If inhibiting S6 kinase decreases protein translation, then could inhibiting S6 kinase could possibly slow down long-term potentiation in neurons?

If inhibiting S6 kinase decreases protein translation, then could inhibiting S6 kinase could possibly slow down long-term potentiation in neurons?

We are searching data for your request:

Forums and discussions:
Manuals and reference books:
Data from registers:
Wait the end of the search in all databases.
Upon completion, a link will appear to access the found materials.


Phosphorylation of S6 induces protein synthesis at the ribosome.

P70S6 kinase is in a signaling pathway that includes mTOR (the mammalian target of rapamycin). mTOR can be activated in distinct ways, thereby activating p70S6K. For example, branched chain amino acids such as leucine are sufficient to activate mTOR, resulting in an increase in p70S6K phosphorylation (and thereby activating it). mTOR is also in a pathway downstream of the kinase Akt. Akt is typically activated upon stimulation of a cell with a growth factor (such as IGF-1). Akt then activates mTOR (by inhibiting the Tsc complex), leading to p70S6K activation.

We also found a paper showing that downregulation of S6 kinase also results in a decrease in protein translation in yeast (but I'm waiting for Matt Kaeberlein to email me the slides from yesterday).

I can't rule it out, but it sounds a lot like trying to tune a piano with sledgehammer.

Neuronal LTP depends on protein translation, but so does absolutely everything else in the cell. Inhibiting protein synthesis at the ribosome will block the formation of all proteins, not just the ones responsible for LTP. Unless there's a link I don't know about between LTP and total levels of protein translation, you're really going to want to look into inhibiting the production of proteins specifically responsible for LTP and not protein synthesis in general.

The genetics of ageing

The nematode Caenorhabditis elegans ages and dies in a few weeks, but humans can live for 100 years or more. Assuming that the ancestor we share with nematodes aged rapidly, this means that over evolutionary time mutations have increased lifespan more than 2,000-fold. Which genes can extend lifespan? Can we augment their activities and live even longer? After centuries of wistful poetry and wild imagination, we are now getting answers, often unexpected ones, to these fundamental questions.

Access options

Get full journal access for 1 year

All prices are NET prices.
VAT will be added later in the checkout.
Tax calculation will be finalised during checkout.

Get time limited or full article access on ReadCube.

All prices are NET prices.


Because it is believed DEPTOR functions predominantly as a modulator of mTOR signaling, reviewing basics of mTOR biology is essential for this review. In the next paragraphs, we review the protein composition of mTORC1 and mTORC2, the signals regulating their activation and the biological processes they control.

mTOR is a highly conserved kinase that belongs to the phosphoinositide 3-kinase (PI3K)-related kinase family. In addition to its kinase domain, this large protein of 289 kDa contains several other conserved structures. An overview of these domains and a complete summary of their functions have been extensively reviewed and are beyond the scope of this review (150). The presence of HEAT domains in mTOR, a domain that promotes protein-protein interaction, was a first indication that mTOR could interact with other proteins. Supporting this possibility, gel-filtration assays revealed a migration profile for mTOR that was much larger than its predicted size. Despite these observations, conventional attempts to purify mTOR-interacting proteins failed for a long time, likely due to the instability of the interactions under standard purification conditions. Using various approaches to circumvent these limitations, independent research groups finally succeeded in identifying mTOR partners (92, 105, 120, 121, 135, 136, 199, 200, 202, 218, 219, 251, 263, 271). These studies revealed that the mTOR kinase forms two distinct protein complexes, namely mTORC1 and mTORC2. It is now well appreciated that the specific architecture of mTORC1 and mTORC2 dictates the ability of each complexes to integrate signals, to target different substrates, and to selectively control specific biological processes. An overview of these complexes and the signaling networks in which they participate is presented in the following sections. We refer the readers to other reviews for more complete description of the mTOR signaling pathway (150, 224, 308).


The mTOR complex 1 (mTORC1) is composed of 1) the mTOR kinase 2) regulatory-associated protein of mTOR (RAPTOR) (105, 135) 3) Dishevelled, Egl-10, and Pleckstrin (DEP) domain-containing mTOR-interacting protein (DEPTOR) (202) 4) mammalian lethal with Sec13 protein 8 (mLST8, also known as GβL) (136) and 5) proline-rich AKT substrate 40 kDa (PRAS40) (218, 251, 263, 271) ( FIGURE 1A, LEFT). mTORC1 is at the center of a signaling hub that integrates signals from different sources to regulate anabolism, growth, and proliferation. An overview of the mTORC1 signaling network is presented in FIGURE 2 . In addition to growth factors and nutrients, which are the best characterized signals that activate mTORC1, this complex can also be regulated by other inputs, including hypoxia, inflammation, energy deficit, cholesterol, and nucleotides ( FIGURE 1A, RIGHT). When active, mTORC1 promotes metabolism and the synthesis of various building blocks to support cell growth and proliferation, including proteins, lipids, and nucleotides (150). This complex also inhibits catabolism by blocking autophagy and lysosome biogenesis. Although the list of proteins that mediates the effect of mTORC1 on all of these processes is constantly growing, its best characterized substrates remains ribosomal protein S6 kinase 1 (S6K1) and eukaryotic translation initiation factor 4E-binding proteins 1 and 2 (4EBP1/2). The phosphorylation states of these proteins are used as gold standards to monitor mTORC1 activity. As briefly mentioned above, rapamycin inhibits mTORC1 by forming a complex with FKBP12, which directly binds to the FRB domain of mTOR (22, 216). This interaction narrows the catalytic cleft to partially occlude substrates from the active site (289). Although rapamycin is extremely potent to block mTORC1 activity toward many substrates, some functions of mTORC1 are insensitive to inhibition by rapamycin (253).

On activation, mTORC1 triggers several negative feedback loops to inhibit growth factor signaling ( FIGURE 3 ). The activation of S6K1 downstream of mTORC1 induces the phosphorylation of insulin receptor substrate-1 (IRS1) on specific serine residues, which destabilizes IRS1 and reduces the ability of growth factors to activate signaling (106, 261). Additionally, mTORC1 per se can directly target IRS1 to promote its degradation (260). Another mTORC1-mediated negative feedback loop involves growth factor receptor-bound 10 (GRB10), an endogenous inhibitor of receptor tyrosine kinases. Studies show that mTORC1 directly phosphorylates and stabilizes GRB10, which, in turn, restricts signaling by binding to receptor tyrosine kinases (113, 296). A description of the feedback loops linking mTORC1 to PI3K signaling have been extensively reviewed by others (82, 214). All of these feedback loops emerging from mTORC1 are thought to have evolved to allow a better cellular control over the energy-consuming processes that are turned when anabolism is activated. It is important to point out that these feedback loops are core components of the mTOR signaling network and that their physiological and clinical impacts are profound, despite being often overlooked.


The mTOR complex 2 (mTORC2) is composed of 1) the mTOR kinase 2) rapamycin-insensitive companion of mTOR (RICTOR) (121, 219) 3) DEPTOR (202) 4) mLST8 (or GβL) (121) 5) protein observed with Rictor-1 and 2 (PROTOR1/2) (165, 199, 251) and 6) mammalian stress-activated protein kinase interacting protein (mSin1) (92, 120). Although some proteins are unique to mTORC2, some components, including DEPTOR, are found in both complexes ( FIGURE 1B, LEFT). mTORC2 integrates signals from growth factors to regulate processes including metabolism, survival, cytoskeletal organization, and cell mobility ( FIGURE 1B, RIGHT) (150, 308). On activation, mTORC2 phosphorylates several AGC kinases, including serum- and glucocorticoid (GC)-induced protein kinase 1 (SGK1), protein kinase C-α (PKC-α), and protein kinase B/AKT (AKT). The best characterized substrate of mTORC2 is AKT, whose activity is directly regulated by mTORC2 via the phosphorylation of serine 473 (221). Unlike mTORC1, mTORC2 is insensitive to amino acids, and its activity is not acutely inhibited by rapamycin (121, 219). Long-term treatment with rapamycin inhibits the activity of mTORC2 in some, but not all, cell lines (203, 220, 289). The inhibition of mTORC2 by rapamycin was also observed in mice (148, 225). An overview of the mTORC2 signaling network is presented in FIGURE 2 .

The strange case of AMPK and cancer: Dr Jekyll or Mr Hyde? †

The AMP-activated protein kinase (AMPK) acts as a cellular energy sensor. Once switched on by increases in cellular AMP : ATP ratios, it acts to restore energy homeostasis by switching on catabolic pathways while switching off cell growth and proliferation. The canonical AMP-dependent mechanism of activation requires the upstream kinase LKB1, which was identified genetically to be a tumour suppressor. AMPK can also be switched on by increases in intracellular Ca 2+ , by glucose starvation and by DNA damage via non-canonical, AMP-independent pathways. Genetic studies of the role of AMPK in mouse cancer suggest that, before disease arises, AMPK acts as a tumour suppressor that protects against cancer, with this protection being further enhanced by AMPK activators such as the biguanide phenformin. However, once cancer has occurred, AMPK switches to being a tumour promoter instead, enhancing cancer cell survival by protecting against metabolic, oxidative and genotoxic stresses. Studies of genetic changes in human cancer also suggest diverging roles for genes encoding subunit isoforms, with some being frequently amplified, while others are mutated.

1. Introduction

The AMP-activated protein kinase (AMPK) is best known as a sensor of both cellular [1–3] and whole body [4,5] energy status. AMPK is activated when ATP bound at a key site on its γ regulatory subunit is displaced by AMP and/or ADP, causing conformational changes that trigger allosteric activation, as well as promoting net phosphorylation (and consequent activation) of the catalytic subunit by upstream kinases. As ADP rises and ATP falls during situations of cellular energy stress, the reaction catalysed by adenylate kinases (2ADP ↔ ATP + AMP) is displaced rightwards, ensuring that AMP rises to an even larger extent than ADP [6], thus activating AMPK in a very sensitive manner. AMPK is also activated by increases in intracellular Ca 2+ [7–9], by glucose starvation [10] and by DNA damage [11–13] via non-canonical, AMP/ADP-independent mechanisms. By phosphorylating downstream targets that switch on catabolic pathways, while switching off anabolic pathways and other ATP-consuming processes such as progress through the cell cycle, AMPK not only promotes ATP synthesis but also restricts cell growth and proliferation in an attempt to restore energy homeostasis and maintain cell viability.

Given this propensity to switch off cell growth and proliferation, and the discovery that the principal upstream kinase phosphorylating and activating AMPK was the well-established tumour suppressor LKB1 [14–16], it seemed likely that AMPK would play a beneficial role (Dr Jekyll!) in cancer and act as a tumour suppressor. There is indeed evidence supporting this, at least in some cancer types, as well as for the obvious corollary that AMPK activators should delay tumorigenesis in those cancers. However, there is contrasting evidence that, in other contexts, the presence of AMPK may play a malevolent role (Mr Hyde!) to promote cancer, most likely by protecting transformed cells against stresses caused either when their growth rate outstrips the ability of their blood supply to deliver nutrients and oxygen or during periods of oxidative stress and/or DNA damage. In such scenarios, the presence of AMPK would increase the viability of the tumour cells and thereby potentially decrease survival of the patient, and in such cases it would be AMPK inhibitors rather than activators that might be therapeutically useful. The purpose of this review is to attempt to reconcile these two apparently conflicting roles of AMPK, and to discuss the different types of situation in which activators or inhibitors of the kinase might be efficacious.

2. AMPK—structure and regulation

AMPK appears to exist universally as heterotrimeric complexes comprising catalytic α subunits and regulatory β and γ subunits. Genes encoding these three subunits are found in the genomes of essentially all eukaryotes, suggesting that the AMPK system evolved very early during eukaryotic evolution [2]. In mammals, there are multiple genes encoding each subunit, generating two α (α1, α2), two β (β1, β2) and three γ subunits (γ1, γ2, γ3). These paralogues appear to have arisen during the two rounds of whole genome duplication that are thought to have occurred during the early development of the vertebrates [3]. The seven gene products (not counting splice and/or start-site variants) can form up to 12 αβγ combinations that display subtle differences in regulation and in tissue and subcellular distribution [3].

Crystal structures of three αβγ combinations from humans, i.e. α2β1γ1 [17], α1β1γ1 [18] and α1β2γ1 [19], as well as partial structures from mammals [20,21], budding yeast [22] and fission yeast [23,24], are now available. The generalized structure of a heterotrimeric AMPK complex is represented in a highly schematic form in figure 1. A current limitation of the existing structures of heterotrimeric complexes is that, in every case, the constructs were crystallized in active conformations, with the catalytic subunit phosphorylated at the activation site and allosteric activators bound at the regulatory sites. Due to the lack of structures in inactive conformations, we still only have a partial understanding of the conformational changes involved in the activation process.

Figure 1. Schematic diagram of the structure of AMPK heterotrimers, with the different subunits colour coded (α, yellow β, lilac γ blue). Based on a structure of the human α1β2γ1 complex [19], although the structures of α2β1γ1 [17] and α1β1γ1 [18] complexes are very similar.

2.1. Structure of the α subunits

Each AMPK-α subunit (coloured yellow in figure 1) has an N-terminal kinase domain with the small N-terminal lobe and larger C-terminal lobe typical of all members of the eukaryotic protein kinase (ePK) family, with the ATP-binding catalytic site in the cleft between the two lobes. Like many other ePKs, AMPK is only significantly active after phosphorylation within the so-called ‘activation loop’ of the C-lobe. In AMPK, the target for phosphorylation is a highly conserved threonine residue, which is conventionally referred to as Thr172 [25] although the exact residue numbering varies with species and isoform (in the view of figure 1, Thr172 is located on the far side of the C-lobe). In other ePKs, phosphorylation within the activation loop changes its conformation to reorient residues involved in both catalysis and protein substrate binding, thus greatly enhancing the reaction rate [26]. The principal upstream kinase phosphorylating Thr172 on AMPK was identified in 2003 to be a complex containing LKB1 and two accessory subunits, STRAD-α or -β and MO25-α or -β [14]. Binding of STRAD-α or -β (which are pseudokinases, structurally related to protein kinases but not active) is required for the kinase activity of LKB1, whereas MO25-α or -β appear to have a structural role to stabilize the complex [27]. The gene encoding LKB1 (called STK11 in humans) had been previously identified as being involved in Peutz–Jeghers syndrome, a rare inherited cancer susceptibility humans with this syndrome are almost always heterozygous for loss-of-function mutations in STK11 [28]. Their major clinical problem is the development of frequent but benign intestinal polyps, which appear to be caused by haploinsufficiency in STK11. However, they also have a greatly increased risk of developing malignant cancers at multiple locations due to loss of heterozygosity in STK11, and often die at a relatively early age from such malignancies [28]. Loss-of-function mutations in STK11 also frequently occur in many sporadic (i.e. non-inherited) cancers, especially in the commonest form of lung cancer, adenocarcinoma [3,29,30] (see also §6). LKB1 is therefore a classical tumour suppressor, and although its sequence showed that it was a member of the ePK family, the downstream target(s) that it phosphorylated were completely unknown until the finding that it phosphorylated and activated AMPK [14–16].

Following the kinase domain on each AMPK-α subunit (figure 1) is a compact bundle of three α-helices termed the autoinhibitory domain (α-AID) [18–20,24]. When ATP rather than AMP is bound at the regulatory site(s) on the γ subunit (see below), the α-AID is thought to interact with the kinase domain to clamp it in an inactive conformation [24]. The α-AID is linked to the globular C-terminal domain of the α subunit (α-CTD) by the α-linker, shown schematically as a yellow chain in figure 1. The α-linker is a region in an extended conformation that contains two conserved segments termed α-regulatory interaction motifs (α-RIM1 and α-RIM2) [31]. These interact with the surface of the γ subunit containing the key regulatory adenine nucleotide-binding site (see §2.3), and movement of this linker is thought to transmit the effects of AMP or ADP binding from the regulatory γ subunit to the catalytic α subunit (see §2.4).

2.2. Structure of the β subunits

The β subunits (coloured lilac in figure 1) contain two conserved regions, the central carbohydrate-binding module (β-CBM) and the C-terminal domain (β-CTD), these being the only regions of the β subunits that are resolved in the current heterotrimer structures. The β-CBM causes a proportion of AMPK in mammalian cells to bind to glycogen particles [32,33]. One function of this may be to co-localize AMPK with glycogen synthase, the key enzyme of glycogen synthesis also found at the surface of the glycogen particle, both isoforms of which are phosphorylated and inactivated by AMPK [34,35]. The β-CBM, however, also has other functions (see §3.2 below). The β-CTD, on the other hand, plays a key structural role as the ‘core’ of the heterotrimeric complex, in that it cross-links the α-CTD and the γ subunit, via interactions that are highly conserved from fungi to mammals [21–23].

