Structural basis for the recruitment and selective phosphorylation of Akt by mTORC2

Abstract

The mechanistic target of rapamycin (mTOR) protein kinase forms two multiprotein complexes, mTORC1 and mTORC2, that function in distinct signaling pathways. mTORC1 is regulated by nutrients, and mTORC2 is a central node in phosphoinositide-3 kinase (PI3K) and small guanosine triphosphate Ras signaling networks commonly deregulated in cancer and diabetes. Although mTOR phosphorylates many substrates in vitro, in cells, mTORC1 and mTORC2 have high specificity: mTORC2 phosphorylates the protein kinases Akt and PKC, but not closely related kinases that are mTORC1 substrates. To understand how mTORC2 recognizes substrates, we created semisynthetic probes to trap the mTORC2 :: Akt complex and determine its structure. Whereas most protein kinases recognize amino acids adjacent to the phosphorylation site, local sequence contributes little to substrate recognition by mTORC2. Instead, the specificity determinants were secondary and tertiary structural elements of Akt that bound the mTORC2 component mSin1 distal to the mTOR active site and were conserved among at least 18 related substrates. These results reveal how mTORC2 recognizes its canonical substrates and may enable the design of mTORC2-specific inhibitors.

Figure 1.. mTORC2 directly and specifically phosphorylates Ser⁴⁷³ of Akt with little contribution from local Akt sequence.

(A) Model for two-step activation of Akt at the plasma membrane. Following PIP3-mediated recruitment, Akt is sequentially (1) phosphorylated at hydrophobic motif Ser⁴⁷³ by mTORC2, inducing a conformational change of the PH domain, and then (2) at activation loop Thr³⁰⁸ by PDK1, resulting in full activity. (B) Purified monodisperse mTORC2 expressed in SF9 insect cells. (C,D) Akt inhibitors and Akt catalytic activity have minimal effect on mTORC2 phosphorylation of Akt Ser⁴⁷³ in vitro. ATP-competitive Akt inhibitors GDC-0068 and AZD-5360, and allosteric Akt inhibitor MK-2206 all have reported Akt IC₅₀ <10 nM; D274A Akt: catalytic null; v/[E] = velocity/[enzyme (mTORC2)]; 5 nM mTORC2, 1 mM ATP, 1 µM Akt, 10 mM MgCl₂. (E) Akt Ser⁴⁷³ phosphorylation assay in transfected serum-starved Akt1/2-null HCT116 cells stimulated for 40 min with insulin and hIGF1. Mutation of Thr⁴⁴³, previously proposed to be critical for Akt phosphorylation by mTORC2, results in only ~50% reduction, and catalytic inactivation via D274A mutation markedly increases Ser⁴⁷³ phosphorylation, together ruling out autophosphorylation (F) mTORC2 and mTORC1 are specific in phosphorylating canonical substrates Akt and 4E-BP1, respectively under identical conditions in vitro; 1 µM substrate, 1 mM ATP, and 10 mM MgCl₂; n=4, representative of 2 experiments. (G) C-tail sequences of human AGC kinases show conservation of Ser⁴⁷³-flanking Phe⁴⁷² and Tyr⁴⁷⁴. (H,I) Local sequence around Ser⁴⁷³ has little impact on Akt Ser⁴⁷³ phosphorylation: neither mutation of Phe⁴⁷² to Ala nor phosphorylation of Tyr⁴⁷⁴, Ser⁴⁷⁷, or Thr⁴⁷⁹ has a significant effect on Ser⁴⁷³ phosphorylation in vitro (quantitative western blotting with compound-modified AktF472A-pSer⁴⁷³ and AktpSer⁴⁷³-pTyr⁴⁷⁴ standards; 5 µM Akt; I, 1 µM Akt). (J) Mass spectrometry shows in vitro phosphorylated Akt tails are predominantly monophosphorylated at pSer⁴⁷³, with some diphosphorylated pSer⁴⁷³-pSer⁴⁷⁵ tails. All comparisons, unpaired t-test.

Figure 2.. mTORC2-Akt complex structure determined using Akt-Torin.

