrient limitation. Indeed, hyperactive mTORC1 signaling mediated by loss of TSC was shown to be a liability in tumor-like stress conditions, rendering cells dependent on desaturated lipids and glucose (33, 39). In addition, our laboratory has previously shown that mTORC1 inhibits degradation and utilization of exogenous proteins as a nutrient source, and that inhibition of mTORC1 is beneficial to cell growth and survival under nutrient-limiting conditions (6). Furthermore, mTORC1 inhibits autophagy, which is required to maintain nucleotide pools and prevent energy crisis during starvation in cancer cells (40). Thus in settings of nutrient-stress, hyperactivation of mTORC1 would be expected to be detrimental, as cells would be unable to inactivate mTOR to pause growth to balance ATP and nutrient supply with demand, nor activate autophagy and alternative nutrient scavenging strategies (28, 29, 33, 40). Supporting this model, our in vitro results showed that activating mTOR mutations, which maintain mTOR activity despite nutrient restriction, were detrimental to cells under amino acid, glucose, or oxygen-limiting conditions. We found that activating mTOR mutants had sustained increases in mTORC1 activity, but not in mTORC2 activity, specifically implicating aberrant mTORC1 activity as the driver of these nutrient vulnerabilities.

The differential effect of activating mTOR mutants in 1% and 0.5% O₂ is particularly instructional. At 1% O₂, cells begin adapting to lower oxygen conditions to limit ATP demand, despite the fact that oxygen is not yet limiting for respiration. In this circumstance, activating mTOR mutations can maintain ATP-demanding growth programs and give cells a growth advantage. However, these same mutations become deleterious as oxygen levels continue to fall and become limiting for respiration and other oxygen-dependent reactions critical for growth (35). Under these circumstances, inappropriately active mTOR becomes a liability as it maintains growth programs that the cell is unable to accommodate, leading to decreased growth or increased cell death.

Activating mTOR mutations provided a growth advantage to cells grown in traditional culture medium, which has excess nutrients to ensure nutrient sufficiency to sustain proliferation for a three-day culture period. However, when grown in HPLM, which mimics human plasma, the mTOR-activating mutations were neutral; likely because nutrients become limiting over a three-day culture period. This highlights the requirement of high or excess nutrient availability for activating mTOR mutations to provide a significant growth advantage, which would be lost as a tumor outstrips its blood supply. The neutrality of activating mTOR mutations in physiologic media may also help explain the limited occurrence of these mutations in human cancers.

RCCC is distinctively hypervascular both due to the physiology of the kidneys—which use fenestrated capillaries and allow the passage of nutrients through the endothelial barrier—as well as the VHL mutations that drive the majority of these cancers via angiogenesis (41, 42). As such, RCCCs are less likely to experience nutrient stress in comparison with other tumor types. This may partially explain why RCCC is an exception, allowing for enrichment of mTOR mutations. The fact that the deleterious effects of mTOR mutants in nutrient-poor conditions is cell-type agnostic further supports the microenvironment-centric view explaining the ability of RCCCs to tolerate activating mTOR mutations. This is also supported by the fact that alterations in mTOR significantly co-occur with alterations in VHL across all TCGA studies. However, there may be other cell-type and tissue-type–specific contributions that allow RCCCs to uniquely tolerate activating mTOR mutations.

The present study also demonstrates that a functional nutrient-sensing arm of mTOR is maintained in nearly all human tumors and is critical for growth and survival in the setting of nutrient stress. A knockout of a component of GATOR1, the nutrient-sensor and negative regulator of mTORC1, phenocopied hyperactivating mTOR mutants by showing a growth detriment under leucine-starved conditions. These results further demonstrate the importance of being able to inactivate mTOR when key nutrients are scarce. Together, the present findings provide potential insights into the vast differences in the observed mutation rates of pathway components and the potential of targeted inhibitors of the RTK/PI3K/Akt/mTOR pathway to disrupt tumor growth.

A.A. Bielska reports grants from American Cancer Society during the conduct of the study. C.B. Thompson reports grants from National Cancer Institute and other support from American Cancer Society, as well as grants from Canadian Institute of Health Research/Province of Ontario, and other support from ACM SIGHPC during the conduct of the study; as well as other support from Agios Pharmaceuticals outside the submitted work. No disclosures were reported by the other authors.

A.A. Bielska: Conceptualization, data curation, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. C.F. Harrigan: Data curation, formal analysis, investigation, methodology. Y.J. Kyung: Investigation. Q. Morris: Resources, supervision, funding acquisition, writing–review and editing. W. Palm: Conceptualization, investigation, writing–review and editing. C.B. Thompson: Conceptualization, resources, supervision, funding acquisition, writing–review and editing.

The results published here are in part based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga. The authors thank members of the C.B. Thompson's laboratory for the critical review of the data presented in this article. A.A. Bielska is the recipient of an American Cancer Society Postdoctoral Fellowship (PF-19-125-01-CSM). This work was performed with assistance from the Donald B. and Catherine C. Marron Cancer Metabolism Center, Memorial Sloan Kettering. C.F. Harrigan was supported by Canadian Institute for Health Research Project Grant #388344, the Province of Ontario and the ACM SIGHPC Computational and Data Science Fellowship. This work was supported by the Cancer Center Support Grant (P30CA008748) to Memorial Sloan Kettering Cancer Center and grants from the National Cancer Institute (to C.B. Thompson) and from Sloan Kettering Institute (to Q. Morris). Q. Morris is a Canada CIFAR AI chair.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.