Initiation

Initiation in mammalian cells starts with the activation of a serine/threonine kinase complex that contains ULK1/2 (orthologues of the yeast protein Atg1), ATG13 and FIP200 (RB1CC1) This complex transfers signals from the nutrient-sensing mTOR kinase (MTOR; mammalian/mechanistic target of rapamycin) to initiate autophagy (Refs Reference Diaz-Troya12, Reference Hara13, Reference Jung14, Reference Ganley15). Inhibition of mTOR-induced phosphorylation of ULK and ATG13 liberates the kinase activity of ULK, which subsequently phosphorylates itself, ATG13 and FIP200. The ULK complex then accumulates at the initiating focus of vesicle formation: the isolation membrane or phagophore (Refs Reference Hara13, Reference Jung14, Reference Ganley15).

Autophagosome formation

Vesicle nucleation – the further development of the isolation membrane/phagophore – critically depends on the activity of the class III phosphoinositide 3-kinase (PI3K-III) hVps34 (PIK3C3; the orthologue of yeast Vps34) and its formation of a complex with Beclin 1 (yeast Atg6) and p150/hVps35 (PIK3R4; yeast Vps15) (Refs Reference Liang16, Reference Liang17).

Vesicle elongation and completion from the isolation membrane/phagophore to a nascent autophagosome and the completed, closed autophagosome is mediated by two ubiquitin-like conjugation systems: the ATG12 and ATG8 conjugation systems (Ref. Reference Ohsumi and Mizushima18). The ubiquitin-like ATG12 is activated by ATG7 and then temporarily binds to the E2-like enzyme ATG10 before being transferred to ATG5. ATG5 further reacts with ATG16 to form a multimeric complex of ATG12–ATG5–ATG16 (Refs Reference Geng and Klionsky19, Reference Kabeya20). In mammalian cells, several orthologues of yeast Atg8 have been identified: MAP1LC3 (LC3), GABARAPL2 (GATE16), GABARAP and GABARAPL1 (ATG8L). They are subjected to modification steps similar to those of their yeast counterpart. The processing of the autophagosome marker LC3 has been investigated most thoroughly, and its stepwise conversion from a cytosolic to a membrane-bound form is exploited for the experimental measurement of autophagy (Ref. Reference Klionsky21). LC3 is synthesised as a precursor protein proLC3 and is immediately processed to LC3-I by ATG4 through cleavage of the C-terminal amino acid. LC3 maturation completes with the reversible conjugation of LC3-I to phosphatidylethanolamine (PE) at the C-terminus by ATG3 and ATG7 to form LC3-II on the surface of autophagosomes (Refs Reference Geng and Klionsky19, Reference Kabeya20).

Both the ATG12–ATG5–ATG16 complex and LC3-II are essential for autophagy. The ATG12–ATG5–ATG16 complex is required for targeting of LC3 to the autophagosomal membrane and accelerated conjugation of LC3 to PE. Atg8, the yeast counterpart of LC3, controls the expansion of the phagophore and the amount of Atg8 directly correlates to the size of the autophagosomes (Refs Reference Geng and Klionsky19, Reference Xie, Nair and Klionsky22). The different stages of autophagosome formation are intertwined and difficult to separate. Therefore the assignment of the different ATGs to corresponding stages and autophagic vesicles is to some degree academic.

Maturation and degradation

The maturation process encompasses the fusion of autophagosomes with lysosomes to form the end-stage vesicle of autophagy: the autolysosome. Molecular mechanisms of autophagosome maturation are only recently emerging and involve the actions of lysosomal proteins LAMP1 and LAMP2, the small GTPase Rab7 (RAB7A), UVRAG (the protein product of the ultraviolet-radiation-resistance-associated gene) and others (Ref. Reference Eskelinen9). The tumour suppressor UVRAG not only regulates the interaction of Beclin 1 and hVps34 at the stage of vesicle nucleation but also plays an important role in the maturation step. UVRAG directs so-called tethering proteins (i.e. proteins that connect the autophagosome to its target) to the autophagosomal membrane and thereby activates Rab7 to facilitate fusion with lysosomes (Ref. Reference Liang17).

