\r\n\tMass spectrometry has a significant potential to further advance the clinical analysis and be a diagnostic tool for pathology or being used in the operation room and enable fast detection of tumor's margin.
\r\n\tThis book aims to provide an insight into different application fields of mass spectrometry for clinical research and routine. Modern mass spectrometry enables analysis and quantitation of very complex samples such as serum or tissue proteome or lipidome and the mass spectrometry imaging complements tissue analysis in pathology departments.
\r\n\tFurthermore, mass spectrometry can be used in an operating room as a device for on-site detection of tumor margins and be a valuable help for the surgeon.
\r\n\tMass spectrometry imaging of tissue is an advanced type of tissue analysis that complements and aids the pathology and the traditional tissue analysis by staining and optical microscopy.
The mammalian target of rapamycin (mTOR) has attracted growing attention during the past decade due to the increase realization of it’s extraordinarily significance in cellular life-sustaining activity on the one hand, and because of its crucial role in a variety of diseases, (including cancer, hamartoma syndromes, cardiac hypertrophy, diabetes and obesity) on the other. mTOR is an atypical serine/threonine protein kinase, belonging to the phosphatidylinositol kinase-related kinase (PIKK) family. Cumulative evidence indicates that mTOR acts as a ‘master switch’ of cellular energy-intensive anabolic processes and energy-producing catabolic activities. It coordinates the rate of cell growth, proliferation and survival in response to extracellular mitogen, energy, nutrient and stress signals [1, 2]. mTOR functions within two distinct multiprotein complexes, mTORC1 and mTORC2, responsible for the different physiological functions. Thus, mTORC1 is considered mostly involved in the regulation of the translation initiation machinery influencing cell growth, proliferation, and survival, while mTORC2 participates in actin cytoskeleton rearrangements and cell survival. mTORC1 and mTORC2 were initially identified in yeast on the basis of their differential sensitivity to the inhibitory effects of rapamycin, mTORC1 being originally considered as rapamycin-sensitive and mTORC2 as rapamycin-insensitive [3-5].
The history of TOR began in the early 1970s when a bacterial strain, Streptomyces hygroscopicus, was first isolated from Rapa Nui island during a discovery program for anti-microbial agents. These bacteria secrete a potent anti-fungal macrolide that was named rapamycin after the location of its discovery [6-9]. Later rapamycin was proven to have anti-proliferative and immunosuppressive properties. In the beginning of 1990s, two rapamycin target genes titled TOR1 (the target of rapamycin 1) and TOR2 were discovered through the yeast genetic screens for mutations that counteract the growth inhibitory properties of rapamycin [10, 11]. Further studies revealed that rapamycin forms the complex with its intracellular receptor, FK506-binding protein 12 kDa (FKBP12), This complex binds a region in the C-terminus of TOR kinase named FRB (FKB12-rapamycin binding) domain, what leads to the inhibition of TOR functions [12-14].
At present, it becomes clear that mTORC1 and mTORC2 activities are mediated through diverse signaling pathways depending on the type of extracellular signal. Thus, signaling from growth factors is mediated predominantly through PI3K-Akt-TSC1/2 pathway and upregulates mTORC1 to stimulate translation initiation, while energy or nutrient depletion and stresses suppress mTORC1 via LKB1–AMPK cascade to trigger off the process of autophagy. In contrast, mTORC2 is insensitive to nutrients or energy conditions. mTORC2 phosphorylates Akt and some other protein kinases regulating actin cytoskeleton and cell survival in response to growth factors and hormones. The physiological functions of mTOR continue to expand. It should be stressed, that the signaling throughout the complicated mTOR network, including branched pathways and feedback loops, is regulated predominantly by phosphorylation and includes myriads of phosphorylation events. Moreover, the complexity of mTOR regulation is amplified by the crosstalk with other signaling pathways, such as MAP kinase- or TNFα-dependent cascades, which activity is also determined by vast number of phosphorylations. The complication of mTOR signaling additionally increases due to the hierarchical character of multiple site-specific phosphorylations of the main mTOR targets. Up to date there are no full clarity, concerning which kinase is responsible for each site phosphorylation as well as functional role and precise mechanisms of each phosphorylation event. The better understanding of underlying molecular mechanisms is now especially essential since inhibitors of mTOR signaling are widely used as drugs in the therapy of cancer and neurodegenerative diseases.
Although mTOR has limited sequence similarities in eukaryotes, it demonstrates a high level of conservation in its key cellular functions. mTOR, also known as FRAP (FKBP12-rapamycin-associated protein), RAFT1 (rapamycin and FKBP12 target), RAPT 1 (rapamycin target 1), or SEP (sirolimus effector protein), is a large 289 kDa atypical serine/threonine (S/T) kinase [15-18] and is considered a member of the phosphatidylinositol 3-kinase (PI3K)-kinase-related kinase (PIKK) superfamily since its C-terminus shares strong homology to the catalytic domain of PI3K [19, 20]. mTOR and yeast TOR proteins share > 65% identity in carboxy-terminal catalytic domains and about 40% identity in overall sequence . At the amino-acid level, human, mouse and rat TOR proteins share a 95% identity [22, 23]. The knockout of mTOR in mice is embryonic lethal, indicating its physiological importance [24, 25].
Structurally, mTOR contains 2549 amino acids and the region of first 1200 N-terminal amino acids contains up to 20 tandem repeated HEAT (a protein-protein interaction structure of two tandem anti-parallel α-helices found in Huntingtin, Elongation factor 3 (EF3), PR65/A subunit of protein phosphatase 2A (PP2A), and TOR) motifs . Tandem HEAT repeats are present in many proteins and may form an extended superhelical structure responsible for protein-protein interactions. HEAT repeats region is followed by a FAT (FRAP, ATM, and TRRAP (PIKK family members)) domain and FRB (FKPB12-rapamycin binding domain), which serves as a docking site for the rapamycin -FKBP12 complex formation. Downstream lies a catalytic kinase domain and a FATC (FAT Carboxyterminal) domain, located at the C-terminus of the protein (Figure 1A). The FAT and FATC domains are always found in combination, so it has been hypothesized that the interactions between FAT and FATC might contribute to the catalytic kinase activity of mTOR via unknown mechanisms [26, 27].
Up to date quite a few phosphorylation sites in mTOR have been reported, namely T2446, S2448, S2481 and S1261 and this list will be probably appended. S2481 is considered to be a site of autophosphorylation . S2481 is the only site the phosphorylation of which is well established for regulating mTOR intrinsic activity [29, 30]; the significance of other phosphorylation sites for mTOR activity are not entirely clear. Recently, S1261 has been reported as a novel mTOR phosphorylation site in mammalian cells and the first evidences of this phosphorylation in regulating mTORC1 autokinase activity has been provided . Although phosphorylation at T2446/S2448 was shown to be PI3K/Akt-dependent, mTORC1 downstream kinase S6K has been also reported to phosphorylate these two sites . The significance of this potential feedback loop is unknown, as it is not yet clear whether and how these phosphorylations influence mTOR activity
Binding of rapamycin–FKBP12 to the FRB domain of mTOR disrupts the association of mTOR with mTORC1-specific component Raptor and thus divide mTORC1 from its targets, blocking mTORC1 signaling. However, whether rapamycin directly inhibits mTOR’s intrinsic kinase activity is still not entirely clear [3, 33, 34].
The TOR complexes mTORC1 and mTORC2
The mammalian mTORC1 and mTORC2 complexes perform non-overlapping functions within the cell. Thus, the best-known function of TORC1 signaling is the promotion of translation. Other mTORC1 functions include autophagy inhibition, promotion of the ribosome biogenesis and of the tRNA production. The main known mTORC2 activity is the phosphorylation and activation of AKT and of the related kinases — serum/glucocorticoid regulated kinase (SGK) and protein kinase C (PKC) . It is also involved in cytoskeletal organization. Although both mTOR complexes exist predominantly in the cytoplasm, some data indicate that they could function in different compartments. Thus, upon nutrients and energy availability mTORC1 is recruited to lysosomes where it could be fully activated  and where it functions to suppress autophagy. Unlike mTORC1, mTORC2 according to the most recent data localizes predominantly in ER compartment where it could directly associate with ribosomes [37, 38]. Additionally, some data evidence that mTOR may actually be a cytoplasmic-nuclear shuttling protein. The nuclear shuttling could facilitate the phosphorylation of mTORC1 substrates under the mitogenic stimulation . The unique compositions of mTORC1 and mTORC2 determine the selectivity of their binding partners. Up to date we know more about mTORC1 rather then mTORC2 probably due to the lack of available and wide-spreaded inhibitors of mTORC2 activity.
TORC1 composition. Within the mammalian cells, TORC1 functions as a homodimer. Each monomer consists of mTOR, regulatory associated protein of mTOR (Raptor), proline-rich AKT substrate 40 kDa (PRAS40), DEP domain TOR-binding protein (Deptor) and mammalian lethal with Sec-13 protein 8 (mLST8, also known as GbL) [40, 41](Figure 1B).
Raptor is a 150 kDa presumably non-enzymatic subunit of mTORC1 that is essential for the kinase mTORC1 activity in vitro and in vivo in response to insulin, nutrient and energy level. [42, 43]. It includes a highly conserved N-terminal region followed by 3 HEAT repeats and 7 WD40 (about 40 amino acids with conserved W and D forming four anti-parallel beta strands) repeats. The Raptor-mTOR interaction is very dynamic, and is thought to require the HEAT repeats of mTOR. It is established that Raptor is indispensable for mTOR to phosphorylate its main effectors p70S6 kinase (S6K1) and eukaryotic initiation factor 4E (eIF4E) binding protein 1 (4E-BP1), but whether Raptor positively or negatively regulates mTOR itself remains controversial . Raptor is essential for mTORC1 complex formation and for the dimerization of TORC1 complexes as it provides direct interaction between TOR proteins from each monomer. Thus it can be considered to be a scaffolding protein that recruits substrates for mTOR thereby demonstrating a stimulating effect on mTOR activity . Alternatively, other study has demonstrated that Raptor negatively regulates mTOR being tightly bound to the kinase . There are also a hypothesis according to which at least two types of interaction exist between Raptor and mTOR depending on nutrients availability. One mTOR-Raptor complex that forms in the absence of nutrients is stable and leads to a repression of the mTOR catalytic activity. In contrast, the other complex that forms under nutrients-rich conditions is unstable, but it is important for in vivo mTOR function  (reviewed in ). Recent studies suggested that the phosphorylation status of Raptor could influence mTORC1 activity . Phosphorylation on S722/792 is mediated by AMPK (AMP-activated protein kinase) and is required for the inhibition of mTORC1 activity induced by energy stress , whereas phosphorylation of Raptor on S719/721/722 is mediated by the p90 ribosomal S6 kinases (RSKs) and contributes to the activation of mTORC1 by mitogen stimulation [45, 46]. Most recently, S863 in Raptor was identified as mTOR-mediated phosphorylation site responsible for the insulin-dependent activation of mTORC1 .
PRAS40, another subunit of mTORC1, has been defined as a direct negative regulator of mTORC1 function . Initially, PRAS40 was identified as a novel substrate of Akt being directly phosphorylated at T246 near its C-terminus . This phosphorylation releases inhibition of mTORC1 by PRAS40. Subsequent studies showed that PRAS40 associates with mTORC1 via Raptor and inhibits mTORC1 activity . A putative TOR signaling motif, FVMDE, has been identified in PRAS40 and shown to be required for interaction with Raptor. Upon binding to Raptor, PRAS40 is phosphorylated on S183 by mTORC1 both in vivo and in vitro  Thus, PRAS40 has been implicated as a physiological substrate of mTORC1. Most recently, two novel sites in PRAS40 phosphorylated by mTORC1, S212 and S221, have been identified . Rapamycin treatment reduced the phosphorylation of S183 and S221 but not S212, indicating that besides mTORC1, other kinases may also regulate the phosphorylation of S212 in vivo .
mLST8 has been identified after Raptor as a stable component of both mTOR complexes . It consists almost entirely of seven “sticky” WD40 repeats, and has been initially shown to bind to the kinase domain of mTOR, leading to the hypothesis that mLST8 positively regulates mTOR kinase activity. It was proposed that mLST8 is essential for a nutrient- and rapamycin-sensitive interaction between Raptor and mTOR . However, there is no substantial evidence to support this idea. It has been speculated, that mLST8 may participate in the amino acids mediated activation of TORC1 being insignificant for other mechanisms of TORC1 activation . Alternatively, recent studies demonstrated functional importance of mLST8 for the Rictor-mTOR interaction, evidencing that mLST8 is involved in mTORC2 rather than in mTORC1 activity.
Deptor binds to mTOR at the FAT domain thus originally proposed to be a part of TORC1. Recently it has been identified as mTOR inhibitor that acts on both TORC1 and TORC2. The upstream regulators of Deptor still remain unknown .
mTORC2 composition and distinctions from mTORC1
In 2004, mTORC2, containing mTOR, mLST8 and Rictor was identified [3, 4]. Since mTORC2 complex was discovered later than mTORC1, its functions and regulatory mechanisms are less understood . TORC2 and TORC1 contain common subunits, as is mTOR itself, mLST8 and Deptor, but instead of Raptor, mTORC2 includes two different subunits, Rictor (rapamycin-insensitive companion of mTOR) and mSin1 (mammalian stress-activated protein kinase (SAPK)-interacting protein 1) [3, 4, 53]. In addition, Protor (protein observed with Rictor) was also considered a component of mTORC2 (Figure 1B) [54, 55]. mTORC2 was originally thought to be rapamycin-insensitive , however, further studies demonstrated that prolonged rapamycin treatment inhibits the assembly of mTORC2 as well as its activity towards Akt phosphorylation in certain cell lines .
Rictor is the first identified TORC2 specific component [3, 4]. It represents a large protein with a predicted molecular weight of about 200 kDa. Although Rictor contains no apparent catalytic domain motifs , knockdown of Rictor results in the loss of actin polymerization and cell spreading, the main known mTORC2 functions . It was shown that the Rictor-mTOR complex does not affect the mTORC1 effectors S6K1 and 4E-BP1, but influence the activities of several proteins known as mTORC2 downstream targets, including phosphorylation of Akt, PKC and the focal adhesion proteins.
mSin1 was recently identified as a novel component of mTORC2, which is important for both the complex assembly and function [57-59]. Sin1 is conserved among all eukaryotic species especially in the middle part of the sequence . A Ras-binding domain and a C-terminal PH domain have been identified recently . The several experimental techniques showed the importance of Sin1 for mTORC2 function . The interaction in vivo between Sin1 and Rictor is more stable than their interactions with mTOR probably due to the ability of Sin1 and Rictor to stabilize each other . Thus knockdown of Sin1 decreases the interaction between mTOR and Rictor, suggesting that Sin1 is important for mTORC2 assembly. Knockdown of Sin1 by RNAi in Drosophila and mammals crucially diminishes the Akt phosphorylation on S473 in vitro. The same effect was observed in Sin1−/− cells .
