Research Article

Architecture of human mTOR complex 1

See allHide authors and affiliations

Science  01 Jan 2016:
Vol. 351, Issue 6268, pp. 48-52
DOI: 10.1126/science.aaa3870

From sensing leucine to metabolic control

The mTORC1 protein kinase complex plays central roles in regulating cell growth and metabolism and is implicated in common human diseases such as diabetes and cancer. The level of the amino acid leucine tells an organism a lot about its physiological state, including how much food is available, how much insulin is going to be needed, and whether new muscle mass can be made (see the Perspective by Buel and Blenis). Wolfson et al. identified a biochemical sensor of leucine, Sestrin2, which connects the concentration of leucine to the control of organismal metabolism and growth. When leucine bound to Sestrin2, it was released from a complex with the mTORC1 regulatory factor GATOR2, activating the mTORC1 complex. Saxton et al. describe the crystal structure of Sestrin2 and show how it specifically detects leucine. Aylett et al. determined the structure of human mTORC1 by cryoelectron microscopy and the crystal structure of a regulatory subunit, Raptor. The results reveal the structural basis for the function and intricate regulation of this important enzyme, which is also a strategic drug target.

Science, this issue p. 43, p. 48, p. 53; see also p. 25


Target of rapamycin (TOR), a conserved protein kinase and central controller of cell growth, functions in two structurally and functionally distinct complexes: TORC1 and TORC2. Dysregulation of mammalian TOR (mTOR) signaling is implicated in pathologies that include diabetes, cancer, and neurodegeneration. We resolved the architecture of human mTORC1 (mTOR with subunits Raptor and mLST8) bound to FK506 binding protein (FKBP)–rapamycin, by combining cryo–electron microscopy at 5.9 angstrom resolution with crystallographic studies of Chaetomium thermophilum Raptor at 4.3 angstrom resolution. The structure explains how FKBP-rapamycin and architectural elements of mTORC1 limit access to the recessed active site. Consistent with a role in substrate recognition and delivery, the conserved amino-terminal domain of Raptor is juxtaposed to the kinase active site.

Since its discovery in 1991 as the target of the immunosuppressant rapamycin, TOR has emerged as a central regulator of cell growth and metabolism. TOR was identified in yeast (1, 2); the mammalian ortholog is mTOR (3, 4). The serine and threonine kinase activity of TOR is tightly regulated in response to physiological conditions, and aberrant mTOR signaling occurs in multiple pathologies, including diabetes, cancer, and neurodegeneration (5, 6).

TOR is the core component of two, functionally distinct signaling complexes, TOR complex 1 (TORC1) and TORC2 (713). TORC1 is sensitive to rapamycin and regulates cell growth by activating protein, lipid, and nucleotide synthesis and by inhibiting autophagy. TORC2 is less well characterized but is rapamycin insensitive and controls diverse cellular processes through phosphorylation of several targets. Mammalian TORC1 (mTORC1) contains, in addition to mTOR, Raptor (regulatory-associated protein of mTOR) (8, 9), mammalian homolog of protein Lethal with Sec Thirteen (mLST8) (7, 14), and possibly several noncore subunits. Whereas mTOR and mLST8 are also found in mTORC2, Raptor is absent. Instead, Rictor (rapamycin-insensitive companion of mTOR) (7, 10, 11) and additional subunits (1517) are required for mTORC2 activity.

The regulation of mTORC1 signaling involves interactions with several binding partners and translocation of mTORC1 within the cell (12, 13). mTORC1 is activated in response to nutrients (amino acids) through the Rag guanosine triphosphatase (GTPase) signaling pathway (18, 19), by growth factors through inhibition of the tuberous sclerosis complex 1 and 2 (TSC1 and TSC2) heterodimer and thereby activation of the small GTPase Rheb (20, 21), and by cellular energy status through adenosine monophosphate–activated protein kinase (AMPK) phosphorylation of TSC2 and Raptor (22, 23).

