Amino Acid Sensing by mTORC1: Intracellular Transporters Mark the Spot

1Central to the ability of a cell to adapt to its microenvironment is mechanistic (formerly mammalian) target of rapamycin (mTOR) complex 1 (mTORC1), a critical signaling hub, which is conserved from yeast to humans, regulating both growth and metabolism (Figure 1A). mTORC1 activity is regulated by a wide range of signals (reviewed in ), including growth factor signaling, cellular energy levels via AMP-dependent kinase (AMPK;, ), oxygen levels, and nutrients, particularly amino acids, as discussed below.

2Genes encoding the key mTORC1 kinase component, mTOR, were first identified in yeast, as TOR1 and TOR2, through molecular genetic studies, which revealed that they were important targets of the drug rapamycin. The term mTOR was initially used to refer specifically to TOR’s mammalian homologs, while nonmammalian TOR was referred to simply as TOR. However, a second definition, mechanistic TOR, has more recently also been employed in articles where mammalian, vertebrate, and invertebrate TOR are considered together, and for simplicity, we will use the latter term in this review.

3Following on from the landmark studies in yeast, mTOR and its regulation by amino acids were shown to be conserved in mammals. A combination of genetic analysis in flies and biochemical work using human cells led to a step change in thinking: growth factor signaling through PI3-kinase (PI3K) and Akt was shown to lie upstream of mTORC1 (reviewed in ), partly, but not exclusively, acting through the heterotrimeric tuberous sclerosis complex (TSC) and the monomeric GTPase, Ras homolog enriched in brain (Rheb).

4The best-characterized downstream function of mTORC1 is the control of mRNA translation. This is achieved via phosphorylation and suppression of the translation initiation inhibitor, eukaryotic initiation factor 4E-binding protein 1 (4E-BP1), in conjunction with activation of S6 kinase (S6K), which controls the transcription of a broad range of ribosome biogenesis genes. However, mTORC1 has numerous other biochemical targets with important metabolic and cellular functions (; Figure 1A).

5This complex metabolic regulatory network, in the classical paradigm, is represented by a model in which microenvironmental inputs funnel through a single mTORC1 hub to give a wide range of outputs (Figures 1 and 2; ). However, there is increasing awareness that many signaling cascades are controlled at a subcellular level, providing the flexibility for a more refined response within a single cell. Several studies of mTORC1 amino-acid-sensing mechanisms, discussed below, support the existence of a multi-hub model with at least two forms of mTORC1 (Figure 1B) controlled by amino acids in different parts of the cell.

6What amino acids regulate mTORC1 activity? Pioneering work in Chinese hamster ovary cells and in the Xenopus oocyte system initially suggested that mTORC1 primarily responds to intracellular levels of leucine. However, subsequent studies have highlighted sensitivities to other amino acids, such as arginine, glutamine, and serine.

7Amino acid transporters were initially implicated in mTORC1 regulation as passageways through the plasma membrane, enabling amino acids to enter cells and activate cytoplasmic amino acid sensors. Consistent with this, the heterodimeric amino acid transporter CD98, solute-linked carrier (SLC)3A2-SLC7A5 (CD98hc-LAT1), in combination with a glutamine transporter (SLC1A5), activates mTORC1 by exchanging leucine for glutamine to increase intracellular leucine levels. Other cell-surface amino acid transporters, e.g., LAT1 and LAT3 and SLC38A2 (SNAT2; ), have also been linked to mTORC1 signaling. The cationic amino acid transporter Slimfast regulates mTORC1 in the adipose-like fat body of the fly larva, controlling growth of the organism in an endocrine manner. These studies, however, leave unanswered the question of how amino acids are sensed once inside the cell.

8Genetic screening of a broad range of amino acid transporters in flies highlighted members of the proton-assisted amino acid transporter (PAT) or SLC36 family as having a particularly potent effect in promoting growth in vivo and activating mTORC1 in a cell-autonomous manner: for example, overexpressing fly PAT family members in the developing eye or wing increases organ growth (Figure 3A; ). The ability of PATs to promote growth increases significantly when growth rates and mTORC1 signaling in the eye are stimulated by activated PI3K (Figure 3B; ), a signaling defect frequently associated with human cancer. Consistent with this context-dependent effect on growth, the fly PAT, Pathetic (Path), is more critical for the growth of neurons with large dendrites than those with small dendrites during normal development. Subsequent analysis in HEK293 and MCF-7 breast cancer cells has shown that the two broadly expressed human PATs, PAT1 and PAT4, are required for amino-acid-dependent mTORC1 activation and cell proliferation. These two human PATs can also promote growth in transgenic flies in vivo.