The final autolysosome is an acidic vesicle wherein the intracellular material gets degraded by lysosomal hydrolases, especially cathepsins. Amino acids and other constituent parts generated in this catabolic process are then released from the autolysosome to fuel cellular resources. In yeast, amino acid efflux is mediated in part via Atg22 (Ref. Reference Yang23). Although mammalian lysosomal amino acid transporters have been described, a clear orthologue of Atg22 in these higher species is yet to be defined.

Beclin 1

Beclin 1 (BECN1) is a critical autophagy-regulating gene and provided the first human genetic link between autophagy and cancer. It is monoallelically deleted in ~50% of human breast, ovarian and prostate cancers and expressed only at low levels in brain tumours (Refs Reference Aita31, Reference Liang32, Reference Miracco33). Subsequent studies of genetically modified mice revealed that while bi-allelic loss of Becn1 is embryonically lethal, mice that retain one copy of the gene are viable, but have a higher incidence of lymphomas, lung and liver carcinomas, as well as mammary hyperplasia and accelerated formation of premalignant lesions induced by hepatitis B virus, compared with wild-type animals (Refs Reference Qu34, Reference Yue35). In addition, reintroduction of BECN1 into human MCF7 breast carcinoma cells, which express only trace amounts of Beclin 1, restores their autophagic capacity and inhibits their tumour-forming potential (Ref. Reference Liang32). Beclin 1 is therefore a haploinsufficient tumour suppressor (Refs Reference Qu34, Reference Yue35).

ATG5 and ATG7

ATG5 and ATG7 are two other critical autophagy regulators that have been deleted in experimental animal models. Unlike complete deletion of Becn1, deficiency in ATG5 or ATG7 is not embryonically lethal and mice are born normal without any apparent abnormalities; however, both Atg5 ^−/−and Atg7 ^−/− mice die within 24 h after birth, presumably as a result of an inability to compensate the neonatal starvation period that occurs after being cut off from the maternal circulation, when autophagy is normally transiently upregulated (Refs Reference Kuma3, Reference Komatsu36). Mice hemizygous for either Atg5 or Atg7 are born viable and develop normally. In contrast to Becn1-hemizygous animals, however, they do not develop tumours (Refs Reference Kuma3, Reference Komatsu36). Intercrossing tumour-prone mice to recently developed conditional knockout mice for Atg5 and Atg7 (Refs Reference Komatsu36, Reference Hara37) will seemingly be required for the spatial and temporal dissection of the contribution of these genes to cancer development.

ATG4

Yeast Atg4 has four mammalian orthologues, of which ATG4C (autophagin-3) is most widely expressed. ATG4C-knockout animals are born viable, survive the neonatal starvation period and develop normally (Ref. Reference Mariño38). Upon nutrient restriction they show organ-specific reduction of autophagy in the diaphragm. Autophagic activity at basal levels is normal. Spontaneous tumours arise at the same frequency as in wild-type controls; however, they do have a higher susceptibility to develop fibrosarcomas after chemical carcinogenesis (Ref. Reference Mariño38). A functional redundancy of the different ATG4 proteins in mammals might explain the notion that ATG4C deficiency does not alter basal autophagic activity in vivo and the fact that ATG4C-deficient mice survive the neonatal starvation period (Ref. Reference Mariño38).

UVRAG

UVRAG is a tumour suppressor gene that is monoallelically deleted in a significant proportion of human colorectal carcinomas, and UVRAG expression in human HCT116 colon carcinoma cells reduces the tumourigenic potential of these cells in xenograft studies (Ref. Reference Liang16). Furthermore, UVRAG is a target for frameshift mutations in colorectal and gastric carcinomas with microsatellite instability (Ref. Reference Kim39).

A role for autophagy deficiency in intestinal tumourigenesis is further supported by the fact that frameshift mutations in other autophagy-related genes – ATG2B, ATG5, ATG9B and ATG12 – have been found in up to 28% of human gastric and colorectal cancers with microsatellite instability (Ref. Reference Kang40). However the significance of this finding is yet to be determined.

BIF1

BIF1 (SH3GLB1) is another protein that positively regulates autophagy, through interaction with UVRAG and Beclin 1 (Ref. Reference Takahashi41). B1F1-knockout mice are born viable and develop normally, without any apparent abnormalities except for an enlarged spleen. However they do have a higher propensity to develop spontaneous cancer (mainly lymphomas and to a lesser extent solid tumours) at an average age of 12 months. BIF1 is thus a positive regulator of autophagy and a potential tumour suppressor (Ref. Reference Takahashi41).