(A) Akt-Torin peptide inhibits mTORC2 phosphorylation of Akt much more potently than Akt-ATP peptide in vitro; 5 µM Akt, 1 mM ATP, 5 nM mTORC2, 10 mM MgCl₂ (n=4). (B) Design of second-generation bivalent mTORC2 ligand Akt-Torin. Torin acrylamide is conjugated to Cys⁴⁷³ of an Akt tail peptide via a Michael addition, and purified Akt-Torin peptide (460–480) is ligated to Akt(1–459) thioester; the expressed protein ligation reaction produces a native peptide backbone, facilitated by Akt Cys⁴⁶⁰. (C) Cryo-EM structures of apo mTORC2 and mTORC2:Akt-Torin by cryo-EM, resolved to 3.12 Å and 2.55 Å, respectively. Diagram shows the components of the complex, their compositions, and structurally resolved regions. Akt binds to mTORC2 in the “substrate arch”, where mLST8, mSin1, and Rictor together form two distinct docking interfaces ~75 Å from the mTOR active site. First, the Akt kinase domain N-lobe binds mSin1 CRIM, which is tethered and stabilized at two points against mLST8. Second, the Akt kinase domain C-lobe binds to the N-terminal region of mSin1, a flexible interaction we name the “mSin1 N-mooring”. The critical component for mTORC2 substrate binding is mSin1, which takes a long and circuitous path, snaking from a pocket in Rictor against which it builds the N-mooring, across the substrate arch, around the back of mLST8, and then back to build the CRIM interface. Subsequent mSin1 domains aRBD (atypical Ras binding domain) and PH are involved in regulation and tether mTORC2 to the plasma membrane, respectively.

Figure 3.. Interplay between Thr450 phosphorylation and mTORC2-Akt Interface 1: Akt N-Lobe – mSin1 CRIM.

(A) The mTORC2:Akt-Torin cryo-EM structure shows a broad docking surface that includes two loops each on mSin1 CRIM and the Akt kinase domain N-lobe that have strong, complementary charges, as seen by the Coulombic electrostatic potential (calculated with ChimeraX). (B) Key Akt residues assayed by transfection in Akt1/2-null HCT116 for impacts on both hydrophobic motif pSer⁴⁷³ and turn motif pThr⁴⁵⁰ phosphorylation show (1) Thr²¹⁹ and Lys¹⁸³ and (2) a basic patch of four residues Lys¹⁵⁸, Lys¹⁶³, Lys¹⁸², and Arg²²² together drive the interaction (all four basic residues mutated in ‘N-lobe 4xAla’ construct). Double K158A-K163A and quadruple 4xAla Akt mutations largely abrogate both pSer⁴⁷³ and pThr⁴⁵⁰ levels despite equal Akt expression. (C) Both CRIM loops are required for Akt phosphorylation, as assayed by complementation transfection experiments in mSin1-null HEK293T cells with endogenous Akt substrate. Deletion of the acidic loop (Δacidic, aa236–245), mutation of all 6 acidic residues in the acidic loop (6xNQ), or mutation of Phe²⁵⁷ + Phe²⁵⁹ on the adjacent loop all reduce pSer⁴⁷³ and pThr⁴⁵⁰. Metap2 and CCM2 are unrelated control human proteins. (D) mSin1-CRIM binds to AGC kinase N-lobes at a shared binding interface that also forms a stabilizing intramolecular interaction with phosphorylated turn motif threonine (pThr⁴⁵⁰ in Akt, pThr⁵⁵⁵ in the PKC-iota crystal structure PDB 3A8W, shown). Remarkably, overlaying the PKC-iota crystal structure in place of Akt bound to mTORC2 highlights that CRIM and phosphorylated turn motif threonine bind the same four key basic residues, and that both could not simultaneously bind. (E) Test of the hypothesis that pThr⁴⁵⁰ may compete with Ser⁴⁷³ phosphorylation by transfection of ‘phospho-mimic’ and control alanine mutants in HCT116 cells. Increased pSer⁴⁷³ with T450D and T450E mutants is consistent with the expected weaker binding of sidechain carbonyl groups than a phosphate to the basic patch. (F) In vitro rates of phosphorylation of Akt Thr⁴⁵⁰ and Ser⁴⁷³ by mTORC2 are similar, within ~30%, supporting a similar direct phosphorylation mechanism of both turn and hydrophobic motifs (note p=0.03 for the rate comparison; full gels for panels B-F in Fig. S25).

Figure 4.. mTORC2-Akt Interface 2: Akt C-Lobe – mSin1 N-mooring.