9Before these studies, human PAT1 and PAT2 had already been shown to transport alanine, glycine, and proline by proton-coupled secondary active transport (Figure 4B;, ). By comparison, Path expressed in Xenopus oocytes (Figure 4B) is not proton assisted and has a much higher amino acid affinity and lower transport capacity, at least for alanine. This and the fact that several PATs are concentrated at the surface of late endosomes and lysosomes (LELs) in many cell types led to the proposal that they might behave as intracellular amino acid sensors that activate mTORC1 through direct signaling, acting as so-called “transceptors” (Figure 4A;, ). The cell-surface amino acid transporter SNAT2 (SLC38A2) may also be a transceptor, since it activates mTORC1 in the presence of the nonmetabolizable amino acid analog Me-AIB.

10A key breakthrough in understanding the subcellular control of mTORC1 by amino acids was the finding that an activated heterodimer of Rag GTPases, RagA or RagB together with RagC or RagD, positively regulates amino-acid-dependent mTORC1 activation. They do this by recruitment of mTOR to the surface of compartments positive for Rab7 and LAMP2 (lysosome-associated membrane protein 2), both markers for LELs. This regulatory role of the Rags is conserved in yeast, flies, and mice, where they are involved in growth control. However, despite these regulatory parallels, the yeast Rag homologs, Gtr1 and Gtr2, seem to play a different role in controlling the subcellular localization of TOR compared to their mammalian counterparts and are not essential for sustained TORC1 activation.

11An amino-acid-regulated mTORC1-containing protein supercomplex was identified on LELs, predominantly by biochemical analyses. It linked the Rags to the LEL membrane via the so-called Ragulator (LAMTOR) complex (Figure 2; reviewed by,, ). Proteins regulating or within this complex have been associated with cancer. For example, components of GATOR1 (Figure 2), which negatively regulates amino acid sensing, act as tumor suppressors, and RagC mutations have recently been linked to follicular lymphoma. Furthermore, amino acid starvation and inactivation of the Rag heterodimer have been linked to recruitment of the tumor suppressor TSC to LELs, while another group has reported that TSC localization is primarily regulated by growth factor signaling.

12But how are amino acids sensed at the LEL surface? Since the V-ATPase proton pump is part of the LEL-located mTORC1 supercomplex and its interactions are modified by amino acids, it was proposed as the sensor. The LEL-localized human PAT1, however, coimmunoprecipitates with RagC; this led to an alternative model in which the amino acid sensing is carried out by amino acid transporters based on their ability to bind amino acids.

13More recently, SLC38A9, a member of an amino acid transporter family with structural similarities to PATs, was highlighted as an amino-acid-sensitive regulator of the LEL-located mTORC1 supercomplex (Figure 2;,, ). Coimmunoprecipitation experiments with different components of this complex consistently pull down SLC38A9, but not PAT1 or other amino acid transporters. This strong interaction is dependent on SLC38A9’s cytosolic N-terminal tail, which has a high affinity for the Rag/Ragulator complex. As discussed later, the stability of this interaction may explain why other transporters were not pulled down under the conditions employed for these studies.

14SLC38A9 expressed in proteoliposomes (Figure 4B;, ) binds to several amino acids with different affinities. It potentially is involved in arginine sensing, though it has higher affinity for other amino acids, such as glutamine. Like some PATs and SLC38A2, SLC38A9 may act as a transceptor because of its close association with the mTORC1 supercomplex. The high degree of evolutionary conservation in other aspects of mTORC1 regulation, however, suggests that in the absence of a fly SLC38A9 homolog, this transporter cannot by itself resolve the amino-acid-sensing puzzle.

