Ageing can be defined as the gradual deterioration of physiological functions, increasing the incidence of age-related disorders and the probability of death. Therefore, the term ageing not only reflects the lifespan of an organism but also refers to progressive functional impairment and disease. The nutrient-sensing kinase mTOR (mammalian target of rapamycin) is a major determinant of ageing. mTOR promotes cell growth and controls central metabolic pathways including protein biosynthesis, autophagy and glucose and lipid homoeostasis. The concept that mTOR has a crucial role in ageing is supported by numerous reports on the lifespan-prolonging effects of the mTOR inhibitor rapamycin in invertebrate and vertebrate model organisms. Dietary restriction increases lifespan and delays ageing phenotypes as well and mTOR has been assigned a major role in this process. This may suggest a causal relationship between the lifespan of an organism and its metabolic phenotype. More than 25 years after mTOR’s discovery, a wealth of metabolic and ageing-related effects have been reported. In this review, we cover the current view on the contribution of the different elements of the mTOR signalling network to lifespan and age-related metabolic impairment. We specifically focus on distinct roles of isoforms and splice variants across the mTOR network. The comprehensive analysis of mouse knockout studies targeting these variants does not support a tight correlation between lifespan prolongation and improved metabolic phenotypes and questions the strict causal relationship between them.
The mTOR pathway in mammals
Target of rapamycin (TOR) was discovered in 1991 in the budding yeast Saccharomyces cerevisiae  and is structurally and functionally conserved in all eukaryotes, including mammals where it is called mammalian TOR (mTOR). mTOR is a serine/threonine kinase that functions as a master regulator of cellular growth and metabolism. mTOR forms two distinct complexes, mTOR complex 1 (mTORC1) and mTORC2 [2–4]. The scaffold protein raptor (regulatory associated protein of mTORC1) [5,6] and the inhibitory protein PRAS40 (proline-rich Akt substrate of 40 kDa) [7–11] are mTORC1-specific binding partners. mTORC2 is formed by rictor (rapamycin-insensitive companion of mTOR) [12,13], mSin1 (mammalian stress-activated protein kinase interacting protein 1) [14,15] and Protor (protein observed with rictor) [7,16]. In addition, mTORC1 and mTORC2 share the binding partners mLST8 (mammalian lethal with SEC13 protein 8) [13,17], DEPTOR (dishevelled, Egl-10 and pleckstrin (DEP) domain containing mTOR-interacting protein)  and the Tti1 (telomere maintenance 2 (TELO2) interacting protein 1)/Tel2 (telomere maintenance 2) complex . Both mTORC1 and mTORC2 are activated by growth factors (e.g. insulin) and amino acids, and mTORC1 is positively regulated by cellular energy (ATP/AMP ratio) (Figure 1) .
The mTOR network
Insulin activates mTORC1 via a well-described signalling cascade that is initiated either by the InR (insulin receptor) or the IGF1-R (insulin like growth factor 1 receptor) (Figure 1). Upon insulin/IGF-1 (insulin like growth factor 1) binding, the receptors dimerize and transphosphorylate their cytoplasmic domains . This leads to the recruitment of IRS (insulin receptor substrate), which is phosphorylated at tyrosine residues by the InR and IGF1-R, and in response acts as a scaffold for many proteins . Class I PI3Ks (phosphatidylinositol 3-kinases) bind to IRS, where they are activated, and thus, phosphorylate PIP2 (phosphatidylinositol-3,4-biphosphate) to form PIP3 (phosphatidylinositol-3,4,5-triphosphate). PIP3 is converted back to PIP2 by the phosphatase PTEN (phosphatase and tensin homolog) . Via binding to PIP3, proteins containing a PH (pleckstrin homology) domain are recruited to the membrane. The AGC (protein kinase A/protein kinase G/protein kinase C (PKC)) kinases PDK1 (3-phosphoinositide-dependent kinase-1) and Akt both contain PH domains . Upon PIP3 formation, they translocate to the plasma membrane, where PDK1 phosphorylates Akt within its kinase domain (KD) . Akt downstream of PDK1 activates mTORC1, by phosphorylating and inhibiting PRAS40 [23,24] and the TSC (tuberous sclerosis complex) complex, formed by TSC1, TSC2 and TBC1D7 (Tre2-Bub2-cell dyvision cycle 16 (Cdc16) (TBC) 1 domain family member 7) [25,26]. The TSC acts as a GAP (GTPase-activating protein) towards the small GTPase Rheb (ras-homologue-enriched-in-brain) . Therefore, TSC inhibition by Akt derepresses Rheb, which directly binds and activates mTORC1 . For Rheb to act on mTORC1, mTORC1 must be localized to lysosomes. Translocation of mTORC1 from the cytosol to the lysosomal surface occurs upon amino acid stimulation and is mediated by the Rag GTPases (Ras-related GTP-binding proteins) [29–32] (reviewed elsewhere [2,4]). In addition, amino acids suppress lysosomal localization of the TSC, leading to derepression of Rheb [33–35]. Amino acids also activate PI3K via an unknown mechanism [36,37], and AMPK (AMP-activated protein kinase) in a CaMKKβ (Ca2+/calmodulin-dependent protein kinase kinase-β)-dependent manner . Next to growth factors and amino acids, the cellular energy status also modulates the mTORC1 activity . When the ATP/AMP ratio is low, AMPK is allosterically activated by AMP and inhibits mTORC1 by activating the TSC  and by directly phosphorylating raptor (Figure 1) . mTORC1 affects virtually all metabolic processes to ultimately regulate cellular growth and survival (reviewed by e.g. Saxton and Sabatini ). We focus here on protein homoeostasis which mTORC1 controls by activating translation and by inhibiting autophagy . mTORC1 positively regulates translation by activation of S6K (ribosomal protein S6 kinase)  and inhibition of 4E-BP (eukaryotic translation initiation factor 4E-binding protein) , leading to increased translation capacity  and enhanced translation initiation , respectively (Figure 1). mTORC1 inhibits autophagy by phosphorylating and inhibiting the kinases ULK1 and ULK2 (unc-51 like autophagy activating kinase 1/2) (Figure 1) , which phosphorylate FIP200 (FAK family kinase-interacting protein of 200 kDa) and thereby enhance autophagosome formation [45–48]. AMPK activates ULK1, thereby acting as an mTORC1 antagonist in autophagy [44,49]. mTORC1 and its substrate S6K reduce insulin sensitivity via negative feedback mechanisms. S6K phosphorylates IRS, leading to IRS degradation [50,51], and mTORC1 phosphorylates and stabilizes Grb10 (growth factor receptor bound protein 10), leading to InR inhibition [52,53]. Both events render cells refractory to insulin and consequently reduce PI3K and Akt activity.