TOR is the founding member of the phosphatidylinositol-kinase–related kinase (PIKK) family (24), members of which share an elaborate domain organization. The kinase domain is situated at the C terminus, following long arrays of first HEAT (Huntingtin, EF3A, ATM, TOR), and then tetratricopeptide (TPR), repeats (25, 26). The crystal structure of a compact C-terminal fragment of human mTOR containing the FAT (Frap, ATM, TRRAP) domain and the kinase domain has been resolved in complex with its obligate accessory protein, mLST8 (27).

Inhibition of mTOR by rapamycin is dependent on the formation of a complex with an intracellular receptor, FK506-binding protein (FKBP) (28, 29), which then binds mTOR in mTORC1. FKBP interacts with mTOR through the hydrophobic rapamycin molecule, which binds pockets in each protein (1, 30, 31). The N terminus of the kinase domain forms the FKBP-rapamycin complex binding (FRB) domain, which juts outward from the N-lobe (27). Binding of the FKBP-rapamycin complex to the FRB domain has been predicted to narrow the active-site cleft (27), suggesting that rapamycin inhibition is due to steric hindrance (32). The FRB domain also binds cognate mTOR targets, and the mutation of key residues in the FRB domain impairs phosphorylation of model substrates (27).

The catalytic activity and substrate specificity of mTORC1 are regulated by Raptor. Raptor is a multidomain protein, predicted by sequence similarity to consist of an RNC (Raptor N-terminal Conserved) domain, a central set of armadillo repeats, and a C-terminal β propeller (8, 9). Raptor functions in substrate binding, being required for phosphorylation of many mTORC1 substrates, the best characterized of which are the translational regulators ribosomal protein S6 kinase (S6K1) and eukaryotic translation initiation factor 4E-binding protein (4EBP). Phosphorylation of these substrates is dependent on their TOR signaling (TOS) motif, which binds Raptor directly (33, 34).

Biochemical investigation of the oligomeric state and composition of TORCs demonstrated that TOR and mTOR complexes are dimeric, and allowed assignment of core components. Owing to difficulties in working with intact mTORC1, however, information on the three-dimensional arrangement of proteins and domains within the complex was limited to crystal structures of fragments (27, 31), and a low-resolution (26 Å) reconstruction through cryo–electron microscopy (cryo-EM) (35) revealing twofold-symmetric, dimeric particles roughly 300 Å by 200 Å by 100 Å in size. In that study, however, the handedness of the reconstruction and the position of individual subunits could not be reliably assigned.

We resolved the structures of human mTORC1 and Chaetomium thermophilum Raptor (CtRaptor) at 5.9 and 4.3 Å resolution by cryo-EM and x-ray crystallography, respectively. We provide a description of the architecture of mTORC1 at a secondary structural level, placing the folded domains of all three mTORC1 subunits and revealing their relative arrangement within the complex.

Results and discussion

Determination of the cryo-EM structure of mTORC1

We coexpressed and copurified human mTOR, Raptor, and mLST8 from insect cells, obtaining a homogeneous mTORC1 complex with a molecular size of ~1.0 MD (fig. S1, A and B). Purified material phosphorylated the mTORC1 substrate 4EBP1 and was inhibited by torin1 or FKBP-rapamycin, as expected for native mTORC1 (fig. S1C). For structural studies, a stable mTORC1 sample was generated by expression in the presence of rapamycin followed by mild glutaraldehyde gradient fixation (36). Single-particle analysis of cryo–electron micrographs from this mTORC1 sample yielded a reconstruction with an overall resolution of 5.9 Å (gold-standard Fourier shell correlation = 0.143 cut-off), within which only a few peripheral regions were ordered to lower resolution (figs. S2 and S3). Secondary structural elements were clearly resolved throughout most remaining regions of the complex, allowing all folded domains predicted within mTORC1 to be assigned. The reconstruction presented here is complemented by the crystal structure of C. thermophilum Raptor that fits the EM density of human mTORC1 and confirms the segregation of the Raptor and mTOR repeat regions. Both the crystal structure of the C-terminal fragment of mTOR (27) and the Raptor structure fit our density, allowing us to determine the handedness of the reconstruction (Fig. 1 and fig. S4). The overall shape of our reconstruction agrees with that of the published reconstruction; however, it appears that the handedness of the previous reconstruction was assigned incorrectly (35).