15Where do transporters like PAT1 and SLC38A9 sense amino acids? The primary focus has been on the LEL lumen, although it is also possible that sensing at the cytosolic side could be involved. Labeled extracellular amino acids rapidly enter the LELs, probably by a combination of endocytosis and uptake across the LEL membrane. For example, LAPTM4b (lysosomal-associated transmembrane protein 4b) recruits the heterodimeric transporter SLC7A5-SLC3A2 (LAT1-CD98hc) to the lysosome, where it can promote leucine uptake. The LEL luminal microenvironment, therefore, provides a complex readout of endocytosed extracellular and transported cytosolic amino acid levels. Indeed, the versatility of LELs as amino acid sensing hubs is well illustrated by the cellular response to starvation. Reduced mTORC1 signaling induces degradation of endocytosed extracellular proteins and autophagy, driving breakdown of cellular proteins via autolysosome formation: the free amino acids produced are critical in maintaining or restoring mTORC1 activity.

16While amino acid transporters potentially provide a direct link between LEL luminal amino acids and mTORC1 activation, cytosolic amino acids have also been implicated in the activation process (Figure 2) and may play an important and complementary role in amino acid sensing. Leucyl-tRNA synthetase (LRS) may act as a leucine sensor that switches on RagD. Cytosolic folliculin (FLCN) and interacting protein, FNIP, mediate activation of RagC and RagD by amino acids by acting as a GTPase-activating complex that converts the Rags into their functionally active GDP-bound forms. Furthermore, Sestrin 2, a member of the Sestrin family of proteins, which activate RagA and RagB either directly or indirectly via GATOR2 (reviewed in ), also senses leucine. Recently, a vertebrate-specific protein, CASTOR1 (or GATS-like protein 3), has been reported as a cytosolic arginine sensor, which blocks GATOR2 function in the absence of arginine.

17Aside from these molecules that directly interact with amino acids to control mTORC1 activity, several other proteins have been implicated in the amino acid sensing process and are briefly reviewed below. Some are of particular interest because they may involve subcellular compartments other than LELs and therefore are consistent with a multi-hub model for mTORC1 control.

18The MAP kinase regulator MAP4K3 was shown to be required for amino-acid-dependent mTORC1 signaling using HEK293T and HeLa cells. This function is conserved in Drosophila and appears to involve interactions with the Rag GTPases. The Class III PI3-kinase, Vps34, which catalyzes the synthesis of the lipid, phosphatidylinositol (3)-phosphate, an important regulator of endocytosis and autophagy, was identified as a potential mediator of mTORC1’s amino acid sensitivity. However, in this case, studies in flies did not suggest a role in normal growth in vivo.

19Two important points can be concluded from these findings. First, genetic manipulation of any molecule that interacts with and/or modifies components of the LEL-localized mTORC1 supercomplex could modulate the amino acid sensing mechanism. More elaborate experiments demonstrating direct amino acid interaction are required to identify genuine amino acid sensors. Second, when mTORC1 is regulated from the surface of intracellular compartments within the secretory and endolysosomal system, its activity will be heavily influenced by changes in membrane trafficking and endolysosomal maturation, themselves controlled by growth factor and mTORC1 signaling. Developing better cell systems to analyze this intricate regulatory interplay will be a critical objective in the coming years.

20Interestingly, the identification of two other molecules involved in membrane trafficking as regulators of an alternative mTORC1 complex has recently provided evidence for a multi-hub model of mTORC1 regulation. The first, Arf1, traditionally implicated in Golgi transport, is involved in glutamine-dependent mTORC1 activation via a Rag-independent lysosomal mechanism in mouse embryonic fibroblasts and HEK293A cells (. The second, Rab1A, involved in ER/Golgi trafficking, affects amino-acid-dependent mTORC1 activation on the Golgi, also in a Rag-independent fashion, in HEK293E and mouse NIH 3T3 cells. Rab1A is overexpressed in colorectal cancer and hepatocellular carcinoma, suggesting that like LEL-localized, Rag-dependent mTORC1, this alternative mTORC1 complex could be an important therapeutic target. This poses the question of how amino acids might be sensed from the Golgi apparatus, particularly since the characterized cytosolic sensors discussed above all act via the Rags.