Summary

In summary, the available data from humans and mouse models where ATGs or their direct positive modulators are lost or expressed at low levels point towards a tumour-suppressive role of autophagy. It is important to keep in mind, however, that in all the above mouse models autophagy inhibition involved deletion of the gene in the target cells embryonically. As a result, all the pathologies observed arose from cells that had always had impaired autophagy. These experiments thus do not clarify the role of acute autophagy loss at a later developmental stage, for instance in an established tumour, a scenario that likely occurs in vivo. In order to address this important question, further analysis is therefore required of conditional autophagy-deficient animals, by using inducible Cre recombinases so that gene deletion can be achieved in the adult animal in a tissue- and tumour-specific manner.

mTOR is a gateway for the control of autophagy

The serine/threonine kinase mTOR is a focal point for the action of many oncogenic and metabolic events as well as the control of autophagy. Many regulators of mTOR signalling, especially the PI3K–AKT–PTEN pathway, are frequently deregulated in human tumours (Ref. Reference Vogt, Gymnopoulos and Hart44). mTOR is part of two different protein complexes: mTORC1 and mTORC2 (Ref. Reference Martin and Hall45). These have different components, most notably raptor (RAPTOR; regulatory-associated protein of MTOR, complex 1) for mTORC1 and rictor (RICTOR; rapamycin-insensitive companion of MTOR, complex 2) for mTORC2 (Refs Reference Kim46, Reference Sarbassov47). Activation of mTORC1 leads to protein synthesis via the actions of S6K1 (RPS6KB1, ribosomal protein S6 kinase, 70 kDa, polypeptide 1) and 4EBP1 (EIF4EBP1, eukaryotic translation initiation factor 4E binding protein 1). By contrast, mTORC2 is less well understood. It activates AKT via phosphorylation and enhances the expression of HIF-1α (HIF1A; hypoxia-inducible factor 1α) (Refs Reference Sarbassov48, Reference Toschi49). While driving translation, mTOR also inhibits autophagy and this effect is mediated via mTORC1 and its interaction with the ULK–ATG13–FIP200 complex. The net effect of mTOR activation is cell growth, proliferation and cell survival (Refs Reference Martin and Hall45, Reference Guertin and Sabatini50).

The Beclin-1–hVps34 complex is a critical target for the regulation of autophagy

mTOR activity is pivotal in determining whether the net effect from signals within the cell is pro- or anti-autophagic. Another important regulator, downstream of mTOR, is the Beclin-1–hVps34 complex (Ref. Reference Pattingre43). Beclin 1 was initially identified as a Bcl-2 (BCL2)-binding protein and was found to have tumour suppressor functions (Ref. Reference Liang32). It has emerged, however, that Beclin 1 function is multifaceted and not restricted to the regulation of autophagy. It is part of a multiprotein complex that regulates autophagy through variation of its composition (Fig. 3). Beclin 1 forms a heterodimer that is stabilised via binding to antiapoptotic members of the Bcl-2 family [Bcl-2, Bcl-X_L (BCL2L1), Mcl-1 (MCL1)] through its Bcl-2-homology domain 3 (BH3) domain (Refs Reference Levine, Sinha and Kroemer51, Reference Maiuri52). Monomerisation of Beclin 1 is required for it to associate with and activate the PI3K-III hVps34, which is critical for the formation of autophagosomes. Therefore, tumour-associated perturbations of antiapoptotic Bcl-2 family members can cause profound inhibition of autophagy by preventing the formation of a Beclin-1–hVps34 complex (Ref. Reference Pattingre53).