Protor-1 and Protor-2 (also known as Proline rich protein 5 (PRR5) [54, 55] and PRR5-like (PRR5L)  are two newly identified mTORC2 interactors which have been identified as Rictor-binding or SIN1 binding proteins . Up to date their functions remain unclear. It is currently accepted that they are dispensable for mTORC2 assembly as well as for its catalytic activity , although Protor stability is dependent on the production of other TORC2 components. It is possible that Rictor and Sin1 act as scaffold proteins for various complexes involving different kinases.
mLST8 and Deptor, as was mentioned above, are the components of both mTORC1 and mTORC2 complexes.
Although this pathway is still considered to be the main way exerting multi-faceted control over mTORC1 activity which sense insulin and growth factors signals to regulate cell growth, at present it becomes clear that at least some of its components also function to mediate responses on other stimulus, such as energy, stress or nutrients which are provided by discrepant signaling pathways, described below.
The TSC1/TSC2 complex (tuberous sclerosis complex 1/2, TSC1/2) has been established as the major upstream inhibitory regulator of mTORC1 [64, 65]. This complex mediates signals from a large number of distinct signaling pathways to modulate mTORC1 activity predominantly via different phosphorylations of TSC2. Functioning as a molecular switch, TSC1/2 suppresses mTOR’s activity to restrict cell growth during the stress, and releases its inhibition under the favorable conditions. The TSC1 and TSC2 genes were identified in 1997 and 1993 respectively as the tumor-suppressor genes mutated in the tumor syndrome TSC 1(tuberous sclerosis complex) [66-68]. TSC is a multisystem disorder characterized by the development of numerous benign tumors (e.g. hamartomas) most commonly detected at the brain, kidneys, skin, heart and lungs. Genetic studies of TSC1 and TSC2 in humans, mice, Drosophila and yeast strongly suggest that these proteins act mainly as a complex. The 140 kDa TSC1 (also known as hamartin) and 200 kDa TSC2 (also known as tuberin) proteins share no homology with each other and very little with other proteins (Figure 1C) TSC1 and TSC2 associate through certain regions  giving a heterodimeric complex. The only known functional domain throughout these two proteins is a region of homology at the C-terminus of TSC2 to the GAP domain of small G-protein Rap1. Searches for a GTPase target regulated by the TSC2 GAP (GTPase-activating protein) domain revealed the small G-protein Rheb. Mammalian TSC2 was shown to accelerate the rate of GTP hydrolysis of Rheb, converting Rheb from the active GTP-bound to the inactive GDP-bound state [69, 70]. This evidences that Rheb is a direct target of TSC2 GAP activity, and TSC2 suppress Rheb function. While the GAP activity of TSC2 is necessary for the complex functionality, TSC1 is required to stabilize TSC2 and prevent its ubiquitin-mediated degradation [71, 72]. Under growth conditions, the TSC1/2 complex is inactive, thereby allowing Rheb-GTP to activate TORC1.
Rheb is a member of the Ras superfamily that appears to be conserved in all eukaryotes and, despite the term ‘brain’ in its name, is in fact ubiquitously expressed in mammals. Whether a GEF protein (guanine-nucleotide exchange factor responsible for reverse process, i.e. change GDP-bound to GTP-bound state) for Rheb exists remains unknown. Several evidences demonstrate that Rheb positively regulates mTORC1. In particular, Rheb overexpression stimulates S6K1 and 4EBP1 phosphorylation, which are indicators of mTORC1 activity. This effect can be reversed by mTOR inactivation or by rapamycin treatment, suggesting that Rheb primarily functions through TORC1 . Although genetic and biochemical studies strongly suggest that GTP-bound Rheb potently activates mTORC1, the molecular mechanism is still unclear. Overexpressed Rheb was shown to bind to mTOR [73, 74]. Associations between endogenous Rheb and mTORC1 components have not been reported. In general, Ras-related small G-proteins bind to their downstream effectors mostly in the GTP-bound state. Surprisingly, Rheb has been found to bind stronger to mTOR in its GDP-bound or nucleotide-free states . At the same time it has been shown that GTP-bound Rheb rather than the GDP-bound stimulates mTOR kinase activity in vitro . Although the mechanism by which Rheb-GTP activates mTORC1 has not been fully understood, it needs Rheb farnesylation and can be blocked by farnesyl transferase (FT) inhibitors. Recently, it was found that Rheb can directly interact with the FKBP12 homologue FKBP38 (named also FKBP8), and this binding seems to be tighter with Rheb-GTP . That study suggests that Rheb-GTP binds to FKBP38 and triggers its release from mTORC1, stimulating mTORC1 activity (Figure 2). In support of this model, an independent study carried out that decreasing FKBP38 expression with antisense oligonucleotides blocked the growth inhibitory effects of TSC1–TSC2 overexpression . Although more studies are needed, these findings suggest that FKBP38 might be a Rheb effector that regulates mTORC1 and, perhaps, unknown targets downstream of the TSC1/TSC2 complex and Rheb.
The responsiveness of mTORC1 signaling to growth factors and insulin is provided through activation of PI3K (phosphatidylinositol-3-kinase) and Akt kinase, but the precise mechanism is still not clear. Through PI3K signaling, Akt also termed PKB (serine/threonine protein kinase B) is activated by most growth factors to phosphorylate several downstream substrates .
PI3K is a heterodimeric protein containing an 85-kDa regulatory and a 110-kDa catalytic subunits (PIK3CA) [78, 79]. PI3K acts to phosphorylate a number of membrane phospholipids to form the lipid second messengers phosphatidylinositol 3,4-bisphosphate (PtdIns(3,4)P2 or PIP2) and phosphatidylinositol 3,4,5-trisphosphate (PtdIns(3,4,5)P3 or PIP3). In response to the upstream inputs, PI3K at the cell membrane is activated through the association of a ligand with its receptor, stimulating p85 to bind phosphorylated tyrosine residues of Src-homology 2 (SH2) domain on the receptor. This association promotes the p110 catalytic subunit to transfer phosphate groups to the membrane phospholipids [78, 80]. Consequently these lipids, particularly PtdIns(3,4,5)P3, attract several kinases to the plasmalemma initiating the signaling cascade [78, 80]. PIP3 accumulation is antagonized by the well-known tumor suppressor, lipid phosphatase PTEN (phosphatase and tensin homolog deleted on chromosome 10), which converts PIP3 to PIP2. One important function of PIP3 is to recruit Akt as well as PDK1 (or PDPK1, 3-phosphoinositide-dependent protein kinase-1)  via their PH (pleckstrin homology) domains to the plasma membrane (Figure 2).
Akt, known as one of the major survival kinases, belongs to the AGC (PKA/PKG/PKC) protein kinase family and is involved in regulating a vast number of cellular processes, including transcription, proliferation, migration, growth, apoptosis and various metabolic processes [3, 82]. Being translocated to the plasma membrane, Akt undergoes partial activation through the phosphorylation of T308 residue within the activation loop by PDK1 and following full activation through the additional phosphorylation at the hydrophobic motif site S473 by PDK2 . After activation Akt quits the cell membrane to phosphorylate intracellular substrates. Particularly, Akt can translocate to the nucleus  where it influences the activity of transcriptional factors, including CREB (cAMP response element-binding), E2F (eukaryotic transcription factor 2), NF-κB (nuclear factor kappa from B cells) through Iκ-K (inhibitor kappa B protein kinase), the forkhead transcription factors, in particular, FOXO1 and FOXO3 and murine double minute 2 (MDM2) which regulates p53 activity [84, 85]. In addition, Akt is able to target some other molecules to influence cell survival including GSK-3β (glycogen-synthase kinase-3β), which regulates β-catenin protein stability, and BAD (the pro-apoptotic molecule Bcl-2-associated death promoter).
Akt was the first kinase demonstrated to phosphorylate directly the TSC1/TSC2 complex in response to growth factors. Human TSC2 contains five predicted Akt sites (S939, S981, S1130, S1132 and T1462 on full-length human TSC2), all of which have been suggested to be subjects of phosphorylation by Akt (Figure 2). Importantly, the two sites were shown definitively to be targeted by Akt in mammalian cells, S939 and T1462 . There is also evidence that either S1130 or S1132 is phosphorylated by Akt in vivo . Finally, Akt can phosphorylate a peptide corresponding to the sequence surrounding S981 in vitro . This residue has been identified as an in vivo phosphorylation site on TSC2 by tandem-MS analyses . However, whether Akt phosphorylates S981 on full-length TSC2 within cells has not been conclusively demonstrated.
The majority of studies postulated that activated AKT promotes TORC1 signaling by phosphorylating multiple sites on TSC2, thereby relieving inhibition of Rheb and activating TORC1 [86, 87, 90, 91]. The data obtained using phosphorylation-site mutants of TSC2 demonstrate that Akt mediated phosphorylation of these sites inhibits the function of the TSC1–TSC2 complex in cells, however the molecular mechanism of this inhibition has been the subject of much debate (reviewed in ). One proposed mechanism involves disruption of the TSC1–TSC2 complex. However, this does not occur rapidly and, although it might contribute to the long-term effects of Akt on mTORC1 signaling, it cannot explain the immediate effects of Akt activation on mTORC1, which are blocked by Akt phosphorylation-site mutants of TSC2. Another proposed mechanism is based on the possibility that phosphorylation of TSC2 alters its subcellular localization, such that it can no longer act as a GAP for Rheb. One study supporting this mechanism found that growth factor stimulation led to increase of the TSC2 levels within the cytosolic fraction . This effect was PI3K-dependent, stimulated by activated Akt and required both S939 and S981 on TSC2. In that study, both TSC1 and Rheb were found exclusively in the membrane fraction, and unlike TSC2, did not show an increase in the cytosolic fraction following growth-factor stimulation. From these findings it was concluded that Akt-mediated phosphorylation of TSC2 on S939 or S981 inhibits the TSC1/TSC2 complex by triggering release of TSC2 from TSC1 at an intracellular membrane also occupied by Rheb. This model points on the significant and rapid dissociation of TSC2 from TSC1 upon phosphorylation – something that has not been detected in the majority of studies to date. Recent studies have suggested that AKT mediated phosphorylation of TSC2 at S939 and S981 creates a binding site for a cytosolic anchor protein, 14-3-3 (tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein, theta polypeptide also known as YWHAQ, 1C5; HS1), a mechanism of regulation shared by several other Akt substrates . Examining interactions between endogenous 14-3-3 proteins and TSC2, another study found that S939 and T1462 were both required for 14-3-3 binding to TSC2 downstream of PI3K signaling. It seems likely that 14-3-3 binding to TSC2 (provided by some combination of phosphorylated S939, S981 and T1462) contributes to Akt-mediated inhibition of TSC2. Binding of 14-3-3 to TSC2 can disrupt binding TSC2 to TSC1 and RHEB, which are associated with endomembranes . However, in 14-3-3 pull-down experiments, both TSC1 and TSC2 were found to bind, and 14-3-3 did not affect the association between TSC1 and TSC2 [94, 95]. It also remains unclear whether TSC2 binding to14-3-3 hindered its GAP activity towards Rheb. Importantly, TSC2 is not an essential target of AKT during normal D. melanogaster development , suggesting the presence of possible additional targets for the AKT mediated regulation of mTORC1.
Growth factors control mTORC1 independently of the TSC complex
As was mentioned above, the PRAS40 binds Raptor and thereby inactivates mTORC1 [48, 50, 57, 63]. In response to growth factors, Akt phosphorylates PRAS40 at T246. This phosphorylation leads to the dissociation of PRAS40 from mTORC1 resulting in a reduced ability of PRAS40 to inhibit TORC1 [48, 49, 57]. This was proposed to be mediated through 14-3-3 binding of the phosphorylated PRAS40 . Thus, bypassing TSC2, AKT phosphorylates PRAS40 and prevents its ability to suppress mTORC1 downstream effectors. The inhibition of PRAS40 by AKT is conserved; in Drosophila, the PRAS40 ortholog Lobe regulates TORC1 signaling . PRAS40 is in turn a substrate of mTORC1, and mTORC1 mediated phosphorylation of PRAS40 S183, [50, 63] has been proposed to negatively regulate mTORC1 signaling by competing with 4EBP1 and S6K for interaction with Raptor. PRAS40 is a direct inhibitor of mTORC1 and antagonizes the activation of the mTORC1 by Rheb•GTP. However, constitutive mTORC1 signaling in TSC2 null mouse embryonic fibroblasts, in which AKT signaling is largely inhibited owing to a negative feedback mechanism (see below), indicates that hyperactive Rheb can overcome PRAS40 mediated inhibition of mTORC1 . Thus, the AKT pathway might stimulate mTORC1 through two interconnected mechanisms: by activating Rheb and/or by inhibiting PRAS40.
It has long been known that mTORC1 signaling is strongly inhibited in cells under the conditions of nutrient deficiency and that the re-addition of amino acids to starved cells can strongly stimulate mTORC1 activity [22, 98]. However, the mechanisms by which amino acids convey signals to mTORC1 remain largely unknown. Earlier studies demonstrated that silencing expression of TSC1/2 confers resistance to amino acid deprivation, indicating that TSC1/2 is involved in the regulation of mTOR function by amino acids . It has been suggested that branched-chain amino acids, (such as leucine), activate mTORC1 by inhibiting TSC1/TSC2 or stimulating Rheb . Consequently, inhibition of Rheb binding to mTOR is critical for the inhibitory effect of amino acid withdrawal on mTOR signaling . However, other studies do not support this idea. Thus, in TSC-null cells (that lack either TSC1 or TSC2), the mTORC1 activity remains sensitive to amino acid deprivation, suggesting that other than TSC2, additional mechanisms may also be involved in the regulation of mTOR by amino acids . Although Rheb is required for the amino acid stimulation of mTORC1, starving of amino acids has no effect on GTP loading [99-102]. Therefore, while there is a requirement for GTP-bound Rheb to induct of mTORC1 by amino acids, amino acids probably do not affect Rheb activity – indicating that regulation of Rheb does not stimulate mTORC1 in response to amino acids.
Recently, Ste20-related kinase MAP4K3 (mitogen activated protein kinase kinase kinase kinase 3) and the class III PI3K hVps34 (human vacuolar protein sorting 34) were proposed to be activated by amino acid and be involved in the transduction of signals from amino acids to mTORC1 [103-107]. While the mechanism by which MAP4K3 regulates mTORC1 remains unknown, a mechanism for hVPS34 was recently proposed (Figure 3). According to this proposed mechanism, amino acids induce an extracellular calcium influx that activates calmodulin, which in turn binds and activates hVps34 . hVps34 then generates PI-3-phosphate (PI-3-P) instead of the PI-3,4,5-tris-phosphate generated by type I PI3Ks , that somehow activates mTORC1. The mechanism also involves the formation of a calmodulin-hVps34-mTORC1 supercomplex. However, the regulation of mTORC1 by hVps34 is thought to be specific to mammalian cells because in flies Vps34 does not regulate TORC1 . This is unexpected because regulation of TORC1 by amino acids is known as very conserved. Furthermore, in certain mammalian cells, amino acids appear to inhibit rather than activate mVps34 . However, additional studies are needed to clarify the roles of these proteins in TORC1 activation.