Also mTORC2 controls central cellular functions, including glucose metabolism [54,55], cell survival  and cytoskeletal organization [12,13]. The mechanisms leading to mTORC2 activation are relatively poorly explored. Previous research has established that insulin activates mTORC2  in a PI3K-dependent manner  at the ribosomes (Figure 1) . However, this PI3K differs from the PI3K upstream of mTORC1 in that mTORC2 is not regulated by mTORC1-driven negative feedback to PI3K or Akt . Different PI3K inputs to the two mTOR complexes could be mediated by distinct isoforms of the PI3K catalytic subunit p110, such as in hippocampal progenitor cells, where p110α activates both mTOR complexes while p110β activates mTORC2 only . Amino acids activate mTORC2 as well, but the exact mechanism remains so far unknown . The best described downstream effector of mTORC2 is Akt, which is phosphorylated by mTORC2 within the so-called hydrophobic motif (HM) . Akt downstream of mTORC2 inhibits the transcription factors FoxO1/3A (forkhead box proteins O1 and O3A) (Figure 1) [61,62] which enhance apoptosis [61,63,64] and mediate stress responses [65–68]. Therefore, mTORC2 promotes cellular survival. mTORC2 also activates the AGC kinase SGK (serum and glucocorticoid-regulated kinase)  that promotes proliferation, migration and cell survival .
Isoforms and splice variants across the mTOR network
Next to the complex wiring of the mTOR network, most of its components occur as different variants, contributing to the network’s complexity and versatility. These variants comprise similar proteins originating from different genes, referred to as ‘isoforms’ (Table 1) and ‘splice variants’ originating from alternative splicing or alternative translation initiation of the same gene (Table 2). Isoforms and splice variants can have overlapping, but also different biological functions. In the following, we focus on the differences between these variants regarding their protein structure, tissue expression pattern, regulation and function within the mTOR network. Isoforms and splice variants of PI3K and FoxO1/3A have been reviewed extensively earlier [70,71] and are therefore not covered here.
This table highlights the differences among the isoforms across the mTOR network. Differences are categorized according to different structure, regulation, expression pattern and biological function. References are indicated in brackets. The colour code refers to the different signalling modules of the mTOR network as indicated in Figure 1.
Abbreviations: ATG13, autophagy related 13; BAD, BCL2-associated agonist of cell death; Cdc42, cell division cycle 42; EMT, epithelial–mesenchymal transition; Erα, oestrogen receptor α; eEF2K, eukaryotic elongation factor-2 kinase; eIF4B, eukaryotic translation initiation factor 4B; GLUT4, glucose transporter type 4; GSK-3, glycogen synthase kinase 3; IL-3, interleukin 3; IRS, insulin receptor substrate; Mdm2, Mdm2 proto-oncogene; MAPK, mitogen-activated protein kinase; PHLPP, PH domain and leucine-rich repeat protein phosphatase; Protor, protein observed with rictor; RORγ, RAR-related orphan receptor γ; Rac1, ras-related C3 botulinum toxin substrate 1; SIRT, sirtuin; YY2, yin yang 2.
Protein domains: pleckstrin homologue (PH), phosphotyrosine-binding (PTB), phosphatidylinositol 3-kinase binding domain (PI3K-BD), binding domain (BD), growth factor receptor-bound protein 2 bidning domain (Grb2-BD), src homology 2 domain-containing phosphatase binding domain (SHP2-BD), kinase regulatory loop binding (KRLB), kinase domain (KD), hydrophobic motif (HM), nuclear localization sequence (NLS), N-terminal domain (NTD), C-terminal domain (CTD), eukaryotic translation initiation factor 4E binding domain (eIF4E-BD), TOR signalling sequence (TOS), Proline/Serine (Pro/Ser).
This table indicates the predicted splice variants in human and mouse (source: uniprot.org, date: 01/2017) and the experimentally validated splice variants including functional differences. References are indicated in brackets. The colour code refers to the different signalling modules of the mTOR network as indicated in Figure 1. *Asterisks designate splice variants that have been predicted, but have not yet been experimentally confirmed. LRefer to primary literature as this variant is not listed in UniProt. The variants arise from alternative splicing unless marked otherwise. Variants marked T originate from alternative translation initiation.
Protein domains: N-terminal domain (NTD), pleckstrin homologue (PH), kinase domain (KD), hydrophobic motif (HM), huntingtin-elongation factor 3-regulatory subunit A of PP2A-TOR1 repeats (HEAT repeats), FRAP-ATF-TTRAP (FAT), FRAP-ATM-TTRAP domain (FATC), raptor N-terminal conserved (RNC), C-terminal domain (CTD), nuclear localization signal (NLS), ras-binding domain (RBD).