Figure 3. Regulation of autophagy by Beclin 1 complexes. Beclin 1 exerts its function on autophagy only within a multimeric protein complex. It has diverse protein-binding domains that allow the formation of different complexes with opposing functions on autophagy. Certain stimuli, for example starvation, disrupt the binding of Bcl-2 proteins from the BH3 domain of Beclin 1. Beclin 1 can then bind and activate the phosphoinositide 3-kinase hVps34. Currently four regulators of the Beclin-1–hVps34 complex have been identified that determine its pro- or anti-autophagic activity. Autophagy-inducing regulators/complexes are depicted in green, inhibiting regulators/complexes in red, and those that have an undetermined role in yellow. Abbreviations: AMBRA1, activating molecule in Beclin-1-regulated autophagy protein 1; ATG, autophagy-related protein; BARKOR, Beclin-1-associated autophagy-related key regulator; Bcl-2, B-cell CLL/lymphoma 2; Bcl-X_L, B-cell lymphoma-extra large; Beclin 1, coiled-coil myosin-like Bcl-2-interacting protein; BH3, Bcl-2 homology domain 3; ECD, evolutionarily conserved domain; Mcl-1, induced myeloid leukaemia cell differentiation protein; Rubicon, RUN domain and cysteine-rich domain containing, Beclin-1-interacting protein; UVRAG, ultraviolet-radiation-resistance-associated gene.

The Beclin-1–hVps34–p150 core complex is complemented by different regulatory proteins that bind to Beclin 1 through domains other than BH3 and define its role in autophagy. To date, three different forms of the complex have been identified, all of which have different functions. They contain either ATG14/BARKOR (Beclin-1-associated autophagy-related key regulator) or UVRAG alone, or both UVRAG and Rubicon (RUN domain and cysteine-rich domain containing). It has been proposed that the ATG14/BARKOR-containing complex targets hVps34 activity towards the isolation membrane and thus functions in autophagosome formation, and that the UVRAG-containing complex promotes autophagosome and endosome maturation (Ref. Reference Itakura54). ATG14/BARKOR positively regulates autophagy, promotes translocation of Beclin 1 to autophagosomes and competes with UVRAG for the same binding site on Beclin 1 (Ref. Reference Sun55). By contrast, the UVRAG- and Rubicon-containing complex suppresses autophagosome and endosome maturation (Refs Reference Zhong56, Reference Matsunaga57). Knockdown of ATG14/BARKOR inhibits autophagy whereas knockdown of Rubicon has the opposite effect, providing further evidence that Beclin 1 has multiple roles in autophagy through the formation of complexes with opposing actions (Fig. 3). Although perturbation of Beclin 1 clearly has implications for cancer, a role for ATG14/BARKOR in cancer has not yet been defined. AMBRA1 is another protein that binds the Beclin-1–hVps34 complex, with a role in the regulation of autophagy in neuronal development (Ref. Reference Fimia58). However, as with ATG14/BARKOR, its role in autophagy regulation has also not yet been extended to cancer.

PI3K-I

PI3K-I (class I phosphoinositide 3-kinase) is a lipid kinase activated via receptor tyrosine kinases (RTKs) by multiple mechanisms (Refs Reference Vogt, Gymnopoulos and Hart44, Reference Yuan and Cantley59). This leads to cell growth and cell proliferation and the concomitant inhibition of autophagy by activation of mTOR. Since RTKs and their downstream effectors are frequently mutated in cancer, activation in the PI3K-I pathway is very common in sporadic tumours (Ref. Reference Yuan and Cantley59). Activated PI3K-I generates the second messenger phosphatidylinositol 3,4,5-trisphosphate [PtdIns(Reference Kuma3,Reference Mizushima4,Reference Maiuri5)P ₃], which then leads to activation/phosphorylation of AKT via 3-phosphoinositide-dependent protein kinase 1 (PDPK1/PDK1). AKT phosphorylates and inhibits tuberous sclerosis protein 2 (TSC2), which as a result permits RheB (RHEB) to activate mTORC1 (Ref. Reference Martin and Hall45). AKT can also activate mTOR independently of TSC2, via PRAS40 (AKT1S1; proline-rich AKT-substrate 40 kDa). In summary, the net effect of PI3K-I activation is activation of mTOR with subsequent inhibition of autophagy (Refs Reference Martin and Hall45, Reference Guertin and Sabatini50).

PTEN

PTEN (phosphatase with tensin homologue) is mutated in multiple cancer types and is the major target for mutation in Cowden disease, a cancer predisposition syndrome (Ref. Reference Krymskaya and Goncharova60). PTEN negatively regulates PI3K-I and thereby alleviates the inhibitory effects of mTOR on autophagy. Consequently, inactivation of PTEN suppresses autophagy. Loss of PTEN is the most common mechanism of PI3K-I activation and is the second most frequently mutated tumour suppressor in human cancer (Ref. Reference Shaw and Cantley61).