Fig. 1 mTORC1 adopts a dimeric, lozenge-shaped architecture.

The reconstructed density obtained without masking peripheral regions, filtered to its global resolution (6.1 Å) and contoured at 3 σ, is shown as a translucent surface, and the corresponding model in cartoon representation. Each successive panel (A to D) is rotated as indicated. Raptor is in shades of green, mTOR in purples (N-terminal HEAT repeats) and blues (C-terminal FAT and kinase domains), mLST8 in orange, and FKBP in red. The active site, at the base of a restrictive cleft, is indicated by a magenta asterisk. A tinted surface representation showing the surface corresponding to one of each of the subunits within the proposed model of mTORC1 is inset.

mTORC1 forms a hollow dimer with minimal subunit contacts

mTORC1 adopts a cage-like, dimeric architecture in which the C2 symmetry axis passes through a large cavity (>60 Å across in places), leading to its characteristic hollow lozenge shape (Fig. 1, A and B). The kinase domains of mTOR are located near the center of the assembly and come close to each other but do not make contact. Raptor and mLST8 contribute peripheral parts of the complex, making up the pinnacles of the longer and shorter axes of the lozenge, respectively. When viewed from an angle perpendicular to the symmetry axis, one face of mTORC1 is characterized by the mTOR kinase domains and mLST8 subunits, with both active-site clefts opening outward from this side of the complex (Fig. 1, A and C). The opposite face comprises the N-terminal HEAT repeat domains of mTOR, which form superhelical α solenoids (Fig. 1, B and D). Raptor binds within the well-ordered juncture of two HEAT repeat domains. The RNC domain abuts the FRB region of the mTOR kinase domain, and the more C-terminal armadillo repeat and β-propeller domains extend outward, becoming less well resolved (Fig. 1, B and C).

The mTOR HEAT repeats contain two α-helical solenoids

The N-terminal portion of mTOR has a bipartite structure consisting largely of α-solenoid helical repeats. The extended, N-terminal region merges directly into the more compact, C-terminal fragment, which wraps around the kinase domain. The repetitive N terminus of mTOR, a prototypical HEAT repeat (25, 26), forms two α solenoids (Fig. 2). The larger section is a highly curved superhelix, which we have dubbed the “horn,” seven repeats of which can be placed at the individual-helix level. The remaining repeats are less clearly resolved, indicating local flexibility. The smaller region adopts a relatively linear arrangement consisting of seven HEAT repeats and a helical linkage to the compact fragment, which we refer to as the “bridge” (Fig. 2). The horn and bridge HEAT domains pack against one another. Nevertheless both sections are predominantly exposed to the environment, supporting a purported role in binding mTOR regulators. The bridge region merges directly into the TPR repeats of the FAT domain (Fig. 2), and the prior crystal structure of this region fits our density with only minor adjustments (27) (Fig. 1 and fig. S4).

Fig. 2 Model showing the suggested domain organization of the mTOR dimer.

(A and B) The C-terminal kinase and FAT domains of mTOR form a compact unit, and the N-terminal HEAT domains adopt an elongated structure with two separate α-solenoids: the “horn” and “bridge.” The proposed model is shown in cartoon representation, rotated as indicated. Symmetry-related mTOR domains are shown in gray. (C) Linear representation of the domain organization of mTOR. The residue numbers indicate the domain boundaries; domain lengths are consistent in residues.

While no linkage between the two HEAT domains is visualized, their termini are in close proximity to one another. Although we cannot definitively assign sequence to regions of density at this resolution, the available evidence from studies of homologs (37) supports an interpretation in which the domains are continuous and contiguous, with the horn representing the N-terminal section and the bridge the more C-terminal part (see Materials and Methods). This suggested topology is shown in Fig. 2.