Most recent studies identified Rag GTPases as activators of mTORC1 by sensing amino acid signals [111, 112]. Rag-mediated activation of TORC1 still requires Rheb, indicating that, during amino acid signaling, Rag complexes act upstream of Rheb. Rag family members (Rag A-D) belong to the Ras superfamily of GTPases. They are unique in their ability to dimerize through long C-terminal extensions. In the presence of amino acids, the dimeric Rag complex, which consists of a Rag A/B monomer and a Rag C/D monomer, binds Raptor and transport mTORC1 to lysosomes, the same intracellular compartment that contains Rheb [36, 111, 112]. Rag complexes are recruited to the lysosomal membrane by the trimeric Ragulator complex , which contains the proteins MP1 (MEK partner 1), p14 and p18. The GTP-loading of Rag A/B appears to be regulated by amino acids, and binding to TORC1 is observed most robustly under nutrient-rich conditions – when Rag A/B is in the GTP-bound state and Rag C/D is in the GDP-bound state [111, 112]. This model answers the question why mTORC1 activity cannot be stimulated by growth factors in the absence of amino acids. It also explains why Rag GTPases are not able to activate mTORC1 activity in vitro . mTORC1 can be fully activated only under the conditions of amino acids availability, Rab-dependent mTORC1 translocation to a Rheb-containing compartment, and Rheb activated by growth factors. However, there are many key aspects that remain to be discovered, such as how branched amino acids are detected by Rag GTPases and the identification of the Rag guanine exchange factor (GEF).
Several data evidence that phosphatidic acid (PA) is essential for mTORC1 activation. The main mechanism for generating PA is the hydrolysis of phosphatidylcholine (PC) by phospholipase D (PLD). In mammals PLD exists as two isoforms (PLD1 and PLD2) possessing different mechanisms of regulation and subcellular distribution . PLD1 is predominantly localized under steady-state conditions at the Golgi complex, endosomes, lysosomes and secretory granules, and is regulated by two major signaling categories: growth factors/mitogens like EGF, PDGF, insulin and serum that implicate tyrosine kinases, and the small GTPase proteins from Arf, Ral and Rho families. PLD2 is largely associated with lipid rafts on the membrane surface. . Both PLD1 and PLD2 have a strong requirement for PIP2 as a co-factor . It has been shown that PLD1 activation stimulates PLD2 by increasing levels of PIP2 (product of PA metabolic modifications) . This makes more clear the involvement of both PLD1 and PLD2 in the mTORC1 activation. The generation of PA by PLD can be suppressed by primary alcohols (such as 1-butanol) through the transphosphatidylation reaction whereby inert phosphatidyl-alcohol is generated instead of PA. This reaction has been widely used to examine PLD significance, and several studies have demonstrated that the activation of mTOR was sensitive to primary alcohols. Thus, 1-butanol was able to block almost completely the serum-stimulated phosphorylation of mTOR downstream targets, S6K1 and 4E-BP1 . From these findings, it can be asserted that PLD production of PA plays an essential role in the mTOR signaling pathway). In skeletal muscle, PA stimulated S6 kinase phosphorylation, and 1-butanol suppressed S6 kinase phosphorylation . Nutrient-dependent multimerization of mTOR was also suppressed by 1-butanol . Therefore, primary alcohols-dependent suppression of PLD activity has been shown to suppress mTORC1 signaling in several cell models .
Several laboratories have shown that mTORC1 is activated in response to exogenously supplied PA. For example, exogenously provided PA stimulated the activation of S6 kinase and phosphorylation of 4E-BP1 in cancer HEK293 cells. The effect of PA was sensitive to rapamycin [115, 118] and was dependent on the presence of amino acids . Coexpression of TSC1/2 was shown to inhibit PA-dependent stimulation of S6K. This indicates that PA-induced S6K activity is mediated through TSC1/2-mTOR signaling. PA was also shown to activate mTOR in macrophages in an Akt-dependent manner .
In addition, several studies have explored the influence of PLD1 and PLD2 expression on mTORC1 activation. Particularly it was reported that PLD2 overexpression increases S6K phosphorylation in MCF7 cells . Overexpression of PLD1 also stimulated S6K phosphorylation in rat fibroblasts . siRNA-mediated knockdown of PLD1 blocked S6K phosphorylation in B16 melanoma cells, and suppression of either PLD1 or mTOR led to melanoma cells differentiation .
At the same time, up to date the precise mechanism of PA-dependent stimulation of mTOR signaling remains unclear. One possibility is that PA binds to mTOR at the FRB domain, the region where the rapamycin-FKBP12 molecule binds mTOR as well. This binding was specific for PA as other phospholipids were unable to bind the FRB with such specificity. It was hypothesized that the competition between the rapamycin-FKBP12 complex and PA for the FRB site may be one of the regulating factors in mTOR activation . According to the other hypothesis the pH locally around mTOR is reduced by PA-generated PLD, which eventually promotes its kinase activity, or allows for interaction with yet unknown promoter substrates . It was shown that PLD1 is an effector of the small GTPase Rheb (see above) within the mTORC1 signaling pathway [123, 124] (Figure 3). It was also reported that PLD2 forms a functional complex with Raptor and mTOR via a TOS (TOR signaling) motif in PLD2, and this interaction was essential for mitogen stimulation of mTORC1 . More recently, dominant negative mutants of both PLD1 and PLD2 were able to suppress the activation of mTORC1 . Therefore, besides PA ability to activate mTORC1, there are several data indicating requirement of PLD itself for the activation of mTORC1. Very recent study provided additional evidence that nutrient stimulation of mTORC1 is dependent on PLD activity which in turn is activated by small GTPases RalA and Arf6 . According to this study, amino acids dependent activation of PLD is mediated trough generated by Vps34 PI-3-phosphate , that could interact with PX domains of PLD1 and PLD2 which are known to be critical for PLD activity . This activation also requires PLD interaction with both RalA and Arf6. Interestingly, these small GTPases have been earlier shown to be implicated in both responding to nutrients and the stimulation of PLD activity. RalA is constitutively associated with PLD1, but does not activate PLD1 by itself. RalA contributes to the activation of PLD1 by recruiting ARF6, which does activate PLD1 activity, into RalA/ARF6/PLD1 complex. While it is still not clear how the presence of nutrients activates RalA and ARF6, the data provided in this study indicate that PLD is a key target of RalA and ARF6 for the stimulation of mTORC1. In concordance with these findings data from our lab evidence that expression of constitutively active Arf6 stimulates PLD activity which leads to the mTORC1 dependent phosphorylation of downstream targets 4E-BP1, S6K1 kinase and its effector ribosomal protein S6 (rpS6). We also show that mTORC1 signaling stimulation contributes to the Arf6 promitogenic activity .
AMPK (the AMP-activated protein kinase, also known as PRKAB1) is activated under the low level of intracellular ATP and found in all eukaryotes. It was initially identified as a serine/threonine kinase that negatively regulates several key enzymes of the lipid anabolism . At present, AMPK is considered to be the major energy-sensing kinase that activates a whole variety of catabolic processes in multicellular organisms such as glucose uptake and metabolism, while simultaneously inhibiting several anabolic pathways, such as lipid, protein, and carbohydrate biosynthesis (reviewed in ). AMPK is upregulated under energy stress conditions in response to nutrient deprivation or hypoxia when intracellular ATP level decreases and AMP increases . In response, AMPK turns on ATP generating pathways while inhibiting ATP consuming functions of the cell . AMPK functions as heterotrimeric kinase complex, which consists of a catalytic (α) subunit and two regulatory (β and γ) subunits. Upon energy stress, AMP directly binds to tandem repeats of crystathionine-β-synthase (CBS) domains in the AMPK γ subunit . Since the ratio of AMP to ATP represents the most accurate way to precisely measure the intracellular energy level, both AMP and ATP are able to oppositely regulate the activity of AMPK. While AMP binding to the γ–subunit allosterically enhances AMPK kinase activity and prevents the dephosphorylation of T172 , ATP is known to counteract the activating properties of AMP . Although ADP does not allosterically activate AMPK, it also binds to AMPK and enhances phosphorylation at T172 [134, 135]. The phosphorylation of the activation loop T172 is absolutely necessary for AMPK activation. At present, several AMPK-phosphorylating kinases have been identified. In addition to the ubiquitously expressed and constitutively active kinase LKB1, Ca2+-activated Ca2+/calmodulin-dependent kinase kinase β (CaMKKβ)  and transforming growth factor β-activated kinase-1 (TAK1) are both known as AMPK activators. Genetic and biochemical studies in worms, flies, and mice have identified the serine/threonine liver kinase B1 (LKB1) as major kinase phosphorylating the AMPK activation loop at T172 residue, under conditions of energy stress . Within the TOR signaling pathway, LKB1 dependent activation of AMPK inhibit mTORC1 activity by two ways (Figure 4). Firstly, AMPK directly phosphorylates the TSC2 on S1387 and T1227 [2, 64, 70, 87, 137, 138]. AMPK phosphorylation of TSC2 has also been reported to act as a primer for the phosphorylation and enhancement of TSC2 function by glycogen synthase 3β (GSK3β). GSK3β dependent phosphorylation of TSC2 on S1341 and S1337 stimulates its GAP activity towards Rheb, leading to the inhibition of mTORC1 . It is possible that GSK3β cooperates with AMPK to fully activate the GAP activity of TSC2. The second, TSC2 independent mechanism by which AMPK can signal to mTORC1,  is a direct phosphorylation of Raptor at two highly conserved residues — S722 and S792. These phosphorylation events induce Raptor direct binding to 14-3-3 protein, which leads to a suppression of mTORC1 kinase downstream activity . Therefore, mTORC1 itself serves as an AMPK substrate for inhibiting phosphorylation.
mTOR signaling pathway is strictly regulated by hypoxia [139, 140], since the sufficiency of oxygen is also essential for cellular metabolism. Hypoxia inhibits mTORC1 signaling via multiple signal pathways, two of them being mediated through activation of the TSC1/TSC2 complex (Figure 4). First, activation of AMPK by hypoxia can enhance TSC complex function. Particularly, it was shown, that brief hypoxia exposure prevents insulin-mediated stimulation of mTORC1 and phosphorylation of its targets p70S6K and 4E-BP1 . Under these conditions mTOR suppression is mediated through a HIF1α (hypoxia-inducible factor 1α)-independent pathway involving AMPK-dependent activation of TSC1/TSC2 [2, 87, 141]. Second way includes the upregulation of TSC1/TSC2 through transcriptional regulation of stress-induced protein REDD1 (Regulated in Development and DNA damage responses, also known as DDIT4 or RTP801) [142, 143]. This response is mediated in part through a mechanism that involves HIF1α, a transcription factor that is stabilized under hypoxic conditions and drives the expression of several genes, including REDD1. Induction of REDD1 can activate the TSC1/2 complex by competing with TSC2 for 14-3-3 proteins binding [142, 144]. Thus, increased REDD1 levels that occur following exposure to hypoxia prevent the inhibitory binding of 14-3-3 to TSC2 , which eventually leads to the inhibition of mTORC1 signaling. Therefore inhibitory effect of REDD1 on mTOR signaling seemed to be dependent on the presence of the TSC1/2 complex, but independent on the LKB1-AMPK signaling [142, 145, 146]. However, most recent studies proposed that hypoxia and the LKB1-AMPK signaling are highly interrelated at least in some type of cells . In response to prolonged hypoxia, REDD1 expression was enhanced by AMPK activation, leading to the inhibition of mTOR pathway. Indeed, it was demonstrated that prolonged hypoxia induced ATP depletion and eventually activate AMPK . Taken together, under hypoxic stress, the inhibition of mTOR activity by REDD1 activation may be mediated either through AMPK-independent or -dependent mechanisms.
Hypoxia may also downregulate mTORC1 through proteins that hinder the the Rheb–mTOR interaction. The PMl (promyelocytic leukaemia tumour suppressor) has been found to bind mTOR during hypoxia and inactivate it via sequestration in nuclear bodies . Likewise, the hypoxia-inducible proapoptotic protein BNIP3 (BCl2/adenovirus E1B 19 kDa protein-interacting protein 3) was found to regulate mTOR by direct association with Rheb . (Reviewed in ).
The protein synthesis stimulation and the inhibition of autophagy are two mostly known biological outputs controlled by this pathway under the favorable conditions, such as nutrient and oxygen availability. By sensing the presence of growth factors and the sufficiency of nutrients, activated mTORC1 mediates the signals to various components of the translation initiation machinery through direct or indirect phosphorylation events . Several data also evidence that mTOR regulates the synthesis of many classes of lipids (such as phosphatidylcholine, phosphatidylglycerol, and sphingolipids, unsaturated and saturated fatty acids) that are required for membrane biosynthesis and energy storage (For the detailed review see . Since the best characterized effectors of mTOR signaling are proteins controlling the translational initiation machinery it is important to understand how mTORC1 signal transduction pathways contribute to protein synthesis regulation (reviewed in ).