In mammals, the InR occurs in two different splice variants, named InR-A and InR-B (Table 2) . InR-A is expressed predominantly in the central nervous system and hematopoietic cells, while InR-B is found in adipose tissue, liver and muscle . Both InR-A and InR-B have α and β domains, but only InR-B has an extra stretch of 12 amino acids in the α domain that changes its binding characteristics towards its ligands . InR-A and InR-B can form homo- or heterodimers and have similar affinities for insulin. However, InR-A binds IGF-1 and IGF-2 with higher affinity than InR-B [74,75]. This difference in affinity may allow them to activate the mTOR network differently in response to the same ligands.
Downstream of the InR, IRS has four different isoforms originating from different genes (IRS1–4), with IRS3 exclusively present in rodents . Therefore, we focus here on the differences among IRS1, IRS2 and IRS4 (Table 1). IRS1 and IRS2 are ubiquitously expressed  and, thus, the most widely studied isoforms. They are highly similar in their domain structure, but IRS2 contains an extra KRLB (kinase regulatory loop binding) domain that binds to the InR and the IGF1-R [78,79]. The KRLB domain limits the tyrosine phosphorylation on IRS2 by the InR and IGF1-R, and thereby inhibits IRS2 function . An in vitro study in skeletal muscle cells has shown that IRS1 and IRS2 tyrosines are dephosphorylated at different rates by phosphatases upon insulin or IGF-1 stimulation . While IRS2 becomes rapidly dephosphorylated after 3–10 min, IRS1 remains phosphorylated for up to 1 h . It is unknown if this difference is mediated by distinct phosphatases or by different dynamic behaviour of the same phosphatase towards IRS1 or IRS2. Such differences in dynamic regulation might contribute to the distinct functional outcomes of IRS1 and IRS2. IRS1 enhances phosphorylation of both Akt1 and Akt2, whereas IRS2 signals mainly through Akt2 . IRS1 but not IRS2 enhances actin remodelling and GLUT4 (glucose transporter type 4) translocation . IRS1 and IRS2 induce the MAPK (mitogen-activated protein kinase) pathway with IRS2 having a stronger effect than IRS1 [82,83]. In addition, IRS2 but not IRS1 allows the maintenance of beta-cell mass at the pancreas [84,85]. IRS1 and IRS2 seem to have opposite effects on metastasis, as IRS2 promotes metastasis in breast cancer cells [86,87], whereas IRS1 suppresses metastatic spread in mice. Only little is known about IRS4, whose protein expression is limited to brain, kidney, thymus and liver . Previous evidence suggests a function of IRS4 during adenoviral infection, where IRS4 up-regulation leads to constitutive Akt activation even in the absence of insulin .
The PI3K-antagonist PTEN exists in three reported variants, termed as PTEN, PTEN-Long  and PTENα , originating from alternative translation initiation (Table 2). PTEN-Long and PTENα both originate from the start codon CUG513 and have the same apparent molecular weight and predicted number of amino acids, suggesting that these two variants could be the same [90,91]. PTEN is ubiquitously expressed while PTENα is predominantly expressed in skeletal and cardiac muscle. All variants contain the functional PTEN domains, including the phosphatase domain, the C2 domain and the tail domain. PTENα and PTEN-Long contain an extra NTD (N-terminal domain) which may lead to different subcellular localizations [90,91]. While PTEN localizes mainly to the cytosol, PTENα resides at mitochondria  and PTEN-Long is secreted into the microenvironment . This difference in localization results in distinct functions, as PTEN is the counterpart and antagonist of PI3K at the plasma membrane, whereas PTENα targets the mitochondrial complex IV (cytochrome c oxidase) to regulate the cellular energy status . PTEN-Long also decreases PI3K signalling in a phosphatase-dependent manner. However, as it is secreted into the microenvironment and able to enter neighbouring cells, it could have its main role in controlling signalling at the tissue level .
There are three Akt isoforms in mammals: Akt1 (PKBα), Akt2 (PKBβ) and Akt3 (PKBγ) (Table 1) which are similar in structure, as they all contain a PH domain, a KD (kinase domain) and the HM. Akt3 is also present as a shorter splice variant Akt3-1 (Table 2), which lacks the mTORC2 phosphorylation site within the HM, and is less responsive to growth factor stimulation [92,93]. The three Akt isoforms display unique tissue expression patterns. While Akt1 is ubiquitously present , Akt2 resides primarily in brown fat, skeletal muscle and liver [94,95], and Akt3 is mainly expressed in brain and testis . Akt1 and Akt2 also differ in their subcellular localization: in rat adipocytes, Akt1 is mainly distributed in the cytosol, whereas Akt2 localizes to GLUT4-containing vesicles . The phosphatases PHLPP1 (PH domain and leucine-rich repeat protein phosphatase 1) and PHLPP2  differently act on the mTORC2 substrate site within the HM of the three Akt isoforms : PHLPP1 specifically dephosphorylates Akt2 and Akt3, while PHLPP2 dephosphorylates Akt1 and Akt3 . The three Akt isoforms also differ regarding their substrate specificities, as only Akt1 and Akt2 phosphorylate TSC2  to promote cell growth, whereas all the three isoforms phosphorylate FoxO1/3A  to inhibit apoptosis. It is unknown if the substrate specificity of the different Akt isoforms also relates to PRAS40. It is likely as a number of further Akt-isoform-specific substrates have been described: Akt1 phosphorylates palladin, an actin-binding protein, to inhibit cell migration  while Akt2 phosphorylates Ankrd2/ARPP (cAMP-regulated phosphoprotein), a muscle-specific protein, which when phosphorylated by Akt, prevents muscle differentiation . Their different substrate specificities grant the Akt isoforms distinct roles in tumour formation. Akt1 specifically promotes cell survival [101,102] and cell growth [103,104], whereas Akt2 is more important for glucose metabolism [105,106], cell migration , tumour metastasis  and EMT . In contrast, Akt3 inhibits migration , and has been linked to neuronal growth and development [108,109]. Interestingly, Akt3 overexpression confers resistance to the Akt inhibitor MK2206  suggesting that Akt3 can be protumorigenic in the context of targeted cancer therapies.