2.3. Structure of the γ subunits

The γ subunits (coloured blue in figure 1) are of particular interest because they contain the regulatory adenine nucleotide-binding sites. In all species, the γ subunits contain four tandem repeats of a sequence motif of around 60 amino acids known as a CBS repeat, so-named by Bateman [36] because they are also present in the enzyme Cystathione β-Synthase and invariably occur as tandem repeats. CBS repeats have been identified in around 75 proteins in the human genome [37], and are also found in archaea and bacteria. Proteins containing them usually have just two tandem repeats, but the AMPK-γ subunits are unusual in having four. A single pair of repeats (known as a Bateman domain or module) forms a pseudodimer with a cleft between the repeats that (due to the approximate twofold symmetry) can provide two ligand-binding sites, although often only one is used. Bateman modules usually bind regulatory ligands containing adenosine (e.g. AMP, ATP, S-adenosyl methionine, NAD, diadenosine polyphosphate) or, less often, guanosine [37,38]. Consistent with this, the CBS repeats in the AMPK-γ subunits provide the critical binding sites for the regulatory nucleotides AMP, ADP and ATP [38]. The four CBS repeats in every AMPK-γ subunit form two Bateman modules that assemble ‘head to head’ to form a flattened disc with the adenine nucleotide-binding sites located close together in the centre, lining a narrow aqueous channel (figure 1) [17–19,21]. Given the presence of four repeats, it might have been expected that AMPK-γ subunits would bind four molecules of nucleotide, but all existing crystal structures suggest that they bind only three. These sites are now numbered according to which repeat in the linear sequence (CBS1 through CBS4) provides residues that bind the adenosine moiety of the nucleotide [39] (the phosphate groups may interact with residues from more than one repeat). Using this nomenclature, adenine nucleotides bind at CBS1, CBS3 and CBS4, while the CBS2 site appears to be always unoccupied. The CBS3 site is primarily accessible to solvent from one side of the disc of the γ subunit (facing the viewer in figure 1), and the CBS1 and CBS4 sites from the other.

2.4. Canonical regulation of AMPK by adenine nucleotides

AMP-activated protein kinase received its name [40] because it is allosterically activated by 5′-AMP [41]. When the assays are performed at physiologically relevant ATP concentrations (5 mM) allosteric activation can be as much as 10-fold [42]. However, even before LKB1 was identified as the upstream kinase and Thr172 as the phosphorylation site, it was realized that increases in the AMP : ATP ratio also promoted net phosphorylation of AMPK in intact cells [43]. This is now known to occur because AMP both enhances phosphorylation by LKB1 [44,45] and inhibits dephosphorylation by protein phosphatases [46]. Both effects are due to the binding of AMP to the substrate, AMPK, and not to the upstream kinase or phosphatase indeed the LKB1 complex appears to have a constant activity in both energy-stressed and unstressed conditions [47]. To summarize, AMP binding has three effects on AMPK: (i) promoting Thr172 phosphorylation (ii) inhibiting Thr172 dephosphorylation (iii) triggering allosteric activation of kinase already phosphorylated on Thr172. These three mechanisms act synergistically and make the system respond to small increases in AMP in a very sensitive manner. It was subsequently reported that the effects of AMP binding on Thr172 phosphorylation [48] and dephosphorylation [20], although not on allosteric activation, could be mimicked by ADP, at least in cell-free assays. Our group has confirmed this, but found that the effect required concentrations of ADP up to 10-fold higher than those of AMP, at least for complexes containing γ1 and γ3 (γ2-containing complexes are more sensitive to ADP) [42,45]. Overall, we believe that increases in the AMP : ATP ratio remain the most important activating signal in vivo, although increases in the ADP : ATP ratio might make a secondary contribution.

CBS4 normally appears to contain a tightly bound ‘non-exchangeable’ molecule of AMP [21]. Similarly, although CBS1 can bind AMP in cell-free assays, it is estimated to have a 10-fold higher affinity for ATP than AMP. Since ATP is usually present in cells at up to 100-fold higher concentrations than AMP, this suggests that, in intact cells, CBS1 would always be occupied by ATP [49]. This leaves CBS3 as the site where ATP and AMP (or ADP) could exchange with each other.

The R531G mutation in the AMPK-γ2 subunit, one of up to 14 mutations that cause an inherited heart disease [50], completely blocks both allosteric activation and increased net Thr172 phosphorylation by AMP [38,51]. Although it is actually located in CBS4, the positively charged side chain of Arg531 interacts with the α-phosphate of AMP bound in CBS3 [21,49] (note that in [21] the CBS3 site was referred to as site 1 see [39] for revised nomenclature of binding sites).

How is the effect of displacement of ATP by AMP at CBS3 transmitted to the catalytic (α) subunit? Consistent with the idea that CBS3 is the critical site for activation, the heterotrimer structures show that, when AMP is bound at CBS3, the α-linker binds to that face of the γ subunit, with α-RIM1 binding across the unoccupied CBS2 site and α-RIM2 physically contacting AMP bound at CBS3 (figure 1). Although there are no crystal structures to confirm this, other biophysical approaches suggest that, when ATP displaces AMP at CBS3, the α-linker dissociates from the surface of the γ subunit containing the CBS3 site [19,52]. This is thought to release the α-AID to rotate back into its inhibitory position behind the kinase domain, with this being prevented when AMP is bound at CBS3 by the interaction of the α-linker with the CBS3 site. Consistent with this model, mutations that would affect the interactions between α-RIM1/α-RIM2 and the γ subunit abolish allosteric activation by AMP [31].

While this model nicely accounts for allosteric activation by AMP, the accompanying conformational changes may also alter the exposure of Thr172 for phosphorylation and dephosphorylation, although that aspect is currently less well understood. It also remains unclear why ADP binding has effects on Thr172 phosphorylation despite the fact that, unlike AMP, it does not cause allosteric activation. Finally, as well as the ‘canonical’ activation by changes in adenine nucleotide ratios discussed above, AMPK can also be activated by several non-canonical mechanisms that will now be briefly described.

2.5. Non-canonical activation by increases in intracellular Ca 2+ , by glucose deprivation and by DNA damage

As well as LKB1, Thr172 can also be phosphorylated by the Ca 2+ /calmodulin-dependent protein kinase CaMKK2 [7–9], which means that AMPK can be activated by increases in intracellular Ca 2+ ions even in the absence of any changes in adenine nucleotide ratios. This occurs, for example, in response to hormones and agonists sensed by G protein-coupled receptors that are coupled via Gq/G11 to release inositol-1,4,5-trisphosphate (IP3) from the plasma membrane, which in turn triggers release of Ca 2+ from the endoplasmic reticulum. Such agonists include, in endothelial cells, thrombin acting at protease-activated receptors and vascular endothelial cell growth factor acting at VEGF receptors [53,54] as well as, in specific neurons of the hypothalamus, ghrelin acting at GHSR1 receptors [55]. The latter effect is important in promotion of appetite during fasting [5], and the role of CaMKK2 in this pathway can explain previous findings that CaMKK2 inhibitors depress appetite in wild-type mice, although not in CaMKK2 knockouts [56].

It has been known for many years that glucose deprivation of mammalian cells activates AMPK [57], and this treatment is often used to switch on AMPK in cultured cells. In fact, genes encoding the budding yeast orthologue of AMPK (the SNF1 complex) were originally identified via mutations that prevented the normal changes in gene expression in response to glucose deprivation [58]. For many years, it was assumed that glucose deprivation activated AMPK by interfering with catabolic ATP production, and thus activated AMPK via the canonical, AMP-dependent mechanism (§2.4). This does indeed seem to be the case in some established tumour cell lines, perhaps because they are highly glycolytic and have a high dependency on glucose for ATP production. However, in other cells such as immortalized mouse embryo fibroblasts (MEFs) it has been found that glucose deprivation activates AMPK without changing AMP : ATP or ADP : ATP ratios, as long as an alternative carbon source such as glutamine is available similar AMP/ADP-independent activation is also observed in rat liver during starvation in vivo [10]. In such cases, activation is thought to occur via a complex mechanism involving the direct sensing of the glycolytic intermediate fructose-1,6-bisphosphate (FBP) by FBP aldolase, and the recruitment of AMPK to a ‘super-complex’ on the lysosomal membrane involving the vacuolar-ATPase, the Ragulator complex, Axin, LKB1 and AMPK. Although this mechanism may operate in tumour cells that are dependent on rapid glucose uptake, a full discussion of it is beyond the scope of this article and interested readers are referred to the original papers [10,59,60] or a recent review [2].

A third type of non-canonical activation of AMPK occurs in response to DNA damage. This was originally reported to occur in response to the topoisomerase II inhibitor etoposide [12], and was later observed following treatment of cells with ionizing radiation [13]. Both treatments cause double-strand breaks in DNA, and are often used in cancer treatment. Double-strand DNA breaks are known to be detected by ATM, a member of the phosphatidylinositol 3-kinase-like kinase (PIKK) family, and the effects of etoposide to activate AMPK were originally claimed to be dependent on ATM, because the effects appeared to be reduced in ATM-deficient cells [12]. In addition, ATM is known to phosphorylate LKB1 at Thr366 [61], and it was reported that AMPK activation by etoposide in cells was reduced by siRNA-mediated knockdown of either ATM or LKB1 [12,62], suggesting the existence of a kinase cascade from ATM to LKB1 to AMPK. However, this cannot be the primary mechanism, because both etoposide [12] and ionizing radiation [13] still activate AMPK in LKB1-null tumour cells. Moreover, our laboratory showed that AMPK activation by etoposide was not blocked by the ATM inhibitor KU-55993, despite the fact that the inhibitor did block phosphorylation of known ATM substrates [11]. We went on to show that AMPK activation by etoposide in LKB1-null cells was mediated by Thr172 phosphorylation catalysed by CaMKK2, and that this was associated with increases in Ca 2+ within the nucleus. Interestingly, only AMPK complexes within the nucleus containing the α1 isoform were activated, even though α2 was also expressed in the cells under study. Perhaps most interesting of all, activating AMPK in LKB1-null cells (using the Ca 2+ ionophore A23187 to activate CaMKK2) provided significant protection against cell death induced by etoposide. The most likely mechanism to explain this was that A23187 caused a G1 cell cycle arrest, thus restricting entry of cells into S phase where they are particularly susceptible to DNA damage. This hypothesis was supported by the fact that the G1 cyclin-dependent kinase inhibitor palbociclib caused a very similar degree of protection against cell death in those cells where it caused G1 arrest, but not in those where it did not [11]. These results are significant, because they suggest that genotoxic treatments such as etoposide and ionizing radiation might be more effective for cancer treatment if they were combined with inhibitors that prevent AMPK activation, and the consequent protection that AMPK can provide against genotoxic stress. This point is addressed further in §6 below.

3. Pharmacological activation and inhibition of AMPK

The realization that AMPK acts as a metabolic master switch, which transforms cellular metabolism from an anabolic to a catabolic state, originally suggested that activators of AMPK might be useful in treating disorders of energy balance such as obesity and type 2 diabetes [63]. Similarly, the discoveries that AMPK inhibited both cell growth and cell proliferation suggested that activators might also be useful in the treatment of cancer [64]. Over the past 20 years, scores of compounds that pharmacologically activate AMPK have been described, a few of which are shown in figure 2. These are discussed in §§3.1–3.3 according to their likely modes of action. There has been much less emphasis on the development of inhibitors, but these are briefly discussed in §3.4.

Figure 2. Structures of a number of AMPK activators. They have been classified according to their mechanisms of activation of AMPK (see §§3.1–3.3). (a) Pro-drugs that are converted inside cells to AMP analogues. (b) Compounds that bind in the allosteric drug and metabolite (ADaM) site. (c) Compounds that activate indirectly by inhibiting mitochondrial ATP synthesis.

3.1. Pro-drugs that are converted inside cells to AMP analogues

The first compound shown to activate AMPK in intact cells was the adenosine analogue 5-aminoimidazole-4-carboxamide riboside (AICAR), which is taken up into cells via adenosine transporters [65] and converted by adenosine kinase into the equivalent monophosphorylated nucleotide, ZMP [66–68] (figure 2a). ZMP mimics all three of the effects of AMP described in §2.4 [66], and AICAR has been much used as an experimental tool to activate AMPK in intact cells and in vivo. It should be noted, however, that ZMP is much less potent as an AMPK activator than AMP, and AICAR only activates AMPK in intact cells because intracellular ZMP accumulates to millimolar concentrations, even higher than the external concentrations of AICAR [66]. The use of AICAR is no longer recommended by the present authors, because ZMP has known off-target effects (e.g. it also mimics the effects of AMP to activate skeletal muscle glycogen phosphorylase [69] and inhibit hepatic fructose-1,6-bisphosphatase [67,70]), and because much more specific activators are now available. One such is C13, a derivative of another adenosine analogue termed C2 that has been esterified on two oxygen atoms of its phosphonate group to make it more cell permeable [71]. C13 is indeed readily taken up by cells, but is then converted into C2 by cellular esterases (figure 2a). Remarkably, C2 is an even more potent activator of AMPK than AMP itself, although it should be noted that it is specific for AMPK complexes containing the α1 isoform, and is inactive on α2 complexes [72]. The high affinity of C2 may arise because it binds, unexpectedly, to the AMPK-γ subunits in a somewhat different orientation than AMP [73].

3.2. Compounds that bind in the allosteric drug and metabolite (ADaM) site

In the structures of AMPK heterotrimers containing either β1 [17,18] or β2 [19], the β-CBM interacts with the N-lobe of the kinase domain of the α subunit via the surface opposite to its glycogen-binding site (figure 1). The cleft between these domains forms the binding site for novel ligands acting on AMPK, which in most cases came out of high-throughput screens that searched libraries of synthetic chemicals for allosteric activators of AMPK. The first to be discovered was the thienopyridone A-769662 [74] but at least 10 have now been reported, including PF-739 [75] and MK-8722 [76] (figure 2b). All activate β1 complexes with higher potency than β2 complexes, and this makes some of them (including A-769662 [77]) highly selective for the former. As well as causing allosteric activation, binding of these compounds also inhibits Thr172 dephosphorylation in cell-free assays [78,79], although in intact cells the predominant effect appears to be allosteric, since large changes in phosphorylation of the AMPK target acetyl-CoA carboxylase in response to these agonists are usually only accompanied by modest changes in Thr172 phosphorylation [78].

One of the curious features of the ligands currently known to bind at this site is that almost all of them are synthetic chemicals rather than natural products. However, many in the field believe that these compounds may be mimicking the effect of some natural metabolite that binds to this site, which is why it has been termed the ‘ a llosteric d rug a nd m etabolite’ (ADaM) site [80]. The only natural product currently known to bind to this site is salicylate, a compound made by plants that acts as a hormone signalling infection by pathogens [81]. In the form of extracts of willow bark, salicylates have been used by humans as medicines since ancient times, and they are still in very wide use as the synthetic derivative acetyl salicylic acid (ASA or aspirin), which is rapidly broken down to salicylate once it enters the circulation. Although aspirin itself is a potent irreversible inhibitor of the cyclo-oxygenases involved in biosynthesis of prostanoids such as thromboxanes [82], salicylate activates AMPK by direct binding at the ADaM site, which occurs at concentrations reached in plasma of patients taking high doses of aspirin and other salicylate-based drugs [81]. Interestingly, regular use of aspirin, usually taken to reduce the formation of blood clots via inhibition of thromboxane synthesis, is associated with a reduced incidence of cancer [83]. Whether this can be explained entirely by inhibition of cyclo-oxygenases, or whether it involves some other target such as AMPK, currently remains unclear.

3.3. Compounds, including biguanides, that activate AMPK indirectly by inhibiting mitochondrial ATP synthesis

Metformin and phenformin (figure 2c) are synthetic biguanides derived from galegine (isoprenyl guanidine) [84], a natural product from the plant goat's rue or Galega officinalis, which was well known as a herbal remedy in seventeenth century England [85]. Both biguanides were introduced for treatment of type 2 diabetes in the 1950s, although phenformin was withdrawn in most countries in the 1970s because its use was associated with the rare but life-threatening side effect of lactic acidosis. The risk of lactic acidosis is much lower with metformin, which has subsequently become the drug of first choice in the treatment of type 2 diabetes worldwide. Although biguanides have been used since the 1950s, the first clues to their mechanism of action did not emerge until 2000, when they were reported to inhibit complex I of the mitochondrial respiratory chain, thus explaining the risk of lactic acid accumulation [86,87] they have subsequently also been shown to inhibit the mitochondrial ATP synthase [88]. Clearly, inhibition of mitochondrial ATP synthesis would be expected to increase cellular ADP : ATP and AMP : ATP ratios and thus activate AMPK by the canonical mechanism. Indeed, activation of AMPK by biguanides in intact cells and in vivo was reported in 2001 [89], and it was subsequently confirmed that this was caused by increases in AMP and/or ADP [51], although metformin may also activate AMPK via the non-canonical lysosomal pathway [90]. Metformin has two major clinical effects: (i) inhibiting glucose production by the liver and (ii) enhancing insulin sensitivity of tissues such as liver and skeletal muscle. Surprisingly, studies with liver-specific double AMPK (α1 −/− α2 −/− ) knockout mice showed that the rapid effects of metformin on liver glucose production were AMPK independent, despite the fact that they were accompanied by increases in cellular AMP : ATP ratios [91]. These acute effects of metformin now appear to be due to direct allosteric inhibition of the gluconeogenic enzyme fructose-1,6-bisphosphatase by AMP [70]. Despite this, studies of mice with double knock-in mutations of the single serine residues that are targeted by AMPK in ACC1 (S79A) and ACC2 (S212A) suggested that the longer term insulin-sensitizing effects of metformin are indeed mediated by AMPK [92]. These mice, in which AMPK no longer acutely inhibits fatty acid synthesis or activates fatty acid oxidation, accumulate excess di- and tri-glycerides in liver and muscle, which is accompanied by insulin resistance. Although wild-type mice developed a similar degree of insulin resistance when placed on a high-fat diet, insulin sensitivity in the knock-in mice did not deteriorate further, possibly because they were already synthesizing so much fat. However, when the high-fat-fed mice were treated with metformin for six weeks, this reversed the insulin resistance of the wild-type mice but had no effect in the knock-in mice. Thus, the longer term effects of metformin on insulin sensitivity, although not its short-term effects on hepatic glucose production, are due to modulation of lipid metabolism by AMPK, most likely by reducing the excessive storage of lipids in tissues such as liver and skeletal muscle [92].

