Targeting mTOR for cancer therapy

The mechanistic target of rapamycin (mTOR) is a dual-specificity protein kinase phosphorylating serine/threonine as well as tyrosine residues [1]. Since the catalytic domain of mTOR resembles that of lipid kinases such as phosphoinositide 3-kinase (PI3K), mTOR is considered as an atypical protein kinase belonging to the PI3K-related kinase family [2]. As a core component of several distinct complexes including mTOR complex 1 (mTORC1), mTOR complex 2 (mTORC2), and a putative mTOR complex 3 (mTORC3), mTOR has critical roles in diverse biological processes, such as cell proliferation, survival, autophagy, metabolism, and immunity [2, 3]. While mTOR and mammalian lethal with SEC13 protein 8 (mLST8) are common members of both mTORC1 and mTORC2, regulatory-associated protein of mTOR (raptor), the 40 kDa proline-rich Akt substrate (PRAS40), and DEP domain-containing protein 6 (DEPTOR) are specific members of mTORC1 [1, 2]. Instead, rapamycin-insensitive companion of mTOR (rictor) and mammalian stress-activated protein kinase-interacting protein 1 (mSIN1 or MAPKAP1) are unique components in mTORC2 but not mTORC1 [1]. Another rapamycin-insensitive complex, mTORC3, consists of ETV7, mTOR, and other undefined components [3]. mTORC1 senses nutrients, growth factors, and cellular energy to orchestrate nucleotide, lipid, and protein synthesis; inhibit autophagy; and stimulate cell growth [2]. mTORC2 is not only regulated by growth factors, but also activates type I insulin-like growth factor receptor (IGF-IR) and insulin receptor (InsR) through the tyrosine kinase activity of mTOR [1]. Besides, mTORC2 regulates the actin polarization and endocytosis [4, 5].

The mTOR signaling pathway has critical roles in mammalian metabolism and physiology. The de-regulated activity of mTOR is involved in many pathophysiological conditions, such as aging, Alzheimer’s disease, diabetes, obesity, and cancer [2]. As a natural inhibitor of mTORC1, rapamycin is able to increase lifespan in mice [6, 7]. mTOR activity is frequently de-regulated in a variety of human cancers, such as breast, prostate, lung, liver, and renal carcinomas. Upregulation of mTOR signaling can promote tumor growth and progression through diverse mechanisms including the promotion of growth factor receptor signaling, angiogenesis, glyolytic metabolism, lipid metabolism, cancer cell migration, and suppression of autophagy [1, 2]. Hence, mTOR is a promising target for cancer therapy. In this review, we discuss the roles of mTOR in human cancer and the rationales and challenges for developing mTOR inhibitors to treat cancer.

The studies of mTORC1 structure demonstrate that mTORC1 adopts a dimeric architecture with an overall size of (280~300) × (200~210) × (100~130) ų [8, 9]. mTOR and LST8 form the core of mTOR complex that contains raptor and other regulatory proteins [8]. The human mTOR contains 2549 amino acids that form several domains including the NH₂-terminal HEAT (N-HEAT), middle HEAT (M-HEAT), FAT, and kinase domain with a FRB insertion (Fig. 1). Raptor also contains a HEAT domain, as well as WD40 and caspase-like domain [8, 9]. Besides, LST8 has WD40 domain. The HEAT motifs have conserved Asp and Arg residues at positions 19 and 25, respectively. A signature motif of WD40 repeats is ~ 40 amino acids often ending with a tryptophan-aspartic acid (W-D) dipeptide [10]. The HEAT repeats 12–13 in one mTOR interact with the HEAT repeats 20–23 in the M-HEAT domain of another mTOR, thereby forming a dimer [8]. Raptor may stabilize the dimer by binding the HEAT repeats 11–13 in one mTOR and repeats 20–22 in another mTOR [8, 11]. In addition, raptor is required for recruiting substrates to mTORC1 [12, 13]. Both mTOR and raptor are subjected to phosphorylation at multiple residues (Fig. 1a), which positively or negatively regulates mTORC1 activity.