(A) Cryo-EM structure of this flexible docking interface shows that Akt residues Ile³⁶⁶-Phe³⁶⁸ abut the mSin1 N-terminal region, especially mSin1 Trp⁷⁶ and Phe⁷⁸, with the unstructured N-terminus of Rictor in close proximity. (B) Mutation of Akt residues 366–368 to glycine or glutamate ablates Ser⁴⁷³ phosphorylation and dramatically limits Thr⁴⁵⁰ phosphorylation in Akt-null HCT116 cells without affecting total protein levels. (C,D): To assay contributions from mSin1, WT and mutant constructs were transfected into mSin1-null HEK293T cells. (C) Residues Trp⁷⁶ and Phe⁷⁸ were mutated to glycines or serines, which show modest effects on both Ser⁴⁷³ and Thr⁴⁵⁰ phosphorylation (n=4–6 biological replicates from n=2–3 independent experiments; unpaired t-test). (D), Arginines 81–83, >15 Å from docked Akt, have been proposed to be important for SGK but not Akt phosphorylation by mTORC2⁶. To test the importance of these residues, they were simultaneously mutated to Ala (RRR81–83AAA, R81–83A for short), which shows large impacts on pSer⁴⁷³ and pThr⁴⁵⁰ levels. Mutations of the individual residues show these effects are mostly due to Arg⁸³, with lesser contribution from Arg⁸¹ and no significant contribution from Arg⁸² (**** p<0.0001, unpaired t-test). To investigate the contribution of unstructured and bridging mSin1 residues 50–76, these were deleted and replaced with a 5-residue GGGGS linker (Δ50–76L); note this deletion removes both Ser⁵⁰ and Trp⁷⁶. This shows minimal effect on Akt phosphorylation, which may be explained by backbone interactions. (Note, these gels are from the same experiment as Fig. 3C, left panels; all lanes shown in Fig. S28C).

Figure 5.. Integrative modeling and membrane association of mTORC2-Akt.

(A) Integrative model centroid solution of mTORC2-Akt co-complex based on cross-linking mass spectrometry and biophysical constraints. Each sphere is a “flexible bead”, representing 10 residues, and localization densities, shown on one symmetric mTORC2 monomer, represent the variability found within the ensemble for each component. The modeling localizes the remaining components of mTORC2, including PH domains of both Akt and mSin1 and the mSin1 aRBD (inset). (B) Model of Akt C-tail phosphorylation. The flexible C-tail (aa 430–480), modeled from the last resolved residue in the Akt C terminus (Ser422), extends ~75Å to the mTOR active site with abundant slack and without violating excluded volumes. The Akt PH domain (purple) is behind the kinase C-lobe in this view and does not encumber phosphorylation. Thr⁴⁵⁰ can be similarly accommodated (see Fig. S35B) (C) Model of mTORC2 membrane engagement: the 4 PH domains (1 from each copy of mSin1 or Akt) in the mTORC2 homodimer can be coplanar within the ensemble, shown interacting with the phospholipid bilayer, e.g., at the plasma membrane. (D) In vitro vesicle kinase assay in which small uni-lamellar vesicles (SUVs, 1 mM total lipid) are pre-incubated with 5 nM mTORC2 and the indicated concentration of WT Akt or Akt lacking the PH domain (ΔPH-Akt, aa 144–480), and then reacted for 10 min at 30°C with 1 mM ATP, 10 mM MgCl₂ (except where indicated) and analyzed by quantitative immunoblotting. Ctrl, control vesicles (PC:PE,1:1); PIP₃ containing vesicles (PC:PE:PS:PIP₃,30:50:19:1). There is a statistically significant 2–3-fold rate increase at 0.2 μM and 1 μM, but not at 5 μM (unpaired t-test). Note that we loaded an equal amount of total Akt on the gel in all lanes, and as such there is less apparent signal for pSer⁴⁷³ with higher concentrations of Akt in the reaction; there is more total phosphorylation, but a smaller proportion of the total Akt is loaded on the gel. (E) Effect of mSin1 PH domain on Akt phosphorylation. mSin1 lacking the C-terminal PH domain (ΔPH) or WT was transfected into mSin1-null HEK293T cells; n=4–16 measurements from 2–8 independent experiments, each containing two biological replicates, p-values from unpaired t-test.