Downstream of Akt, TSC2 is present in healthy individuals in several variants originating from alternative splicing of exons 25 and 31 (Table 2) [111–113]. TSC2 transcripts lacking exons 25 (Uniprot ID: P49815-2, P49815-3, P49815-5, P49815-6 and P49815-7) and/or 31 (Uniprot ID: P49815-4, P49815-5, P49815-6 and P49815-7) occur at higher levels as compared to the full length variant (Uniprot ID: P49815-1) . Exon 25 is detectable in lymph node, muscle and thyroid tissue, whereas exon 31 is found in adipose, adrenal, brain, breast, colon, kidney, lung, prostate, skeletal muscle, testes and thyroid tissue . Deletion of exons 25 and 31 does not affect the TSC dependent inhibition of mTORC1 . However, it may affect TSC2’s properties as a substrate for Akt and AMPK. Akt phosphorylates TSC2 at residues Ser939, Ser981 and Thr1462 [26,114,115] and variants missing exon 25 do not contain the target site at Ser981. The TSC2 residues phosphorylated by Akt serve as a scaffold for 14-3-3 protein binding to and subsequent degradation of TSC2 . This suggests that TSC2 containing exon 25 has a higher 14-3-3 affinity and higher turnover than TSC2 variants without that exon . AMPK activates TSC2 by phosphorylation at Ser1387 and Thr1271 . The AMPK target site at Thr1271 is contained in exon 31 and, therefore, TSC2 variants that lack exon 31 may be less sensitive to AMPK activation. Further studies are required to delineate the biological consequences of the different structures of the TSC2 splice variants upon Akt and AMPK activation. TSC1 variants originating from alternative splicing have so far not been identified experimentally under healthy conditions. However, mutations in either TSC1 or TSC2 that lead to aberrant splicing and generation of multiple splice variants are commonly found in a wide range of pathologies such as cancer and TSC . TSC is a rare genetic disorder caused by mutations in the TSC1 and TSC2 genes, leading to benign tumours in multiple organ systems . The TSC Leiden Open Variation Databases (TSC LOVD, www.lovd.nl) displays all variants reported for TSC1 and TSC2, being currently 870 and 2473, respectively (date: 05/2017). For TBC1D7, the third member of the TSC protein complex, no isoforms or splice variants have been found experimentally. Also Rheb, the target of the TSC, does not present any isoform or splice variant.
Components of the mTOR complexes occur in different variants (Table 2). mTOR itself exists in two splice variants, mTORα and mTORβ, of which mTORβ is the shorter version . mTORα contains the protein–protein interaction domains HEAT (huntingtin-elongation factor 3-regulatory subunit A of PP2A-TOR1 repeats), FAT (FRAP–ATM–TTRAP) and FATC (FAT-carboxy terminal domain) and a KD, while mTORβ lacks the HEAT and FAT domains . Studies in rats suggest that mTORα is ubiquitously expressed whereas mTORβ is mainly expressed in lung, heart, stomach, intestine and liver . mTORβ can still form the two mTORC1 and mTORC2 complexes, and phosphorylate the mTORC1 substrates S6K1 and 4E-BP1 as well as the mTORC2 substrate Akt. Cells overexpressing mTORβ have a shorter G1 phase, suggesting a role in cell-cycle progression . This may be due to a stronger association of mTORβ with c-Myc, a transcription factor that controls cell-cycle progression. Also raptor exists in two splice variants, termed as raptor and raptor-v2. Raptor-v2 (Table 2) is the shorter variant lacking the HEAT repeats, which are needed for binding to mTORC1 substrates . Raptor is ubiquitously expressed but levels are higher in brain, immune cells, gastrointestinal tract and kidney . Raptor-v2 mainly localizes to the pituitary, nasal mucosa and muscle . Raptor-v2 binds mTORC1 but cannot bind S6K1, as it lacks the HEAT repeats . This suggests that raptor-v2 inhibits mTORC1 by sequestering mTOR away from the functional complex. However, further studies are required to characterize the biological function of raptor-v2.
Regarding the members of mTORC2, no isoform or splice variant for rictor has been so far described. However, five mSin1 splice variants have been experimentally identified, mSin1.1–mSin1.5 (Table 2) . No information is available on their expression pattern. Only mSin1.1, mSin1.2 and mSin1.5 (also termed as mSin1, mSin1α and mSin1β ) are found in mTORC2 and mediate Akt phosphorylation. mSin1.1 is the full length protein and contains a RBD (ras-binding domain) and a PH domain. mSin1.2 and mSin1.5 lack part of the RBD, and mSin1.5 also lacks the PH domain . mSin1.1 and mSin1.2 overexpression increases Akt phosphorylation in response to insulin, whereas mSin1.5 overexpression enhances Akt phosphorylation independent of insulin . This suggests that mTORC2 containing mSin1.1 or mSin1.2 is activated by insulin, whereas mTORC2 with mSin1.5 is constitutively active or responds to signals other than insulin. Protor has two isoforms, Protor-1 and Protor-2 (Table 1) [7,16], that are ubiquitously expressed . They do not contain any known functional domain, but their N-terminal sequence is highly conserved and they both bind mTORC2 via rictor [7,16]. Protor-1 also exists in three splice variants, Protor-1α, Protor-1β and Protor-1γ, of which only Protor-1α and Protor-1β can form a part of mTORC2 . In kidney, Protor-1 in mTORC2 is required for the phosphorylation of SGK and activation of sodium transport . Protor-2 interacts with and activates the RNA binding protein TTP (tristetraprolin) to ensure mRNA turnover under stressed conditions . Further research is required to better understand the differences in expression, function and regulation among these variants.