The assembly of mTORC2 and Saccharomyces cerevisiae TORC2 follows a similar principle to mTORC1. The human mTORC2 structure reveals a hollow rhombohedral fold with overall dimensions of ~ 220 × 200 × 130 (ų) [14]. A dimer of mTOR is located in the core of this complex, while each mTOR or TOR heterodimerizes with rictor and mSIN1 [14, 15]. Rictor has an NH₂-terminal armadillo (ARM) repeat cluster (~ 900 residues), and the rest of the rictor is largely unstructured (Fig. 1b) [16]. Interestingly, ARM and HEAT domains have similar conserved residues that form the hydrophobic domain core and may have a common phylogenetic origin [17]. In addition, mSin1 has a CRIM, a Ras-binding domain (RBD), and a pleckstrin homology (PH) domain [18]. During the assembly of mTORC2, the FRB domain of mTOR binds to mSin1 and the carboxy terminal region of rictor, while the NH₂-terminal portion (residues 506–516) of rictor interacts with the COOH-terminal region (residues 1186-1218) of M-HEAT of mTOR [14]. In addition, mSin1 directly binds to rictor. Both rictor and mSin1 are responsible for recruiting substrates to mTORC2. Of note, both rictor and mSin1 have mTOR-independent partners. For example, rictor interacts with integrin-linked kinase and promotes its phosphorylation of Akt [19], while mSin1 interacts with Ras and inhibits ERK1/2 phosphorylation [20]. Thus, the outcome from the manipulation of rictor or mSin1 alone may not exactly reflect the function of mTORC2.

The activity of mTORC1 is regulated by growth factors, cellular energy, stresses and nucleotides, etc. The lysosomes are primary sites for mTORC1 activation. The activation of mTORC1 by growth factors is dependent on Ras homolog enriched in the brain (RHEB), a lysosomal GTPase that directly interacts with mTOR and activates it [21]. Upon binding to growth factors such as epidermal growth factor (EGF) and insulin-like growth factor (IGF), the growth factor receptors (EGFR, IGFR, etc.) are activated, which in turn activate PI3K-PDK1-Akt signaling pathway. Active Akt phosphorylates tuberous sclerosis complex 2 (TSC2) and inhibits the TSC complex, a GTPase-activating protein (GAP) complex consisting of TSC1/2 and TRE2-BUB2-CDC16 domain family member 7 (TBC1D7) [22, 23]. The TSC complex can inactivate RHEB thereby inhibiting mTOR [24]. Therefore, the activation of Akt leads to the depression of RHEB and then activates mTORC1. Moreover, the ubiquitination of RHEB regulates its ability to activate mTORC1 [21]. The E3 ubiquitin ligase RNF152 catalyzes RHEB ubiquitination, leading to an increase in the interaction between RHEB and TSC [21]. In contrast, Akt can phosphorylate the deubiquitinase USP4 that promotes RHEB deubiquitination thereby releasing RHEB from TSC [21].

Downstream of the growth factor receptors, the mitogen-activated protein kinase (MAPK) also upregulates mTORC1 activity. Mechanistically, MEK1/2 promotes raptor phosphorylation through ERK1/2 and p90 ribosomal S6 kinase (RSK1/2). ERK1/2 directly phosphorylates raptor at S8, S696, and S863, while RSK1/2 phosphorylates raptor at S719/722 [25, 26]. Meanwhile, the intestinal cell kinase (ICK), a MAPK-related kinase, phosphorylates raptor at T908 [27]. Phosphorylation of raptor by ERK/RSK/ICK promotes the activation of mTORC1.

mTORC1 not only senses growth factors, but also responds to cellular energy. Low cellular energy results in an increase in AMP/ATP ratio, which activates the energy sensor AMP-dependent kinase (AMPK). AMPK stimulates the GAP activity of TSC and then promotes the inhibition of RHEB by TSC, leading to the downregulation of mTORC1 [28]. In addition, the TCA cycle metabolite ketoglutarate inhibits mTORC1 through repressing ATP synthase, increasing AMP/ATP ratio and activating AMPK [29]. Cellular energy deficiency usually leads to endoplasmic reticulum stress, which in turn induces the unfolded protein response (UPR). Ire1, ATF6, and PERK are three major mediators of the UPR. Upon ER stress, ATF6 can induce RHEB expression, which in turn promotes mTORC1 activation and cell survival [30]. However, overactivated mTORC1 is also harmful to cell survival under ER stress. Mutations in TSC1/2 or activation of RHEB renders cells hypersensitive to ER stress-induced apoptosis, which may be due to the downregulation of ATF4/6 by mTOR [31]. Therefore, mTORC1 may have versatile effects on cell survival under ER stress.