mTOR HEAT repeats bridge the kinase domains

Although Raptor is proposed to mediate mTOR dimerization by binding between the mTOR subunits (35), our structure reveals that mTOR itself forms a complete dimer (Fig. 2). The horn and bridge HEAT domains pack against one another, and the first HEAT repeat of the mTOR horn region is buried against the base of the adjacent mTOR FAT domain, completing an interlocking interaction between the two subunits (Fig. 3A). Dimerization changes the conformation of the FAT domain of mTOR relative to that in the monomeric crystal structure (fig. S4) (27). TPR helices 4 to 12 rotate to accommodate the interaction, presenting the interface and stacking neatly across the first repeat of the horn (Fig. 3B). Notably, the conformation of the kinase domain appears unaffected by dimerization. This is not unexpected, as the kinase domain can maintain an active state even in the absence of Rheb (27). Thus regulation is probably mediated largely by controlling substrate access to the active site.

Fig. 3 The horn and bridge of mTOR complete a full dimeric interaction, linking the two FAT-PIKK units.

(A) The complete interaction is shown intact, and the reconstruction, filtered to its global resolution (5.9 Å) and contoured at 6 σ, is indicated as a translucent surface; the corresponding model is shown as a secondary structural cartoon. (B) The interaction surface is shown with the point of view rotated as indicated. The symmetry-related horn domain is denoted horn′.

Structure of Raptor

Raptor determines substrate specificity of mTORC1, by mediating recruitment (18, 19) and recognition (8, 9). Density corresponding to the RNC domain and the proximal part of the armadillo repeat of human Raptor is well defined in our reconstruction, whereas the distal portion of the armadillo repeat and the C-terminal β-propeller domain are flexible (Fig. 1). To resolve the entire Raptor structure, we purified insect cell-expressed Raptor from C. thermophilum (CtRaptor), which exhibits 44% sequence identity to human Raptor (fig. S5). To facilitate crystallization, CtRaptor was subjected to limited proteolysis, which removed a large N-terminal extension absent in human Raptor (fig. S6). Several smaller proteolytic fragments remained associated with the Raptor core (fig. S6). Experimental phases for CtRaptor crystals were determined with a mercury derivative, and a backbone model was traced at a resolution of 4.3 Å (fig. S7). CtRaptor adopts an extended Z shape with the RNC and β-propeller domains arranged at opposite ends of the armadillo domain (Fig. 4).

Fig. 4 CtRaptor adopts a Z shape with the RNC domain and the β propeller at opposing ends of the armadillo domain.

(A and B) CtRaptor is shown in cartoon representation, and domain borders are indicated in a sequence schematic. The putative substrate-binding site, inferred from homology to CASPases, is indicated by a blue sphere. Polypeptide linkers are shown in gray; stretch 1 (dark gray) originates from the N terminus of the RNC domain and spans the armadillo (ARM) domain to the β propeller; stretch 2 (medium gray) is associated with the C-terminal region of the armadillo domain; stretch 3 (light gray) is bound along the concave surface of the armadillo domain. (C) Linear representation of the domain organization of Raptor. The residue numbers indicate the domain boundaries; domain lengths are consistent in residues.

Fig. 5 Raptor binds to and organizes the N terminus of mTOR through a horn-bridge-armadillo α-solenoid stack.

(A) The complete interaction is shown with the reconstructed density, filtered to its global resolution (5.9 Å) and contoured at 6 σ, indicated as a translucent surface; the corresponding model is shown as a secondary structural cartoon. (B) The antiparallel α-solenoid stack is shown as a secondary structural cartoon; the image generated with the point of view rotated as indicated and the threefold interaction surface are shown with dotted magenta lines.

The N-terminal RNC domain exhibits an α-β-α sandwich fold that is structurally related to that found in CASPases (fig. S8A), consistent with conserved motifs in both families (38). The RNC domain is connected to the armadillo domain through four helices, three of which originate from a large insertion (relative to the CASPase fold) between strand one and a shortened helix four, and the fourth from the C terminus of the RNC domain. Two of the active-site residues of CASPases are conserved in the RNC domain (38). Our model of CtRaptor reveals that there is also structural conservation of the corresponding regions. However, the N terminus of helix 1, which contributes a key arginine residue to the CASPase active site, is displaced by more than 13 Å in CtRaptor.