The earliest identified and best-studied mTORC1 targets are S6K kinases (p70 ribosomal protein S6 kinase 1 and 2) and 4EBP1 (eIF4E binding protein 1); both proteins involved in the translation initiation process  (Figure 5A). Protein synthesis is one of the most energy consuming processes within the cell and translation rates are strictly regulated mostly through modification of the eukaryotic initiation factors (eIFs). In eukaryotes, several mRNAs are translated in a cap-dependent manner. The cap structure, m7GpppN (where N is any nucleotide), is present at the 5’ terminus of the majority cellular eukaryotic mRNAs (except those in organelles) . The cap structure is bound by the eIF4F (eukaryotic initiation factor 4F) complex, which contains three initiation factors — the mRNA 5’ cap-binding protein eIF4E, an ATP-dependent RNA helicase eIF4A and a large scaffolding protein eIF4G, which provides docking sites for the other proteins. Briefly, to assemble the eIF4F complex, eIF4E binds the 5′ cap and recruits eIF4G and eIF4A. eIF4A along with eIF4B acts to unwind the mRNA 5’ secondary structure to facilitate ribosome binding . It is especially essential, since stable secondary structures are often found in the 5′ UTR of specific mRNA species, many of which encode proteins that are involved in promoting cell growth and proliferation, and significantly suppress their translation efficiency . As the translation preinitiation complex is recruited near the 5′ end of mRNA, this requires the structured UTR to be ‘linearized’ — not only for the initial binding of the 40S ribosome but also for subsequent searching for the downstream initiation codon. Although eIF4A alone exhibits low levels of RNA helicase activity the last one is substantially stimulated by its regulatory cofactor, eIF4B. Thus, eIF4B enhances the affinity of eIF4A binding to ATP, which, in turn, increases the processivity of the eIF4A helicase function . eIF4G recruits the small ribosomal subunit to the mRNA (and the poly(A)-binding protein, PABP) through the ribosome associated large multisubunit factor eIF3. As a result the assembly of the 48S translation preinitiation complex takes place, allowing for the ribosome scanning and translation initiation [22, 26]. The translation initiation factors and cofactors that are regulated by mTORC1 signaling include eIF4G, eIF3, eIF4B, eIF4E and 4EBP1, of which 4EBP1 is considered to be the most well-known mTORC1 direct effector protein.
eIF4G serving as a modular scaffold for the translation preinitiation complex formation, is phosphorylated in response to growth factor stimuli at multiple sites, some of which are dependent on mTORC1. These sites are clustered in a hinge region of eIF4G that joins two structural domains, and it has thus been predicted that the modification might induce conformational changes in the protein that affect its activity . Nevertheless, the precise molecular mechanism by which eIF4G phosphorylation regulates its function remains to be determined. Regulation of the mRNA cap binding protein eIF4E is mediated mainly in two ways, firstly, through phosphorylation at S209 in its C-terminus by MAP kinase signaling integration kinases 1 and 2 (Mnk1/2)  and, secondly, through the sequestration by small, heat stable phosphoproteins termed 4E-binding proteins, 4E-BPs  belonging to the 4E-BPs translation repressors family. One of these proteins, 4E-BP1 is a direct mTORC1 phosphorylation target. In quiescent cells, hypophosphorylated 4EBP1 binds tightly to eIF4E. As 4EBP1 and eIF4G share the same eIF4E-binding motif 4EBP1 competes with eIF4G for an overlapping binding site on eIF4E, and prevents eIF4G from interacting with eIF4E. On mTORC1 activation, hyperphosphorylated 4EBP1 dissociates from eIF4E, allowing for the recruitment of eIF4G and eIF4A to the 5’ end of an mRNA. Thus, the effects of 4E-BP1 on protein translation are not limited to switching ‘off ’ or ‘on’ protein synthesis; they can also alter the range of nascent proteins by mediating a switch between cap-dependent and cap-independent translation. Indeed, during specific stress conditions, such as nutrient depletion, hypoxia or metabolic stress, the cell can reduce the activity of mTORC1, resulting in the cessation of cap-dependent translation and the concomitant promotion of cap-independent translation of essential pro-survival factors. Rapamycin inhibits mTORC1-dependent 4E-BP1 phosphorylation, stimulating the interaction between eIF4E and 4E-BP1, what leads to cap-dependent translation inhibition .
Control of the 4E-BPs by mTOR
Upon the stimulation (by growth factors, mitogens and hormones), human 4E-BP1 is phosporylated at 7 sites, 4 of which are involved in mTOR signaling [157, 158]. These are T37, T46 and T70, and S65. The 4E-BP1 phosphorylation is proceeded in a hierarchical manner (first T37 and T46, then T70 and last S65) . S65 and T70 are located near the eIF4E-binding site. Often phosphorylation of these residues is stimulated by insulin in a rapamycin-sensitive manner. Some data evidence that phosphorylation of S65 is required for release of eIF4E from 4E-BP1, however the role of phosphorylation of this site is unclear . Molecular dynamics findings  and earlier biophysical data suggest that phosphorylation of S65 and T70 is insufficient to bring about release from eIF4E. Phosphorylation of both S65 and T70 depends upon the prior phosphorylation of the N-terminal threonines, and modification of T46 is considered to be essential for phosphorylation of T37 [157, 161]. The phosphorylation of T70 and S65 in human 4E-BP1 depends upon a further site, S101 . The phosphorylation of the N-terminal threonin residues in 4E-BP1 depends upon a certain sequence in the N-terminus, which includes the Arg-Ala-Ile-Pro (‘RAIP’ motif) [91, 163]. This phosphorylation is not significantly influenced by TOS motif inactivation and according to some data is rather insensitive to rapamycin . This suggests that it could be mediated independently of mTORC1. However, several data evidence that phosphorylation is mediated by mTORC1: (i) it is inhibited by starvation of cells for amino acids; (ii) it is activated by Rheb; (iii) it is suppressed by TSC1/2; (iv) it is sensitive to inhibitors of the kinase activity of mTOR and (v) it is decreased in cells in which mTOR levels have been knocked down . Therefore, further study of this process needs to clarify the molecular mechanisms of mTOR downstream signaling.
Control of the S6Ks by mTOR
Another important target of mTORC1 is the S6 kinases family, including S6K1 and S6K2. Ribosomal protein S6 (rpS6) is highly phosphorylated protein containing at least five phosporylating sites in its C-terminus. There are two main classes of protein kinases which are responsible for rpS6 phosphorylation in vitro, namely the p70 S6 kinases (S6Ks) and the p90 ribosomal protein S6 kinases, also known as RSKs [164, 165], (reviewed in ). The observed sensitivity of rpS6 to rapamycin lead to the speculation that S6K are responsible for rpS6 phosphorylation as their activation is mediated by mTOR. Unlike S6Ks, the RSKs are not influenced by rapamycin since they are known to be activated through the classical MAP kinase (ERK) pathway (see below). There are two similar S6 kinase proteins, S6K1 and S6K2, in mammals , which show 70% of amino acid homology. Each p70S6K gene encodes two distinct proteins due to alternative splicing of the mRNAs. Several data confirm that activation of both the S6K1 and S6K2 are regulated by mTORC1 [118, 167, 168]. S6K1, which was discovered earlier than S6K2, is ubiquitously expressed and appears to be more critical for the control of cell growth. S6K1 can be activated by a wide variety of extracellular signals and is known as the major rpS6 kinase in mammalian cells and key player in the control of cell growth (cell size) and proliferation [169, 170].
Earlier it was thought that activated S6K1 regulates translation of a class of mRNA transcripts that bear a 5’-terminal oligopolypyrimidine (5’-TOP). Particularly, it was shown that S6K1 phosphorylates eIF4B on S422, which is located in the RNA binding region that is necessary for promoting the helicase activity of eIF4A . Few data indicate that eIF4B phosphorylation by S6K1 is both sufficient and necessary for its recruitment to the translation preinitiation complex . However, there are also some data that disprove this model. In S6K1/2–/– cells, 5’-TOP mRNA translation is intact and still rapamycin-sensitive . These results are in concordance with earlier data showing that mitogenic-stimulated or amino acid dependent 5’-TOP mRNAs translation is dependent on PI3K mediated signaling, and does not require S6K1 activity and ribosomal protein S6 phosphorylation [174, 175]. Instead, a role for the S6 kinases in controlling the cell size has been suggested as deletion of S6K leads to animal size decrease . Studies performed on ‘knock-in’ mice in which all sites phosphorylated by the S6 kinases were mutated also indicated a role for S6 phosphorylation in cell growth . These knock-in cells still demonstrated faster rates of protein synthesis at the same time being decreased in size. This could be explained by elevated access of the S6K to other substrates involved in translation, such as eIF4B and eEF2 kinase (see below) .
Another pool of data connecting S6K1 activity and translation initiation occurs from the study of potential tumor suppressor, Programmed cell death 4 (PDCD4) protein (Figure 5A). PDCD4 binds to eIF4A and is thought to inhibit its helicase activity . PDCD4 is also thought to prevent eIF4A from incorporating into the eIF4F complex by competing with eIF4G for eIF4A binding . S6K1 phosphorylates PDCD4 on S67 in response to growth factor stimulation resulting in its subsequent degradation through the ubiquitin ligase βTrCP101. Therefore, S6K1-dependent phosphorylation of PDCD4 prevents the inhibitory effect of PDCD4 towards eIF4A helicase function within the eIF4F complex.
Recent data give new evidence on interconnection of the mTOR/S6K1 pathway and translation preinitiation complex assembly [172, 179]. Under the poor growth conditions, S6K1 but not mTORC1 binds with multisubunit initiation factor eIF3 that was identified as a dynamic scaffold for mTORC1 and S6K1 binding  (Figure 5B). Upon growth factors or nutrients availability, the mTORC1 is recruited to eIF3 and phosphorylates S6K1. Based on polysome analysis and cap-binding assays, it is thought that the mTORC1–eIF3 complex associates with the mRNA 5′ cap, bringing mTORC1 into proximity with 4EBP1. Phosphorylation of S6K1 at T389 leads to its dissociation from eIF3. T389-phosphorylated S6K1 binds to PDK1 (Figure 5A), which phosphorylates S6K1 at T229. The fully activated S6K1 is able to phosphorylate eIF4B and S6. Phosphorylation of eIF4B by S6K1 at S422 promotes its association with eIF3 [172, 180]. The interaction of mTOR with eIF3 also strengthens the association between eIF4G and eIF3 . Described interactions cooperate to enhance the assembly of translation initiation complex and facilitate cap-dependent translation.
The S6 kinases are activated by phosphorylation at multiple sites. Several of them lie within the C-terminus, while two others lay immediately C-terminal to the catalytic domain. One of these, T389 in the shorter form of S6K1, which is located at a hydrophobic motif carboxyterminal to the kinase domain, is directly phosphorylated by mTOR as part of the mTORC1 complex [182, 183]. Phosphorylation here is required for the consequent phosphorylation of S6K1 by PDK1 at a T229 in the activation loop of the catalytic domain. Phosphorylation at T229 allows full activation of S6K1. S6K2 is likely regulated in a similar manner. Both S6K1 and 2 contain a TOS motif within their N-terminus region, which interacts with Raptor, promoting phosphorylation of S6Ks by mTORC1 . The phosphorylation within the C-terminal region seems to open access to the sites T389/T229, phosphorylation of which provides the complete activation. It is not known exactly which kinase is responsible for C-terminal phosphorylation sites. Nevertheless mTOR also indirectly contributes to the phosphorylation of the C-terminal sites. A motif RSPRR exists in this region probably plays a significant role in the inhibitory effect of the C-terminal region. It has been speculated that a negative S6K1 regulator binds S6K via this motif and that mTOR could broke this binding . The C-terminal region of S6K1 also determines whether S6K1 can be phosphorylated by mTORC2. Mutant S6K1 with deletion of this region is a substrate for mTORC2 . Some data indicate that for S6K1 activation, mTOR can directly phosphorylate S371 in vitro, and this event modulates T389 phosphorylation by mTOR [186, 187].
In addition to the discussed above mTORC1 targets, S6Ks and 4E-BP1, both of which modulate translation initiation, mTOR signaling also regulates the translation elongation process through the phosphorylation of eukaryotic elongation factor 2 (eEF2). eEF2 is a GTP binding protein that mediates the translocation step of elongation . eEF2 is phosphorylated at T56 within the GTP-binding domain and this phosphorylation impedes its ability to bind the ribosome, thus inhibiting its function [188, 189]. Insulin and other stimuli induce the dephosphorylation of eEF2, and this effect is blocked by rapamycin, indicating that this effect is also mediated through mTOR . The eEF2 phosphorylation function is attributed to a highly specific enzyme called eEF2 kinase (eEF2K) . Phosphorylation of eEF2 at T56 by eEF2 kinase impedes the eEF2 binding of to the ribosome and the translocation step of the elongation . The calcium/calmodulin-dependent protein kinase eEF2K is an atypical enzyme since the sequence of its catalytic domain differs substantially from that of other protein kinases, and it is not a member, e.g., of the main Ser-Thr-Tyr kinase superfamily . The C-terminal half of the eEF2K polypeptide contains several sites of phosphorylation including the binding site for the substrate eEF2 at the C-terminus . mTOR negatively regulates eEF2 kinase and consequently activates eEF2. mTOR is considered being able to phosphorylate 3 sites, as was determined by their rapamycin- and/or amino-acid starvation sensitivities [188, 193]. S366 in the C terminus of the catalytic domain has been identified as the site being phosphorylated by S6K and by p90RSK . The phosphorylation at S359 has been shown to be also regulated in a rapamycin-sensitive manner in response to insulin-like growth factor 1 (IGF1) and inactivates eEF2K , but the kinase responsible for this phosphorylation remains to be determined. Recently, a novel phosphorylation site located immediately adjacent to the CaM-binding site in eEF2K that is regulated markedly in response to insulin in an mTOR dependent manner has been identified. This site (S78) is not known to be phosphorylated by any known protein kinase in the mTOR pathway. Phosphorylation at this third site also causes the inactivation of eEF2 kinase, in this case by inhibiting the binding of CaM, which binds immediately C-terminal to S78 . eEF2K is thought to be a target of signaling from mTOR independently of other known targets of this pathway, which implies the existence of a novel (probably mTOR-controlled) protein kinase that could acts upon S78 in eEF2K. These data provide a molecular mechanism by which mTOR could regulate peptide chain elongation.
Since the protein synthesis depends on the amount of ribosomes and transfer RNAs (tRNAs) it is important to know that mTOR signaling also contributes to the regulation of tRNA production, promotion of rRNA synthesis and ribosome biogenesis. Thus mTOR signaling tightly regulates transcription of ribosomal RNAs (rRNAs) and tRNAs by RNA polymerases I and III . mTOR can associate with general transcription factor III C (TFIIIC) and relieve its inhibitor Maf1, leading to increased tRNA production. mTORC1 activity also promotes association between transcription initiation factor 1A (TIF-1A) and polymerase I (PolI), thereby promoting rRNA synthesis . The activity of several other transcription factors, such as signal transducer and activator of transcription-1 and -3 (Stat-1 and Stat-3) is also suggested to be regulated by mTORC1-mediated phosphorylation in a rapamycin-sensitive manner .
Autophagy is a lysosomal-dependent cellular degradation process that generates nutrients and energy to maintain essential cellular activities upon nutrient starvation. A term autophagy appeared from Greek “auto” (self) and “phagy” (to eat), refers to an evolutionarily conserved, multi-step lysosomal degradation process in which a cell degrades long-lived proteins and damaged organelles. Three forms of autophagy have been identified, namely macroautophagy, microautophagy and chaperone-mediated autophagy that differ with respect to their modes of delivery to lysosome and physiological functions . Macroautophagy (hereafter autophagy) is the major regulated catabolic mechanism that involves the delivery of cytosolic cargo sequestered inside specific intracellular double-membrane vesicles, called autophagosomes to the lysosomal compartment and subsequent fusion with lysosomes to form single-membrane-bound autophagolysosomes, in which the sequestered material is degraded by acidic lysosomal hydrolases. On one hand, autophagy is crucial for cell survival under extreme conditions through degradation of intracellular macromolecules, which provides the energy required for minimal cell functioning when nutrients are deprived or scarce. Also, autophagy-mediated elimination of altered cytosolic constituents, such as aggregated proteins or damaged organelles, preserves cells from further damages, indicating that autophagy plays a protective role in early stages of cancer . On the other hand, autophagy plays a death-promoting role as type II programmed cell death (type II PCD), compared to apoptosis (type I PCD), as a bona fide tumor suppressor mechanism in cancer .