While the regulation of mTORC1 by growth factors is dependent on RHEB and the TSC complex, amino acids can stimulate mTORC1 independent of TSC. The regulation of mTORC1 by amino acids is very complicated, involving multiple amino acid sensors and protein machinery [32]. The lysosomal Ragulator (RAG) guanosine triphosphatases (GTPases) play key roles in the activation of mTORC1 by amino acids. RAGA or RAGB heterodimerizes with RAGC or RAGD [33]. Further, RAG proteins form a large complex with LAMTOR1/2/3/4/5, which recruit RAG and mTORC1 to the lysosomal surface [34]. The activity of RAG is regulated by two complexes, GATOR1 and GATOR2. GATOR1, which is composed of DEPDC5, NPRL2, and NPRL3, inhibits the GTPase-activated protein (GAP) activity of RAGA/B thereby repressing the activation of mTORC1 by amino acids [35]. Instead, GATOR2, a protein complex consisting of MIOS, WDR24, WDR59 SEH1L, and SECB, negatively regulates GATOR1 by inducing DEPDC5 degradation [35]. Furthermore, KICSTOR, a large complex consisting of KPTN, ITFG2, C12ORF66, and seizure threshold 2 (SZT2), recruits GATOR1 to the lysosomal surface and mediates the interaction between GATOR1 and RAG [36, 37].

Sestrin (SESN) is another category of negative inhibitors of amino acid-induced mTORC1 activation. Mechanistically, SESNs interact with GATOR2, leading to the release of GATOR1 from GATOR2. The released GATOR1 in turn inhibits RAG and mTORC1 [38‐40]. Of note, SESN2 is known as a leucine sensor in mTORC1 signaling. Leucine directly binds to SESN2, leading to the dissociation of SESN2 from GATOR2. The released GATOR2 binds to GATOR1 and then prevents the inhibition of RAG by GATOR1. These sequential processes result in RAG-mediated mTORC1 activation [41]. To prevent the overactivation of mTORC1 by amino acids, there are negative feedback pathways to RAG-mediated mTORC1 activation. Two E3 ubiquitin ligases, RNF152 and SKP2, reportedly induce RAGA ubiquitination and potentiate the binding of RAGA to GATOR1 [42, 43]. While leucine sufficiency is sensed by SESN2, the stimulation of mTORC1 by arginine is mediated by SLC38A9 [44]. Moreover, the ubiquitin ligase TRAF6 can catalyze K63 ubiquitination of both Akt and mTOR thereby promoting the activation of Akt and mTORC1 by amino acids [45, 46].

In addition, mTOR may be activated by lipid and cholesterol. Fatty acid metabolism leads to the de novo synthesis of phosphatidic acid (PA), which stabilizes both mTORC1 and mTORC2 [47]. Moreover, cholesterol can stimulate mTORC1 activation and growth signaling. Mechanistically, SLC38A9 acts as a lysosomal cholesterol sensor to stimulate the activation of mTORC1 by RAG complex [48]. Recently, it was reported that mTORC1 is also responsive to the levels of purine nucleotides [49]. While adenylate stimulates mTORC1 by inhibiting TSC, guanylate downregulates RHEB and then inhibits mTORC1 [49]. The mechanisms underlying the regulation of TSC and RHEB by adenylate and guanylate remain to be known.

Although mTORC1 and mTORC2 are distinct complexes, there is a crosstalk between these two complexes. On one hand, mTORC2 can activate IGF-IR-Akt axis thereby upregulating mTORC1 [1]. On the other hand, mTORC1 feeds back to inhibit mTORC2 via S6K1, one of the substrates of mTORC1. Once activated by mTORC1, S6K1 phosphorylates rictor and mSin1 on T1135 and T86/398, respectively, leading to the impairment of mTORC2 integrity [50‐52].