The Raptor armadillo domain comprises seven helical hairpins, five of which are canonical armadillo repeats (fig. S8B). The β-propeller domain of Raptor associates with the armadillo repeats through the interaction of blades two and three with repeats five, six, and seven. The interaction takes place at the rim of the β propeller, leaving the face accessible for putative interacting proteins (Fig. 4 and fig. S8B). Three additional stretches of extended CtRaptor peptide are visible in the electron density (Fig. 4 and fig. S7). Stretch 1, belonging to the N terminus, originates at the β propeller, spans the entire armadillo domain, and joins the N terminus of the RNC domain. Stretch 2 forms a V-shaped helical segment at the C terminus of the repeat domain beneath the β propeller. Stretch 3 runs along the concave face of the armadillo domain. Stretches 2 and 3 likely represent the linker region between the armadillo and β-propeller domains, corresponding to the fragments observed by mass spectrometry (fig. S6). The corresponding linker region of human Raptor (residues 700 to 900) is frequently phosphorylated (39); based on our structural data, such posttranslational modification could affect interactions with the Raptor core in vivo.

The conformation of CtRaptor in crystals is close to that of human Raptor in the mTORC1 EM reconstruction, allowing us to fit the crystal structure as a rigid body into the EM density. The projection of the β-propeller domain of Raptor away from the core of mTORC1 is consistent with a role for this domain in the recruitment of regulatory proteins (Figs. 1 and 4) (18, 19). The Raptor RNC domain is positioned directly at the mTOR active-site cleft, implying its involvement in Raptor-mediated substrate recognition (Figs. 1 and 4) (33, 34).

Raptor stabilizes the mTOR HEAT domains

Raptor is necessary for formation of the mTORC1 complex. The principal interaction between Raptor and mTOR consists of an α-solenoid stack formed between the horn and bridge domains of mTOR and the Raptor armadillo domain (Fig. 5A). The horn and bridge domains run antiparallel to each other, forming a contact offset by half their depth (Fig. 5B). The N-terminal helices of the Raptor armadillo repeat and the base of the RNC domain occupy this “step,” making up the remaining interaction surface with the loops of the HEAT repeats within the bridge and helices of those in the horn (Fig. 5B). Given that Raptor provides roughly two-thirds of the interaction surface stabilizing the HEAT domains, the stability of the N-terminal regions of mTOR would be weakened in its absence. The formation of the mTORC1 dimer is also dependent on the interaction between the (flexible) horn N-terminal HEAT domain of mTOR and the C-terminal FAT domain. Without stabilization of the mTOR N-terminal HEAT repeats in a single conformation, dimers could not be readily formed. Raptor binding thus may favor the dimerization of mTOR molecules without directly engaging in dimer formation (7, 35).

Implications of mTORC1 architecture for substrate selectivity and delivery

The structure of the C-terminal mTOR fragment revealed that the kinase domain is held in a catalytically active conformation by the surrounding FAT domain and segments inserted within the kinase domain itself (27). The FRB domain and mLST8 limit access to the adenosine 5′-triphosphate (ATP)–binding cleft, preventing activity toward noncognate substrates, which would otherwise be problematic for a constitutively active enzyme. In our assembled complex, access to the active site is further restricted by Raptor. Whereas the FRB domain and mLST8 narrow the active-site cleft from the N- and C-lobe sites, respectively, the RNC domain additionally restricts access from the solvent-exposed surface below the FRB domain. This results in the enclosure of the active-site cleft from all directions, reducing its width to ~20 Å (Fig. 6A).

Raptor directly interacts with substrate proteins through their TOS motifs, FDIDL in S6K1 (34) and FEMDI in 4EBP1 (33), which include a conserved aspartate at position four. Given the structural similarity between the RNC domain and CASPases, which recognize four-residue sequences with an aspartate at position four, equivalent residues in CASPase and Raptor may function in substrate recognition (Fig. 6, C and D). The corresponding site is located directly within the mTORC1 active-site cleft, ~50 Å from the kinase center and optimally positioned for recruitment and delivery of subtrate to the active site (Fig. 6). Considering that the peripheral Raptor β propeller has been proposed to bind proteins involved in regulating mTORC1, it is possible that this regulation occurs via the peptide stretch connecting the β propeller to the RNC domain of Raptor.