Protein variants downstream of mTORC1 have been extensively characterized, and we discuss them here in the context of cell growth, translation and autophagy. Firstly, S6K has two isoforms encoded by different genes, S6K1 and S6K2 (Table 1) [123–125]. They both have an NTD, a CTD (C-terminal domain) and a KD. In addition, S6K2 has an NLS (nuclear localization sequence) next to its CTD . Both S6K1 and S6K2 are ubiquitously expressed  and, although they are both activated via phosphorylation by PDK1 [125,127] and mTORC1 [40,125], different isoform-specific signalling inputs contribute to their activation . S6K1 is inhibited by leucine starvation  and activated by the actin-binding protein neurabin , phospholipase D1 , the Rho family G proteins Rac1 and Cdc42 (cell division cycle 42)  and the deacetylases SIRT1 (sirtuin 1) and SIRT2 . S6K2 is activated in response to the cytokine IL-3 (interleukin 3) , by the MAPK pathway  and PKC . Both S6K1 and S6K2 phosphorylate S6, a protein that forms a part of the ribosomal 40S subunit  and, thus, the distinct activators of S6K1 and S6K2 may be a means to specifically control translation in different cellular conditions, such as growth factor stimulation, different cell-cycle phases or during the immune response. In addition to acting on S6, S6K1 also activates translation by phosphorylating eIF4B (eukaryotic translation initiation factor 4B)  and eEF2K (eukaryotic elongation factor-2 kinase) . Furthermore, S6K1 promotes cell survival by inhibiting BAD (BCL2-associated agonist of cell death) , Mdm2  and GSK-3 (glycogen synthase kinase 3) , and induces transcription by activating Erα (estrogen receptor alpha)  and CREM (cAMP responsive element modulator) . S6K2 activates transcription by binding to the transcription factors YY1 (yin yang 1)  and RORγ (RAR-related orphan receptor gamma) . Both S6K isoforms exist as several variants (Table 2). The S6K1 mRNA yields two differently translated variants originating from different translational start sites, p70-S6K1 and p85-S6K1 , the latter containing an N-terminal NLS. There is also a shorter S6K1 with a truncated KD originating from alternative splicing, termed p31-S6K1 in mice, and hS6K1-h6A and hS6K1-h6C in humans . The tissue expression of the S6K1 variants is unknown but within the cell p70-S6K1 localizes to the cytoplasm , p85-S6K1 to both cytoplasm and nucleus [148,149] and p31-S6K1 localizes to the nucleus . Both p70-S6K1 and p85-S6K1 have been shown to be targeted by mTORC1, as rapamycin treatment decreases both phosphorylation of p70-S6K1 at pThr389 and of p85-S6K1 at pThr412 . However, in cells arrested in prometaphase, p85-S6K1 is phosphorylated at pThr412 in an mTORC1-independent manner  and hence, p85-S6K1 might not be regulated by mTORC1 under all the circumstances. There are no conclusive data confirming if p31-S6K1 is targeted by mTORC1 . Further studies are needed to discriminate the biological functions of each of these variants. Alternative translation also occurs for Rps6kb2 (S6K2's gene), giving rise to two variants, p56-S6K2 and p54-S6K2, that differ by the presence of an NLS in p56-S6K2 in the NTD, but not in p54-S6K2 . Hence, they reside in different compartments, p56-S6K2 in the membranous fraction and p54-S6K2 in the soluble fraction . Further work is required to establish the exact subcellular localization and distinct functions of the S6K2 variants.
4E-BP, another direct mTORC1 substrate, exists in three isoforms (4E-BP1–3) (Table 1). All 4E-BP isoforms share an eIF4E (eukaryotic translation initiation factor 4E) binding domain (BD) and a TOS motif. In addition, 4E-BP1 and 2 contain an RAIP motif, named after its amino acid sequence. Whereas 4EB-P2 is ubiquitously expressed , 4E-BP1 and 4E-BP3 are primarily found in the pancreas and skeletal muscle [152,153]. In addition, 4E-BP1 is expressed in adipose tissue  and 4E-BP3 in the kidney . mTORC1 inhibits 4E-BP1 and 2 by phosphorylation at their RAIP motifs. Phosphorylated 4E-BP1 and 2 are released from eIF4E, and consequently eIF4E can bind the 5′ cap of mRNAs to initiate translation [43,154]. 4E-BP3 does not contain a RAIP motif and therefore is not inhibited by mTORC1, although mTORC1 phosphorylates 4E-BP3 at other sites . Interestingly, prolonged mTORC1 inhibition enhances 4E-BP3 expression at the transcriptional level, and this is mediated by the transcription factor TFE3 (transcription factor binding to Ig heavy constant μ enhancer 3) . In addition to inhibiting eIF4E binding to the cap, 4E-BP3 regulates eIF4E at the nucleus to regulate nuclear mRNA export .
mTORC1 regulates autophagy by phosphorylating ULK. ULK has four isoforms, ULK1–4, but only ULK1 and ULK2 are inhibited by mTORC1 (Table 1) . ULK1 and ULK2 are ubiquitously expressed  and both induce autophagy [157,158]. ULK1 has a stronger affinity than ULK2 towards the other members of the autophagy-inducing complex, Atg13 (autophagy related 13) and FIP200 . This suggests different functions of the ULK1/Atg13/FIP200 and ULK2/Atg13/FIP200 complexes during autophagy, but further studies are needed to test this hypothesis.
mTOR and ageing
Ageing can be defined as the gradual deterioration of the physiological functions necessary for survival . This concept relates both to the lifespan of an individual and to the manifestation of age-related disorders, such as obesity, diabetes and myopathy. In other words, increased longevity can reflect either an increase in lifespan or a reduction in age-related diseases. The role of mTOR in the ageing process has been a topic of research over the last decades, and we give here an overview of the key findings in invertebrate and vertebrate model organisms and humans.