Following the initial findings that AMPK was activated by biguanides [89], and that the tumour suppressor LKB1 acted upstream of AMPK [14], the question of whether biguanide use had any influence on cancer was addressed. Retrospective studies suggested that the use of metformin in patients with type 2 diabetes in the Tayside region of Scotland was associated with a significant (around 30%) reduction in the incidence of cancer [93]. This association has since been confirmed in studies of many other diabetic cohorts [94–96], although its validity has been challenged due to the possibility of time-related biases [97] and it remains just a correlation, with no proof of direct causation. In addition, even if the association is valid, it does not necessarily imply that metformin acts directly on AMPK within the tumours themselves, rather than indirectly via AMPK-dependent or -independent effects on other tissues or organs. For example, since metformin is currently only used to treat type 2 diabetes, we do not know whether its use would be associated with reduced cancer incidence in subjects without diabetes (although there have been small trials in patients with breast or endometrial cancer, these were only ‘window-of-opportunity’ trials to assess various markers in the short period prior to surgery [98–101]). Note also that the different cancer incidence in patients with type 2 diabetes taking metformin is observed when comparing with those on other medications [93–96]. Metformin enhances insulin sensitivity and thus reduces insulin release, but many of the other commonly used medications, such as sulfonylureas and glucagon-like peptide-1 agonists, work in part by enhancing insulin secretion, while some subjects are even treated directly with insulin. Insulin is, of course, a growth factor that promotes proliferation of cells by activating the Akt pathway. One explanation of the apparent protective effect of metformin against cancer in patients with diabetes is therefore that, unlike most other treatments, it reduces rather than increases the levels of insulin, with high insulin levels being responsible for increased cancer incidence in patients on other medications [102]. Indeed, a related phenomenon is seen in patients with cancer who are treated with phosphatidyl-inositol 3-kinase (PI3 K) inhibitors, who often secrete extra insulin to compensate for the insulin resistance induced by the drugs, thus reducing their anti-cancer efficacy. Experiments with mouse models suggest that this effect can be overcome by additional dietary or pharmacological treatments that reverse the insulin resistance induced by these drugs [103].

There are many other compounds that activate AMPK by inhibiting mitochondrial ATP synthesis, one example being resveratrol [51], which inhibits the mitochondrial ATP synthase [104]. Another is sorafenib, originally developed as an inhibitor of receptor-linked tyrosine kinases such as the VEGF and platelet-derived growth factor (PDGF) receptors and used to treat some liver, kidney and thyroid cancers [105]. However, it also activates AMPK at therapeutically relevant concentrations by inhibiting the respiratory chain [106]. Remarkably, more than 100 natural products derived from traditional Asian medicines have within the last few years also been shown to activate AMPK in intact cells [107], and the effects of at least two of them, i.e. berberine [51] and arctigenin [108], appear to be due to inhibition of complex I of the mitochondrial respiratory chain. We suspect that many of the others may also work through inhibition of either complex I or the ATP synthase, which are both large, membrane-bound complexes containing no less than 44 and 14 protein subunits, respectively. It is perhaps not surprising that many hydrophobic compounds might find inhibitory binding sites within these complexes. This class of AMPK activator is particularly diverse in structure (e.g. those in figure 2c), indicating that they may interact with distinct sites. Many of the natural products that activate AMPK may be produced by plants to provide a chemical defence to deter grazing by insects or other animals, or infection by pathogens, and poisoning of complex I or the ATP synthase would seem to represent good ways to achieve those aims. Interestingly, many of these toxic plant products are stored within the plants that synthesize them either in the vacuole or in the cell wall [109], where they would not come into contact with the plant's own mitochondria.

3.4. AMPK inhibitors

At present, no specific AMPK inhibitors are available. The only AMPK inhibitor that has been widely used in the literature is compound C (also known as dorsomorphin). Although developed as an AMPK inhibitor, the claim that it was specific for AMPK came from the original report that it did not inhibit a panel of just five other protein kinases [89]. However, in a screen of 70 protein kinases, nine were inhibited to a greater extent than AMPK [110], while in a more recent screen of 120 kinases documented in the MRC Kinase Inhibitor Database ( no less than 30 were inhibited to a greater extent than AMPK. The use of compound C cannot therefore be recommended, even as an experimental tool. Other AMPK inhibitors have been reported [111,112], but have not yet been widely used.

4. Downstream targets of AMPK

Once activated, AMPK phosphorylates numerous downstream proteins, with at least 60 being identified as well-established targets in a recent review [113]. The core recognition motif for AMPK is well defined: it requires a basic residue (R, K or H) either three or four residues N-terminal to the phosphorylated serine/threonine (referred to as the P-3 and P-4 positions) as well as hydrophobic residues (L, M, I, F or V) at P-5 and P+4 [113–115]. The ACC1 isoform of acetyl-CoA carboxylase, which is a particularly good substrate for AMPK, has additional specificity determinants N-terminal to this core motif, which are not present in all downstream targets. These are another basic residue at P-6, and an amphipathic α-helix running from P-5 to P-16 that binds in a hydrophobic groove on the surface of the C-lobe of the AMPK kinase domain [116]. We discuss some of these targets below, focusing on those that may be particularly relevant to the role of AMPK in cancer.

4.1. Proteins and genes involved in catabolic pathways

Catabolic processes switched on by AMPK are summarized in figure 3. In many cell types, depending on the expression of specific glucose transporters (GLUTs), AMPK activation enhances glucose uptake. In skeletal muscle, AMPK acutely promotes translocation of vesicles containing GLUT4 from intracellular vesicles to the plasma membrane, in part by a mechanism involving phosphorylation of the Rab-GAP protein TBC1D1 [117]. In the longer term, AMPK also increases expression of GLUT4 protein via a mechanism that may involve direct phosphorylation of class IIa histone deacetylases (e.g. HDAC5) [118], which appears to cause their exclusion from the nucleus [119] and therefore promotes net acetylation and transcriptional activation at the GLUT4 promoter. AMPK activation also acutely activates glucose transport by the more widely expressed glucose transporter GLUT1 [120], in part via phosphorylation and consequent degradation of TXNIP, an α-arrestin family member that appears to promote internalization of GLUT1 as well as reduced levels of its mRNA [121]. In some but not all cells, AMPK acutely stimulates glycolytic flux via a mechanism involving direct phosphorylation of 6-phosphofructo-2-kinase/fructose-2,6- bisphosphatase, the enzyme that makes and breaks down fructose-2,6-bisphosphate via distinct domains of a bienzyme polypeptide [122]. Phosphorylation by AMPK increases the kinase activity, thus increasing the cellular concentration of fructose-2,6-bisphosphate, a potent allosteric activator of the glycolytic enzyme 6-phosphofructo-1-kinase [122]. However, this mechanism is limited to specific cell types, because only the PFKFB2 [123] and PFKFB3 [124] isoforms are targets for AMPK. PFKFB2 is expressed in cardiac myocytes and some other tissues, while alternative splicing or differential promoter usage yields two main isoforms of PFKFB3 that differ by a short C-terminal sequence these are the so-called ubiquitous (or constitutive) isoform and the inducible isoform [122]. The expression of the inducible form is very low in most adult tissues, but is increased by pro-inflammatory stimuli in monocytes and macrophages [124] and it is constitutively expressed in many tumour cells [125].

Figure 3. A ‘wheel’ of downstream targets and the pathways they regulate, focusing on catabolic processes that are activated by AMPK.

Although AMPK can therefore acutely activate ATP production by glycolysis in some cell types, in the longer term it tends to promote mitochondrial oxidative metabolism instead, which is much more efficient in terms of ATP production per glucose consumed (≈36 ATP per glucose by oxidative metabolism, as opposed to only two by glycolysis). Oxidative metabolism is, however, less compatible with providing precursors for cell growth, so it tends to be used to a greater extent in quiescent rather than proliferating cells [126]. In the short term, AMPK activates the uptake of fatty acids into mitochondria via phosphorylation of the acetyl-CoA carboxylase isoform ACC2 [127]. While ACC1, the first AMPK target to be identified, is thought to produce the cytoplasmic malonyl-CoA used in fatty acid synthesis, ACC2 localizes to mitochondria [128] and is thought to produce the mitochondrial malonyl-CoA that inhibits uptake of fatty acids into mitochondria via the carnitine:palmitoyl transferase system. Phosphorylation of ACC2 lowers malonyl-CoA and therefore relieves inhibition of carnitine:palmitoyl-CoA transferase-1 (CPT1), thus causing acute promotion of mitochondrial fatty acid oxidation [127].

In the longer term, AMPK activation has several effects on mitochondria that enhance their capacity to produce ATP at a rapid rate. Firstly, AMPK activation promotes mitochondrial biogenesis itself, involving increased replication of mitochondrial DNA as well as expression of many nuclear-encoded mitochondrial proteins, by activating the transcriptional co-activator PGC-1α [129]. This is effected either by direct phosphorylation of PGC-1α [130] or by increasing the cellular concentration of NAD + , a cofactor required for deacetylation and activation of PGC-1α by SIRT1 [131]. Secondly, being the major site of cellular production of reactive oxygen species, mitochondrial components are particularly prone to oxidative damage, and if this affects their function mitochondria need to be removed and their contents recycled by the targeted form of autophagy known as mitophagy. Relevant to this, AMPK has been shown to promote both autophagy and mitophagy either by phosphorylation of the protein kinase that triggers autophagy, ULK1 [132,133], or by phosphorylation of the Ca 2+ /calmodulin-dependent kinase DAPK, generating a Ca 2+ /calmodulin-independent form that phosphorylates the key autophagy protein Beclin-1 [134]. Thirdly, mitochondria are now known to exist, especially in quiescent cells, not as small separate organelles, but as branching networks of tubules that can be almost as long as the cell containing them [135]. If any regions of such a network become damaged, they need to be segregated off from healthy regions via the process of mitochondrial fission, so that they become small enough to be recycled by mitophagy. Intriguingly, AMPK activation has been shown to promote mitochondrial fission by direct phosphorylation of mitochondrial fission factor (MFF) [136]. These findings are consistent with one aspect of the phenotype of skeletal muscle-specific double AMPK knockouts (either α1 and α2 [137] or β1 and β2 [138]), in which muscle accumulates abnormally shaped and apparently malfunctioning mitochondria. Overall, AMPK appears to play several crucial roles in mitochondrial homeostasis. Since mitochondria are the main source of cellular ATP in most cells, this makes perfect sense for a signalling pathway that is activated by energy stress and/or glucose deprivation.

4.2. Proteins and genes involved in anabolic pathways

As well as switching on catabolic pathways that generate ATP, AMPK also switches off almost all major anabolic pathways (figure 4). AMPK was originally defined via its ability to phosphorylate and inactivate both acetyl-CoA carboxylase (ACC1) and 3-hydroxy-3-methylglutaryl-CoA reductase (HMGR), the key regulatory enzymes of fatty acid and sterol synthesis, respectively [139,140]. Fatty acid synthesis is a significant energy-consuming pathway in rapidly dividing tumour cells, being a major consumer of both ATP (consumed in the ACC1 reaction) and NADPH (consumed in the two reductive steps catalysed by the fatty acid synthase complex). Indeed, ACC1 remains one of most rapidly phosphorylated substrates for AMPK, and the phosphorylation of Ser80 (human numbering) on ACC1, monitored using a phosphospecific antibody, remains the most reliable and widely used cellular marker of AMPK function.

Figure 4. A ‘wheel’ of downstream targets and the pathways they regulate, focusing on anabolic and other processes that are inhibited by AMPK.

As well as acutely inhibiting fatty acid synthesis by direct phosphorylation of ACC1 (which catalyses the first two steps of fatty acid synthesis from acetyl-CoA), AMPK activation also downregulates expression of the genes encoding ACC1 (ACACA) as well as the gene (FASN) encoding the fatty acid synthase complex, a dimeric multienzyme polypeptide that catalyses the remaining seven reactions leading to a saturated C16 fatty acid (palmitate). The AMPK targets responsible for these effects may be the transcription factors sterol response element binding protein-1c (SREBP1c) [141] and/or the carbohydrate response element binding protein (ChREBP) [142], which have both been reported to be directly phosphorylated by AMPK.

As well as inhibiting de novo synthesis of fatty acids, AMPK inhibits synthesis of triglyceride and phospholipid synthesis by inactivating the first enzyme (glycerol phosphate acyl transferase, GPAT) involved in the synthesis of the common intermediate diacylglycerol [143], although whether this is due to direct phosphorylation of the enzyme remains unclear. Two direct targets for AMPK are the muscle (GYS1) [34] and liver (GYS2) [35] isoforms of glycogen synthase, which catalyse the transfer of glucose from UDP-glucose to the growing non-reducing ends of the glycogen particle. Both are inactivated by phosphorylation at equivalent N-terminal sites by AMPK, although this inactivation is over-ridden by high concentrations of the feed-forward allosteric activator of glycogen synthase, glucose-6-phosphate [144]. The need to co-localize AMPK with glycogen synthase may be one reason why AMPK-β subunit isoforms in all species carry a carbohydrate-binding module (β-CBM see figure 1).

Another key pathway in growing cells is biosynthesis of nucleotides. The ribose or deoxyribose components of nucleotides are derived from ribose-5-phosphate generated in the pentose phosphate pathway. It has recently been reported that PRPS1 and PRPS2, the two major isoforms of phosphoribosyl pyrophosphate synthetase that metabolize ribose-5-phosphate in the first step of nucleotide biosynthesis, are phosphorylated and inactivated by direct phosphorylation by AMPK [145]. The pentose phosphate pathway also generates NADPH that is used for fatty acid biosynthesis, as well as for regenerating reduced glutathione used to combat oxidative stress. The large requirement for NADPH and nucleotide biosynthesis in rapidly proliferating cells may be one reason why they exhibit rapid glucose uptake to provide input of glucose into the pentose phosphate pathway.

Nucleotides are, of course, the building blocks for RNA and DNA. In rapidly proliferating thymocytes, the addition of actinomycin D (a general inhibitor of RNA synthesis) reduces oxygen uptake by as much as 15% [146], suggesting that RNA synthesis accounts for at least that percentage of total ATP turnover. Since up to 80% of the total RNA in a typical cell is ribosomal RNA (rRNA), synthesis of the latter is a major anabolic pathway and consumer of energy in proliferating cells. Consistent with this, AMPK activation has been found to inhibit rRNA synthesis by direct phosphorylation of the transcription factor for RNA polymerase I, TIF-1A (encoded by the RRN3 gene) [147].

Arguably the most important biosynthetic pathway in proliferating cells is translation (protein synthesis). Over 50% of the dry weight of most cells is protein while, in the proliferating thymocyte system mentioned above, inhibition of protein synthesis reduced oxygen uptake by more than 20% [146]. AMPK switches off translation by at least two mechanisms. Firstly, it inactivates the target of rapamycin complex-1 (TORC1), which is known to promote the initiation step of ribosomal protein synthesis by triggering the phosphorylation of multiple proteins, including eukaryotic initiation factor-4E binding protein-1 (EIF4EBP1) and ribosomal protein S6 kinase-1 (RPS6KB1) [148]. Phosphorylation of EIF4EBP1 leads to selective translation of mRNAs containing 5′-terminal oligopyrimidine (5′-TOP) sequences, which often encode mRNAs encoding proteins required for rapid cell growth, including most ribosomal proteins as well as other proteins involved in translation [149]. AMPK inactivates mTORC1 by at least two mechanisms, i.e. inhibitory phosphorylation of the Raptor subunit that targets the complex to downstream targets and to the lysosome where it is activated, and activatory phosphorylation of TSC2, which forms a key part of the TSC1:TSC2:TBC1D7 complex. The latter has a Rheb:GTPase activator protein (Rheb:GAP) domain on TSC2 that converts the mTORC1-activating G protein Rheb to its inactive GDP-bound form [150]. Secondly, AMPK inhibits the elongation step of ribosomal protein synthesis by promoting phosphorylation of elongation factor-2 at Thr56. This residue is phosphorylated not by AMPK itself but by elongation factor-2 kinase (EF2 K), a member of the atypical protein kinase (aPK) family that is activated by Ca 2+ /calmodulin. AMPK appears to activate EF2 K in part by direct phosphorylation [151] and in part by inactivating mTORC1, with EF2 K being phosphorylated and inactivated by p70S6K1 downstream of mTORC1 [152,153].