The ability of mTORC1 to regulate autophagy is as highly conserved as well as the process of autophagy itself. AMPK has been indicated as a main upstream regulator of mTORC1 mediated autophagy inhibition.
The mechanism by which TORC1 negatively regulates the autophagic machinery has first been described in yeast. Genetic screenings for autophagy defective mutants led to the identification of more than 30 essential autophagy-related genes (Atg).These proteins can be classified into several groups depending on their function and interdependency. Most upstream is a protein complex that comprises the serine/threonine kinase Atg1, as well as two accessory proteins Atg13 and Atg17. In mammals, two homologs of Atg1, uncoordinated 51-like kinase 1 (ULK1) and ULK2 have been identified. Accumulating evidence suggests that ULK1 is a key regulator of autophagy initiation. ULK1 is directly phosphorylated by TORC1 [200-202]. Recently, it has also been shown that mTORC1-mediated phosphorylation of ULK1 impairs its activation by AMPK and results in an overall decrease in autophagy . ULK1 and ULK2 are found in a stable complex with mammalian autophagy-related protein 13 (mAtg13), the scaffold protein FAK-family interacting protein of 200 kDa (FIP200)   and Atg101, an additional binding partner of Atg13 that has no ortholog in yeast . In contrast to yeast, the composition of this complex does not change with the nutrient status. Several data evidence that the phosphorylation status within the Ulk1/2-Atg13-FIP200 complex dramatically changes with the nutrient availability. Under rich growth conditions, mTORC1 associates with the Ulk1/2-Atg13-FIP200 complex, via direct interaction between Raptor and Ulk1/2 (37). The active mTOR phosphorylates Atg13 and Ulk1/2 , thereby downregulating Ulk1/2 kinase activity and suppressing autophagy (Figure 6). In response to starvation, the mTORC1-dependent phosphorylation sites in Ulk1/2 are rapidly dephosphorylated by yet unknown phosphatases, what stimulates Ulk1/2 autophosphorylation and phosphorylation of both Atg13 and FIP200. Several serine and threonine residues in human Ulk1 whose phosphorylation was decreased after starvation have been recently identified from which S638 and S758 have been proposed to be most probable mTORC1 negative phosphorylation sites [203, 206]. Ulk1/2 autophosphorylation and following FIP200 and Atg13 phosphorylation in turn leads to translocation of the entire complex to the pre-autophagosomal membrane and to autophagy induction [200, 201, 205]. However, the functional relevance of Ulk1/2-mediated phosphorylation of Atg13 and FIP200 for this recruitment and the relevant phosphorylation sites has not been verified yet. Interestingly, another Ulk1-dependent phosphorylation site in human Atg13 (S318) has been identified recently . The authors of that study could show that the Hsp90-Cdc37 chaperone complex selectively stabilizes and activates Ulk1.
In yeast, autophagosomes originate from a single preautophagosomal structure. Although an equivalent structure seems to be absent from mammalian cells, a special subdomain in the endoplasmic reticulum (ER) termed the “omegasome” has been suggested as a putative origin of autophagosomes. This structure is enriched in PI(3)P, a product of the phosphatidylinositol 3-kinase (PI3K). A hierarchical analysis of the mammalian Atg proteins could recently confirm the recruitment of Ulk1 proximal to these omegasomes . The translocation of Ulk1, presumably in a complex with Atg13 and FIP200, is the initial step of autophagosome biogenesis and is completely abrogated in FIP200–/– cells . The subsequent recruitment of the PI3K depends on Ulk1 and its kinase activity . Recently, two groups found evidence for the mechanism by which Ulk1 and Ulk2 in turn negatively regulate mTORC1 signaling. Particularly, the phosphorylation of Raptor at numerous sites was strongly enhanced after overexpression of Ulk1. Interestingly, one of these residues (T792) is the abovementioned effector site through which AMPK negatively regulates mTORC1 activity . The multiple Ulk1-dependent phosphorylation of Raptor either results in direct inhibition of mTORC1 kinase activity , or interferes with Raptor-substrate interaction , thus finally leads to reduced phosphorylation of mTORC1 downstream targets.
Apart from mTOR, Ulk1/2 is phosphorylated (probably on S317 and S777 or S555 according to different studies) by AMPK under glucose starvation and thereby activated [112, 203, 211, 212]. Under nutrient sufficiency phosphorylation of ULK1 S758 by active mTORC1 disrupts ULK1 interaction with, and hence activation by, AMPK . Although the data concerning the role of ULK1/2 certain sites phosphorylation is rather discrepant it is clear that in mammals, phosphorylation of ULK1 by AMPK is strongly required for ULK function in the response to nutrient deprivation. Therefore, AMPK could control ULK1 via a two-pronged mechanism, ensuring activation only under the appropriate cellular conditions – firstly, by direct phosphorylation and secondly, by suppression of mTORC1-mediated ULK1 inhibition . Several studies demonstrated that Ulk1 in addition directly interferes with mTORC1 downstream signaling and negatively regulates S6K1 activity, both in Drosophila and mammalian cells . Taken together these data evidence that mTOR subnetwork occupy the key position in autophagic pathways.
In contrast to mTORC1, very little is known about the upstream regulation of TORC2. Rapamycin–FKBP12 complex does not bind directly to mTORC2, but long-term rapamycin treatment disrupts mTORC2 assembly in ~20% of cancer cell lines through an unknown mechanism . It remains to be determined why rapamycin-mediated inhibition of mTORC2 assembly only occurs in certain cell types. One hypothesis suggests that some mTORC2 subunits could prevent the binding of rapamycin/FKBP12 complex to the mTOR FRB domain by the competing mechanism (reviewed in ). However, there are no enough data to support this model.
It seems that mTORC2 is activated in response to growth factors but is insensitive to nutrients and energetic stress, . Thus, like TORC1, TORC2 can be stimulated by growth factors through PI3K . Consequently, treatment with PI3K inhibitors can inhibit TORC2-mediated target phosphorylation . Thus it was suggested that mTORC2 lies downstream of PI3K signaling . Rheb which is known as a key upstream activator of mTORC1 showed negative and indirect effect on the regulation of mTORC2 both in Drosophila and mammalian cells . Some data pointed on TSC1-TSC2 function in mTORC2 regulation . Moreover, the TSC1/TSC2 complex was found to physically associate with mTORC2, but not with mTORC1. The molecular mechanism through which the TSC1/TSC2 complex promotes mTORC2 activation is unclear. It is also currently unknown whether some pathways that regulate TSC1/TSC2 ability to inhibit mTORC1, also influence on mTORC2 activation.
The best-characterized target of mTORC2 is AKT, which is phosphorylated at S473 upon TORC2 activation [53, 58, 85]. Numerous studies attempted to identify the crucial kinase(s) (often referred to as PDK2) responsible for the phosphorylation of S473 in Akt. Several enzymes are in the candidate list, including PDK1, integrin-linked kinase (ILK), Akt itself, DNA-dependent protein kinase catalytic subunit (DNA-PKcs) and mTORC2 . Since mTORC2 complex fulfills the role of the Akt S473 kinase, mTORC2 has been identified as the PDK2 . Akt is a member of the AGC kinase family (see above), which also includes S6Ks, serum glucocorticoid-induced protein kinase (SGKs), RSKs, and PKCs [62, 216]. mTORC2 has been shown to phosphorylate AKT, SGK and PKCα . mTORC2 seems to regulate Akt by phosphorylation of its two different sites. The mTORC2-mediated Akt hydrophobic motif phosphorylation on the regulatory S473 site is dependent on growth factor signaling, whereas a basal activity of mTORC2 maintains the constitutive phosphorylation of Akt on T450 site in its turn motif . This difference indicates that phosphorylation of the T450 and S473 sites on Akt by mTORC2 are separate events and might take place at different locations. It has been proposed that translocation of Akt to the plasma membrane coupled with its phosphorylation on T308 and S473 is a critical step in activation of Akt by growth factor signaling . Phosphorylation of AKT on S473 enhances the activation phosphorylation motif at T308, which is absolutely required for AKT activity.
The major functions of mTORC2 include the regulation of cytoskeletal organization and the promotion of cell survival. If the last one is mediated apparently through AKT activation, the mechanisms, which realize mTORC2 function in cytoskeletal reorganization, are not obvious. Paxillin, which functions as a docking protein, localizing to the focal adhesions of adherent cells  has been shown to be highly phosphorylated at Tyr118. Knockdown of mTORC2 inhibited the phosphorylation of paxillin . Rho, Rac and Cdc42, three best-characterized members of the Rho family of small GTPases, were demonstrated to be involved in actin cytoskeleton assembly and disassembly . It was reported that mTORC2 may function as upstream regulator of Rho GTPases to regulate the actin cytoskeleton .
Interestingly, the TORC2-mediated activation of AKT places TORC2 upstream of TORC1 in the TOR signaling cascade. A most recent publication has highlighted a role for ribosomes in the activation of TORC2 . The authors have found that active mTORC2 was physically associated with the ribosome, and insulin-stimulated PI3K signaling promoted mTORC2-ribosome binding. Interaction of mTORC2 with NIP7 (nuclear import 7, a protein responsible for ribosome biogenesis and rRNA maturation) was shown to be required for full activation of mTORC2 by insulin. Noteworthy, inhibition of protein translation had no effect on mTORC2 activation, supporting the notion that mTORC2 is activated by the ribosome, but not translation. Ribosome associated mTORC2 displays kinase activity toward AKT in vitro. Inhibition of PI3K activity blocks the interaction between the ribosome and mTORC2, as well as inhibits mTORC2 activation in response to insulin, confirming that NIP7-ribosome assembly activates mTORC2 downstream from PI3K. It appears that the mTORC2 components, Rictor and/or Sin1, which are not found in mTORC1, interact with the 60S subunit of ribosome. Interestingly, another study  has also recently reported the association of mTORC2 with the ribosome and proposed that the ribosomal association is important for the cotranslational phosphorylation of the AKT turn motif. These findings are coherent with very recent data that point on endoplasmic reticulum (ER), the cellular organelle highly enriched with ribosomes, as a major compartment of mTORC2 localization. Moreover, the signaling from growth factor does not change the ER localization of mTORC2 as well as its translocation to the plasma membrane. Besides it was suggested that the mTORC2-dependent phosphorylation of Akt on S473 occurs on the surface of ER . These observations raise many interesting questions regarding the regulation of TORC2 and its ribosomal interactions, but it also indicates that additional levels of interplay between TORC2 and TORC1 may exist, as both complexes are linked to the process of ribosome biogenesis.
In addition to the PI3K–Akt pathway, activation of Ras-MAPK signaling can also stimulate mTORC1 activity. The Ras–mitogen-activated protein kinase (MAPK) pathway is a key signaling pathway that is involved in the regulation of normal cell proliferation, survival, growth and differentiation. This pathway includes the whole number of kinases, being regulated through phosphorylation in consecutive order. The Ras–MAPK signaling network has been the subject of intense research because mutations in (or overexpression of) many of the signaling components from this pathway are a hallmark of several human cancers and other human diseases . The Ras–ERK (extracellular signal-regulated kinase-1 and -2) pathway has an established role in regulating transcription , but a connection between this pathway and translational regulation is less clear. Over the past few years, mitogen activated Ras–ERK pathway has been shown to trigger the activation of mTORC1 signaling. This is mediated by ERK and RSK dependent phosphorylations of TORC1 pathway components.
p90RSKs (also known as MAPKAP kinase 1 (mitogen-activated protein kinase-activated protein) kinase-1) are a family of Ser/Thr kinases that lies downstream of the Ras–MAPK cascade and has overlapping substrate specificity with Akt. The RSK isoforms are directly activated by ERK1/2 in response to growth factors, many polypeptide hormones, neurotransmitters, chemokines and other stimuli. RSKs phosphorylate several cytosolic and nuclear targets and they are involved in the regulation of different cellular processes, including cell survival, cell proliferation, cell growth and motility. Following the stimulation of cells with growth factors, RSKs are phosphorylated at multiple Ser and Thr residues by several kinases; these phosphorylation events are directly or indirectly initiated by the activation of the ERK/MAPK cascade . Six different phosphorylation sites have been mapped in RSK1/2 (and are conserved in RSK3/4), of which four have been shown to be important for their activity (S221, S363, S380 and T573 in human RSK1). Following mitogen stimulation, ERK1/2 phosphorylates T573 at the C-terminal domain (CTKD) activation loop of RSK, resulting in CTKD activation. ERK1/2 might also phosphorylate S363 (the turn motif) and T359 (unknown function) in the RSK linker region. The activated CTKD of RSK then autophosphorylates S380 in the hydrophobic motif, creating a docking site for PDK1. After binding, PDK1 phosphorylates the NTKD activation loop S221, leading to the complete activation of the protein and following phosphorylation of the substrates by the NTKD. The NTKD also phosphorylates S749 in the CTKD domain of RSK, differentially modulating the interaction of RSK isoforms with ERK1/2 and thereby completing a sequence of coordinated phosphorylation events and protein–protein interactions that culminate in RSK activation and downstream signaling throughout the cell . Other factors that have been shown to be involved in the activation of RSK include the p38 MAPK, the ERK5 MAPK and fibroblast growth factor receptor-3 (FGFR3). RSK was found to phosphorylate TSC2 at the C-terminus S1798  and, to a lesser extent, the two conserved Akt sites (S939 and T1462) and inactivates its suppressor function, thereby promoting mTOR signaling and translation (Figure 7). RSK mediated phosphorylation of TSC2 is additive to AKT mediated inhibitory modifications of TSC2, but how these phosphorylation events lead to TSC2 inhibition remains unclear.
Erk1/2 kinase itself also impacts the mTORC1 regulation. Thus, a number of additional sites on TSC2 were found to be weakly induced by PMA , including an ERK consensus site S664. This site, and a second site on TSC2, S540, was independently found to be directly phosphorylated by ERK and to contribute to ERK-mediated activation of mTORC1 signaling . Strikingly, phosphorylation of S540 and/or S664 by ERK was found to disrupt the association between TSC1 and TSC2. This effect was also detected following phosphorylation of the TSC1/TSC2 complex in vitro, suggesting that it is direct and does not require other proteins.
In addition, it was recently shown that RSK also directly impacts the mTORC1 complex phosphorylating Raptor, and thereby promotes mTORC1 kinase activity . RSK phosphorylates at least two evolutionarily conserved Raptor serine residues that lie within a region with no homology to known functional domains. Whereas S721 lies within a classical RSK consensus sequence (RXRXXpS/T), S719 is located within a minimal phosphoacceptor sequence (RXXpS) that was found to be sufficient in other RSK substrates, such as DAPK, c-Fos, and CREB. Although the underlying molecular mechanism of this was not fully defined, this study provided new insights into Ras–ERK signals to mTORC1. As tumor promoting phorbol esters and some growth factors activate mTORC1 signaling independently of AKT, phosphorylation of Raptor by RSK might provide a mechanism to overcome the inhibitory effects of PRAS40 inhibitory phosphorylation of TSC2 at S664 and S1798, respectively [89, 226, 227]. Collectively, these data suggest that ERK signaling activates mTORC1 through multisite phosphorylation events by both ERK and its downstream target RSK.