Ras–MEK–ERK

The Ras–MEK–ERK pathway is deregulated in many cancers and has a central role in mTOR activation independent of AKT. The oncogenic small GTPase Ras has an ambiguous role in autophagy. Oncogenic HRAS-V12 has been reported to suppress protein degradation via activation of PI3K-I in NIH3T3 cells (Ref. Reference Furuta62), whereas in nontransformed IMR90 human diploid fibroblasts, HRAS-V12 induces senescence with concomitant increased autophagy (Ref. Reference Young7). The role of ERK (MAPK1, mitogen-activated protein kinase 1) in autophagy is also multifactorial. ERK has been shown in different in vitro settings to induce autophagy (Refs Reference Ogier-Denis63, Reference Pattingre, Bauvy and Codogno64, Reference Corcelle65). Recent findings indicate that this may occur through upregulation of Beclin 1 and destabilisation of mTOR via induction of TSC. Interestingly in this context, basal activity of MEK (MAP2K1)/ERK conferred basal levels of mTOR and Beclin 1 that were incapable of triggering autophagy (Ref. Reference Wang66). By contrast to these reports, however, it has also been shown that ERK can inhibit TSC2 and promotes tumourigenesis in vitro and in vivo (Ref. Reference Shaw and Cantley61).

STK11

STK11 (serine/threonine kinase 11; also known as LKB1) is a tumour suppressor that is mutated in patients with hereditary intestinal polyposis (Peutz–Jeghers syndrome), which leads to a cumulative lifetime risk of cancer at age 70 of ~90% (Ref. Reference Rustgi67). In addition, mutations in STK11 are found in a high percentage of sporadically occurring lung malignancies. STK11 is a positive regulator of autophagy through the actions of AMPK (AMP-activated protein kinase) and mTOR. Treatment of a mouse model for Peutz–Jeghers syndrome (heterozygous for Stk11) with the mTOR inhibitor rapamycin significantly reduced the tumour burden (Refs Reference Liang68, Reference Wei69).

DAPK1

DAPK1 (death-associated protein kinase 1) has tumour-suppressive functions and is epigenetically silenced in a variety of human tumours (Ref. Reference Gozuacik and Kimchi70). DAPK was believed to be a positive regulator of autophagy through interaction with LC3-interacting microtubule-associated protein MAP1B and disruption of the Beclin-1–Bcl-X_L complex (Ref. Reference Zalckvar71). Recent data, however, showed that DAPK is a positive regulator of mTORC1 through interaction with TSC2 (Ref. Reference Stevens72). How this finding relates to the net effect of DAPK on autophagy has not been examined.

ARHI

The tumour suppressor ARHI (aplasia Ras homologue member I; also known as DIRAS3 for DIRAS family, GTP-binding RAS-like 3) is downregulated in more than 60% of ovarian cancer (Ref. Reference Yu73). It induces autophagy by upregulation of ATG4 and blocking PI3K-I and therefore mTOR (Ref. Reference Lu74). In addition, its expression leads to tumour dormancy in ovarian cancer xenograft models, which can be reversed upon treatment with chloroquine, a pharmacological inhibitor of lysosomal function and, as a result, autophagy (Ref. Reference Lu74).

p53

p53 (TP53) is a major tumour suppressor, being mutated or lost in ~50% of all human cancers (Ref. Reference Hainaut and Hollstein75). In response to various forms of cellular stress, p53 becomes activated, leading to multiple phenotypic effects including induction of cell-cycle arrest and cell death (Ref. Reference Crighton and Ryan76). Since p53 has been shown to respond to, and also to modulate, metabolic stress, it is perhaps no surprise that p53 has also been found to regulate autophagy (Refs Reference Crighton77, Reference Feng78, Reference Tasdemir79). The way in which p53 regulates autophagy, however, appears to be dependent on context. When elevated by cellular stress, p53 accumulates in the nucleus and activates target genes that mediate its tumour-suppressive effects. One of these genes, DRAM1 (damage-regulated-autophagy modulator), encodes a phylogenetically conserved lysosomal protein that promotes autophagy (Refs Reference Crighton77, Reference Crighton, Wilkinson and Ryan80, Reference O'Prey81). Interestingly, DRAM1 was also found to be required for p53-induced programmed cell death, although the mechanism underlying the interplay between DRAM1, autophagy and apoptosis is yet to be determined (Ref. Reference Crighton77). p73 (TP73), a closely related protein to p53, has also been shown to modulate DRAM1 and autophagy, although autophagy induced by p73 was found to be independent of DRAM1 (Ref. Reference Crighton82). Since this points to other target genes of p53/p73 being involved in autophagy regulation, it is interesting to note that sestrin 2 (SESN2) has also been linked to both p53 and p73 (Refs Reference Budanov83, Reference Rosenbluth84). In an independent study, sestrins have been shown to modulate mTOR through AMPK (Ref. Reference Budanov and Karin85). A link has now also been reported between p53, sestrin 2 and autophagy (Ref. Reference Maiuri86), although the role of sestrin 2 in p73-driven autophagy is yet to be explored.