4.3. Progress through the cell cycle

As well as inhibiting most major biosynthetic pathways, AMPK activation can also cause cell cycle arrest (figure 4). As long ago as 2001, it was reported that the AMPK activator 5-aminoimidazole-4-carboxamide riboside (AICAR) caused arrest in the G1 phase of the cell cycle in HepG2 cells, which was attributed to phosphorylation of the transcription factor p53 at Ser15, and consequent increased expression of the G1 cyclin-dependent kinase inhibitor p21 CIP1 (CDKN1A) [154]. This was followed by a demonstration that both AICAR and low glucose caused cell cycle arrest in MEFs [155]. These effects were at least partially dependent upon p53, because they were reduced in p53 −/− MEFs. Although both AICAR and glucose deprivation can have off-target, AMPK-independent effects, the effects of low glucose also appeared to be AMPK dependent because they were reduced by expression of a dominant negative AMPK mutant (a kinase-inactive AMPK-α mutant that inhibits endogenous AMPK-α subunits by competing for available β and γ subunits). This group also reported that an activated kinase domain construct of AMPK could directly phosphorylate p53 at Ser15 [155]. However, the sequence around Ser15 is not a good fit to the AMPK recognition motif, lacking a basic residue at P-4 or P-3 (in fact, with an acidic residue at P-4 instead). It seems more likely that the phosphorylation of p53 observed in response to AMPK activation is indirect.

Another potential mechanism by which AMPK causes G1 arrest involves phosphorylation of another cyclin-dependent kinase inhibitor, p27 WAF1 (CDKN1B). In breast cancer cells (MCF-7), p27 was found to be phosphorylated at its C-terminal residue (Thr198), and this appeared to stabilize the protein, thus increasing its expression and causing cell cycle arrest as well as appearing to promote autophagy. Phosphorylation of Thr198 increased in response to AICAR treatment or glucose starvation of cells. Although some evidence was presented that Thr198 is directly phosphorylated by AMPK, the site is not a perfect fit to the AMPK recognition motif (for example, being the C-terminal residue, there is no hydrophobic residue at P+4), and mutation of Thr198 only had a modest effect on phosphorylation by AMPK in cell-free assays [156]. Further studies are therefore required to confirm that this is a direct effect of AMPK.

Although AMPK activation by both AICAR [14] and low glucose [60] requires the presence of LKB1, AMPK could still cause G1 arrest in three different LKB1-deficient tumour cell lines if it was activated by the addition of a Ca 2+ ionophore to activate the alternative upstream kinase, CaMKK2. This effect was abolished either by expression of a dominant negative AMPK-α2 mutant or by a double knockout of AMPK-α1 and -α2 [157]. Thus, AMPK can cause G1 arrest even in the absence of its tumour suppressor upstream kinase, LKB1. Interestingly, treatment of cells with the Ca 2+ ionophore A23187 caused G1 arrest without affecting the expression of CDKN1A or CDKN1B, despite the fact that the overall expression of both was reduced by expression of the dominant negative mutant or the double knockout [157]. Thus, in this case changes in CDKN1A or CDKN1B expression cannot be the sole explanation for cell cycle arrest.

5. AMPK and cancer—evidence from mouse models

We will now discuss the evidence that, depending upon the context, AMPK can act either as a tumour suppressor or as a tumour promoter in mouse models.

5.1. AMPK—a tumour suppressor?

With the discovery that AMPK activation requires the tumour suppressor LKB1, the realization that AMPK inhibits cell growth and proliferation, and the epidemiological evidence that the AMPK activator, metformin, provides protection against cancer, it seemed increasingly likely that AMPK would also be a tumour suppressor. One caveat was that, soon after the discovery that LKB1 acted upstream of AMPK, LKB1 was found to be required for the phosphorylation and activation of at least 12 other kinases closely related to AMPK (now referred to as the AMPK-related kinase or ARK family), which share very similar sequences within their activation loops [158,159]. Although none of the ARKs (unlike AMPK) are known to inhibit cell growth and proliferation, knockdown of LKB1 using RNAi was reported to enhance expression of SNAIL, a protein that promotes the epithelial-to-mesenchymal transition, and hence metastasis of tumour cells, by reducing the phosphorylation of DIXDC1 by two of the ARKs, MARK1 and MARK4 [160]. It therefore remains possible that at least some of the tumour suppressor effects of LKB1 might be mediated by one or more of the ARKs, rather than AMPK.

Confirmation of a tumour suppressor role for AMPK in vivo required the study of tumorigenesis in AMPK knockout mice. However, there are two isoforms of the catalytic subunit (α1 and α2) and a global double knockout is embryonic lethal [161], thus necessitating the use of tissue-specific double knockouts. While this approach is now possible, it is also very time-consuming. However, a shortcut arose with the realization that cells of the haematopoietic lineage, including lymphocytes, exclusively express AMPK-α1 [162]. Thus, to study the role of AMPK in lymphomas and/or leukaemias, it was only necessary to knock out a single gene, i.e. the Prkaa1 gene encoding AMPK-α1.

The first study to suggest that AMPK was a tumour suppressor used a Eµ-Myc model [163], in which B-cell lymphoma is induced by transgenic over-expression of the Myc oncogene from a B-cell-specific promoter. Consistent with the idea that AMPK-α1 is a tumour suppressor, loss of both alleles of Prkaa1 in this model markedly accelerated development of B-cell lymphomas, whereas loss of a single allele had an intermediate effect. Eµ-Myc lymphoma cells and other tumour cells expressing shRNAs targeted at AMPK-α1 were also studied in vitro. In general, the AMPK knockdown cells exhibited mTORC1 hyper-activation and increased glucose uptake and lactate production compared with controls, and this appeared to be due to increased expression of hypoxia-inducible transcription factor-1α (HIF-) [163]. The 5′-UTR of mRNA encoding HIF-1α contains 5′-TOP sequences [164] and their translation is thus enhanced by mTORC1 activation ([165] see §4.2). Thus, loss of AMPK in the tumour progenitor cells enhances glucose uptake and glycolysis even under normoxic conditions. This is an example of the well-known ‘Warburg effect’, in which tumour cells display high levels of glucose consumption in order to generate precursors for biosynthesis derived from the pentose phosphate pathway and glycolysis. For example, ribose-5-phosphate for nucleotide biosynthesis and NADPH for lipid synthesis are generated via the pentose phosphate pathway, while serine (required for one-carbon metabolism used in purine nucleotide biosynthesis) is generated by a pathway that branches off from the glycolytic intermediate 3-phosphoglycerate [126].

A drawback with this B-cell lymphoma model was that Prkaa1 was knocked out globally [163], so it was not possible to conclude that the effect was due to a cell-autonomous loss of AMPK-α1 in the B-cell progenitors themselves, rather than an indirect effect of loss of AMPK-α1 in some other cell type. In an attempt to address this, wild-type mice were irradiated to inactivate their endogenous immune system, and were then reconstituted with haematopoietic stem cells from either Eµ-Myc/Prkaa1 −/− or Eµ-Myc/Prkaa1 +/+ mice. Interestingly, all of the mice receiving AMPK knockout cells developed lymphomas, but only 20% of those receiving the AMPK wild-type cells [163], thus supporting the idea that the effect of AMPK loss was at least partly cell autonomous.

Another study involved crossing mice with global knockouts of the genes encoding p53 (Trp53) and AMPK-β1 (Prkab1), the latter being the principal β subunit isoform expressed in T-cell precursors in the thymus [166]. Knockout of Prkab1 caused earlier onset of T-cell lymphomas in both homozygous and heterozygous p53 knockouts, suggesting that β1 had a tumour suppressor role in T-cell lymphoma. However, once again the knockout of Prkab1 in this model was global rather than T-cell specific, so it was not possible to conclude whether this was a cell-intrinsic effect on AMPK in the tumour progenitor cells themselves.

A specific loss of AMPK in the tumour progenitor cells has recently been achieved using a model of T-cell acute lymphoblastic leukaemia/lymphoma (T-ALL) [167]. As reported previously [168], mice with a T-cell-specific knockout of PTEN (using Cre recombinase expressed from the Lck promoter) started to develop lymphomas at about 50 days of age, and essentially all of the mice had developed T-ALL by 150 days. While knocking out the Prkaa1 gene using the same Lck-Cre system had no effect on its own, when combined with PTEN knockout the lymphomas arose earlier and overall tumour-free survival was greatly reduced (figure 5) [167]. These results suggested that basal AMPK activity in developing T cells is sufficient to provide protection against T-ALL. However, this model also provided an excellent opportunity to test whether treatment with biguanides would protect against this type of cancer (see §3.3). Since the expression of AMPK-α1 would be absent in lymphoma cells and their progenitors but normal everywhere else, it would also be possible to determine whether any effect of biguanides was a cell-intrinsic effect to activate AMPK in the tumour progenitor cells themselves. Rather disappointingly, metformin had no effect, which correlated with a lack of AMPK activation and a failure to detect metformin by liquid chromatography–mass spectrometry (LC:MS) in the thymus of mice with lymphomas. By contrast, phenformin significantly enhanced tumour-free survival, and this correlated with AMPK activation, and detection of phenformin by LC:MS, in the thymus of mice with lymphoma. Intriguingly, protection against T-ALL by phenformin was only observed when the tumours expressed AMPK, with no effect in the AMPK knockouts (figure 5). Thus, protection against T-ALL by phenformin was dependent upon the expression of AMPK in the tumour progenitor cells, and was cell autonomous, while the failure of metformin to provide protection was due to lack of uptake of the drug by thymocytes. Phenformin has also been shown recently to slow growth of murine breast cancer cells in vivo in a mouse allograft model, although the role of AMPK was not examined [169].

Figure 5. Effect of T-cell knockout of AMPK (AMPK KO) and oral phenformin on tumour-free survival in mice bearing T-cell knockout of PTEN (PTEN KO). Where indicated, phenformin was administered in drinking water starting from 30 days of age. Original data from [167].

Another mouse model suggesting a tumour suppressor role for AMPK used prostate epithelial-specific knockouts of the Pten and Prkab1 genes [170]. Although the knockout of Prkab1 as well as Pten did not affect prostate size, it did result in a higher proliferative index and pathological grade. A drawback with this model was that the prostate gland also expresses AMPK-β2, which might have partially compensated for lack of β1 and might be why the effects on tumorigenesis were relatively modest.

Other evidence supporting the idea that AMPK is a tumour suppressor comes from studies of ubiquitin ligases involved in cellular degradation of AMPK subunits. MAGE-A3/-A6 are closely related members of the melanoma antigen family of proteins, which are normally only expressed in testis but become re-expressed in many tumours, hence their designation as tumour antigens [171]. MAGE-A3/-A6 bind to the ubiquitin E3 ligase TRIM28, and a screen revealed AMPK-α1 to be a target for polyubiquitylation by this complex, with consequent proteasomal degradation. Consistent with this, knockdown of MAGE-A3/A6 or TRIM28 in tumour cells increased the expression of AMPK-α1 and triggered the expected changes in metabolism and signalling, including inhibition of mTORC1. Finally, various human tumour cells that express MAGE-A3/-A6 have reduced levels of AMPK-α1 protein [171].

Another cancer-associated ubiquitin ligase, UBE2O, targets degradation of α2, the other catalytic subunit isoform of AMPK [172]. Knockout of Ube2o attenuated tumour development in mouse models of both breast and prostate cancer, supporting the idea that the protein has tumour-promoting functions. A search identified AMPK-α2 as an UBE2O-interacting protein that is targeted for polyubiquitylation and proteasomal degradation, and the levels of α2 but not α1 were upregulated in tissues from Ube2o −/− mice. A human colon carcinoma cell line also grew less rapidly in mouse xenografts when UBE2O was knocked down using shRNA, and this was reversed by concurrent knockdown of AMPK-α2 but not -α1. The UBE2O gene is located in humans at 17q25, a region amplified in up to 20% of breast, bladder, liver and lung carcinomas. Using immunohistochemistry of human breast tumours, there was a negative correlation between expression of UBE2O and AMPK-α2, but a positive correlation between UBE2O expression and S6 phosphorylation, a marker for the mTORC1 pathway [172].

5.2. AMPK—a tumour promoter?

Despite the evidence discussed in the previous section that AMPK-α1 and -β1 are tumour suppressors that protect against the development of B- and T-cell lymphomas as well as prostate cancer, other studies suggest that AMPK may protect the tumour cells (rather than the patient), and thus promote tumour formation, at least when disease is already established. Rathmell's group used a different model of T-ALL in which oncogenic NOTCH1 was expressed in vitro in murine haematopoietic stem cells that carried a floxed AMPK-α1 gene and Cre recombinase driven by a tamoxifen-inducible promoter. These were multiplied in irradiated mice, and then injected into irradiated secondary recipient mice. After a period of 10 days to allow disease to become established, the mice were then treated with tamoxifen to acutely delete AMPK-α1 in the T-ALL cells. In this model, knocking out AMPK-α1 reduced the recovery of T-ALL cells in spleen, lymph nodes and bone marrow, and enhanced survival of the mice [173]. Thus, once T-ALL tumour cells have developed the presence of AMPK-α1 appears to enhance T-ALL cell viability and reduce mouse survival. While AMPK therefore acts as a tumour suppressor during the development of T-ALL [167], once the tumours have occurred it appears to paradoxically switch to being a tumour promoter instead.

Another study using a mouse model of acute myeloid leukaemia (AML) also concluded that AMPK acted as a tumour promoter [174]. Here, mice carrying floxed alleles of Prkaa1 and Prkaa2, as well as Cre recombinase expressed from the Mx1 promoter, were injected with poly(I:C) to delete AMPK-α1 and -α2 from haematopoietic cells, with mice lacking Mx1-Cre as controls. Haematopoietic progenitor cells from these mice were then transduced with retroviruses expressing three different cancer-promoting gene fusions (MLL-AFP, MOZ-TIF2 or BCR-ABL) and were then transplanted into irradiated recipient mice. The absence of AMPK from the leukaemia-initiating cells either delayed the onset of disease (BCR-ABL) or enhanced mouse survival (MLL-AFP or MOZ-TIF1). Thus, the presence of AMPK was required to maintain full leukaemogenic potential of the cells in these models. Evidence was provided that this was because the lack of AMPK increased the recovery of reactive oxygen species (ROS) in leukaemia-initiating cells from bone marrow, correlating with decreased ratios of reduced : oxidized NADP and glutathione, and increased DNA damage. This was ascribed to a reduced glucose uptake via GLUT1, which is regulated by AMPK via phosphorylation of TXNIP (see §4.1). The authors also proposed that the leukaemia-initiating cells lacking AMPK were particularly vulnerable to stress in the bone marrow, because the glucose concentrations were lower than in peripheral blood, especially under conditions of dietary restriction of the mice [174].

Consistent with these findings, reduced survival of AMPK-deficient human tumour cells undergoing stress has been observed in several in vitro studies. For example, LKB1-null tumour cells, or LKB1-expressing tumour cells with AMPK-α1 knocked down using shRNA, were more susceptible to cell death induced by glucose starvation or extracellular matrix detachment, suggesting that AMPK activation protected against these insults [175]. In another example, a synthetic lethal siRNA screen was carried out to detect protein kinases required for survival of U2OS cells that over-expressed the Myc oncogene from a tamoxifen-inducible promoter. One of the top hits was AMPK-α1, which was also shown to be activated during Myc over-expression [176].

Evidence that AMPK can promote tumours was also obtained recently using a mouse model of lung cancer in which the tumours develop in situ at their site of origin, and in which the authors had ‘bitten the bullet’ by knocking out both AMPK-α1 and -α2. Here, mice expressing Lox-STOP-Lox alleles of the KRAS G12D oncogene and firefly luciferase were crossed with mice expressing floxed alleles of Tp53 (encoding p53) and/or Stk11 and/or Prkaa1 plus Prkaa2. To model non-small cell lung carcinoma, Cre-recombinase was delivered to the lungs by nasal inhalation of lentiviral vectors. This procedure triggers recombination at twin loxP sites in a small subset of lung epithelial cells, in which expression of KRAS G12D and luciferase would be switched on, and p53 and/or LKB1 and/or AMPK-α1/-α2 would be knocked out expression of luciferase also allowed tumours to be imaged by bioluminescence, and thus their growth to be monitored in vivo. Knockout of LKB1 enhanced growth in tumours expressing mutant K-Ras as reported previously [177] but, by contrast, knockout of both AMPK-α1 and -α2 was found to cause reductions in the size and number of lung tumours, especially in tumours expressing mutant K-Ras and lacking p53. Overall, these results confirmed that LKB1 is a tumour suppressor in non-small cell lung cancer as expected, while the presence of either AMPK-α1 or -α2 promoted tumour growth [178].