TOR and lifespan in invertebrates
The process of ageing has been widely studied in model organisms such as the budding yeast S. cerevisiae, the nematode Caenorhabditis elegans and the fruitfly Drosophila melanogaster. Much research on ageing has been performed in these organisms due to their short lifespans, their easy manipulation and the availability of powerful genetic tools. Studies in S. cerevisiae have shown that inhibition of TORC1 with rapamycin increases the chronological lifespan (duration of time in which the cells in stationary phase remain viable) . In addition, the longevity phenotype induced by dietary restriction was found to be TOR dependent . By generating yeast deletion collections, several long-lived mutants were identified. Among them were strains with mutations in genes of the TOR signalling axis  and the TOR substrate Sch9/S6K , as well as transcription factors that up-regulate genes encoding amino acid biosynthetic enzymes and amino acid permeases .
Studies in C. elegans provide evidence that rapamycin  and dietary restriction  increase the lifespan also in multicellular organisms. CeTORC1 (TORC1 in C. elegans) inhibition by mutation or RNAi (RNA interference) of let-363 (mTOR) , daf-15 (raptor) , raga-1 (RagA GTPase) [162,166] or rheb-1 (Rheb)  causes longevity phenotypes (mammalian ortholog protein names indicated in brackets). Also mutations in the CeTOR upstream regulators daf-2 (InR)  and age-1 (PI3K)  dramatically extend the lifespan of C. elegans. In contrast to the increased longevity of worms in which let-363 (mTOR) or daf-15 (raptor) are targeted by RNAi or mutation, rict-1 (rictor) deficiency causes short-lived worms , suggesting that the two CeTOR complexes have opposite roles in regulating the lifespan of C. elegans.
Further evidence on the evolutionary conservation of TOR’s role in lifespan and ageing arose from the confirmation of these phenotypes in D. melanogaster. Rapamycin  or dietary restriction  increases lifespan in flies as well. Also inhibition of the insulin pathway by mutations in the InR gene  or in the InR substrate CHICO  results in increased longevity. Additionally, overexpression of Tsc1 (TSC1 in D. melanogaster) or Tsc2 (TSC2) or dominant-negative forms of dTOR (mTOR) and its substrate dS6K (S6K), causes lifespan extension . Furthermore, the TORC1 substrate d4E-BP (4E-BP) has a pivotal role delaying fly ageing . Hence, inhibition of the signalling axis converging on TOR and its substrates prolongs lifespan in non-vertebrate model organisms.
mTOR and ageing in mice
The before-mentioned studies provide strong evidence that signalling through insulin and TOR restricts the lifespan of invertebrates. Evidence that this mechanism is also conserved in mammals came from mouse studies in which rapamycin or dietary restriction increased lifespan [177,178]. However, dissecting the mechanisms underlying mTOR’s role in mammalian ageing proved to be even more challenging than in invertebrates. One reason is the higher complexity of the mammalian insulin-mTOR axis with several isoforms and often various splice variants at almost all levels of the pathway. As detailed earlier, tissue and subcellular distribution varies greatly for the different variants, leading to tissue-specific biological outcomes of mTOR. It is therefore not surprising that knockouts of the same gene in different tissues lead to divergent phenotypes .
Several knockout studies in mice suggest functions of components of the insulin-mTOR network in ageing (Table 3). Lifespan extension has been observed in a female mouse model with heterozygous whole body double knockout of the mTOR and mLST8 genes, Mtor and Mlst8  or with a homozygous whole body knockout of Irs1 . Also, a homozygous whole body knockout of Rps6kb1 (ribosomal S6 kinase 1), the S6K1 gene, extends lifespan in mice . Lifespan extension has been also reported for heterozygous knockouts of Irs2 in the whole body or in brain only , but the reproducibility of this phenotype has been questioned . In agreement with the studies in C. elegans , mTORC2-specific knockouts enhance ageing also in mice, as a whole body knockout of Rictor decreases lifespan .
This table highlights the phenotypes of knockout mouse models that display delayed or accelerated ageing phenotypes. Each group is subdivided depending on whether the phenotype relates to the lifespan of the mice or to their metabolic profiles. For each mouse model, the target gene is indicated. Whole body knockout was performed if not indicated otherwise. References are indicated in brackets. The colour code refers to the different signalling modules of the mTOR network as indicated in Figure 1. As Mlst8 and Mtor are part of both mTOR complexes, the knockout models are indicated for mTORC1 and mTORC2. A more detailed comparison of the phenotypes with knockouts of the complex-specific components Raptor and Rictor is provided in the text. LLess adipose tissue, GImpaired glucose tolerance, IInsulin resistance, OObesity, MMyopathy, TAppearance of tumours, ?Phenotype not reproduced.
Abbreviations: Ppdk1, 3-phosphoinositide-dependent kinase-1; Eif4ebp1, eukaryotic translation initiation factor 4E-binding protein 1; Insr, insulin receptor; Irs1, insulin receptor substrate 1; Irs2, insulin receptor substrate 2; Irs3, insulin receptor substrate 3; Mlst8, mammalian lethal SEC13 protein 8; Mtor, mammalian target of rapamycin; Rictor, rapamycin-insensitive companion of mTOR; Rptr, regulatory associated protein of mTORC1; Rps6kb1, ribosomal S6 kinase 1; Tsc1, tuberous sclerosis complex protein 1; Tsc2, tuberous sclerosis complex protein 1.