While mTORC2 directly activates IGF-IR and InsR, receptor tyrosine kinases such as EGFR, PDGFR, and IGF-IR can activate mTORC2 via PI3K. Mechanistically, PI3K-induced PtdIns (3,4,5) P3 (PIP3) binds to the PH domain of mSin1 and then disables the inhibition of mTOR kinase domain by mSin1, thereby activating mTORC2 [18]. In addition, PI3K promotes the association of mTORC2 with ribosome, where mTORC2 is activated [53]. Therefore, mTORC2 also responds to growth factors. Notably, another study suggests that mTORC2 activity is localized in the plasma membrane, mitochondria, and endosomal vesicles, and the activity of mTORC2 via the mSin1-PH domain at the plasma membrane is PI3K- and growth factor-independent [54]. In addition, IKKα interacts with mTORC2 and enhances its kinase activity towards Akt [55]. These data suggest that the activation of mTORC2 involves multiple location and different mechanisms.

How does mTORC2 respond to cellular energy and nutrients? The energy sensor AMPK inhibits mTORC1 and then releases the suppression of mTORC2 by mTORC1, leading to the activation of mTORC2 [56]. Thus, upregulation of mTORC2 may help cells adapt to low levels of cellular energy. Moreover, mTORC2 is activated by glutamine starvation. Activated mTORC2 upregulates the expression and phosphorylation of glutamine:fructose-6-phosphate amidotransferase 1 (GFAT1), the rate-limiting enzyme of the hexosamine biosynthesis pathway (HBP) [57, 58]. A study of budding yeast demonstrates that the LKB1-ELM1-GIN4/HSL1 axis is required for coordinating TORC2 signaling to the changes in carbon source [59]. It remains to know if similar pathway works in human cancer cells.

Similar to mTORC1, mTORC2 is also stabilized by phosphatidic acid (PA), a central metabolite in the synthesis of membrane phospholipids [60]. The generation of PA is catalyzed by the phospholipase D, diacylglycerol kinases, and lysophosphatidic acid acyltransferases. Moreover, the activity of mTORC1 and mTORC2 is regulated by mLST8 ubiquitination. It has been reported that the E3 ubiquitin ligase TRAF2 positively regulates K63-linked polyubiquitination of mLST8, which impairs its interaction with mSin1 and compromises the mTORC2 integrity, but enhances the assembly of mTORC1 [61]. On the contrary, the deubiquitinase OTUDB7 removes polyubiquitin chains from G_βL to promote G_βL interaction with mSin1 and the integrity of mTORC2 [61]. Besides, the exchange factor found in platelets, leukemic, and neuronal tissues (XPLN) interacts with mTORC2 and negatively regulates mTORC2 activity [62]. Lastly, mTOR is a target of proteasomal degradation when it is ubiquitinated by FBXW7 [63].

As a protein kinase, mTOR catalyzes the phosphorylation of its targets and regulates their activity. mTORC1 and mTORC2 have different substrates. While the repertoire of mTOR substrates keeps increasing, there are more targets remaining to be identified. S6K1 and 4E-BP1 are two well-known mTORC1 targets. mTORC1 phosphorylates S6K1 at T389 and 4E-BP1 at multiple residues [64]. Phosphorylation of S6K1 by mTORC1 leads to increased protein and nucleotide synthesis. While 4E-BP1 is a negative regulator of 5′cap-dependent mRNA translation, phosphorylation of 4E-BP1 by mTORC1 induces its dissociation from eIF4E, thereby relieving its inhibition of protein synthesis [65]. To cope with increased protein synthesis, mTORC1 also promote ribosome biogenesis by inducing ribosomal RNA transcription. Mechanistically, mTORC1 may translocate to the nucleus, where it binds to ribosomal DNA promoter [66‐68]. Nuclear mTOR also phosphorylates TFIIIC and Maf1, thereby promoting tRNA gene transcription [69]. In fact, nuclear mTOR regulates RNA polymerase 1/2/3-driven transcription. In addition, mTORC1 phosphorylates the E3 ubiquitin ligase SKP2 at S64 and then inhibits SKP2 ubiquitination and degradation [70]. Given that SKP2 promotes the degradation of many proteins, mTORC1 may regulate the turnover of SKP2 substrates indirectly. Thus, mTORC1 not only promotes protein synthesis, but also regulates protein degradation.