Our cryo-EM maps of mTORC1 reveal additional density next to the FRB domain (Fig. 6). This density represents a copurified complex of Spodoptera frugiperda FKBP (SfFKBP) and rapamycin, which was used to enhance mTORC1 expression. SfFKBP shares 78% identity to human FKBP12, and its presence in purified mTORC1 was confirmed by mass spectrometry. The FKBP-rapamycin density further reduces the active-site cleft to ~10 Å. It also lies between the RNC domain and the active site, supporting steric hindrance as the mechanism by which rapamycin may inhibit mTOR toward certain substrates, but not others (32) (Fig. 6B). Although it has been proposed that FKBP-rapamycin binding compromises mTORC1 stability, possibly by displacing Raptor (35), we do not observe this effect. Furthermore, the FRB domain that binds FKBP-rapamycin (30, 31) is not near the dimerization interface; nor does it contact the RNC domain, making it unlikely that the binding of FKBP-rapamycin would destabilize mTORC1 by steric hindrance (Fig. 6B).

Fig. 6 The RNC domain of Raptor is positioned adjacent to the FRB domain of mTOR, complementing the active-site cleft of the kinase domain.

(A and B) View into the active-site cleft of one half of an mTORC1 dimer. The reconstructed density, filtered to its global resolution (5.9 Å) and contoured at 3 σ, is indicated as a translucent surface, and the corresponding model is shown in cartoon representation. The active site is indicated with a magenta asterisk. Panel (B) is shown with the perspective rotated, as indicated. (C and D) Schematic illustrating the proposed mode of substrate (gray) binding and delivery to the mTOR active site.

Implications for mTORC2 and other kinases

The architecture of mTORC1 provides a structural basis for studying mTORC1 function. Given the similar overall appearance of our structure and that of TORC2 (37, 40), we anticipate that the architecture of mTORC1 might be conserved in parts of mTORC2. In particular, we anticipate that the mode of dimerization, in which a complete dimer is formed through mTOR-mTOR contacts alone, will be conserved. An interaction of Rictor in mTORC2 with the bridge and horn HEAT domains, similar to that of Raptor in mTORC1, could stabilize the N terminus of mTOR and facilitate mTORC2 dimerization. Raptor forms this interaction through its armadillo repeat domain; a large domain of Rictor is also predicted to form an α-helical solenoid. Other members of the PIKK kinase family, including ATM, require accessory proteins that interact with HEAT repeats and are thought to dimerize in vivo. Although the N-terminal HEAT-repeat domains of the family are their least conserved parts, it appears possible, given the residual sequence similarity, that PIKK family members may have common modes of interaction (24).

Supplementary Materials

Materials and Methods

Figs. S1 to S8

Table S1

References (4156)

References and Notes

  1. Acknowledgments: We thank the ETH Zürich Scientific Center for Optical and Electron Microscopy (ScopeM) for access to EM facilities, P. Tittmann for technical support, T. Sharpe for size exclusion chromatography multi-angle laser light scattering analysis, the Biozentrum Proteomics Core Facility for mass spectrometric protein identification, the beamline staff at Swiss Light Source beamlines X06SA and X06DA for crystallographic support, and H. S. T. Bukhari for 3D modelling in Maya. C.H.S.A. was supported by an ETH Zürich postdoctoral fellowship and a European Molecular Biology Organization long-term fellowship. E.S. was supported by a fellowship from the People Program (Marie Curie Actions; REA grant agreement number 328159). This work was supported by the European Research Council funding to M.N.H. and N.B., and by the Swiss National Science Foundation via the National Centre of Excellence in RNA and Disease, project funding 138262 and R’Equip 145023. The cryo-EM density maps representing mTORC1 have been deposited in the EM Databank as EMD-3212 and EMD-3213 and the corresponding model in the Protein Data Bank as PDB ID 5FLC. The crystal structure of CtRaptor has been deposited in the Protein Data Bank as PDB ID 5EF5.
View Abstract

Navigate This Article