Longevity studies in mice are relatively rare due to their comparatively long lifespan. In contrast, studies that focus on the role of the mTOR network in age-related, metabolic phenotypes and diseases are more common (Table 3). Adipose tissue-specific knockouts of the InR gene Insr  or the raptor gene Rptr  result in mice with substantially less fat that are protected against obesity and hypercholesterolaemia. In addition, knockout of either Tsc1 or Tsc2 enhances tumour formation [188,189]. This suggests that inhibition of the insulin-mTORC1 axis protects higher organisms against age-related metabolic and tumour disorders. However, this concept is challenged by the fact that most other knockout models of the insulin-mTOR network in mice develop phenotypes that positively link with age-related diseases and hence, could be considered as phenotypes of accelerated ageing (Table 3). Such phenotypes encompass, for example impaired glucose tolerance, insulin resistance, obesity and myopathy, which correlate with increased age in mice and men [190–192]. Impaired glucose tolerance has been observed in whole body knockouts of Irs4  and Akt2 , in tissue specific knockouts of the Insr in muscle [194 ], β cells  and brain  or of the PDK1 gene, Pdpk1, in the liver . Insulin resistance occurs in mice lacking Irs2  or Akt2 , and in tissue-specific knockouts of the Insr in liver  or brain . Obesity is also observed for whole body Irs2 knockout mice , as well as for specific knockouts of the Insr in muscle  or brain . Myopathy is a characteristic phenotype of mice with muscle-specific knockouts of Pdpk1 , Mtor , Rptr  and in a double knockout model of Rps6kb1 and Rps6kb2 .
The observation that so many mouse knockout models of the insulin-mTOR network develop metabolic phenotypes which can be linked with age-related metabolic impairment may seem at odds with the studies in invertebrates where a large majority of insulin-TOR network mutants display a prolonged lifespan. However, metabolic phenotypes in mouse studies must be interpreted with caution regarding their relationship with lifespan and ageing. The reason is that it is not possible to discriminate whether a detrimental metabolic phenotype in a mouse knockout model is due to an age-inhibitory role of the targeted gene or due to a potentially essential role of this gene in metabolic processes. A well-known example of such a seemingly contradictory phenotype has been observed for rapamycin in mice. Rapamycin does increase lifespan via mTORC1 inhibition but, when chronically administered, rapamycin causes secondary effects leading to mTORC2 inhibition and substantial impairment of glucose tolerance and insulin action . Hence, rapamycin extends lifespan and severely impairs metabolism at the same time via distinct mechanisms. The fact that lifespan extension can be observed concomitantly with metabolic impairment suggests that metabolic alterations and lifespan are not always strictly causally linked.
The notion that lifespan and metabolic outcomes of genes can be separated is further strengthened while taking a closer look at isoform-specific effects of genes in the insulin-mTOR axis on ageing and metabolic phenotypes in mice (Table 3). For example, different IRS isoforms govern metabolism and ageing in distinct, often opposite ways. A homozygous whole body knockout of Irs1 prolongs lifespan in female mice , although they develop insulin resistance . In contrast, a whole body knockout of Irs2 shortens the lifespan of both male and female mice , and leads to a diabetic phenotype [198,205]. Irs4 knockout mice have a milder phenotype than Irs1 or Irs2 knockouts, with regard to insulin sensitivity defects , and no effects on lifespan have been reported. Distinct phenotypes are also observed for knockouts of the different Akt isoforms. Mice with heterozygous Akt1 knockout display a prolonged lifespan . In contrast, Akt2-deficient mice are insulin resistant with elevated plasma triglycerides and diabetes in males . An Akt3 knockout does not seem to have an ageing-related metabolic phenotype . Downstream of mTORC1, a whole body knockout of Rps6kb1 prolongs lifespan , whereas a Rps6kb2 knockout shows no obvious phenotypic abnormalities . More extensive characterization of the Rps6kb2 knockout model would allow for better understanding of S6K2’s potential role in ageing. Mice with a knockout of Eif4ebp1, 4E-BP1’s gene, exhibit no difference in lifespan, although they present an increased metabolic rate and a reduction in adipose tissue , again questioning the strict causal relationship between beneficial metabolic features and lifespan extension. The role of 4E-BP2 has only been studied by double knockout of Eif4ebp1 and Eif4ebp2, and these mice are obese and insulin resistant . However, no information about the lifespan of this model is available and no 4E-BP3 knockout has been so far reported.
mTOR and ageing in humans
Most research on mTOR in humans focuses on age-related diseases such as cancer and diabetes. Regarding ageing itself, studies in long-lived primates, specifically in Rhesus monkeys, have shown that calorie restriction delays disease onset and possibly mortality . The effect of rapamycin on primate lifespan has not yet been reported, but rapamycin improves immune function in elderly humans . Given that intervention studies on longevity and ageing in higher primates and humans are scarce, candidate gene and genome-wide association studies (GWAS) are the main tools to understand the relationship between genetic makeup and human lifespan and disease susceptibility. Such studies use cohorts of advanced age and focus on genetic factors that correlate with exceptional longevity and healthy ageing.
GWAS in long-lived humans have so far yielded only very few gene associations that correlate with ageing. Indeed, only associations of ApoE (apolipoprotein E) and FoxO3a genes have been replicated in several studies . It is surprising that components of the insulin-mTOR network have not been identified in these studies. Possibly the GWAS does not have enough power to detect such correlations, as this approach only allows the detection of common genetic variations. Development of new techniques such as whole-genome sequencing permits the detection of rare potentially functional genetic variants and raises much hope for the detection of further genetic correlations with lifespan and age-related processes. In a recent study, whole-genome sequencing was used to analyse human healthy ageing, defined as disease-free ageing without medical intervention . However, no major genetic contributors to healthy ageing could be identified.