In the absence of cellular stress, basal levels of p53 have, in contrast, been shown to inhibit autophagy. This effect, however, does not involve gene activation and occurs through a cytoplasmic mechanism at the endoplasmic reticulum (Ref. Reference Tasdemir79). Pharmacological inhibition or genetic ablation of p53, both in vitro and in vivo, was shown to induce autophagy. Although the induction and inhibition of autophagy by p53 are mechanistically distinct and therefore seemingly separable, it would be interesting to study the balance of these two effects when both are active – are they opposing or mutually exclusive? In this regard, it is interesting to note that certain tumour-derived mutants of p53 can inhibit autophagy through the cytoplasmic mechanism, but they do not cause the activation of DRAM1 or sestrin 2 that would be required for the induction of autophagy (Ref. Reference Morselli87). In tumours with mutant p53, therefore, it would be conceivable that autophagy would be reduced through both lack of induction as well as enforced inhibition.

ARF

ARF [‘alternative reading frame’ of the CDKN2A (INK4A) locus] is a bona fide tumour suppressor that is known to induce p53 levels following oncogenic stress and as a result has been shown to induce p53-dependent autophagy and thereby to enhance cell death. ARF-induced autophagy might therefore contribute to its tumour-suppressive function (Ref. Reference Abida and Gu88). ARF can also induce autophagy in the absence of p53, probably through interaction with Bcl-X_L which normally inhibits the Beclin-1–hVps34 complex (Ref. Reference Pimkina89). A smaller truncated form of ARF, smARF (resulting from initiation from an alternative internal methionine), has also recently been found to modulate autophagy in a p53-independent manner. In this context, smARF is considered to induce autophagy at mitochondria and causes induction of a cell death dependent on autophagy genes (Ref. Reference Reef90).

Others

In addition to the tumour suppressors and oncogenes discussed above, which have been shown to have an active role in autophagy regulation, several other tumour-associated factors could be postulated to modulate autophagy given their connection to pathways that have been previously implicated in autophagy regulation. As an element of speculation is involved, an exhaustive list of such factors could be generated; we feel, however, that the factors listed below are worthy of discussion even if further studies are required to clarify a connection to autophagy regulation and the extent to which this may be important for the role of the factor in tumour development.

Summary

In summary, the mTOR and Beclin-1–hVps34 complexes are targeted by a multitude of different oncogenes and tumour suppressors, which thereby regulate autophagy. New players are constantly being identified and new links are being made, making the regulation of autophagy in cancer incredibly complex.

Autophagy and cell death

The role and mechanisms of autophagy in promoting cell survival in contexts such as nutrient deprivation are unequivocal. It is also without question that autophagy accompanies cell death in certain scenarios, although the exact contribution of autophagy is complex and less well defined. In systems where apoptosis was inhibited, either genetically or pharmacologically, cell death dependent on ATGs was reported (Refs Reference Shimizu96, Reference Yu97); in other words, inhibition of certain ATGs (ATG5, Beclin 1, ATG7) suppressed cell death. In one report, where cells were treated with the caspase inhibitor zVAD-fmk to inhibit caspase-dependent death, it was proposed that the cell death was mechanistically driven via the selective autophagic degradation of catalase, which subsequently overloads the cell with damaging ROS (Ref. Reference Yu98). A more direct role of ATGs in cell death has also been shown: cleavage of ATG5 has been reported to occur following cell death induction and this cleaved form has a direct proapoptotic role (Ref. Reference Yousefi99). This could indicate, therefore, that in studies where cell death has been compromised by ATG5 knockdown, incorrect conclusions may have been drawn that the cell death observed was dependent on autophagy. The generality of these observations, however, is yet to be determined.