Figure 6.. mTORC2 similarly phosphorylates AGC kinases at both turn and hydrophobic motifs.

(A) Domain diagram of Akt (Akt1), PKCα, and PKN1 and alignment of turn motif (TM) and hydrophobic motif (HM). The HM residue for PKN1 is instead a ‘phosphomimic’ aspartate. Domains: C1, C2, conserved regions; HR1, heptapetide repeat 1; these domains of PKCα and PKN1 are involved in regulation and membrane targeting. (B) PKC phosphorylation assay in HCT116 cells mirrors Akt results. Cells were transfected with PKCα constructs, serum-starved, and treated with DMSO or Torin1 for 1 hour before stimulation with insulin and hIGF1 and immunoblotting. Catalytic-null D463A PKCα is indistinguishable from WT, and interfaces mirroring Akt Lys¹⁸³ (PKCα Lys³⁷²) and Akt Thr²¹⁹ (PKCα Thr⁴⁰⁹) are similarly required for HM phosphorylation at Ser⁶⁵⁷. (C,D) Effect of mTORC2 PH domain on transfected substrate PKCα HM (Ser⁶⁵⁷) and PKN1 TM (Ser⁹¹⁶) phosphorylation; comparison to endogenous Akt serves as an internal control. Quantified blots highlight that PKCα phosphorylation does not depend on the mSin1 PH domain, whereas PKN1 TM shows a similar dependence on the mSin1 PH domain as the Akt HM; n=4–16 measurements from 2–8 independent experiments, each containing two biological replicates; p-values from unpaired t-test. (E,F) mSin1 and mSin1 subdomain contribution to PKCα and PKN1 phosphorylation in mSin1-null HEK293T cells transfected with WT mSin1 or the indicated mSin1 mutants. Δacidic lacks the CRIM acidic loop, 6xNQ mutant replaces the six acidic residues in the acidic loop with uncharged counterparts, and R81–83A and Δ50–76L constructs are in the mSin1 N-terminus. (E) mSin1 presence and acidic loop are required for PKC phosphorylation; other constructs that affect Akt have less effect on PKCα. (F) PKN1 TM Ser⁹¹⁶ phosphorylation depends on mTORC2 status and is not detected in the S916A mutant. mSin1 mutations have nearly identical effects on PKN1 as Akt. (G) AGC kinases phosphorylated on the C-tail by mTORC2. Both TM and HM are directly modified, except for kinases marked †, in which HM Ser/Thr is Asp/Glu and only TM is targeted. P-values from unpaired t-tests.

Figure 7.. Models for mTORC2 recognition of Akt.

(A) The same surface in the Akt N-lobe binds to both P-Thr⁴⁵⁰ and mTORC2, through mSin1 CRIM. We propose that Akt P-Thr⁴⁵⁰ binds to and shields the N-lobe, simultaneously stabilizing Akt and limiting mTORC2 recognition. P-Thr⁴⁵⁰ is in dynamic equilibrium with an open conformation that mSin1 CRIM can intercept, leading to complete Akt docking at both interfaces, between (1) Akt N-lobe :: mSin1 CRIM and (2) Akt C-lobe :: mSin1 N-mooring. Once the Akt kinase domain is docked, Akt Ser⁴⁷³, like a scorpion tail, can then find its way into the mTOR active site, where direct Ser⁴⁷³ phosphorylation is catalyzed. FAT and Spiral are domains of mTOR, labeled to emphasize orientation. (B) Comparison of substrate-bound mTORC2 at the plasma membrane vs mTORC1 at the lysosome surface reveals that the orientation of mTOR is reversed in the two complexes. Our mTORC2 integrative model shows PH domains of both mTORC2 and Akt directly engaged to the plasma membrane via PIP₃. In contrast, multiple lipid anchors on the small G-protein Rheb (PDB ID 6BCU)(1) and the Rag-Ragulator complex (PDB ID 6U62)(2) anchor mTORC1 to the lysosome membrane; both Rag-Ragulator and Rheb are bound to mTORC1-specific component Raptor, which also binds substrate 4E-BP1 TOS peptide. mTOR and mLST8, shared between the two complexes, are 180 degrees “flipped” relative to the membrane in the two models. This opposite orientation may help explain some of the complexes’ unique substrate specificities.