6. Evidence from analysis of human cancer genomes

Although most of the evidence discussed in §5 was obtained in mouse models of cancer, comparison of genetic alterations in genes encoding the LKB1-AMPK pathway in biopsies of human cancers, compared with normal tissue, can also provide useful clues about roles of the pathway in human cancer. The cBioPortal database ( [179,180]) provides a particularly user-friendly way to analyse the many studies of human cancer of this type that have been performed to date. Figure 6 summarizes genetic changes in the STK11 gene encoding LKB1, and all seven genes encoding subunit isoforms of AMPK, extracted from cBioPortal in April 2019. Each vertical bar represents an individual cancer genome project, with the height of the bar representing the percentage of cases where genetic alterations were seen (only studies with changes in greater than or equal to 3% of cases are shown). Since LKB1 is a known tumour suppressor, one would expect to observe mainly mutations (green bars) or deletions (blue bars) when analysing STK11. This is indeed generally the case (figure 6a), although there are some anomalous cancer studies where gene amplification was observed instead (red bars), particularly in pancreatic and prostate cancers. By contrast, changes in the PRKAA1 gene, encoding AMPK-α1, were mostly amplifications (note preponderance of red bars in figure 6b), which is more consistent with the idea that AMPK-α1 can act as a tumour promoter. An important caveat is that gene amplifications in cancer usually involve whole segments of chromosomes rather than individual genes. It was therefore possible that the PRKAA1 gene is located close to an oncogene for which amplification was being selected, with PRKAA1 simply accompanying it as an innocent bystander. However, an argument against that possibility comes from analysis of concurrent genetic changes in STK11 and PRKAA1 in single cancer studies, such as the 230 cases of lung adenocarcinoma in The Cancer Genome Atlas (figure 7) [181]. In that study, the STK11 gene was either deleted or mutated (mostly truncations or missense mutations predicted to cause loss of function) in 43 cases (19%) and PRKAA1 was amplified in 22 (10%). However, these changes never coincided (p = 0.005), which would be expected to occur by random chance if they were occurring independently. The most frequent mutations in this study of lung adenocarcinoma were in the KRAS (36%) and TP53 genes (47%), encoding K-Ras and p53. Interestingly, amplification of PRKAA1 was almost mutually exclusive with mutations in KRAS (p = 0.005), but co-occurred with mutations in TP53 (p < 0.001).

Figure 6. Summary of genetic alterations in human cancer in genes encoding (a) LKB1 (STK11), and (b–h) the seven genes encoding AMPK subunit isoforms. Based on analysis of the ‘curated set of non-redundant studies’ in the cBioPortal database in early April 2019, using the gene names shown.

Figure 7. Co-occurrence or mutual exclusion of genetic alterations in the PRKAA1 (AMPK-α1), STK11 (LKB1), KRAS (K-Ras) and TP53 (p53) genes in human lung adenocarcinoma. Results were generated using cBioPortal from the results of a single study [181]. Each column of vertically aligned bars represents a single case cases with no alterations in any of the genes are not shown.

Why should amplification of the AMPK-α1 gene be mutually exclusive with mutations in the LKB1 gene? The answer to this seems obvious, because there would be little point in over-expressing AMPK-α1 if LKB1 was not present to phosphorylate and activate it. These considerations suggest that PRKAA1 amplification is being selected for, rather than just being an innocent bystander. However, why amplification of the AMPK-α1 gene should co-occur with mutations in p53 is less obvious. The classical role of p53 [182] is to become stabilized or activated in response to DNA damage, and to cause a G1 cell cycle arrest in order to allow time for the damage to be repaired, which it achieves by inducing transcription of genes such as the G1 cyclin-dependent kinase inhibitor p21 CIP1 (CDKN1A). Intriguingly, as already discussed in §2.5, AMPK complexes containing α1 are also activated by genotoxic agents such as etoposide, and can trigger a similar G1 cell cycle arrest [11]. It therefore seems possible that PRKAA1 amplification may be selected for in p53-null tumours because over-expression of AMPK-α1 can compensate to some extent for p53 loss, and could thus enhance survival of p53-null tumour cells undergoing genotoxic stress.

In marked contrast to the frequent amplification of the PRKAA1 gene in cancer, the PRKAA2 gene encoding AMPK-α2 is much more often mutated (note preponderance of green bars in figure 6c). Interestingly, all six of the cancer studies where the gene was most frequently mutated (10–23% of cases) were of skin cancer or melanoma. The reasons for this are not clear, but separate analysis showed that in all of the skin cancer/melanoma studies listed in cBioPortal there were 80 mutations affecting AMPK-α2 and just 10 affecting α1. Although it is not yet clear how many of the former cause loss of function in α2 complexes, these results suggest that AMPK-α2 may play a tumour suppressor role in skin cancer and melanoma.

When it comes to the AMPK-β subunits, there was a striking difference between the behaviour in human cancers of the PRKAB1 and PRKAB2 genes, encoding β1 and β2 (figure 6d,e). While genetic changes in PRKAB1 were detected in just a very small number of cancer studies and were quite variable in genetic type, the PRKAB2 gene was frequently amplified in numerous different cancers (note preponderance of red bars in figure 6e), suggesting, if anything, a tumour promoter role. Since the C-terminal domain of the β subunit (β-CTD) forms the ‘core’ of the heterotrimeric AMPK complex (see §2.2), over-expression of β2 may perhaps help to stabilize and increase expression of the α and γ subunits, even when the genes encoding those subunits lack genetic alterations. However, why it should only be the gene encoding β2, and not β1, that is amplified remains unclear.

Alterations in the genes encoding the three γ subunits tend to occur at a lower frequency than those encoding the α and β subunits, and are more mixed in genetic type (figure 6f–h). However, there were some interesting findings, such as the 41% of cases (albeit only five out of 12) in which the PRKAG1 gene was deleted in adenoid cystic breast cancer [183].

Looking at the genetic alterations occurring in the genes encoding LKB1 and AMPK subunits in human cancer, one striking observation is that all eight genes are frequently amplified in neuroendocrine prostate cancer (labelled NE prostate in figure 6). This is a subset of prostate cancer that has become resistant to anti-androgen treatment [184]. The significance of this intriguing observation remains unclear at present.

7. Conclusion—is AMPK a tumour suppressor or a tumour promoter, or both?

In this final section we will attempt to reconcile the apparently conflicting reports that AMPK can variously act to promote or suppress tumorigenesis. Our view is that AMPK can act either as a tumour suppressor or a tumour promoter, depending on the context. It can be argued that in all of the mouse studies where a tumour suppressor role was supported (e.g. in the Eµ-Myc model of B-cell lymphoma [163], the p53-null [166] and PTEN-null [167] models of T-cell lymphoma and the PTEN-null model of prostate cancer [170]), AMPK function had been knocked out prior to tumorigenesis. For example, in the Eµ-Myc model [185], loss of AMPK-α1 would have occurred during embryogenesis whereas, although over-expression of Myc in pre-B cells has certainly occurred by 35–50 days of age [186], lymphomas do not start to arise until 50 days and their median onset is ≈80 days [187]. Thus, events additional to Myc over-expression must occur before lymphomas are generated. The same applies to the PTEN knockout model of T-ALL, where the Lck promoter-driven knockout of PTEN and/or AMPK-α1 would have occurred by 30 days of age but lymphomas did not start to arise until later (figure 5).

On the other hand, in those mouse models of cancer where AMPK appeared to be acting as a tumour promoter, it can be argued that the knockout of AMPK usually occurred either simultaneous with, or even after, tumorigenesis had been initiated. For example, in the study of T-ALL by Kishton et al. [173] (§5.2), transformation was generated in vitro by forced expression of an oncogenic mutant of Notch1, and the T-ALL cells were then transferred to irradiated recipient mice and disease allowed to become established prior to AMPK being knocked out by treatment with tamoxifen. It is particularly instructive to compare this model with our own more recently published model of T-ALL [167], where AMPK-α1 had been specifically knocked out in T-cell progenitors prior to lymphomas starting to occur, in which basal AMPK was clearly protecting against development of lymphomas, and in which activation of AMPK using phenformin provided further protection.

Coming to other mouse studies that support a tumour promoter role for AMPK, in the autochthonous model of non-small cell lung cancer [178], knockout of AMPK would have occurred simultaneously with expression of mutant K-Ras and loss of p53, which may have been sufficient to trigger tumorigenesis on their own. The only study that supported a tumour promoter role but where AMPK had been knocked out prior to disease onset was the model of AML by Saito et al. [174]. However in that case (as in the study of T-ALL by Kishton et al. [173]) transformation had been achieved by enforced expression of oncogenes in vitro in haematopoietic progenitor cells, and the real test of the role of AMPK was in the survival and/or proliferation of the leukaemia cells in vivo in irradiated recipient mice. It can be argued that these two studies, by carrying out transformation in vitro, may have been less likely to detect a tumour suppressor role of AMPK.

Overall we propose that, when loss of AMPK occurs prior to initiation of tumorigenesis in vivo, this would remove the restraints on the mTORC1 pathway and unleash other biosynthesis processes and the cell cycle, thus transforming the cells into a metabolic and proliferative state that is primed for tumour formation. Under these circumstances, AMPK acts as a tumour suppressor, and AMPK activators may provide additional protection against tumorigenesis, such as the effect of phenformin in T-ALL [167]. These results suggest that AMPK activators might one day find a place in providing protection against cancer, perhaps in individuals who are at high risk of developing the disease. If biguanides are used, it might also make sense to use phenformin which, being more membrane permeable than metformin even in the absence of a transporter, is much more likely to activate AMPK in the tumour progenitor cells. Although phenformin was withdrawn for treatment of type 2 diabetes because of the risk of life-threatening lactic acidosis, the risk of this complication was actually quite low (≈64 cases per 100 000 patient-years [188]), and might be more acceptable in the context of cancer rather than diabetes. Alternatively, some of the other AMPK activators discussed in §3 might perhaps be developed for this purpose.

We also propose that, once the cancer cells have started to grow in vivo, AMPK switches from being a tumour suppressor to a tumour promoter (like the transformation of the benevolent Dr Jekyll into the malevolent Dr Hyde in Stevenson's novel!). Under these circumstances, the role of AMPK is to protect the cell in which it is expressed, irrespective of whether that cell is a cancer cell or a normal cell. By protecting cancer cells against stresses such as shortage of oxygen or nutrients, or oxidative or genotoxic stress, AMPK would enhance their survival and thus, in the long term, promote growth of tumours. Under these circumstances, AMPK is acting as a tumour promoter, which suggests that AMPK inhibitors might be efficacious in treatment of cancer. They may be particularly effective: (i) in cases where the PRKAA1 or PRKAB2 genes are amplified, causing AMPK over-expression (ii) when given in combination with genotoxic treatments such as etoposide or radiotherapy, thus reducing the viability of tumour cells during such therapies. At present we do not have any well-characterized and specific inhibitors of AMPK (see §3.4), but future work can be directed at correcting that deficiency.


Recent studies in our laboratory involving live animals were approved by the Ethics Review Committee of the University of Dundee in accordance with the UK Animal (Scientific Procedures) Act 1986.


Adrenoceptors belong to the GPCR family, a conserved family of seven transmembrane receptors that is one of the largest protein classes to be targeted for drug therapy (Sriram & Insel, 2018 ). These receptors are classified as α- or β-adrenoceptors, based on differences in responses to various catecholamines such as adrenaline, noradrenaline and isoprenaline. The α-adrenoceptors have been classified into two major families: α1 and α2 and the β-adrenoceptors are subdivided into β1-, β2-, and β3-subtypes . All adrenoceptor subtypes have common primary structures comprising one extracellular N-terminal domain, seven α-helical transmembrane spanning regions, and one intracellular C-terminal tail. Recent studies have shown that α1- and β1-adrenoceptors in the heart, β2-adrenoceptors in skeletal muscle, and β3- adrenoceptors in brown adipose tissue (BAT) can link to a protein called mechanistic target of rapamycin (mTOR), which plays a significant role in physiological and metabolic responses.

mTOR is an atypical serine/threonine kinase with a molecular weight of

289 kDa, belonging to the PI3K-related kinase family. mTOR interacts with other molecular components to form two physically and functionally distinct complexes, namely, mTOR complex 1 (mTORC1) and mTOR complex 2 (mTORC2). In mTORC1, the mTOR protein interacts with regulatory-associated protein of mTOR (Raptor Kim et al., 2002 ), proline-rich PKB (Akt) substrate of 40 kDa (PRAS40 Sancak et al., 2007 ), mammalian lethal with SEC13 protein 8 (MLST8 Kim et al., 2003 ), DEP domain-containing mTOR-interacting protein (DEPTOR Peterson et al., 2009 ), Tel two interacting protein 1 (Tti1), and telomere maintenance 2 (Tel2 Kaizuka et al., 2010 ). On the other hand, mTORC2 comprises mTOR the scaffold protein rapamycin-insensitive companion of mTOR (Rictor Sarbassov et al., 2004 ) mammalian stress-activated protein kinase interacting protein 1 (mSIN1, Jacinto et al., 2006 ) protein observed with Rictor 1 and 2 (Protor1/2, Pearce et al., 2007 ) MLST8 (Kim et al., 2003 ) DEPTOR (Peterson et al., 2009 ) inhibitor of nuclear factor κ-B kinase (IKK, Xu et al., 2013 ) Sestrin 3 (Tao, Xiong, Liangpunsakul, & Dong, 2015 ) exchange factor found in platelets, leukemic, and neuronal tissues (Xpln, Khanna, Fang, Yoon, & Chen, 2013 ) tuberous sclerosis complex 2 (TSC2 Huang, Dibble, Matsuzaki, & Manning, 2008 ) and Tel2 and Tti1 (Kaizuka et al., 2010 ).

While remarkable progress has been made in understanding the role of mTORC1, the contributions of mTORC2 are less well understood. Collectively, many studies have demonstrated that mTORC1 plays a vital role in the regulation of cellular homeostasis, growth and response to stress. mTORC1 activated under nutrient-replete conditions promotes protein synthesis by several complementary mechanisms. First, mTORC1 activates the ribosomal protein S6 kinase 1 and ribosomal protein S6 kinase 2 (S6K1/2), which in turn activates the protein translation process (Laplante & Sabatini, 2013 Saxton & Sabatini, 2017 ). In parallel, mTORC1 inhibits eukaryotic translation initiation factor 4E (eIF4E)-binding protein-1 (4E-BP1) and thus allows the formation of the eIF4F complex that triggers cap-dependent translation (Kennedy & Lamming, 2016 Laplante & Sabatini, 2013 Saxton & Sabatini, 2017 ). Finally, mTORC1 boosts translation by phosphorylation and consequent inactivation of the target La-related protein 1 (LARP1 Fonseca et al., 2015 ). When active, LARP1 represses translation of terminal oligopyrimidine mRNAs that encode ribosomal proteins and positive regulators of translation.

mTORC1 modulates cell metabolism, as it increases glycolysis by promoting transcription and translation of hypoxia-induced factor 1α (TNFα, Hudson et al., 2002 ). It also activates the transcription factor sterol regulatory element-binding proteins 1 and 2 (SREBP1/2), which promote lipogenesis (Kennedy & Lamming, 2016 Laplante & Sabatini, 2013 Saxton & Sabatini, 2017 ). Further, mTORC1 plays a role in mitochondrial biogenesis through PPAR-γ-mediated activation of the transcription factor Ying-Yang 1 (Cunningham et al., 2007 Laplante & Sabatini, 2012 ). When cells are subjected to stress or nutrient starvation, they undergo a regulated catabolic process termed autophagy. Another well-characterized role of mTORC1 is the inhibition of autophagy under nutrient replete conditions. mTORC1 phosphorylates Unc-51 like autophagy activating kinase (ULK1), preventing its activation via 5′AMP-activated protein kinase (AMPK), which in turn inhibits autophagy (Kim, Kundu, Viollet, & Guan, 2011 ). The phosphorylation and nuclear translocation of the transcription factor EB (TFEB), which regulates the expression of proteins governing autophagy and lysosomal biogenesis, is inhibited by mTORC1 (Settembre et al., 2012 ). Recent studies have demonstrated that mTORC1 also contributes to protein turnover via the ubiquitin–proteasome system. Acute inhibition of mTORC1 increases proteasome-dependent proteolysis (Rousseau & Bertolotti, 2016 Zhao et al., 2015 a). Interestingly, long-term activation of mTORC1 in mouse embryonic fibroblasts due to deletion of inhibitory Tsc2 also increases proteasome activity (Zhang et al., 2014 ). This finding was replicated in a mouse model of neuronal Tsc2 deletion and in the liver of a wild-type mice subject to fasting then 6 hr refeeding. The authors suggest that longer term activation of proteasomal pathways by mTORC1 is an adaptive response that supports protein synthesis by replenishing the cellular amino acid pool (Zhang et al., 2014 ). Two further mTORC1 targets have been identified: (a) The γ isoform of phosphatidylinositol-5-phosphate 4-kinase (PIP4K2C) maintains basal mTORC1 signalling during starvation (Mackey, Sarkes, Bettencourt, Asara, & Rameh, 2014 ) and (b) repressed by TOR (REPTOR) is a downstream effector of TORC1 in Drosophila melanogaster (Tiebe et al., 2015 ). When TORC1 phosphorylates REPTOR, it leads to cytoplasmic retention in contrast, upon inhibition of TORC1, REPTOR is dephosphorylated, translocates into the nucleus, and activates transcription of target genes involved in energy homeostasis and cellular survival under conditions of nutrient starvation (Tiebe et al., 2015 ).