Clearance of dying cells during salivary gland development in Drosophila melanogaster involves autophagy and in this context it appears that the full-blown cell death response is dependent on caspases and ATGs (Ref. Reference Berry and Baehrecke100). Increasing evidence also suggests a non-cell-autonomous role of autophagy in cell death, which extends the housekeeping functions of autophagy to a multicellular level. In an in vitro mouse model of embryonic development, inhibition of autophagy prevented the upregulation of two important cell-surface engulfment signals – phosphatidylserine and lysophosphatidylcholine – in cells destined to die (Ref. Reference Qu101). At first impression, inhibition of autophagy in this system seems to result in cell death, but it is really the energy-dependent expression of engulfment signals that is impeded, which causes the accumulation of cellular corpses by default, that is at play. Exogenous methylpyruvate is an alternative source of energy and its addition compensates for the low levels of ATP in autophagy-deficient cells, leading to expression of the aforementioned engulfment signals. In an vivo context, it is notable that embryonic Atg5-null mice display reduced phagocytosis of dead cells in lung and retinal tissue, pointing to the wider implications of this phenomenon and to the caution that should be employed when dying cells seemingly arise following autophagy inhibition (Ref. Reference Qu101).

In all the above-mentioned examples, ATGs facilitate cell death, but it is arguable that a situation has not yet been observed where a cell actually dies solely from autophagy. Furthermore, as discussed above, the role of autophagy in cell death is not restricted to cell-autonomous effects. Therefore, the term ‘autophagic cell death’ (or programmed cell death type II) has to be used with caution. It has been suggested it be replaced with the term ‘cell death with autophagy’ as it is now widely considered that autophagy can be a component in cell death, perhaps as a permissive signal, but not its sole perpetrator (Ref. Reference Kroemer and Levine102).

Autophagy in cell cycle arrest, dormancy and senescence

Sustained growth arrest, like programmed cell death, is an important mechanism to counteract tumour growth in vivo and has recently been shown to be critically linked with autophagy (Ref. Reference Berry and Baehrecke100). Senescence is a critical barrier against malignant transformation and an important effector programme of chemotherapy. It is defined as an irreversible cell cycle arrest of viable and metabolically active cells following various endogenous and exogenous stresses (Ref. Reference Prieur and Peeper103). In HRAS-V12-infected human diploid fibroblasts, autophagy is activated and its inhibition resulted in delayed onset of the senescent phenotype (Ref. Reference Young7). In a reciprocal fashion, overexpression of the autophagic protein ULK3 induced autophagy and accelerated the onset of senescence (Ref. Reference Young7). However, the exact role of ULK3 in autophagy is currently unclear.

Tumour dormancy generally describes a phase of apparent stasis in tumour growth and is mechanistically subdivided into two models. Single-cell dormancy is a phase of prolonged cell cycle arrest that can be overcome upon cell-autonomous genetic changes or alterations in the microenvironment. According to the micrometastasis model, tumour dormancy is a balanced state of cell renewal and cell death (Ref. Reference Fehm104). (Both dormancy and senescence are not to be confused with quiescence, which is a physiological normal resting state of a cell.) ARHI expression induces autophagy and leads to tumour dormancy in ovarian cancer xenograft models, which can be reversed upon treatment with chloroquine, a pharmacological inhibitor of lysosomal function and therefore autophagy (Ref. Reference Lu74).

When taken together, these reports clearly emphasise that growth arrest and autophagy are intertwined, and tentatively posit that autophagy and its role in senescence and dormancy may be a facet of tumour suppression in multiple cancers.