21In addition to PAT1, the other widely expressed human PAT transporter required for mTORC1 activation, PAT4 (SLC36A4;, ), has recently been shown to be predominantly localized on the trans-Golgi network in several cell types. In HCT116 colorectal cancer cells, PAT4 regulates a form of mTORC1 that appears to be resistant to the drug rapamycin and seems to more strongly affect 4E-BP1 phosphorylation than S6K, in contrast to rapamycin-sensitive mTORC1. High PAT4 levels in colorectal cancer patients are associated with poor prognosis after surgery. PAT4 interacts with Rab1A, Raptor, and mTOR on the Golgi, as shown by proximity ligation assay (PLA), which detects closely apposed antigens in situ in whole-mount cells. These data support a model in which an mTORC1 signaling hub is assembled on the Golgi. The levels of PAT4 determine the resistance of mTORC1 to either glutamine or serine starvation, two amino acids that are rapidly metabolized in many cancer cells. It remains unclear whether there is any functional relationship between PAT4-regulated mTORC1 and the lysosomal form of Rag-independent, glutamine-sensitive, and Arf1-regulated mTORC1 mentioned above.

22Additional Rab1A biochemical and PLA data support the existence of a Rag-independent, Golgi-localized form of mTORC1. The LELs are, therefore, not the only platform for amino-acid-dependent mTORC1 signaling. The alternative mTORC1 signaling hubs potentially respond to different subsets of amino acids or different subcellular cues. For example, while LEL-located mTORC1 can detect extracellular-derived or autophagic amino acids in the LEL lumen, Golgi-localized mTORC1 could be controlled by amino acids trafficked back in a retrograde fashion from the endosomal system or brought into the Golgi lumen by Golgi-localized amino acid transporters. Significantly, the different amino acid and target preferences of specific mTORC1 hubs suggest distinct functions.

23Although more studies are required to determine how different mechanisms of amino acid sensing might work together to control the mTORC1 hubs, there is a common thread in the amino acid transporter story. PAT (SLC36) and SLC38 transporter families are structurally related. The region of greatest homology is in the 11 transmembrane domain core of the transporter, which in SLC38A9 has been shown to interact specifically with components of the mTORC1 supercomplex (Figure 2; ), albeit with lower affinity than the N-terminal SLC38A9 domain. One explanation for the various, and sometimes contradictory, findings from different groups studying transporter-mediated sensing mechanisms is that the experimental approaches employed focus on different aspects of the mTORC1 regulatory puzzle. Furthermore, depending on the cell type or culture conditions employed, certain sensing mechanisms may dominate over others. There are in fact unanswered questions relating to all the mechanisms proposed to date. Consideration of how each study approaches the problem could help piece together the puzzle in a more integrated way.

24Although for some years leucine was highlighted as the key amino acid in mTORC1 regulation, other important amino acids have since been implicated. In some cases, these amino acids promote cellular leucine uptake, but in others, they more directly influence mTORC1 activity. What led to the emergence of these multiple sensing mechanisms?

25One factor is that different cell types, particularly cancer cells, metabolize amino acids differently, so that even nonessential amino acids, like serine or glutamine, become limiting upon starvation. Some studies have by design, or inadvertently, tended to focus on a specific starvation regime or one mTORC1-regulatory amino acid transporter with distinct amino acid specificity, thereby probing only one sensing mechanism. Moreover, in many cancer cell lines, oncogenic pathways may selectively affect the activity of specific sensing pathways; this is illustrated by the effect of PI3K on PAT-induced growth in vivo (Figure 3B; ).

26Approaches for studying amino acid sensitivity also differ. Cells are often starved of amino acids and serum, and then specific amino acids are added back, or alternatively one amino acid is removed in the absence of serum. Other groups have used dialyzed serum to mirror physiological conditions more closely. These types of starvation experiments highlight requirements for particular amino acids but do not clarify whether mTORC1 is sensing different levels of specific amino acids or merely responding to amino acid presence or absence. This point can be addressed by varying sensor and amino acid concentrations. One important point is that although a few molecules involved in amino acid sensing have been tested in different models such as mice or flies (e.g.,,, ), most have not. Therefore, the physiological relevance of cell-culture starvation experiments remains uncertain when organisms employ multiple homeostatic mechanisms in vivo to maintain extracellular amino acid levels.

27For those amino acid transporters that act as sensors for mTORC1, an important question has been whether they interact with the amino acid(s) that they are believed to sense. Current assays often rely on measuring amino acid transport into Xenopus oocytes or proteoliposomes. In the transceptor signaling model for mTORC1 regulation (Figure 4A), however, even amino acids that compete with substrates, but are not transported, are potential regulators. The known substrates of PAT1, proline and glycine, which have been implicated in cancer growth and metabolic control, may therefore be sensed by this transporter (Figure 2), or other competing amino acids could be involved. In addition, neither the oocyte plasma membrane nor a proteoliposome (Figure 4B) mirrors the complex limiting membranes of LELs or Golgi, making it difficult to be certain that amino acid specificities observed are biologically significant. Amino acid affinities are typically calculated by labeling external amino acids. While the medium around an oocyte may be topologically equivalent to the LEL/Golgi lumen, the proposed site for sensing, it can represent the cytosolic side in proteoliposomes (Figure 4B;, ) and potentially provide a very different affinity profile.