Compared to mTORC1, studies and knowledge of mTORC2 regulation and function have lagged behind. One well-characterized role of mTORC2 is its response to growth factors and insulin via PI3K-dependent mechanisms (Gan, Wang, Su, & Wu, 2011 ). mTORC2 directly phosphorylates Akt at Ser 473 , which is facilitated by prior phosphorylation of Thr 308 by phosphoinositide-dependent kinase 1 (PDK1), as part of the insulin cascade (Sarbassov, Guertin, Ali, & Sabatini, 2005 ). mTORC2 can modulate PKCα activity and thereby play a role in remodelling of the actin cytoskeleton (Sarbassov et al., 2004 ). Similarly, a study by Jacinto et al. ( 2004 ) demonstrated that mTORC2 regulates cell polarity and cytoskeletal organization through the regulation of PKCα and Ras homolog gene family member A. mTORC2 has also been demonstrated to regulate other PKC family members, including PKCδ (Gan et al., 2012 ) and PKCζ (Li & Gao, 2014 ). Hydrophobic motif phosphorylation and activation of PKCδ plays a vital role in fibroblast migration and pulmonary fibrosis development (Gan et al., 2012 ) whereas mTORC2 modulation of PKCζ activity is involved in organization of the actin cytoskeleton (Li & Gao, 2014 ). Sciarretta et al. ( 2015 ) conducted a study showing that mTORC2 negatively regulates the activity of macrophage stimulating 1 (MST1), as disruption of Rictor/mTORC2 leads to a significant activation of MST1. This marked MST1 activation promotes cardiac dilation, cardiac dysfunction, and impaired cardiac growth and adaptation in response to pressure overload.

While mTORC1 and mTORC2 both have distinct functions, there is evidence that these two complexes are interconnected. S6K1, downstream of mTORC1, directly phosphorylates rictor of mTORC2 and promotes a negative regulatory effect on the mTORC2-dependent phosphorylation of Akt-Ser 473 (Dibble, Asara, & Manning, 2009 ). mTORC2-activated Akt, in contrast, enhances mTORC1 activity through the inactivation of tuberous sclerosis complex (TSC1/2), a complex that inhibits mTORC1 via GTPase-activating protein activity towards Ras homologue enriched in the brain (Dibble et al., 2012 ).

We will discuss the manner in which current knowledge of mTOR relates to recent studies demonstrating that adrenoceptor agonists increase activation of mTORC1-mediated cell growth and also mTORC2-mediated glucose uptake and cell survival in vivo and in vitro (Olsen et al., 2014 Sato et al., 2014 Sato et al., 2018 ). We have focused this review on the interplay between adrenoceptors and mTOR in skeletal and cardiac muscle as well as adipose tissue, in light of our own expertise and the need to assimilate considerable information that is now available for these tissues. However, given the ubiquitous expression of mTOR and its partner proteins, as well as widespread expression of different adrenoceptor subtypes, it is highly likely that adrenoceptor–mTOR pathways are important in additional cell types. For example, there are a number of studies linking activation of hippocampal β-adrenoceptors with mTOR-dependent increases in protein translation (Connor, Wang, & Nguyen, 2011 Gelinas et al., 2007 ). These mechanisms are critical for long-term potentiation and memory consolidation.

1.1 Role of β2-adrenoceptor-mediated mTOR activation in skeletal muscle

Skeletal muscle comprises up to 50% of total body mass, consumes a significant proportion of metabolic fuel, and has a major role in whole-body metabolic homeostasis, being responsible for 75% of insulin-mediated glucose uptake and utilization in the fed state. There is evidence showing that the sympathetic nervous system promotes glucose uptake in active skeletal muscle (e.g., during exercise and fight-or-flight responses), which results primarily from noradrenaline release from adrenergic nerve terminals, acting on β-adrenoceptors at the cell surface (Nonogaki, 2000 ). Skeletal muscle expresses abundant β-adrenoceptors that are predominantly β2-adrenoceptors, with 7–10% β1-adrenoceptors and no detectable β3-adrenoceptors (Nevzorova, Bengtsson, Evans, & Summers, 2002 ). Stimulation with the β-adrenoceptor agonist isoprenaline promotes glucose uptake in L6 myoblasts and myotubes, and intact skeletal muscle in vitro and in vivo (Nevzorova, Evans, Bengtsson, & Summers, 2006 Sato et al., 2014 ). Notably, isoprenaline increases glucose uptake to a greater extent than insulin in vivo in wild-type mice, but not in β1/β2-adrenoceptor knockout mice (Sato, Dehvari, Oberg, Dallner, et al., 2014 ), consistent with another study showing that mice lacking all three β-adrenoceptors display glucose intolerance (Asensio, Jimenez, Kuhne, Rohner-Jeanrenaud, & Muzzin, 2005 ).

Insulin stimulates skeletal muscle glucose uptake by activating signalling steps that increase the translocation of glucose transporter 4 (GLUT4) from intracellular vesicles to the cell surface. Following insulin-mediated increases in PI3K activity, phosphatidylinositol 3,4,5-trisphosphate (PIP3) recruits PDK1 and inactive Akt to the plasma membrane via N-terminal PH domains, facilitating Akt phosphorylation at Thr 308 by PDK1. In parallel, mTORC2 is phosphorylated via unknown mechanisms. A conformational change in Akt associated with phosphorylation of Thr 308 enables mTORC2 to phosphorylate Akt at Ser 473 , leading to full activation. Akt promotes subsequent phosphorylation of the Rab GTPase-activating protein Akt substrate of 160 kDa (AS160) at Thr642, which is critical for insulin-increased GLUT4 translocation (Figure 1). Our previous studies showed that isoprenaline-stimulated glucose uptake in L6 muscle cells was markedly reduced by the PI3K inhibitors PI-103, wortmannin, and LY294002 (Sato, Dehvari, Oberg, Dallner, et al., 2014 ), suggesting that insulin receptor and β2-adrenoceptor-mediated glucose uptake may share a common signalling pathway. Unlike responses, we observed to insulin however, there was no Akt phosphorylation at Thr 308 or Ser 473 , or AS160 phosphorylation at Thr 642 upon isoprenaline treatment, nor any increase in PIP3 levels, and glucose uptake was not inhibited by Akt inhibitor X (Nevzorova et al., 2002 Nevzorova et al., 2006 Sato, Dehvari, Oberg, Dallner, et al., 2014 ). Earlier studies demonstrated that PI-103 and other widely used PI3K inhibitors including wortmannin and LY294002 have substantial affinity for related kinases including mTOR (Brunn et al., 1996 Knight et al., 2006 ). It is thus clearly important to consider the involvement of mTOR as well as PI3K when interpreting inhibitory effects of LY294002, wortmannin, or PI-103 on downstream signalling outputs. In light of this, we found that the highly specific mTOR inhibitor KU0063794 (Sato, Dehvari, Oberg, Dallner, et al., 2014 ) inhibited both isoprenaline and insulin-stimulated glucose uptake indicating that mTOR is involved in adrenoceptor-stimulated glucose uptake. The combined results show that the pathways shared by insulin and isoprenaline overlap at a more downstream point leading to mTOR activation, and the β2-adrenoceptor-associated pathway does not include PI3K or Akt. siRNA knockdown of mTORC2 (rictor), but not mTORC1 (raptor), markedly inhibits both insulin-mediated and β2-adrenoceptor-mediated glucose uptake (Sato, Dehvari, Oberg, Dallner, et al., 2014 ). In addition, in muscle lacking rictor, insulin-stimulated Akt phosphorylation at Ser 473 and AS160 at Thr 642 are dramatically decreased, and muscle-specific rictor knockout mice display glucose intolerance and decreased insulin-stimulated glucose uptake (Kumar et al., 2008 ). This confirms mTORC2 as a key regulator of glucose uptake in skeletal muscle. Confirming this, we found that KU0063794 also inhibits β2-adrenoceptor-mediated skeletal muscle glucose uptake ex vivo and in vivo (Sato, Dehvari, Oberg, Dallner, et al., 2014 ).

The β2-adrenoceptor couples primarily to Gαs proteins, activating adenylyl cyclase to increase intracellular cAMP levels, resulting in PKA activation. β2-Adrenoceptor stimulation can also cause cellular effects independently of this classical cAMP–PKA pathway. After agonist stimulation, β2-adrenoceptors are rapidly phosphorylated by G protein receptor kinases (GRKs), allowing recruitment of β-arrestins (which uncouple the receptor from its Gα protein partners), receptor internalization, and activation of β-arrestin-mediated signalling pathways (Tobin, Butcher, & Kong, 2008 ). The signalling effectors linking the β2-adrenoceptor with activation of mTORC2 are still unknown but may be downstream of PKA as the selective PKA inhibitor PKI decreases isoprenaline-induced mTORC2 phosphorylation, and 8-bromo-cAMP increases mTORC2 phosphorylation (Sato, Dehvari, Oberg, Dallner, et al., 2014 ). Interestingly, β2-adrenoceptor-stimulated glucose uptake is only partly dependent on cAMP (Nevzorova et al., 2002 Nevzorova et al., 2006 Sato, Dehvari, Oberg, Dallner, et al., 2014 ), suggesting contributions from alternative effectors that are cAMP-independent. Involvement of GRK2 in β2-adrenoceptor-mediated glucose homeostasis has been suggested as one possible mechanism (Dehvari et al., 2012 ). CHO-K1 cells stably expressing the human GLUT4 carrying an exofacial c-Myc epitope (CHO-GLUT4myc) were transfected with wild type or a truncated β2-adrenoceptor lacking the entire C-terminal tail, or co-transfected with wild-type β2-adrenoceptor and βARKct, which sequesters Gβγ subunits required for GRK2 recruitment to the plasma membrane. Cells expressing wild-type β2-adrenoceptor plus βARKct, or the truncated receptor alone, showed markedly reduced isoprenaline-stimulated glucose uptake compared with cells expressing the wild-type β2-adrenoceptor only. In addition, CHO-GLUT4myc cells expressing a kinase-dead GRK2 K220R mutant displayed significantly decreased GLUT4 translocation to the cell surface (Dehvari et al., 2012 ). Collectively, our studies indicate the potential role of GRK2 and PKA as upstream kinases of mTORC2 following activation of β2-adrenoceptors.

Glucose uptake mediated by β2-adrenoceptors is blocked by GLUT inhibitors and by pretreatment with GLUT4 siRNA (Sato, Dehvari, Oberg, Dallner, et al., 2014 ). Type 2 diabetes is closely associated with defects in insulin signalling mechanisms involving insulin receptor substrates (Chakrabarti et al., 2013 ), PI3K activity, and Akt phosphorylation (Cusi et al., 2000 ), but β2-adrenoceptors expressed in skeletal muscle could bypass these defects through mTORC2-mediated regulation of GLUT4 trafficking, providing a compensatory pathway following loss of insulin sensitivity (Sato et al., 2014 Sato, Dehvari, Oberg, Dallner, et al., 2014 ). This is of particular interest considering that β2-adrenoceptor expression is unaltered in skeletal muscle from diabetic patients (Frederiksen et al., 2008 ).

Apart from being important for glucose uptake, β2-adrenoceptor-stimulated cAMP accumulation can have long-term effects on muscle phenotype (Pearen, Ryall, Lynch, & Muscat, 2009 ). Chronic stimulation of skeletal muscle β2-adrenoceptors utilizing agonists such as clenbuterol, fenoterol, and formoterol can activate anabolic signalling pathways, leading to increased muscle mass and force-producing capacity (Lynch & Ryall, 2008 ) The anabolic and anti-catabolic processes in response to β2-adrenoceptor agonists occur via protein translation and synthesis mediated by the Akt–mTOR–S6 kinase signalling axis (Hagg et al., 2016 Figure 1). Chronic stimulation of β2-adrenoceptors increases the transcription of PPAR-γ coactivator 1-α, which is associated with the suppression of myostatin, and these effects are blocked by ICI-118,551, a highly selective β2-adrenoceptor antagonist (Jesinkey, Korrapati, Rasbach, Beeson, & Schnellmann, 2014 ). Treatment of mice with formoterol stimulates small but significant increases in the phosphorylation of Akt and mTOR in gastrocnemius muscle after 8 hr, differing in time frame from more acute measurements of Akt/mTOR phosphorylation and glucose uptake (10 min to 2 hr Sato, Dehvari, Oberg, Dallner, et al., 2014 ). Dexamethasone-induced muscular atrophy and slow-to-fast myosin heavy chain isoform transition is antagonized by the β2-adrenoceptor agonist clenbuterol, which stimulates Akt and mTORC1 activity, and insulin-like growth factor 1 expression (Jesinkey et al., 2014 ). These findings could potentially provide a new basis for a pharmacological approach to target mTOR for the treatment of conditions involving muscle loss.

1.2 The role of adrenoceptor-mediated mTOR activation in the heart

1.2.1 α1-adrenoceptors and mTOR in the heart

Cardiac function is tightly regulated via both α1- and β-adrenoceptors, due to release of noradrenaline from sympathetic nerve terminals innervating the heart and by circulating adrenaline released from the adrenal gland in response to danger or stress. The β-adrenoceptors comprise roughly 90% of total cardiac adrenoceptors, and the α1- adrenoceptors account for the remaining 10%. In heart failure, unlike β1-adrenoceptors, α1- adrenoceptors are not down-regulated and may therefore play an enhanced role in regulating cardiac contractility (Skomedal, Borthne, Aass, Geiran, & Osnes, 1997 ). Although mRNAs for all three α1- adrenoceptor subtypes are detected in the heart of mice and rats, cardiomyocytes express only the α1A- and α1B-subtypes (O'Connell et al., 2003 ) while α1D-adrenoceptors are confined to the coronary vasculature (McCloskey et al., 2003 O'Connell et al., 2003 ). Due to the putative enhanced role in heart failure, α1- adrenoceptor function and signalling are therefore of particular interest.

A series of knockout mouse studies indicate that neither α1A- nor α1B- adrenoceptors are required for basal contractile function (O'Connell et al., 2003 Vecchione et al., 2002 ). However, cardiomyocyte-specific overexpression of the α1A- adrenoceptor enhances basal contractile function (Lin et al., 2001 ) and reduces adverse remodelling following pressure overload (Du et al., 2004 Du et al., 2006 ). These results are consistent with an in vitro study by Mohl et al. ( 2011 ), identifying an α1A- adrenoceptor-mediated signalling pathway that increases calcium entry and cardiomyocyte contractility. In contrast, overexpression of the α1B-adrenoceptor causes depressed contractile function and pathological remodelling in the heart (Lemire et al., 2001 Wang, Du, Autelitano, Milano, & Woodcock, 2000 ). The capacity of α1A- adrenoceptors to increase contractile function may have important compensatory roles in the failing heart.

In addition to maintaining myocyte contractility, activation of α1-adrenoceptors promotes glucose uptake (Shi, Papay, & Perez, 2016 ), receptor-mediated preconditioning, cardiac hypertrophy, and inhibition of cardiomyocyte apoptosis (Jensen, O'Connell, & Simpson, 2011 O'Connell, Jensen, Baker, & Simpson, 2014 ). α1-adrenoceptors are expressed in human myocardium and are not down-regulated in heart failure (Jensen, Swigart, De Marco, Hoopes, & Simpson, 2009 ), and blockade of α1- adrenoceptors worsens heart failure (Dhaliwal et al., 2009 Jensen et al., 2011 ). In murine cardiac myocytes that express endogenous α1A- and α1B- adrenoceptors, long-term agonist treatment increases the abundance of α1A- adrenoceptors without desensitization of inotropic effects, while increased stimulation or expression of the α1A- but not the α1B- adrenoceptor in vivo limits global cardiac remodelling and reduces mortality from heart failure (Du et al., 2006 Rorabaugh et al., 2005 ). In a transgenic rat model that overexpresses the cardiomyocyte α1A-adrenoceptors, animals are protected from heart failure by increased angiogenesis associated with secretion of VEGF from cardiomyocytes (Zhao, Zhai, Gygi, & Goldberg, 2015 ).