Following the identification of mTORC2, it was found that protein kinase C (PKC) α/β were the substrates of mTORC2 that regulates the actin cytoskeleton [4, 71]. Moreover, mTORC2 phosphorylates and activates other AGC kinases, such as serum and glucocorticoid-induced kinase (SGK) and Akt. mTORC2 phosphorylates Akt at S473, leading to allosteric activation of Akt in cooperation with the catalytic activation by PDK1, which phosphorylates Akt at T308 [72]. During the synthesis of nascent proteins, mTORC2 can co-translationally phosphorylate some polypeptides while they are attached to the ribosome. IGF2 mRNA-binding protein (IMP) is responsible for the splicing and translation of IGF2 mRNA. mTORC2 co-translationally phosphorylates IMP1 at S181 and then promotes IMP1 binding to the untranslated region of IGF2 mRNA and enables translational initiation by internal ribosomal entry [73]. mTORC2 not only enhances the production of IGF2 protein, but also phosphorylates and activates IGF-IR and insulin receptor [1]. In contrast to mTORC1’s activity as a ser/thr kinase, mTORC2 has tyrosine kinase activity towards IGF-IR/InsR [1].

The activity of mTOR is frequently upregulated in human cancer. The aberrant activation of mTOR in human cancer may be attributed to mTOR pathway-activating mutations, amplification, or overexpression of the components of mTOR complexes and mutations or loss of negative regulators of mTOR. PIK3CA mutations are frequently detected in human cancer. Activation of PI3K promotes both mTORC1 and mTORC2 activation. In addition, mutations in KRAS and BRAF may lead to mTORC1 activation. Especially, KRAS can directly bind to PIK3CA (p110α) and activates PI3K pathway, leading to mTOR activation [74]. mTOR-activating mutations are observed in kidney cancer. While mTOR activity is usually upregulated by growth factors and amino acids, activating mutations in mTOR may result in RAG- and RHEB-independent mTOR hyperactivation, thus loss of the dependency on growth factors and amino acids [75]. Point mutations in RHEB and GATOR1 were also detected in renal cancer and endometrial cancer [76]. RHEB1 is overexpressed in acute myeloid leukemia (AML) and promotes AML progression [77]. Whereas mTOR amplification is rare in human cancer, rictor amplification is detected in various kinds of cancer, such as breast cancer, gastric cancer, and liver cancer [78, 79]. Moreover, rictor is overexpressed in human cancers of the brain, breast, lung, gastric, colon, liver, and tongue [80, 81].

Given that mTOR has critical roles in tumor progression, mTOR inhibitors hold promise in cancer therapy. Indeed, rapamycin analogs (rapalog) have been approved for treating cancer in the clinic. In addition, many mTOR inhibitors with different mechanisms of action have been developed, some of which are undergoing clinical trials in variety types of human cancer.

Drug resistance is a serious problem in treating cancer. Although there may be an initial response, long-lasting treatment with chemotherapeutic or molecular-targeted drugs often faces the challenge of drug resistance. Due to the tumor heterogeneity, some tumors do not respond to a given drug at all. Clonal selection, adaptive evolution, and resistance to cell death are general mechanisms for drug resistance. Due to the complexity and crosstalk in signaling networks, cancer cells may adapt to an inhibitor that targets a given signaling pathway via the compensatory activation of other pathways. Although mTOR inhibitors exhibit potent anti-cancer effects in many preclinical models, resistance does occur. As described below, there are multiple mechanisms underlying the resistance to mTOR inhibitors (Fig. 2).

Given that preclinical studies demonstrate the anti-cancer efficacy of mTOR inhibitors alone or in combination with chemotherapy, radiotherapy, and targeted therapy, there are many completed or ongoing clinical trials to test the efficacy of mTOR inhibitors for treating various types of human cancer (Table 1). In general, most of mTOR inhibitors are well tolerated, while there are some common adverse effects including fatigue, rash, mucositis, and metabolic complications. mTOR inhibitors are associated with a significantly increased risk of hyperglycemia, hypertriglyceridemia, and hypercholesterolemia [187]. Other adverse events of everolimus are thrombocytopenia, anemia, nausea, and stomatitis [188]. Ridaforolimus is orally bioavailable and better tolerated in children than the adults [189]. Deforolimus was well tolerated and showed encouraging anti-tumor activity across a broad range of malignancies when administered intravenously, and a dose of 12.5 mg/day is being evaluated in phase II trials [190].