Erk1/2-RSK pathway also contributes to the mTORC1 downstream signaling, this includes RSK dependent in vivo and in vitro phosphorylation of eukaryotic translation initiation factor-4B (eIF4B) and rpS6 . Although early studies indicated S6K1 and S6K2 as the major rpS6 kinases in somatic cells [228, 229], the role of RSK in regulating site-specific rpS6 phosphorylation and translation in somatic cells has been recently readdressed . Particularly, in vitro and in vivo evidence suggests that S6Ks phosphorylate every site in rpS6, while RSK predominantly phosphorylates S235 and S236 . Studies from S6k1/S6k2-knockout mice showed that there was minimal phosphorylation of rpS6 at S240/244, but there was persistent phosphorylation at S235/236 . In accordance with this finding, RSK1 and RSK2 were shown to phosphorylate rpS6 on S235/236 in response to Ras–MAPK-pathway activation, using an mTOR-independent pathway . The RSK mediated S235/236 phosphorylation correlated with formation of the translation pre-initiation complex and increased cap-dependent translation, pointing that RSK provides an additional mitogen- and oncogene-regulated input that links the ERK pathway to the regulation of translation initiation . Translation initiation factor-4B is also phosphorylated by RSK and S6K on S422 [171, 180]. Therefore, phosphorylation and regulation of eIF4B function by RSK and S6K exemplifies the convergence of two major signaling pathways that are involved in translational control. Together, these findings suggest that the mitogen-activated Ras–ERK–RSK signaling module, in parallel with the PI3K–AKT pathway, contains several inputs to stimulate mTORC1 signaling.
Although the activation of mTORC1 downstream of most cytokines including insulin and growth factors is likely to occur through the Akt and ERK signaling mechanisms described above, accumulating evidence suggests that other cytokines, such as tumor necrosis factor α (TNFα), can also induce mTORC1 activity. TNFα is a proinflammatory cytokine that is involved in many human diseases, including cancer [231, 232]. Early studies implicated the TNFα pathway in mTORC1 activation . Recently, it has been shown that IKKβ (inhibitor of nuclear factor κB (NFκB) kinaseβ; also known as IKBKB), a major downstream kinase in the TNFα signaling pathway, can associate with and phosphorylates TSC1 at S487 and S511, resulting in the inhibition of TSC1–TSC2 and, therefore, the activation of mTORC1 . Tsc1−/− mouse embryo fibroblasts expressing TSC1 mutants lacking these sites lose their responsiveness to TNFα for activation of mTORC1, whereas phosphomimetic mutation lead to a basal increase in mTORC1 signaling. Authors proposed a mechanism involving rapid dissociation of the complex and increased degradation of TSC1. However, the results suggest minimal effects on the stability of the TSC1–TSC2 complex, and the precise mechanism of acute complex inhibition by phosphorylation of these sites is not known. In certain human cancers, TNFα promotes vascular endothelial growth factor (vEGF) expression and angiogenesis through activated mTORC1 signaling as a result of IKKβ mediated suppression of TSC1 . This has provided a plausible mechanism that could link inflammation to cancer pathogenesis. Moreover, TNFα also signals to AKT . Activated AKT in turn induces IKKα (also known as CHUK), . It has been shown that IKKα associates with mTORC1 in an AKT dependent manner . Importantly, IKKα is required for efficient induction of mTORC1 activity by AKT in certain cancer cells [235, 236]. It remains unclear, however, how the association of IKKα with mTORC1 can result in the activation of mTORC1.
Maintenance of cellular energy homeostasis and life-sustaining activity requires their appropriately adaptation to the continually varying surrounding environment. This adaptation is provided by the differential expression of genes that is strictly controlled at the levels of transcription and translation. To provide the rapid response to the environmental cues cells switch vast number of intracellular signaling cascades that define the activity of key proteins responsible for transcription and translation regulation. This implies operative and directed changes in the activities of proteins mediating the signal transduction. Phosphorylation represents one of the most important intracellular regulatory molecular mechanisms since it provides the rapid and reversible activation or downregulation of protein activities. Not surprisingly, mTOR signaling network, which integrates and promotes the prompt respond to environmental changes is mainly regulated through this type of posttranslational modification. mTOR is known as a “switch master” that converts vast array of nutrient-, cytokine-, energy- and stress-sensitive signals into the alterations of cellular metabolism including protein and lipid biosynthesis and autophagy. Indeed such resources consuming processes as growth and proliferation could occur only under the conditions of nutrient and energy sufficiency. When energy or amino acids become limiting, cell growth needs to be restricted and protein production needs to be downregulated so that cells can use their limited resources to survive. mTORC1 contributes to overall cap-dependent translation including initiation and elongation steps by several different pathways. Significantly, all of these pathways use phosphorylation as common molecular mechanism of regulation. Most of the proteins from mTOR-dependent pathways (for instance TSC2, Akt, S6K, etc.) contain multiple phosphorylating sites, which mediate stimulating or negative effect on their activity. Some of mTOR partners are characterized by the hierarchical mode of phosphorylation (rpS6, Akt as the examples), whereby each previous phosphorylation opens the opportunity for the subsequent ones. Interestingly, one certain site could serve as phosphorylation target for more than one kinase, therefore implementing the competitive mechanism of regulation (for instance, eIF4B S422). The complexity of mTOR signaling increases due to the presence of positive or negative feedback loops as well as crosstalk with other pathways. The number of reported phosphorylation sites throughout mTOR pathways constantly increases although the precise molecular meaning of several already discovered phosphorylation events remains unclear. This is also true to some molecular mechanisms of mTORC1 and especially mTORC2 functioning. About mTORC1 signaling, a number of issues remain unresolved. For example, aside from S6K, 4EBP1 and ULK1, the downstream direct targets that mediate the cellular effects of TORC1 signaling are largely unaccounted for. In addition, how the specificity of TORC1 signaling is achieved and how multiple signals are integrated is not known. Concerning TORC2, the upstream regulators are poorly defined. This knowledge seems to be of great significance since mTOR is considered a central node of intracellular signaling network and deregulation of its activity strongly contributes to the wide spectrum of human diseases. Further studies will give us a better understanding of the whole picture of mTORC1/mTORC2 functioning that could be applied to the development of new approaches to the treatment of mTOR-associated diseases.
Life cycle assessment (LCA) is the quantification of potential environmental impacts and the resource use throughout a product’s life cycle: from raw material acquisition, via production and use phases, to waste management . It has been frequently applied by consultants, researchers, industry, and authorities for the past 30 years. It has proven useful for gaining knowledge on the life cycle, for communication of environmental information, and for various kinds of decision-making.\n
Meanwhile, it was clear almost from the start that results from different LCAs can contradict each other. This is still true, despite many attempts to harmonize, standardize, and regulate LCA. From history, we learn that it is not realistic to expect LCA to deliver a unique and objective result. It should not be regarded as a single unique method; it is more fruitful to consider it a family of methods.\n
Attributional LCA (ALCA) and consequential LCA (CLCA) are important groups within this family of methods. The choice between ALCA and CLCA guides other methodological decisions in the LCA, such as the choice of input data and the modeling of processes with multiple products. However, within ALCA and CLCA, there are still many decisions to be made—many versions or members within each group in the LCA family.\n
The purpose of this chapter is to discuss and clarify key concepts in relation to ALCA and CLCA and to guide the reader through the necessary and subjective methodological choices. The example used often relates to the supply of electricity in the life cycle, because much of the methodological debate has been on how to model electricity. The chapter is still relevant to all kinds of LCA, because energy supply is part of virtually all LCAs and because most of the discussion is valid also for modeling other parts of the life cycle. Furthermore, the chapter is relevant to other, similar types of quantitative environmental and sustainability assessments—for example, carbon footprint, which essentially is an LCA except that it is limited to emissions of greenhouse gases .\n
To structure the discussion on the pros and cons of different methodological choices, I start by establishing a set of criteria for what an LCA, or a quantitative environmental systems analysis in general, should be and do (Section 2). The ALCA and CLCA approaches are outlined in Section 3, and their implications for the choice of data and allocation problems are discussed in some detail in Sections 4 and 5, respectively. Section 6 includes an assessment of the two approaches based on previous discussions. The chapter concludes with a few recommendations for the LCA practitioner.\n
The LCA methodology is diverse, and the interpretation of the key concepts also varies between researchers. This chapter presents my view on the matter, which is subjective but based on knowledge gained from more than three decades of research in LCA and energy systems analysis. I present my arguments for this view but leave it to you, the reader, to accept my view or to choose another perspective.\n
Environmental systems analysis is different from traditional science in that the aim is not just to systematically gather knowledge; it has the specific aim to gather and communicate knowledge that results in actions that reduce the negative environmental impacts of human activity in total or at least per functional unit, that is, per unit of utility that the studied system generates. The more a method for environmental assessments can be expected to contribute to this purpose, the better it is.\n
For a method to benefit the environment, it must be possible to apply. The results need to be reasonably accurate, possible to communicate, and perceived as relevant by decision-makers (Figure 1). Furthermore, the method should be resistant against abuse. Each of these criteria is briefly discussed below.\n\n
Different methods meet the criteria to varying degrees, but no method is ideal from all aspects. There will always be a trade-off between, for example, feasibility and accuracy. Hence, the set of criteria is not sufficient as a tool for objective selection of the best methods; however, it can be used for structured discussions on the pros and cons of available methods.\n
To have any effect, the method must be used. The more often it is used, the more results it will generate. How often environmental assessments are made depends on how useful the results are (see Section 2.2–2.4). But it also depends on how easy the methods are to apply and how expensive the studies become. This in turn depends on how complex the methods are and on the extent to which the data and models needed are available. The method becomes more cost-efficient and can potentially have a greater impact if the results and conclusions it generates can be generalized and reused in multiple decision situations. Hence, the method should ideally be easy and cheap to apply and generate results and conclusions that can be generalized.\n
An environmental assessment is sometimes designed to guide a specific decision. To have a positive effect, the results must guide such decisions in the right direction more often than not. The greater the chance that results will point in the right direction, the better. Hence, the method should ideally generate results that are as comprehensive, accurate, and precise as possible.\n
Besides guiding specific decisions, an environmental assessment can contribute to increasing the knowledge of experts and decision-makers. If accurate, such knowledge not only contributes to deliberate immediate actions but can also have a positive impact on future decisions. To educate decision-makers and other stakeholders, the environmental assessment must be transparent and possible to understand. Decision-makers receive a large amount of information and have limited capacity for information processing. For this reason, the method and the results it generates should be easy to communicate and understand. Communication is easier when the concepts used in the method are clear and intuitively easy to understand. Communication is more challenging when the study is very comprehensive or conceptually complex. Hence, the method should ideally result in studies that are transparent, have a simple structure, and use intuitively clear concepts.\n
In order for environmental assessment to have a positive effect, the information and knowledge they generate must result in actions. Decision-makers often have conflicting goals, and decisions are often not rational in the sense that they are based on documented facts only. To convince and inspire decision-makers, the study should be perceived as relevant, legitimate, and credible and the recommendations clear. A study can be perceived as more relevant if it focuses on things that the decision-makers can influence and/or have a clear connection to. Legitimacy increases if the study is perceived as impartial and fair. Credibility can be obtained, for example, through sensitivity analyses. The conclusions and recommendations are clear when the uncertainties are not too great.\n
Relevance and legitimacy are highly subjective. They both increase if the design of the study accounts for the need for knowledge as perceived by the decision-makers. This means that the choice of methods should ideally be adapted to the situation and may vary depending on the decision-makers involved.\n
Robustness here means that the method gives roughly the same results regardless of who applies it. This makes the method more difficult to abuse, that is, to apply in environmental assessments with the purpose to stop or delay decisions with positive consequences for the environment or to defend decisions with poor consequences. The method becomes more robust if it does not require the user to make assumptions or subjective choices that greatly affect the results. It is also more robust if there are detailed guidelines for how the method is to be applied and/or an established good practice for the application.\n
As clear from the Introduction, we can distinguish between ALCA and CLCA. The distinction between two types of LCA was suggested in the beginning of the 1990s [4, 5]. It was established toward the end of the decade  to resolve debates on what type of input data to use in LCAs (cf. Section 4) and on how to deal with the allocation problems that occur when, for example, a process produces more than one type of product (Section 5). Various names were used on the two types of LCA , but the terms attributional/consequential have been used since 2001 .\n
Several different definitions of attributional and consequential LCA have been suggested [9, 10]. I prefer the definitions of Finnveden et al., in what is probably the most cited scientific paper on LCA :
Attributional LCA: LCA aiming to describe the environmentally relevant physical flows to and from a life cycle and its subsystems
Consequential LCA: LCA aiming to describe how environmentally relevant flows will change in response to possible decisions
These definitions clearly connect ALCA/CLCA not only to methodological choices but also to the goal of the study, because they respond to different questions (Figure 2). An ALCA gives an estimate of how much of the global environmental impact belongs to the product studied. A CLCA gives an estimate of how the global environmental impact is affected by the product being produced and used.\n\n
Note that the latter can include both increases and reductions in the environmental impact. It is not unusual that an increase in the production of a product leads to increases in emissions as well as to environmental benefits. The production of district heating in a combined heat-and-power (CHP) plant in Sweden, for example, generates emissions from the CHP plant but reduces emissions in other parts of the electricity system, when electricity from the power plant replaces other electricity production.\n
There are thus two types of LCAs, carbon footprints, etc.:
Attributional assessments, which give an estimate of what part of the global environmental burdens belongs to the study object
Consequential assessments, which give an estimate of how the production and use of the study object affect the global environmental burdens
The choice between ALCA and CLCA affects system boundaries. In the example of district heating from a CHP plant, a CLCA includes both the emissions from the CHP plant and the reduction in emissions from the electricity production displaced by electricity from the CHP plant. In general, when a production process delivers more than one type of products, the CLCA should take into account how the process is affected by a change in the of the product investigated. If it affects the production of other products from the process, the system should be expanded to include the effect of that change.\n
A more advanced CLCA can also include other types of consequences. An increased use of a material in the studied system can, for example, lead to less material being used in other systems. This reduction can be quantified with a partial equilibrium model of the market . The alternative use most likely to be affected can be identified through an econometric analysis .\n
An investment in a relatively new energy technology can contribute to improvements in that technology and thus to more such investments being made in the future. Such an indirect effect can in some cases be very large . In an advanced CLCA, the effect could be roughly estimated using an energy system model with so-called experience curves .\n
An ALCA, in contrast, does not include environmental benefits or other indirect consequences that arise outside the life cycle of the investigated product. Instead, the raw material use and emissions of a co-production process are partitioned between the products of that process. In the cogeneration example above, the environmental burdens of the CHP plant are divided between the electricity and the heat. Such a partitioning is called allocation and can be done in several different ways (see Section 5.1).\n
The choice between attributional and consequential LCA also affects the choice of input data to the calculations. An ALCA estimates how much of the world’s environmental impact belongs to a product. If electricity is used in the product’s life cycle, the calculations must include the product’s share of the environmental burdens of the electricity production system. This is calculated by multiplying the product’s electricity consumption by the average environmental burden of the electricity system per unit of electricity delivered. The figures describing the average environmental burdens are called average data. The electricity described by these average data is called average electricity.\n
Average data is used not only to model electricity production in ALCA. If the product investigated contains steel, average data is used to model steel production. The same applies to other input goods. In order to calculate the average environmental impact of a production system, the boundaries of the production system must be defined. This can also be done in different ways (see Section 4.1).\n