In vitro and in vivo studies have indicated that stimulation of α1-adrenoceptors reduces cardiomyocyte cell death. Hypoxia-, serum starvation-, and isoprenaline-induced apoptosis can be inhibited by exposure of cardiomyocytes to phenylephrine, a non-selective α1-adrenoceptor agonist. This phenylephrine cytoprotective effect was blocked by phentolamine and prazosin (Iwai-Kanai et al., 1999 ). Cardiomyocytes from α1A-/α1B-adrenoceptor knockout mice display significantly increased necrosis and apoptosis when subject to toxic stimuli such as doxorubicin or H2O2 (Huang et al., 2007 O'Connell et al., 2006 ), and this sensitivity can be reduced by re-expression of α1A- adrenoceptors but not α1B-adrenoceptors (Huang et al., 2007 ). The chemotherapeutic agent doxorubicin produces cardiotoxic effects in patients and in animal models. In mice, long-term in vivo infusion of the α1A- adrenoceptor agonist A61603 protects cardiomyocytes against apoptosis and reduces adverse ventricular remodelling and myocardial fibrosis following doxorubicin treatment, thereby improving cardiac function (Chan, Dash, & Simpson, 2008 Montgomery et al., 2017 ). These protective effects of A61603 are not observed in α1A- adrenoceptor knockout mice. Another study showed that dabuzalgron, an orally available, selective α1A- adrenoceptor agonist also increases survival and preserves fractional shortening in wild-type but not in α1A- adrenoceptor knockout mice (Beak et al., 2017 ). All of these studies indicate that α1- adrenoceptors could be an important target in the failing heart.

The non-selective α1- adrenoceptor agonist phenylephrine is a well-known hypertrophic agent in the heart and has been linked to activation of the mTORC1 target S6K1 (Boluyt et al., 1997 ). Treatment of neonatal rat ventricular myocytes (NRVMs) with phenylephrine stimulated the activity of S6K1, increased protein synthesis, and produced a 50% increase in cardiomyocyte area. Phenylephrine-induced S6K1 activity and hypertrophy were significantly reduced by the mTORC1 inhibitor rapamycin and by the PI3K inhibitor LY294002 however, the authors acknowledge that compounds such as LY294002 affect other PI3K-related kinases (Boluyt et al., 1997 ). As outlined in the skeletal muscle section of this review, PI3K inhibitors including wortmannin and LY294002 have substantial activity at mTOR (Brunn et al., 1996 Knight et al., 2006 ). Thus, studies in which LY294002 is used as a sole PI3K inhibitor should be regarded with caution. Taken together, these results suggest that phenylephrine activates S6K1 and promotes cardiomyocyte hypertrophy via mTORC1 and possibly PI3K. We have shown recently that treatment of NRVMs with the highly selective α1A-AR agonist A61603 increases phosphorylation of S6 ribosomal protein, a downstream target of mTORC1 and S6K1, and this is inhibited by rapamycin. NRVM hypertrophy observed in response to A61603 was prevented by the mTOR inhibitor KU0063794, which blocks the phosphorylation and activation of both mTORC1 and mTORC2 (Sato et al., 2018 ). It is thus clear that α1-adrenoceptors stimulate mTORC1 and that this could be an important player in the ability of α1-adrenoceptors to protect the heart.

Phenylephrine stimulates activation of S6K1 and phosphorylation of 4E-BP1 in adult cardiomyocytes (Wang & Proud, 2002 ). The latter protein interacts with eIF4E and represses translation. Phosphorylation of 4E-BP results in its dissociation from eIF4E and activation of mRNA translation. The response to phenylephrine was blocked by MEK inhibitors, and adenoviral expression of constitutively active MEK caused activation of S6K1, phosphorylation of 4E-BP1, and activation of protein synthesis in a rapamycin-sensitive manner. This study provides insight into a signalling pathway involving Ras, MEK, and mTOR (Wang & Proud, 2002 ). Phenylephrine also activates S6K2 in adult rat ventricular cardiomyocytes. Both MEK1/2 inhibitors and rapamycin abolished phenylephrine-induced activation of S6K2, and the expression of constitutively active MEK1 activated S6K2. This indicates that MEK/ERK1/2 in combination with mTOR signalling plays a role in regulating phenylephrine-induced S6K2 activation (Wang, Gout, & Proud, 2001 ).

Although the classic α1- adrenoceptor signalling pathway includes Ca 2+ -dependent PKC, phenylephrine also regulates S6K1/2 and 4E-BP1 (downstream substrates of mTORC1) leading to protein synthesis in a Ca 2+ -independent PKC manner in adult cardiomyocytes (Wang, Rolfe, & Proud, 2003 ). The classical Ca 2+ -dependent PKCα and the Ca 2+ -independent PKCδ and PKCε are readily detected in adult cardiomyocytes (Puceat, Hilal-Dandan, Strulovici, Brunton, & Brown, 1994 Steinberg, Goldberg, & Rybin, 1995 ). In addition, Ca 2+ -independent PKC is also required for the phenylephrine-induced ERK1/2 activation demonstrated by the significantly reduced ERK1/2 activation in the presence of the broad-spectrum PKC inhibitor BIM I (Toullec et al., 1991 ). Rottlerin (Gschwendt et al., 1994 ), a selective inhibitor of PKCδ, almost completely inhibited the phenylephrine-induced ERK1/2 phosphorylation, while Gö6979 (Martiny-Baron et al., 1993 ), an inhibitor of Ca 2+ -dependent PKC has no obvious effect on ERK1/2 activation. Furthermore, Rottlerin prevented phenylephrine-induced S6K activation whereas Gö6979 had no apparent effects. Phosphorylation of 4E-BP1 was also inhibited by rottlerin in a similar manner (Wang et al., 2003 ). These data suggest Ca 2+ -independent PKC isoforms play a vital role in α1-adrenoceptor-mediated mTOR signalling in adult cardiomyocytes.

While mTORC1 plays an important role in cardiomyocyte hypertrophy, there is convincing evidence that mTORC2 promotes cardiomyocyte development and survival (Gonzalez-Teran et al., 2016 Shende et al., 2016 Xu & Brink, 2016 ). For example, mice with cardiomyocyte-specific knockdown of rictor and thus disruption of mTORC2 display abnormalities by the age of 6 months, including cardiac dilation, fibrosis, and exacerbated heart failure in response to pressure overload (Sciarretta et al., 2015 Yano et al., 2014 ). Following ischaemic preconditioning, activation of mTORC2 promotes cardiomyocyte survival in part by suppressing activity of the kinase Mst1 (large tumour suppressor kinase 2), a key component of the Hippo pathway that promotes apoptosis and inhibits cell growth (Sciarretta et al., 2015 Yano et al., 2014 ). Importantly, cardiomyocytes that are rictor-deficient or overexpress Mst1 display increased cell death. In the study by Shende et al. ( 2016 ), tamoxifen-inducible cardiomyocyte-specific rictor knockdown was used to allow normal cardiac development. Mice in which Cre recombinase expression was induced at 4 or 10 weeks of age displayed normal cardiac size and echocardiography up to 44 weeks after tamoxifen treatment, but transverse aortic constriction and resultant pressure overload caused more pronounced cardiac dysfunction than in wild-type mice, indicating the importance of mTORC2 in the failing heart (Shende et al., 2016 Volkers et al., 2013 ).

Cardiac α1-adrenoceptorshave been linked with mTOR in exerting cardioprotective effects. Serum and glucocorticoid-responsive kinase-1 (SGK1) is a downstream substrate of mTORC2 (Garcia-Martinez & Alessi, 2008 ) that regulates cardiomyocyte survival and hypertrophy in response to the non-selective α1-adrenoceptor agonist phenylephrine, both in vivo and in vitro (Aoyama et al., 2005 ). Cardiomyocytes infected with an adenoviral vector encoding constitutively active SGK1 show reduced apoptosis after serum- or oxygen-deprivation and increased [ 3 H]-leucine incorporation in response to phenylephrine, while expression of kinase-dead SGK1 increases apoptosis. SGK1 has also been placed downstream of PI3K (Park et al., 1999 ), although again inhibition of mTOR may have confounded the interpretation of these experiments involving the use of LY294002 as a PI3K inhibitor.

We have demonstrated that noradrenaline and the α1A-adrenoceptor agonist A61603 increase glucose uptake in NRVMs by parallel activation of AMPK and mTORC2 but do not promote phosphorylation of Akt at Thr 308 or Ser 473 (Sato et al., 2018 ). The lack of Akt phosphorylation mirrors similar findings by Wang et al. ( 2001 ), who demonstrated using adult cardiomyocytes that phenylephrine does not produce Akt phosphorylation at Ser 473 and that adenoviral expression of a dominant-negative Akt mutant fails to block activation of S6K2 by phenylephrine. We found that the mTORC1/2 inhibitor KU0063794 partly reduced α1A- adrenoceptor and insulin-stimulated glucose uptake in cardiomyocytes, whereas the mTORC1 inhibitor rapamycin had no effect. A61603 stimulated the phosphorylation of mTOR at Ser 2448 and Ser 2481 . Overall, the data suggest that α1A-adrenoceptors stimulate mTORC2 to increase glucose uptake and mTORC1 to promote protein synthesis and hypertrophy in NRVMs (Sato et al., 2018 Figure 2), but the detailed mechanism whereby α1A-adrenoceptors activate mTORC2 is still not known.

1.2.2 β- adrenoceptors and mTOR in the heart

Activation of β-adrenoceptors plays an important role in the regulation of cardiovascular function, including positive inotropic and chronotropic effects (Bristow et al., 1993 Brodde, 1991 ). Noradrenaline exerts its effects on the heart nearly exclusively via β1-adrenoceptors (Kaumann, Hall, Murray, Wells, & Brown, 1989 ). Thus, under normal physiological conditions β1-adrenoceptors are the predominant cardiac adrenoceptors responsible for regulation of heart rate and contractility. The β1-adrenoceptors activate the canonical Gαs-adenylate cyclase-cAMP-PKA signalling cascade. In cardiomyocytes, the activation of PKA promotes phosphorylation of multiple proteins that increase calcium mobilization primarily from the sarcoplasmic reticulum and, to a lesser extent, from the extracellular milieu, leading to increased rates of contraction and relaxation and to increased force of contraction (Sirenko et al., 2014 Figure 2). In the early stages of heart failure, cardiac output is increased via overstimulation of β1-adrenoceptors as a compensatory mechanism for the insufficient blood and oxygen supply (Brodde, 1993 ), but this leads to longer term structural damage, including ventricular remodelling, cardiomyocyte apoptosis and fibrosis, and cardiac hypertrophy (Engelhardt, Hein, Wiesmann, & Lohse, 1999 O'Connor et al., 1999 ). In addition, recent evidence has shown that β1-adrenoceptors decrease myocardial autophagy that maintains cellular homeostasis (Wang et al., 2013 Wang et al., 2015 ). Inhibition of autophagy causes the accumulation of denatured proteins and damaged organelles, contributing to cardiac dysfunction (Magnusson, Wallukat, Waagstein, Hjalmarson, & Hoebeke, 1994 ), and up-regulation of autophagy by the mTORC1 inhibitor rapamycin can improve impaired cardiac function (Wang et al., 2015 ). The β1-adrenoceptor-mediated inhibition of autophagy occurs via PKA phosphorylation of Ser 12 in the autophagy-related protein LC3 (Kroemer, Zamzami, & Susin, 1997 ). mTORC1 is overactive in the early stages of heart failure and plays a role in the β1-adrenoceptor-mediated inhibition of autophagy (Wang et al., 2015 Figure 2).

1.3 The role of β-adrenoceptors and mTOR in adipose tissue

There are two types of adipose tissue with distinct physiological functions: white adipose tissue (WAT) that stores chemical energy as triacylglycerol and BAT that releases chemical energy as heat (thermogenesis). BAT is responsible for sympathetically mediated non-shivering thermogenesis in mammals and is activated by members of the adrenoceptor family (Cannon & Nedergaard, 2004 ). In addition, many groups have described the existence of brown adipocytes in depots thought to be primarily WAT, both in animal models and in humans (Petrovic et al., 2010 Wu et al., 2012 ). These cells differ from prototypical BAT found in rodents or human infants and have been termed “brite” (brown in white) or “beige” adipocytes (Petrovic et al., 2010 Wu et al., 2012 ). The appearance of brite adipocytes per se is insufficient to promote increased energy expenditure, as these cells must also be activated by environmental, hormonal, or pharmacological stimuli such as drugs acting at GPCRs (Merlin et al., 2016 ). The expression of adrenoceptors in brown, white, and brite adipocytes and their contribution to adipocyte function is described in detail in an accompanying review (Evans, Merlin, Bengtsson, & Hutchinson, 2019 ). We will focus here on the interplay between adrenoceptor signalling and the role of mTOR complexes in adipocyte browning and glucose metabolism.

1.3.1 β-adrenoceptors and mTOR in WAT

When nutrients are plentiful, insulin is released from the pancreas and stimulates the uptake of glucose and fatty acids by adipose tissue, where they are stored as triacylglycerol forming lipid droplets. Insulin signalling in adipocytes is mediated by the PI3K–Akt–mTOR pathway, producing anabolic effects including cell growth and inhibition of lipolysis (Chakrabarti et al., 2013 ). During periods of fasting or stress, catecholamines are released by the sympathetic nervous system to activate β-adrenoceptors. Stimulation of the β3-adrenoceptors in WAT activates adenylyl cyclase, leading to increased cAMP levels and PKA activity. PKA phosphorylates and regulates several important targets in adipocytes, including hormone-sensitive lipase and the lipid droplet-associated perilipins, which collectively promote triglyceride hydrolysis and liberation of free fatty acids (Granneman & Moore, 2008 Figure 3).

Two studies have suggested that adrenoceptor-stimulated lipolysis inactivates mTOR in WAT (Mullins et al., 2014 Scott & Lawrence, 1998 ). Mullins et al. ( 2014 ) demonstrated that β-adrenoceptor-mediated lipolysis suppresses glucose uptake because lipolysis causes both mTORC1 and mTORC2 complexes to dissociate (Figure 3). This is in agreement with the proposal that in white adipocytes, cAMP indirectly prevents activation of mTOR, since there is a decrease in p70S6K, a downstream target of mTORC1 (Scott & Lawrence, 1998 ). Conversely, there are new studies indicating that stimulation of β3-adrenoceptors in WAT does not inhibit mTOR complexes but instead activates mTORC1 through PKA (Liu et al., 2016 ), resulting in browning of WAT depots. This variance in results might be due to the fact that β-adrenoceptor stimulation interacts differently with mTOR in different WAT depots. Nonetheless, these results suggest that β-adrenoceptor regulation of mTOR could have an important role in WAT function.

1.3.2 β-adrenoceptors and mTOR in BAT

Binding of noradrenaline to BAT β-adrenoceptors activates intracellular signalling cascades leading to increased expression of uncoupling protein 1 (UCP1) and breakdown of triglycerides to free fatty acids that activate UCP1 in the inner mitochondrial membrane (Figure 3). Activated UCP1 collapses the proton gradient that drives ATP synthesis and energy storage thus, β-adrenoceptor signalling increases mitochondrial respiration and non-shivering thermogenesis (Cannon & Nedergaard, 2004 ). The metabolic capacity of BAT potentially allows it to influence whole-body energy homeostasis. For instance, BAT has been shown to play an important role in the regulation of glucose homeostasis and insulin secretion (Guerra et al., 2001 ). Cold exposure of animals increases glucose uptake into BAT due to activation of the sympathetic nervous system (Shibata, Perusse, Vallerand, & Bukowiecki, 1989 Shimizu, Nikami, & Saito, 1991 ), and this response is mimicked by administration of β-adrenoceptor agonists in vivo (Liu, Perusse, & Bukowiecki, 1994 Olsen et al., 2014 ). Mouse brown adipocytes cultured in vitro also display increased glucose uptake upon treatment with β-adrenoceptor agonists (Chernogubova, Hutchinson, Nedergaard, & Bengtsson, 2005 Dallner, Chernogubova, Brolinson, & Bengtsson, 2006 Merlin et al., 2018 Olsen et al., 2014 ). While a role for β3-adrenoceptor-mediated glucose uptake in rodents is well established, the contribution of these receptors in human adipose tissue is less clear. It has been demonstrated, however, that cold exposure increases 18 F-2-deoxyglucose uptake in human BAT depots, and this effect can be mimicked by administration of the β3-adrenoceptor agonist mirabegron that is used clinically for overactive bladder (Baskin et al., 2018 Cypess et al., 2015 ).