Autophagy and the management of oxidative stress

Metabolic stress, as a result of either insufficient nutrient/oxygen supply or increased energetic demands of rapidly dividing tumour cells, induces autophagy as an alternative source of energy and metabolites. Enhanced autophagy is frequently found in hypoxic regions of tumours and contributes to cell survival (Ref. Reference Degenhardt105). Initially, it is therefore illogical how loss of autophagy contributes to tumour formation (Refs Reference Liang32, Reference Qu34, Reference Yue35). The first contribution to solve this conundrum came from the observation that metabolic stress promotes necrotic cell death in vivo and in vitro within cells that are defective in both autophagy and apoptosis. Necrotic cell death was associated with a strong inflammatory response, which ultimately enhanced tumour growth (Ref. Reference Degenhardt105). This therefore provides one explanation of how autophagy may be tumour suppressive – through its capacity to impede necrosis. Whether this is a factor, however, downstream of chemotherapeutic drugs that induce cell death through necrosis is an interesting question and worthy of further investigation (Refs Reference Amaravadi and Thompson106, Reference Zong107). Meanwhile it is becoming increasingly substantiated that autophagy reduces oxidative stress, maintains protein and organelle quality control and thereby limits cellular damage (Refs Reference Karantza-Wadsworth27, Reference Mathew28).

Metabolic stress leads to accumulation of ROS, damaged proteins and damaged organelles such as mitochondria (which are in turn additional sources of ROS) and impedes genetic stability when compensatory mechanisms fail. Autophagy is a critical countermeasure following metabolic insults and attenuates these changes, thereby promoting cellular survival. In different studies, autophagy-defective tumour cells were indeed more susceptible to cell death but, importantly, at the same time were also more susceptible to genomic damage and possessed greatly enhanced tumourigenic potential when compared with autophagy-competent cells in vivo (Refs Reference Karantza-Wadsworth27, Reference Mathew28).

Sequestosome 1 (SQSTM1; p62) has been identified as the critical molecular link between defective autophagy, genomic instability and tumourigenesis (Ref. Reference Mathew26). p62 is an adapter protein that links polyubiquitinated proteins or aggregates to ATG8/LC3 on the surface of autophagosomes. Importantly, it is also a decisive element in modulating different molecular pathways, such as the nuclear factor NF-κB pathway through recruitment of different signalling molecules and the extrinsic cell death pathway by promoting aggregation of caspase 8 (Ref. Reference Jin108). p62 is frequently upregulated in human tumours and of utmost importance in controlling oxidative stress and tumour growth (Refs Reference Mathew26, Reference Moscat and Diaz-Meco109). Tumour cells, therefore, with impairment in both apoptosis and autophagy preferably accumulate p62 under stress, which thereby promotes tumourigenesis (Ref. Reference Mathew26). In a self-supporting feedback loop, metabolic stress leads to increased production of ROS that in turn were responsible for accumulation of p62. p62 itself was also causative for further oxidative damage, accumulation of damaged mitochondria, an enhanced induction of the protein-folding machinery in the endoplasmatic reticulum, and a DNA-damage response. Elevated levels of p62 suppressed the activation of target genes of NF-κB. In Ras-induced lung tumours, p62 is essential for tumour formation, but in contrast to the observation by Mathew et al. (Ref. Reference Mathew26), inhibition of p62 led to impaired NF-κB signalling (Ref. Reference Duran110). These differences in NF-κB activation are not readily explained and may be due to the versatility of p62 in activating different pathways (Ref. Reference Moscat and Diaz-Meco109) or the different cellular contexts. Nonetheless, p62 accumulation is clearly causative for tumourigenesis and exemplifies how cancer progression can be affected by the selective autophagic degradation of a specific protein. Figure 4 illustrates how the metabolic capacity of autophagy might be the decisive factor for proliferation control in tumourigenesis.

Figure 4. Autophagy manages cellular stress to counteract tumour growth. In autophagy-competent cells, cellular stress leads either to cell death, growth arrest in the form of premature senescence, or survival. If the autophagic capacity falls below a threshold that is necessary to maintain cell integrity, for example through genetic alterations or pharmaceutical intervention, cells are unable to compensate metabolic stress. As a result, necrotic cell death occurs and causes attraction of tumour-promoting macrophages. In surviving cells, however, failure to clear p62 (sequestosome 1) and p62-associated aggregates results in further accumulation of ROS (reactive oxygen species), damaged proteins and organelles, altered cell signalling and DNA damage. Clinically, this phase might present as an initial tumour remission. However, the acquisition of growth-promoting mutations in a subset of cells might potentially lead to more aggressive tumour cells. Abbreviation: NF-κB, nuclear factor κB.