A CLCA aims to generate information on how the study object affects the environmental burdens of the world. If electricity is used in the system investigated, the CLCA should include data that reflects how the environmental burdens of the electricity production system are affected by this electricity use. In a few cases, the system investigated has a significant impact on the electricity production—for example, in a study of a future electric car fleet. In such cases, the CLCA should ideally be based on input data that reflects how such a large change in production volume would affect the production system’s environmental burdens. Such data are called incremental data. With incremental data, the environmental burdens per kWh electricity often depend on the size of the change in power generation (compare the slope of the two lines representing incremental data in Figure 3).\n\n
In most cases, however, the electricity use in the system investigated is so small it has only a marginal impact on the electricity system. A change can be described as marginal when it occurs within a range where the environmental burdens as a function of the production interval can be approximated with a straight line (see Figure 3). Within this range, the slope of the line represents the approximate increase in environmental burdens per unit increase of electricity produced. Since the line is straight, the environmental impact per kWh is approximately constant, and the environmental impact of an additional electricity demand is proportional to the size of this demand. Data that reflects the environmental impact per kWh change within this range is called marginal data. The electricity described by marginal data is called marginal electricity.\n
A CLCA should, if possible, include marginal data not only on electricity production but also on the production of other inputs where the study object only has a marginal impact on the total production volume. There are different types of marginal effects and different ways of identifying marginal production. This is further discussed in Section 4.2.\n
A CLCA can be made to describe and estimate the consequences of a given decision but also to investigate what a specific decision-maker can influence. If this decision-maker can completely shut down or replace a production system, the CLCA should include the entire production system. The environmental burdens per unit produced in this system are then the total burdens of the system divided by the total product output. This is identical to the average data.\n
If marginal or average data are to be used in the LCA depends on whether the study is attributional or consequential, as discussed above. However, there are several types of average and marginal data. The next question to ask is therefore what average or marginal values should be used as input in the calculations.\n
An ALCA is based on average data on the production systems in the product life cycle. In order to calculate the average environmental impact of the production systems, they must be identified, and their boundaries must be defined.\n
When the supplier of a material or component is known, this supplier is linked to the product through contracts and through the economic and physical flows resulting from the contracts. Established good ALCA practice is then to use as specific data as possible. These are data representing the average environmental performance of the supplier or, when possible, of the individual processes in the production plant.\n
In many cases the supplier is unknown, for example, because the product is not yet being produced or because the material or component is bought on a market where the actual supplier shifts over time. Here, established ALCA practice is to use average data for the relevant geographical area. Ideally, this is the area from where the good is bought and/or the area covered by the market, which might be global or regional.\n
Energy carriers like electricity, gas, or district heat are distributed in networks. When the suppliers are known, there are contractual links and economic flows to the supplier, but there is no clear physical flow from the production process to the user. If the contract specifies the producer, it is rather uncontroversial to use data representing a weighted average over the production plants that the supplier has in the network.\n
Contracts might also specify that the electricity bought is produced with a specific technology, such as wind power. In such cases, it is reasonable to use data for wind power in the ALCA. To be more specific, it is reasonable to use average data for the wind power of the producer or supplier to which the contract applies. If the deal is on wind power from a specific plant or site, average values for that plant/site should ideally be used. Of course, similar rules apply if the contract specifies that the electricity is hydro or some other specific technology, or green electricity in general.\n
When the electricity supplier is unknown, many influential LCA guidelines (e.g., [18, 19, 20]) recommend the use of national average data or, for very large countries, average data for regional electricity grids. This might be because electricity supply has traditionally been a responsibility of national authorities. For the past decades, electricity production has been privatized in many countries, power producers have become international companies (e.g., EDF, Vattenfall, E.ON), electricity grids have become more integrated nationally and between countries, and electricity trade and transfer between countries have increased. This means that most electricity systems are no longer isolated national or regional grids. There are strong arguments for using average data for a larger geographical area instead. However, there are various ways to define this area. I here discuss them with a focus on Northern Europe, where I have my expertise:\n
Although production of electricity is increasingly privatized, the electricity sector is still to a large extent regulated by national authorities. One way to defend the use of national average data is to define the electricity system by the geographical scope of regulating authorities. Note, though, that electricity production is affected not only by national authorities but also by local authorities and by international cooperation, for example, within the European Union (EU).\n
Another approach is to define the geographical area by the electricity market. Since the establishment of the Nordic electricity exchange, NordPool, there is a well-established Nordic market, and the corresponding electricity system is often perceived as Nordic, including Sweden, Norway, Denmark, and Finland. As NordPool expands and the transmission capacity to other parts of northern Europe increases, it becomes increasingly relevant to regard the market as North European. There is also an EU directive aiming toward a common European electricity market, with provisions to remove bottlenecks in the electricity transfer between countries. In the future, the electricity market may be described as pan-European.\n
The electricity system can also be defined based on physical facts, for example, the transmission capacity between or within countries. This can be insufficient at times when a lot of electricity is produced at one place and used elsewhere. As a result, there will often be a difference in electricity price, for example, between North and South Sweden and between North and South Germany. The boundaries of the system can be defined where the transfer of electricity is limited by the transfer capacity in the grid, for example, between northern and southern Germany.\n
Alternatively, the electricity system can be defined as the area where the electricity network is synchronized, allowing for transfer of electricity without conversion to direct current. Conversion of electricity is a bottleneck because it is associated with energy loss. Based on this physical bottleneck, a system boundary is between Jutland and Zealand in Denmark, where the former is synchronized with continental Europe but the latter with the rest of Scandinavia.\n
Regardless of the geographical boundaries of the electricity system, the question remains as to whether data should apply to the average of the electricity produced in this area or whether they should apply to the average of the energy used in the area. In the latter case, imports and exports of electricity must be accounted for in the calculation of the average.\n
The difference between short- and long-term marginal effects is important in a CLCA . The distinction between short and long term is well-established within economic theory. Short-term effects in economics are effects on the utilization of existing production capacity that occurs before the production capacity has been able to adapt to, for example, a change in demand. The capacity itself is thus assumed to be unaffected in a short-term perspective.\n
When long-term effects are examined, the production capacity is assumed to completely adapt to the change in demand, to the extent that the risk of capacity shortage is the same as before the change. For the production of most goods, this means that the utilization rate of the capacity is assumed not to change. However, for electricity the long-term marginal effect of increased electricity use may include the construction of, for example, new wind turbines that have lower utilization rates than other power plants. This reduces the total utilization rate in the electricity system, although the risk of capacity shortage is unchanged.\n
If the electricity use in the life cycle is small, the probability is very small that it will affect the energy system’s production capacity. Electricity for lighting in a single house is, for example, a drop in the sea, compared to the total production capacity of the electricity system. The sea, on the other hand, does not consist of much else than drops. If a change in the lighting of a house happens to be what triggers an investment in a new power plant, the effect of the lighting becomes much greater than the electricity demand of the lighting. The long-term marginal effect is calculated as the expected value, i.e., the small probability times the large outcome. This expected value is 1 kWh/year changed production capacity per kWh/year change in the consumption of electricity.\n
The short- and long-term marginal effects can be difficult to communicate, as they are easily confused with the effects of changes made in the near or far future. However, short-term effects can arise far into the future, and long-term effects can occur in the coming decades. As an example, the long-term marginal technologies in 2020 are the technologies whose production capacity is affected by energy use in 2020. These effects may occur in 2025–2035. Meanwhile, the short-term marginal effects in 2050 relate to how a change in energy use in 2050 affects the utilization of the production facilities that exist in 2050. These effects occur during that same year and the years immediately thereafter. Short-term marginal effects of a disruption in 2050 thus arise later than the long-term effects of a disruption in 2020.\n
To make communication easier, the concepts short- and long-term marginal effects are sometimes replaced by “operating” and “built” margins. A draw-back of this terminology is that the term built margin is somewhat misleading: changes in production capacity are not always the construction of new facilities; it may instead be the closure of existing production facilities. The long-term marginal effects of a change in energy use in the year 2020 can include technologies in energy plants that are constructed during the period 2025–2035, but they can also include technologies in energy plants that are shut down during the years 2020–2030.\n
Which concepts to use depends on the context. In communication with the general public, the rough meaning of the concepts should be easily understood. Operating and built margin are good terms to use in this context. In communication with researchers in the field, however, the precision of the concepts is important. Then it is probably better to talk about short- and long-term effects. In communication with policy-makers and professional actors in the industry, the appropriate choice of words may depend on the situation and the level of knowledge of the audience.\n
Changing demand for a product often gives rise to both short- and long-term marginal effects: the utilization rate is affected first, and after a while the change also contributes to new power plants being built or old ones being shut down. Changing demand can also affect investments in several different technologies, and these investments can in turn affect both the utilization rate of existing plants and other, future investments. This means that the full marginal effect is complex. The complex margin in an energy system can be estimated in an optimizing, dynamic model that can account for both the short-term and long-term margin changes . The complex marginal effect is then defined as the difference between the results of two model runs: one with the change in energy demand and one without it.\n
The complex margin is, in theory, the most correct to use for CLCAs whether the possible decisions involve changes in the short term (e.g., putting out a lamp) or the long term (e.g., changing the heating system in the house). This is because even short-term changes can produce long-term marginal effects. Investment decisions are based on assessments of the future demand and price of the product. These assessments are, in turn, affected by the current market situation. If we increase electricity consumption this year, we might contribute to investment decisions being taken next year or the year after that.\n
In practice, the complex marginal effects are very difficult to estimate. It requires model calculations over the relevant time period. Model runs suggest that this time period never ends, because indirect effects occur when new production plants must be replaced far into the future . Unfortunately, the uncertainty very far into the future is too great for modeling to be meaningful. The choice of time horizon in the model is subjective and depends on the time resolution in the model. If each year is modeled as a single or a handful of time slots, the model usually extends a couple or a few decades into the future [21, 22, 23, 24]. An hour-by-hour model is more likely to cover just a single year , although it can still be possible to model a few years where each model year represents, for example, a decade .\n
Identifying marginal effects with an energy system model requires special expertise. There are rarely resources to develop an energy system model within the framework of a specific LCA. With the right expertise, the marginal effects can be studied in an existing model. It is, of course, even easier to use results from published model runs as a basis for assumptions about the marginal effects. Assumptions about marginal effects of electricity use in Sweden can be based on results from, for example, Hagberg et al. . However, the simpler the method used to generate complex marginal data, the greater the risk that they do not reflect the marginal effects caused by the specific electricity use being studied.\n
Perhaps the biggest problem is that the uncertainty in complex marginal data is extremely large. Optimizing dynamic energy systems models indicate that the complex marginal effects of Swedish electricity use vary greatly depending on assumptions on, for example, investment costs, future fuel prices and policy instruments—where the two latter are highly uncertain . Completely different marginal effects can occur in a single electricity scenario, depending on whether the expansion of wind power in the scenario is assumed to be driven by an increased electricity demand or by other motives . A small change in the use of district heating can change the optimum development of an entire district heating system completely . This illustrates that the actual effects of a small change in demand are and will remain basically unknown. An optimizing dynamic systems model can remind us of the great uncertainty, but not give much knowledge of the actual marginal effects.\n
Referring to the criteria in Section 2, input data on complex marginal effects make the CLCA results more accurate, but just a little—particularly if these data are from previously published model runs. Generating case-specific complex marginal data leads to a method that is difficult to use. The use of complex marginal data also makes the study less comprehensible: it is a challenge to explain marginal results from an energy system model. This makes it more difficult for decision-makers to assess the relevance and validity of the results.\n
If complex marginal effects are to be introduced at all in a CLCA depends on the context. In many cases, it is probably better to use a method that is easier to use and explain. The LCA practitioner and the decision-makers should then be aware that the method used is simplified and that the actual marginal effects remain unknown.\n
A simplified method can be limited to focusing on short- or long-term marginal effects only. Since investment and closure decisions have consequences for the environment during a long time, such effects are typically more important for the environment than changes in the use of existing production capacity. In other words, the long-term marginal effects are typically more important for the environment than the short-term marginal effects .\n
In some cases, however, a change in demand cannot be expected to have any effect at all on the production capacity. This applies if the existing production system has a significant overcapacity and closure of existing plants is not a reasonable option. It also applies if the production capacity is expanded for political or other strategic reasons, rather than to cover an expected demand for the product. A change in current Swedish electricity use might, for example, not have any effect on new investment decisions, because there is an overcapacity in the North European electricity system and because wind and solar power is still being expanded for policy and strategic business reasons. On the other hand, a change in electricity use can contribute to keeping electricity prices up or down, which can make decisions on continued investments more or less difficult. There is also a long-term political ambition to phase out coal and nuclear power. A change in electricity demand can contribute to a quicker or slower closure of such power plants. This discussion reminds us that the actual marginal effects are difficult to foresee. Different assumptions are possible, even if the environmental assessment is limited to long-term marginal effects.\n
Another way to simplify things is to use the five-step procedure presented by Weidema et al.  to identify the production technology that is affected by a marginal change in demand. This procedure involves responding to five questions:
Is short or long term the relevant time perspective?
What market is affected? Here, both a geographical delimitation and a delimitation in different market segments may be required, for example, in base- and peak-load electricity or in eco-labeled and non-ecolabel products.
What is the trend in demand in this market? If demand declines faster than the natural turnover rate in production capacity, long-term marginal effects are assumed to consist of closure of existing plants; otherwise they are assumed to consist of investments in new facilities.
Which production techniques are flexible, that is, can vary their production volume in response to market demand?
Which technology will be affected? If the marginal effect is an investment, it is assumed to be in the technology that is cheapest to expand. If the marginal effect is a closure, it is assumed that it is in the technology that is most expensive to utilize.