There is strong evidence that glucose uptake in response to β3-adrenoceptor agonists occurs via a Gαs–cAMP–PKA pathway, based on the use of pharmacological inhibitors (Chernogubova, Cannon, & Bengtsson, 2004 Olsen et al., 2014 ). In addition, 8-bromo-cAMP and upstream activation of Gαs by cholera toxin both increase glucose uptake in primary brown adipocytes (Chernogubova et al., 2004 Olsen et al., 2014 ). Other mechanisms involved in β3-adrenoceptor-mediated glucose uptake include localization of the β3-adrenoceptors in lipid-rich microenvironments in the plasma membrane (Sato et al., 2012 ), conventional and novel PKC isoforms (Chernogubova et al., 2004 ), and AMPK (Hutchinson, Chernogubova, Dallner, Cannon, & Bengtsson, 2005 Inokuma et al., 2005 ). As demonstrated in skeletal muscle, mTORC2 plays a pivotal role in adipocyte glucose uptake stimulated by β-adrenoceptor agonists, as well as insulin.

The contributions of mTORC1 and mTORC2 have been examined in mice with specific ablation of raptor or rictor in all adipocytes, as these cells express Cre recombinase under control of the adiponectin promoter (Kumar et al., 2010 Polak et al., 2008 ). Ablation of raptor (mTORC1) in adipose tissue increases mitochondrial uncoupling but has no effect on insulin-mediated Akt phosphorylation or glucose tolerance profiles in chow-fed mice (Polak et al., 2008 ). In contrast, adipocytes isolated from mice with fat-specific ablation of rictor (mTORC2) display reduced insulin-stimulated Akt-Ser 473 phosphorylation, GLUT4 translocation to the cell surface, and glucose uptake, and these mice have impaired glucose tolerance profiles in vivo (Kumar et al., 2010 ). These studies indicate that like in skeletal muscle, the mTORC2 complex is involved in glucose homeostasis in adipocytes.

We have demonstrated using brown adipocytes that mTORC2 is involved in β3-adrenoceptor-mediated glucose uptake (Olsen et al., 2014 ). Overall inhibition of mTOR by Torin-1 or KU0063794 reduces glucose uptake, but two lines of evidence demonstrate the involvement of mTORC2 rather than mTORC1: (a) 24-hr, but not 2-hr, rapamycin treatment attenuates β3-adrenoceptor-mediated glucose uptake (rapamycin acutely inhibits mTORC1, whereas long-term treatment prevents mTORC2 assembly), and (b) siRNA against rictor, but not raptor, reduces glucose uptake by β3-adrenoceptors (Mohl et al., 2011 Olsen et al., 2014 ). In brown adipocytes, β3-adrenoceptor-mediated glucose uptake depends on de novo synthesis and translocation of GLUT1 (Dallner et al., 2006 ), which are both cAMP-dependent (Figure 3). mTORC2 is specifically involved in the translocation of newly synthesized GLUT1 to the plasma membrane, but is not required for de novo synthesis of GLUT1 (Olsen et al., 2014 ). In brown adipocyte cultures, inhibition of PI3K by compound 15e, or of Akt by inhibitor X, reduced insulin- but not isoprenaline-stimulated glucose uptake. Akt was phosphorylated at Thr 308 and Ser 473 in response to insulin but not isoprenaline (Olsen et al., 2014 ).

A recent study has also shown that mice lacking rictor in adipose tissue are hypothermic, show increased susceptibility to cold, and have impairment of cold-induced glucose uptake and glycolysis (Albert et al., 2016 ). This study indicates that mTORC2 plays a central role in adipose tissue metabolism and translocation of GLUT 1/4 in vitro and in vivo. Interestingly, the GLUT 1/4 content in the plasma membrane of brown adipocytes was not altered by cold exposure in that study. Also in contrast to our previous findings (Olsen et al., 2014 ), both immortalized mouse brown adipocytes treated with noradrenaline (1 μM) for 5 min and native BAT from wild-type mice treated for 30 min in vivo with noradrenaline (1 mg·kg −1 ) showed phosphorylation of Akt at Ser 473 , known to be downstream of mTORC2. There is no clear explanation for the disparity with our brown adipocytes however, an emerging view is that adipose depots display considerable heterogeneity in cell composition (Shinoda et al., 2015 ). This would account for differences between in vivo and in vitro data and may also be consistent with phenotypic differences between primary brown adipocyte cultures that are representative of the starting population of stromal vascular pre-adipocytes and immortalized adipocytes that have been selected for the presence of plasmid encoding SV40 T antigen (Klein, Fasshauer, Klein, Benito, & Kahn, 2002 ) and therefore represent only a small subset of the starting cell population. In addition, noradrenaline may activate the α2-adrenoceptors present in BAT or brown adipocytes, promoting signalling via a Gαi/o-Gβγ-PI3K-PDK1-Akt axis. In immortalized human multipotent adipose-derived stem (hMADS) brown adipocytes treated with low concentrations of isoprenaline, glucose uptake is blocked by the mTOR inhibitor KU0063794, as seen in mouse brown adipocyte primary cultures (Olsen et al., 2014 ). It would be interesting to determine whether hMADS cells display Akt phosphorylation at Ser 473 in response to isoprenaline treatment.

1.3.3 mTORC1 mediates browning of brite adipocytes

In addition to BAT, there is increasing evidence for the existence of brown adipocytes in depots thought to be primarily WAT, both in animal models and in humans (Petrovic et al., 2010 ). These cells differ from prototypical BAT found in rodents or human infants and have been termed “brite” (brown in white) or “beige” adipocytes. Two studies indicate that brite adipocytes contribute significantly to whole-body energy expenditure: Mouse models that have increased brite adipocytes in WAT are protected from diet-induced obesity (Seale et al., 2011 ), and browning of WAT contributes to non-shivering adaptive thermogenesis in the absence of classical brown adipocytes (Schulz et al., 2013 ). Our in vitro results show that stimulation of the β3-adrenoceptors increases glucose uptake in brown and brite adipocytes, but not white adipocytes, in contrast to insulin, which increases glucose uptake in all three adipocyte cultures (Merlin et al., 2018 ).

Separate studies have shown that the β-adrenoceptor–cAMP–PKA pathway can lead to mTORC1 activation (Figure 3) and is necessary for the induction of adipose tissue browning and BAT development (Liu et al., 2016 ). In addition, wild-type mice treated with the mTORC1 inhibitor rapamycin or mice with adipocyte-specific deletion of raptor are cold-intolerant and show impaired expression of UCP1 and other mitochondrial components in inguinal WAT, suggesting that there may be a role for mTORC1 even in the early development of inguinal WAT brite adipocytes (Liu et al., 2016 Tran et al., 2016 ). Several downstream target genes of PPAR-α and oestrogen-related receptor α (ERRα) are similarly under the control of mTORC1. PPAR-α is a master nuclear receptor for fatty acid β-oxidation and has been shown to participate in UCP1 expression either directly or indirectly through ERRα (Morganstein et al., 2010 ). Therefore, mTORC1 appears to have an important role in the catabolic process of adipose tissue browning and the dissipation of chemical energy by thermogenesis.


Multiple pieces of evidence suggest the involvement of signaling molecules regulating cytoskeleton remodeling and actin dynamics in the structural and synaptic phenotypes displayed by FXS model mice (67, 72, 74). It has been reported that FMRP and Rac1 are connected in a common signaling pathway and that Rac1 functions are compromised in the absence of FMRP (67, 68). Moreover, either a dominant-negative form of PAK1, which inhibits PAK activity, or pharmacological blockade of PAK1 rescues structural and behavioral alterations displayed by FXS model mice (6971). Our discovery of increases in Rac1 activity, phosphorylation of PAK1/2 and cofilin, and F-actin/G-actin ratio is in line with previous studies, suggesting an interaction between Rac1 and FMRP (67, 68, 72). Our results may help to explain why reduction of PAK phosphorylation/activation is effective in rescuing several phenotypes in FXS model mice (6971). It should be noted that in contrast with our findings, a decrease in phosphorylation of cofilin and increased levels of PP2Ac, which is the phosphatase that dephosphorylates cofilin, were reported in murine fibroblast lacking FMRP (68). The difference between these findings is most likely due to the difference in the models (cell culture versus brain slices) and/or cell types (fibroblasts versus neurons) used in the two studies. In the future, it will be interesting to address whether Rac1-PAK1/2-cofilin signaling is elevated in the brains of FXS patients.

We found that activation of group 1 mGluR receptors triggers the activation of the PAK1/2-cofilin signaling pathway that regulates actin dynamics in area CA1 of the hippocampus in wild-type mice. In contrast, when we examined the effect of mGluR activation on the activity of the PAK1/2-cofilin pathway in FXS model mice, we did not observe further increases, suggesting that this signaling pathway is already at plateau. This result is consistent with several previous reports demonstrating the decoupling of mGluRs from downstream signaling pathways in FXS model mice. For instance, baseline phosphorylation of p70 ribosomal protein S6 kinase 1 (S6K1) and eIF4E-eIF4G interactions, which are downstream effectors of the mTORC1 and ERK pathways and link mGluRs to the protein synthesis machinery, are already elevated in FXS model mice and are not stimulated further by activation of group 1 mGluRs (38, 39). Moreover, stimulation of group 1 mGluRs leads to enhanced hippocampal LTD in FXS model mice but no longer requires activation of ERK and mTORC1, or de novo protein synthesis (27, 82, 83, 88). Consistent with our findings, it was demonstrated that Rac1-PAK1 activation induced by synaptic activity is defective in the hippocampus of FXS model mice (54). In this context, our results indicate that the Rac1-PAK1/2-cofilin pathway is altered and uncoupled to group1 mGluR activity in FXS mice, as was previously shown for mTORC1 and ERK signaling, as well as de novo protein synthesis.

Notably, we found that blocking eIF4E-eIF4G interactions with 4EGI-1, which is increased in FXS model mice (38, 40), normalizes Rac1-PAK1/2-cofilin signaling (Fig. 4), synaptic plasticity (Fig. 2), context discrimination, and spine density (Fig. 1) phenotypes in FXS model mice. 4EGI-1 is an inhibitor of cap-dependent protein synthesis, which, by blocking the association of eIF4E with eIF4G, prevents the formation of the eIF4F initiation complex (47, 48, 75, 76). The effects of 4EGI-1 on Rac1-PAK1/2-cofilin signaling are observed both in the baseline condition for FXS model mice and after DHPG stimulation in both wild-type and FXS model mice, suggesting that interfering with eIF4E-eIF4G interactions is beneficial in the absence of FMRP. From a molecular standpoint, it is possible that these multiple phenotypic rescues mediated by 4EGI-1 in FXS model mice have more than one explanation.

Many of the phenotypes displayed by FXS model mice are rescued by interfering with components of the mTORC1 and ERK pathways, both of which regulate protein synthesis downstream of group 1 mGluRs (39, 41, 43, 44). For example, either genetically or pharmacologically inhibiting the mTORC1 substrate S6K1 (41, 43) or the ERK-dependent phosphorylation of eIF4E (39) can rescue multiple phenotypes in FXS model mice. Notably, eIF4E-eIF4G interactions are enhanced in FXS model mice (38, 40) and both mTORC1 and ERK activation increase the levels of eIF4E-eIF4G interactions to initiate protein synthesis (8991). At the same time, normalization of multiple phenotypes displayed by FXS mice can also be achieved by either genetically or pharmacologically inhibiting the activity of PAK1 (69, 70).

Our data are consistent with a model describing CYFIP1 as a molecule regulating protein synthesis and actin dynamics by shuttling between the FMRP/eIF4E and Rac1/WRC complexes (9, 67, 72). Our results indicate that, at baseline, CYFIP1 is preferentially associated with Rac1, possibly in the WRC complex, rather than eIF4E, in FXS model mice (Fig. 6B). Administration of 4EGI-1 inhibits the association of eIF4E to eIF4G and creates free eIF4E that competes with Rac1 to bind CYFIP1, thus restoring the balance between these signaling pathways (Fig. 6C). Although the precise molecular details linking the interaction of Rac1 to CYFIP1 (with the effect on PAK1/2-cofilin signaling by 4EGI-1) are unclear, it is possible that Rac1 becomes less effective in activating the downstream PAK1/2-cofilin in the absence of CYFIP1. In the future, it will be interesting to investigate the molecular details of these interactions.

(A) In area CA1 of the hippocampus in WT mice, CYFIP1 is present in two macromolecular complexes: CYFIP1-FMRP-eIF4E, which represses translation, and CYFIP1-WRC-Rac1-GTP, which regulates actin remodeling. Activation of mGluR1/5 changes the balance between these two complexes by increasing the activation of Rac1 and downstream signaling molecules and inducing protein synthesis via the relocation of CYFIP1 to the WRC-Rac1-GTP complex. The concomitant change in these signaling pathways ensures normal physiological synaptic plasticity and higher brain function. (B) In FXS model mice, the signaling molecules regulating protein synthesis and actin dynamics are no longer regulated by activation of mGluR1/5. Moreover, the absence of FMRP leads to exaggerated protein synthesis, perhaps via enhanced association of eIF4E to eIF4G. The lack of FMRP also results in an alteration in actin dynamics via enhanced Rac1-GTP abundance and increased association of CYFIP1 to WRC-Rac1-GTP. The disruption of these two signaling modules results in aberrant synaptic plasticity, spine morphology, and brain function that are characteristic of FXS. (C) 4EGI-1 restores the balance between protein synthesis and actin dynamics by creating free eIF4E that competes with Rac1-GTP to bind CYFIP1. This normalizes the aberrant synaptic plasticity, spine morphology, and cognitive function exhibited in FXS model mice. Dashed arrows indicate an indirect phosphorylation mechanism. Thick and thin arrows indicate enhanced and reduced interactions, respectively. Red arrows indicate the effect of 4EGI-1.

In summary, our results support a model that links dysregulated protein synthesis with altered actin dynamics in FXS, and underscore the importance of a balanced regulation of these two pathways for a normal brain function (Fig. 6). Moreover, our results suggest an alternative way to counteract the abnormal phenotypes displayed by FXS model mice, which can be pursued by using drugs that normalize both molecular pathways.

Concluding remarks and perspectives

The mTOR pathway is emerging as a critical player in the etiology of cancer and metabolic diseases, including diabetes and obesity. The recent breathtaking advances in the understanding of the upstream and downstream targets of mTOR provide rational explanations for the origins and progression of these diseases. For example, insulin is a major upstream effector of mTOR that increases protein synthesis as part of the modulation of anabolic processes in response to glucose. Thus, deficiencies in mTOR signaling might play a role in the development of glucose- and insulin-resistant type II diabetes (Pende et al. 2000). As discussed above, a link between the mTOR pathway and cancer is also clearly evident, as most of the upstream and downstream components of mTOR are directly implicated in cancer initiation and progression. The enhanced understanding of the mTOR signaling pathway should lead to the design of drugs to treat diabetes and cancer. The success of rapamycin in clinical trials for cancer, restenosis in heart valves (Marks 2003), and arthritis (Forre et al. 2000) highlights the multitude of diseases whose origins stem from aberrant proliferation and that are linked to mTOR. Other drugs that act on other components of the pathway are also sought. Studies are in progress to develop drugs that inhibit upstream mTOR effectors such as Akt/PKB or downstream targets such as eIF4E.

Several important details related to the regulation of mTOR activity remain unresolved. In particular, the mechanism by which Rheb activates mTOR is still elusive, and future studies will be likely directed toward resolving this link. In addition, as discussed earlier, there needs to be clarification on the interplay between the regulation of mTOR activity by growth factors and by nutrients. Another important and unresolved question concerns the identity of the downstream target(s) of S6K, which activates the translation of TOP mRNAs to stimulate cell growth. As described above, eIF4B could be a candidate (Raught et al. 2004), although other as yet undiscovered proteins could also play a role (e.g., see Fingar et al. 2004).

An important avenue for future studies is the understanding of the cross-talk between the PI3K-Akt/PKB-mTOR signaling pathway and the signaling pathway leading to the activation of ERK. It is clear that both pathways cooperate to effect many cellular functions. These interactions have critical consequences for the control of cell growth, memory, and learning. These two signaling pathways activate key components of the translational machinery involved in recruiting ribosomes to mRNA. The ERK pathway is responsible for phosphorylating eIF4E (Waskiewicz et al. 1997 Pyronnet et al. 1999 Radimerski et al. 2002), a modification that is thought to increase its activity whereas, as described above, the mTOR pathway phosphorylates 4E-BPs, which, in turn, stimulate eIF4E activity and enhance ribosome recruitment. Recent experiments show that the ERK and mTOR pathways cooperate to stimulate translation and induce glioblastomas in a mouse model (Rajasekhar et al. 2003). As both pathways become activated in neurons in response to experience, they likely cooperate to promote new protein synthesis required for learning and memory.

Watch the video: Long Term Potentiation and Memory Formation, Animation (September 2022).


  1. Nikolrajas

    I apologise, but, in my opinion, you are not right. I am assured. Let's discuss. Write to me in PM.

  2. Jozy

    I can advise you on this matter. Together we can come to the right answer.

  3. Alphonso

    I apologize for interrupting you, but I need a little more information.

  4. Matthan

    m ... yes dirt, violence, cruelty.

  5. Arne

    Also that without your we would do very good idea

Write a message