Moreover, MLN0028-treated patients may suffer from anorexia, dyspenea and macunopapular rash [191]. In clinical trials of solid tumors, the PI3K/mTOR inhibitor NVP-BEZ235 (twice daily) is poorly tolerated, which leads to treatment discontinuation in some patients and limits its efficacy in treating cancer [192, 193]. Apitolisib (GDC-0980), another dual pan-PI3K/mTOR inhibitor, also has grade 3–4 adverse effects and is less effective than everolimus [194]. GSK2126458 (GSK458) plus trametinib has poor tolerability, due to skin and gastrointestinal toxicities such as diarrhea [195]. Daily oral administration of PF-04691502 (8 mg/day) has adverse events including fatigue, nausea, vomiting, hyperglycemia, and rash [196]. The occurrence of the above-mentioned adverse effects following treatment with mTOR inhibitors may be due to the critical roles of mTOR in metabolism and immunity.

The discovery of TOR in yeast and mTOR in mammals is a fundamental breakthrough in understanding cell and organism growth, metabolism, and diseases. In-depth studies to clarify the regulators and effectors of mTOR signaling have revealed multiple networks that work together to integrate growth factors, nutrients, sterols, and nucleotides signaling. The identification of the critical roles of mTOR and its regulators in tumorigenesis has driven the development of the ever-growing list of mTOR inhibitors. While some of the mTOR inhibitors have been approved to treat cancer patients, more mTOR inhibitors are under check to fulfill their promise for cancer therapy.

It appears that mTOR inhibitors have mixed efficacy in patients with distinct kinds of cancer and among patients with the same kind of cancer. Recent studies reveal that tumor organoids may help drug testing [225, 226]. Tumor organoids may be used to test the response of a given tumor to mTOR inhibitors. Alternatively, patient-derived tumor grafts may be transplanted to animals, followed by testing their response to mTOR inhibitors [227]. It would be of interest to determine if these emerging technologies are clinically relevant.

In the era of precise medicine, it needs to determine if there are predictive biomarkers that may guide the stratification of patients in clinical trials or help identify the patients who most likely benefit from treatment with mTOR inhibitors in a clinical setting. Gene testing is a promising approach to achieve this goal. The candidates for gene testing may include mTOR, PIK3CA, GATOR, KRAS, and BRAF. Mutations in PIK3CA and GATOR have been associated with higher sensitivity to mTOR inhibition in preclinical studies. Hence, PIK3CA mutations may be potential sensitive markers. In contrast, KRAS/BRAF mutations may be resistant biomarkers. Both DNA from tumor samples and ctDNA from the blood may be subject to testing of gene mutations. In addition, gene mutations in the tumors may be dynamic during cancer evolution or regression [228]. It remains to determine if dynamic testing of ctDNA during the course of therapy may monitor cancer evolution and better predict drug resistance, thereby adjusting the treatment regimen in time. Recent progress in liquid biopsy may help address this critical issue [229, 230]. In addition to gene testing, the solvable factors in the blood may be potential biomarkers as well. Of particular note, the mechanisms underlying the varied responsiveness to mTOR inhibitors in cancer patients may be complex. Rather than a single or few biomarkers, a set of biomarkers may be more powerful and accurate to meet the challenge.

Moreover, toxicity is a critical problem that precludes the clinical administration of drugs. Although mTOR inhibitors exhibit a promising efficacy in preclinical studies, some inhibitors have serious adverse effects in patients and have to be discontinued. Hence, elucidation of the mechanisms underlying these adverse effects may help manage them in the clinic.

Drug resistance is a serious challenge to successful cancer therapy. As discussed above, the mechanisms for mTOR inhibitor resistance are complex. Further studies to elucidate the diverse mechanisms may help design strategies to overcome the resistance to mTOR inhibition. Mechanism-based combination of mTOR inhibitors with chemotherapeutic agents or molecular-targeted drugs may be practical in the clinic. We expect the results from many ongoing clinical trials to validate the most powerful regimens that include mTOR inhibitors.