In contrast with these unbiased approaches, analyses of specific sets of candidates in the insulin-mTOR signalling network to unravel their potential roles in ageing have yielded more success. A study that included 122 Japanese ‘semisupercentenarians’ (older than 105 years) found polymorphic variations of the InR and IRS1 genes (INSR and IRS1) that are more frequent than in the control group . In addition, a polymorphism in Akt1 that significantly associates with lifespan has been found in three independent Caucasian cohorts . In the Leiden Longevity Study, gene expression analysis of nonagenarians show that expression of EIF4EBP1 and AKT1S1 (4E-BP and PRAS40's genes), two negative effectors of the mTORC1 pathway, is higher in the aged group compared with the middle-aged control group . Moreover, raptor is expressed at a lower level in middle-aged members of the longevity families as compared with similarly aged controls . Finally, low insulin signalling has been associated with improved old-age survival in women . Of note, these targeted studies do not address isoforms or splice variants of the mTOR signalling network and, therefore, some relevant candidates may have been overlooked.
mTOR signalling is widely recognized as a key element in ageing and age-related metabolic conditions in a wide range of organisms from yeast to rodents [160,162,171,177]. In invertebrates, inhibition of the insulin-TOR network by mutations or RNA interference extends lifespan, suggesting a positive link between TOR activity and ageing progression. While studying the same genes in higher organisms such as mammals, the relationship between mTOR and ageing becomes more complex. Although there are some clear examples of inhibition of the mTOR pathway that lead to lifespan extension, most knockouts result in the development of metabolic conditions that could rather be considered as a sign of accelerated ageing. A limitation is that most of these mice studies only analyse metabolic parameters and not lifespan, and conclusions cannot be drawn from metabolic phenotypes on shortened or prolonged lifespan. The concomitant occurrence of prolonged lifespan and detrimental metabolic phenotypes or beneficial metabolic features with no lifespan effect in some of these models challenge the idea of a strict and direct relationship between metabolic alterations and lifespan in knockout models of genes of the insulin-mTOR axis.
This review emphasizes the need to identify and characterize the isoforms and splice variants across the mTOR network to achieve a better understanding of the contribution of these different elements to metabolism and ageing and the interrelationship of both. Although many of these isoforms have been identified long ago, they have been considered as proteins with overlapping function for a long period. This is reflected by the fact that isoforms with the number 1 in their names are often better studied than their counterparts with higher numbers. Hence, the extent of knowledge on these variants often relates to their arbitrary numbering in databases.
For splice variants, we lack even more knowledge as they are ignored by most experimental studies even though the majority of the genes that code for mTOR network components have been predicted to produce several splice variants. Furthermore, RNA splicing is required for longevity downstream of dietary restriction and the CeTORC1 pathway in C. elegans . The gap in our knowledge becomes even more apparent when considering that additional, abnormal splice variants occur in many genetic diseases and cancers . A recent global study on the interactomes of splice variants has demonstrated that splice variants share only half of their interaction partners and have distinct tissue expression and, therefore, should be considered as distinct proteins . Thus, further work is required to experimentally identify and functionally characterize both naturally occurring and disease-causing variants in the mTOR network, and to better understand the relationship of these genes and their splice products with metabolic regulation, ageing and lifespan, and age-related diseases.
mTOR inhibition prolongs lifespan and delays the appearance of age-related metabolic diseases, suggesting that mTOR is a major determinant of ageing and metabolic balance.
The mTOR network in higher organisms harbours many different isoforms and splice variants, which exert complex effects on lifespan and metabolism.
A comprehensive analysis of mouse knockout studies targeting these variants does not support a tight causal relationship between an improved metabolic phenotype and lifespan prolongation.
We thank Dr. Mark Nellist for his expert opinion on TSC2 variants. We also thank Alexander Martin Heberle, Dr. Laura Corbett, Dr. Antje Thien, Marti Cadena Sandoval and Ulrike Bosch for critical reading of this manuscript. The authors sincerely apologize to those whose work could not be cited in this review because of space limitations.
The authors declare that there are no competing interests associated with the manuscript.
K.T. is supported by the BMBF (Bundesministerium für Bildung und Forschung) e:Med project GlioPATH [grant number 01ZX1402B]; the BMBF e:Med project MAPTor-NET [031A426B (to K.T.)]; the Rosalind–Franklin Fellowship program of the University of Groningen; the Ubbo Emmius Fund; the Stichting Michelle (calls 2015 and 2017); and the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) - [grant number TH 1358/3-1].
P.R.N. and K.T. wrote the manuscript.
protein kinase A/protein kinase G/protein kinase C
AMP-activated protein kinase
autophagy related 13
cell division cycle 16
cell division cycle 42
cAMP responsive element modulator
dishevelled, Egl-10 and pleckstrin domain
DEP domain containing mTOR-interacting protein
eukaryotic elongation factor-2 kinase
eukaryotic translation initiation factor 4B
oestrogen receptor α
FAT-carboxy terminal domain
FAK family kinase-interacting protein of 200 kDa
Forkhead box O
glucose transporter type 4
growth factor receptor-bound protein 2
glycogen synthase kinase 3
genome-wide association study
Huntingtin-elongation factor 3-regulatory subunit A of PP2A-TOR1 repeat
insulin like growth factor 1
insulin like growth factor 1 receptor
insulin receptor substrate
kinase regulatory loop binding
mitogen-activated protein kinase
mammalian lethal with SEC13 protein 8
mammalian stress-activated protein kinase interacting protein 1
mammalian target of rapamycin
mTOR complex 1
nuclear localization sequence
PH domain and leucine-rich repeat protein phosphatase
protein kinase C
proline-rich Akt substrate of 40 kDa
protein observed with rictor 1/proline-rich protein 5
phosphatase and tensin homologue
ras-related C3 botulinum toxin substrate 1
ras-related GTP-binding protein
regulatory associated protein of mTORC1
retinoic acid receptor
ras-homologue enriched in brain
rapamycin-insensitive companion of mTOR
RAR-related orphan receptor γ
ribosomal S6 kinase
ribosomal protein S6
ribosomal protein S6 kinase
serum and glucocorticoid-regulated kinase
TBC 1 domain family member 7
telomere maintenance 2
target of rapamycin
TOR signalling sequence
tuberous sclerosis complex
TRAF and TNF receptor-associated protein
unc-51 like autophagy activating kinase
eukaryotic translation initiation factor 4E-binding protein