28The methodologies employed to demonstrate that amino acid transporters are functioning within an active mTORC1-regulatory complex influence the resulting conclusions. Coimmunoprecipitation data defined components of the mTORC1 supercomplex on LELs (e.g.,, ) but relied on the membrane complex involved remaining stably associated during the pull-down procedure. The PLA approach highlights less stable interactions in situ but can detect molecules separated by up to 40 nm, potentially identifying nearby proteins that are not part of a single complex. mTOR recruitment to LELs has been used as a simple assay for amino-acid-dependent activation. However, it concentrates at the surface of LELs independently of amino acids in the absence of SLC38A9. This suggests that the factors affecting recruitment are more complex than current models indicate.

29Amino acid transporter overexpression has been used as a complementary approach to knockdown to demonstrate functional roles in mTORC1 activation. But the results need careful interpretation. For example, SLC38A9 overexpression makes mTORC1 signaling amino acid insensitive. On the other hand, overexpression of several PATs can promote growth and mTORC1 signaling in vivo and in culture or may, when expression levels are raised further, inhibit these processes, presumably via a dominant-negative mechanism (Figure 3A;, ).

30These issues highlight the importance of using a range of experimental approaches when dissecting out the mechanisms of transporter-mediated amino acid sensing, particularly when considering a multi-hub model for mTORC1 control.

31As discussed above, in vivo and cell-culture evidence strongly implicates multiple intracellular amino acid transporters in determining the amino acid sensitivity of mTORC1. Equally important, recent findings encourage thinking beyond a model where a single mTORC1 hub acts from the LELs to control metabolism and growth. Instead, mTORC1 should be considered as providing a sensory readout from more than one subcellular compartment with varying sensitivity to amino acids, potentially leading to different outputs (Figure 5).

32This view complements and extends findings suggesting that additional mTORC1 signaling inputs may also be sensing metabolically important changes in other subcellular compartments. For example, buildup of reactive oxygen species (ROS) in peroxisomes appears to negatively regulate mTORC1 via activation of a peroxisome-associated TSC complex, ultimately promoting autophagy (Figure 5; ). In the multi-hub model, the proposal that the amino acid microenvironment inside specific subcellular compartments could provide different readouts of a cell’s metabolic state, to which specific mTORC1 hubs respond, introduces a new layer of complexity to mTORC1 control and function. However, at present, except in the case of LELs, we have very limited understanding of how amino acids might accumulate inside different mTORC1-regulatory compartments and whether as yet unidentified compartmentalized amino acid transporters, like CD98 on lysosomes, might be involved. Moreover, even for LEL-localized mTORC1, the picture is complicated by the existence of transporter-independent sensing mechanisms for cytosolic amino acids.

33The concept of different mTORC1 complexes in a multi-hub model having different target specificities, and therefore possibly distinct cellular functions, flags up the importance of testing multiple downstream targets in studying each sensing mechanism. Given the roles of mTORC1 in normal metabolism, growth, aging, and also pathological processes, such as neurodegeneration and cancer, such analysis should dissect out more specialized functions for particular amino-acid-sensing transporters and possibly reveal selective effects of different amino acids, amino acid analogs, and mTORC1 inhibitory drugs. For example, rapamycin-resistant mTORC1 preferentially affects specific mTORC1-dependent substrates and therefore potentially specific pathologies. Ultimately, the development of better in vivo systems will be required to assess the physiological and disease relevance of the alternative mTORC1 hubs (e.g., ). The use of positron emission tomography (PET)-labeled amino acids, specific inhibitors, and/or inducible knockdown of selective regulators may be helpful in pursuing this goal. In this context, amino acid transporters now offer a selective tool to unpick the multi-hub mTORC1 model, providing new strategies for targeting of this central metabolic pathway.

34D.C.I.G., C.W., and A.L.H. wrote the manuscript.