This five-step procedure can be used in CLCAs of a wide range of products. The procedure points at a single technology where the marginal effect occurs. This contributes to making the CLCA approach feasible and comprehensible—but at the cost of simplifying assumptions: that the relevant effects are either short-term or long-term rather than both, that markets and market segments can be clearly distinguished and do not affect each other, that the production volume of a technology is either completely flexible or not at all flexible, and that decisions are based solely on economic rationality. Each of these simplifications reduces the accuracy of the CLCA results. The LCA practitioner and the user of the LCA results should both be aware of this. The five-step procedure can be described as a structured way to arrive at an assumption of the marginal effects, rather than a method of identifying the actual marginal effects.\n
Another approach is to collect information on plans to close and/or expand the production capacity and assume that the built margin is the mix of technologies in these plans. This is also an assumption, because plans do not always come true  and because some of the closure and investment decisions might be driven by policy or business strategies rather than by the demand for the product.\n
Assumptions about the marginal effects can, of course, be made even without a structured or formal procedure. Long-term marginal effects in the electricity system can, as the first approximation, be assumed to be electricity production in new natural gas-fired power plants, as they have an environmental performance that is better than some possible marginal techniques but worse than others. A possible sensitivity analysis can be based on data from old coal power or old nuclear power, as the closure of such power plants can be included in the long-term marginal effects and because they are near opposite ends of the scale for several important environmental impacts. Similarly, a first approximation and the extreme values can be identified for marginal production of other products.\n
To simply make an assumption is likely to be the easiest method to produce marginal data for the environmental assessment. On the other hand, pure assumptions make the study less accurate. They can also make the study less comprehensible in the sense that the basis for the assumptions can be difficult to communicate. If the assumptions appear arbitrary, the study also becomes less credible, which reduces the likelihood that the results inspire decisions and actions.\n
A single production process often serves many different life cycles: diesel from a single refinery and steel from a steel mill can be used in almost any life cycle. If the production process generates a single type of product (e.g., steel), this is not considered a problem in LCA. We obtain input data to the calculations by simply dividing the total environmental burdens of the process by the total production, the functional output, of the process. The resulting input data are an average for that process and, hence, most suited for an ALCA. In a thorough CLCA we should ideally instead use input data that reflect how the environmental burdens of the process change as a result of a change in the total functional output. This is still a straightforward process, at least in theory.\n
A methodological problem occurs when the process generates more than one type of product or function, which are used in different life cycles. A refinery, for example, produces many different fuels and materials. A steel mill might produce residual heat besides the steel. A CHP plant produces electricity and heat. Waste incineration serves the function of treating many different waste flows and might, at the same time, generate electricity, residential heating, and/or process steam. The problem is to decide on how quantify the total functional output of the multifunctional process and, hence, how to allocate the environmental burdens of the process to the various life cycles it serves. The approach to this problem depends on whether the LCA is an ALCA or a CLCA.\n
An ALCA aims to estimate what share of the global environmental burdens belongs to the product investigated. Faced with the allocation problem, the task is to estimate what share of the burdens of the multifunctional process belongs to the product investigated and also what share of input materials, energy, etc. The basis for this allocation has to be a property that the products and/or functions of the process have in common: mass, energy content, economic value, etc. The total output of the process can be quantified in terms of this property, and the burdens of the process can be partitioned and allocated to the different products/functions in proportion to this property.\n
What properties the products and functions have in common varies between multifunctional processes:
A refinery: mass, energy, exergy, and price
A CHP plant producing electricity and heat: energy, exergy, and price
A steel mill with residual heat: price
Waste incineration with energy recovery: price
As indicated from this short list, the price is sometimes the only possible basis for allocation. In many ALCAs, it is the only allocation key that can be consistently used throughout the life cycle. Economic value can also be considered a valid basis for the allocation, since the economic value of the products is a proxy for their contribution to the expected profit from the process. The expected profit is typically the reason for investing and running the process and, hence, the cause of its impacts on the environment [12, 29].\n
Economic allocation is often criticized because it will make the LCA results vary as prices change over time. However, the LCA results can be made more stable by using the average price over a period of several years as basis for the allocation. This will also more precisely reflect the causality, because the expected profit is more likely to depend on the average price than on the price at a specific point in time.\n
There are cases where the economic value does not reflect a causality, because the processes are not driven by the expected profit but by concern for, e.g., the environment. These include noncommercial processes such as municipal wastewater treatment plants  and landfills. In these cases, the economic value is less valid and might not even be possible to use as basis for the allocation.\n
When we choose an allocation key, we might account for what the intended audience considers to be fair. This increases the legitimacy of the study in their eyes, which increases the chance of the LCA leading to decisions.\n
The choice of allocation method also depends on how feasible it is. If the allocation problem is not important for the results and conclusions of the ALCA, the easiest methods can be used to keep the cost of the study down. This can include allocating all burdens to the main product of the process—for example, to the steel from the steel mill with residual heat.\n
A CLCA aims to estimate how the global environmental burdens are affected by the production and use of the product investigated. Faced with a multifunctional process, the task is to estimate how the flows of the process are affected: the flows of input materials and energy, the emissions and waste flows, and the output of each product and function. When the output of products and functions for use in other life cycles are affected, the CLCA system should ideally be expanded to include the processes that are affected by this change in flows.\n
A change in the demand for one of the products from a multifunctional process can affect decision-makers running the process and other actors in various ways that are difficult to predict and model. To make the CLCA approach feasible, we can choose to divide the multifunctional processes into three idealized cases :
Independent production: a change in the demand for the product investigated affects the output of this product but not the flow of other products and functions from the process.
Use of main product in joint production: an increase in the demand for the product investigated drives the process and increases the output of all its products and functions proportionally.
Use of by-product in joint production: a change in the demand for the product investigated does not affect the process or any of its outputs; instead it affects the alternative use of the by-product.
The idealized cases are simplifications of reality: products from a multifunctional process are rarely produced completely independent of each other , and the process is rarely driven by only one of the functional outputs.\n
If the products of the multifunctional process are independently produced, the input data for each of the products should reflect how the environmental burdens of the process change when the production of this product changes while the production volume is constant for the other products.\n
If the CLCA includes the use of the main product from a joint multifunctional process, the LCA model should include this process and also the processes affected by a change in the volume of by-products. The latter are typically assumed to be the production of products that compete with and are substituted by by-products from the multifunctional process (see Figure 4). Since the study is a CLCA, the competing production should ideally be modeled based on marginal data (cf. Sections 3 and 4.2).\n\n
If the CLCA instead includes the use of a by-product, the operation of the multifunctional process is assumed to be unaffected by the demand for this product. The use of such a by-product does not affect its production; instead, it affects how much of the by-product is available for other purposes. The CLCA model should include affected processes only, which means it should not include the multifunctional process. Instead, the model ideally includes the marginal, alternative use of the by-product. This is the use affected by a (usually marginal) change in supply of the by-product (Figure 5).\n\n
In some cases, the by-products are not fully utilized: for example, part of the residual heat from a steel mill might be cooled off, and part of a residual material might be disposed as waste. In such cases, a change in the use of the by-product is not likely to affect the alternative use but instead how much of the by-product needs to be cooled off or disposed of in some other way. The CLCA should include the affected disposal process.\n
Note that the “expanded” system in Figure 5 is not necessarily larger than the original system. It does not include the multifunctional process or the production of fuel and other raw materials for that process. Instead, it includes the disposal or alternative use of the by-product and any foreseeable consequences thereof.\n
The easiest method, such as ignoring the production of by-products, can be applied in the CLCA if the choice of approach is not important for the results and conclusions of the study. However, more information is required to decide on such a cut-off in a CLCA, compared to an ALCA. Even if a multifunctional process has little environmental burdens, making it unimportant in an ALCA, a change in this process might have environmentally important consequences elsewhere, hence making it significant for the CLCA.\n
Attributional and consequential LCA have both advantages and disadvantages [9, 32]. This section discusses the choice between ALCA and CLCA using the criteria described in Section 2. The intention is not to determine what kind of LCA is superior but to discuss and explain their strong and weak aspects. The intention is also to show how the criteria in Section 2 can be used systematically to structure a discussion and assessment of methodological options.\n
In a CLCA, the system model often needs to be expanded (Section 5.2), which requires environmental data on more processes and also economic data on the markets affected by the production and use of the product investigated (cf. Section 4.2). The databases that exist today usually include average data, but few include marginal data—Ecoinvent 3 is a notable exception, although its marginal data are rough. All of this means that a CLCA risks becoming unfeasible or at least significantly more expensive than an ALCA. On the other hand, the CLCA can exclude parts of the life cycle that are not affected by the production of by-products (cf. Section 5.2). The cost of CLCAs can also be reduced by limiting the study to the consequences expected to be the most important for the conclusions.\n
With time, CLCAs may become easier to carry through if future databases include more of marginal data.\n
A CLCA generates information on the environmental impact of a specific decision or information on how a decision-maker can affect the environment. This is just the accurate information to have as a basis for decisions that contribute to reducing the total negative environmental impacts or, at least, the impact per functional unit.\n
An ALCA might be more precise and comprehensive, because a detailed and comprehensive CLCA might be too expensive or even unfeasible to carry through (see Section 6.1). As an ALCA is refined, it becomes more detailed, and the results converge toward an exact response to the attributional question: how much of the world’s environmental impact belongs to the product studied? However, even a very precise answer to this question will in some cases guide decisions in the wrong direction, because the impacts belonging to a product are not the same as the consequences of producing and using this product (see Figure 2).\n
Refining a CLCA can involve accounting for more causal relationships. This makes the CLCA more comprehensive, but it does not necessarily mean that the results converge toward a final answer. On the contrary, as an additional causal relationship is included in the calculations, the results might shift completely and point in another direction.\n
The CLCA provides, by definition, more information on how decisions affect the environment; however, if the CLCA results are highly uncertain and do not converge toward a final, true result, the CLCA might not guide decisions in the right direction more often than an ALCA.\n
An ALCA is based on the concepts “life cycle” and “value chain” which are intuitively clear and easy to communicate. The system model in an ALCA usually has a simple structure, which means that it can easily be presented in a way that is transparent, at least in principle. The high level of detail that can be achieved, however, makes the study bulky and can make it a challenge in practice to communicate to decision-makers and other stakeholders.\n
The basic concept in a CLCA is “consequences.” This is also intuitively easy to understand. However, other concepts required to understand the study (marginal production, partial equilibrium, etc.) are more difficult to grasp. The system model is also more complex with environmental burdens, avoided burdens, and additional, indirect burdens and with models of markets between the models of production processes. Making such a study comprehensible to decision-makers and stakeholders can be very difficult.\n
An ALCA can be interpreted to distribute responsibility and guilt for environmental impact, and recognition and goodwill for environmental improvements in the value chain, a part of the technological system that is linked to the production and use of the product through contracts and/or physical flows. An LCA model based on such clear links can be perceived as a relevant basis for choosing between products and for decisions on changes in the product. If the choice of allocation methods and system boundaries is accepted by the decision-maker, the results will also be perceived as fair and legitimate. However, they can be questioned by actors who have other, subjective perspectives on what is fair and right.\n
The fact that a CLCA provides information on how possible decisions affect the environment can also be perceived as very relevant to the decision-maker. Rational decision-making requires information on the consequences of the decision. However, the CLCA typically include indirect consequences occurring in processes to which the product is not linked through physical flows or contractual obligations. The decision-maker might not want to be held responsible for such consequences. In order to account for them anyway, the decision-maker probably needs to be driven by the desire to actually improve the environment, rather than simply getting recognition for good environmental performance.\n
The ALCA practice is more well-established than CLCA. Environmental product declarations, a specific application of ALCA, also have detailed guidelines specifying the method . In other applications, ALCA requires subjective choices of system boundaries (Section 4.1) and allocation methods (Section 5.1). However, the ALCA results are somewhat less sensitive to subjective choices than CLCA where the results might shift from positive to negative depending on system boundaries and assumptions. All this implies that ALCA is more robust and more resistant to abuse in the sense that the results depend less on who is doing the study.\n
The actual consequences of a decision are almost always highly uncertain. If the sensitivity analysis of a CLCA takes full account of the great uncertainty, the study will rarely reach clear conclusions. This increases the risk of decisions and actions not being taken, especially if the actions are expensive or undesirable in other ways. The large uncertainty in the actual consequences makes it easy to misuse CLCA results to cast in doubt environmentally desirable decisions.\n
However, when the ALCA is completed, the results can be abused if presented as a basis for decisions. This is because the ALCA does not aim to investigate the consequences of the decision on the environment. In a country with little fossil-based power production, such as Norway or Sweden, an ALCA can, for example, conclude that energy efficiency is not important for electric appliances. It can also indicate that residential heating should be provided through heat pumps rather than district heating from CHP plants fired with natural gas and perhaps even biofuel. A CLCA would not be likely to produce such results. If and when CLCA practice becomes more established, it will also become somewhat more difficult to abuse.\n
Attributional and consequential LCA respond to different questions: what part of the global environmental impacts is associated with the product investigated, and how does the product affect the global environmental impacts? In most applications and for most study objects, the choice between ALCA and CLCA is open. Since the two types of LCA have different advantages and disadvantages, it cannot be unequivocally stated that one is better than the other . Roughly stated, the CLCA is more accurate, while ALCA have advantages when it comes to all other criteria. However, what kind of study is easiest to understand and most inspiring will vary between different decision-makers.\n
To ensure that the study is perceived as relevant, it is a good idea to, if possible, discuss the goal and scope of the study with the client before deciding on what type of LCA to carry through. To make the study as legitimate as possible, it might also be useful to discuss with other stakeholders. In such a discussion, it is important to carefully explain what type of information is provided by an ALCA and a CLCA. Figure 2 can be used in that explanation. It is also important to make clear the limitations of the different methods. Only then can the client and other stakeholders decide on the type of study they want.\n
As should be clear from this chapter, the actual effects of a decision on the global environmental impacts are in most cases highly uncertain. We will never know how close the CLCA results are to reflect the actual consequences. For this reason, CLCA should probably not be presented as a method to estimate the actual consequences. Instead the results are the consequences foreseeable within the methodological framework we choose to use in the study.\n
The risk that the study will be abused will also vary from case to case. Here, it does not help to consult the client. The LCA practitioners must instead use their own judgment and decide what kind of LCA is the most appropriate. In this decision, it may be good to consult with colleagues and/or to discuss with other stakeholders. The decision to make an ALCA or a CLCA should therefore be taken by the LCA practitioner after discussions with the client and possibly with other stakeholders and colleagues.\n
This chapter was completed with financial support through the EU ERA-Net Sumforest project BenchValue (Formas Grant Number 2016-02113). It is based on translated parts of a report in Swedish, “Miljöbedömning av energibärare – vägledning för livscykelanalyser,” produced within the framework of the North European Energy Perspectives Project (NEPP). Jenny Gode, Mathias Gustavsson, Erik Dotzauer, and Lisa Hallberg contributed valuable input to this report. The methods and concepts presented in this chapter were to a large extent developed by or in collaboration with Bo Weidema, Göran Finnveden, and other international LCA experts. However, the author is solely responsible for the final content of the chapter.\n