Autophagy in malignant transformation and cancer progression

Macroautophagy (herein referred to as autophagy) is a mechanism that mediates the sequestration of intracellular entities within double‐membraned vesicles, so‐called autophagosomes, and their delivery to lysosomes for bulk degradation (He & Klionsky, 2009). Autophagosomes derive from so‐called phagophores, membranous structures also known as ‘isolation membranes’ whose precise origin remains a matter of debate (Lamb et al, 2013). Indeed, the plasma membrane, endoplasmic reticulum (ER), Golgi apparatus, ER‐Golgi intermediate compartment (ERGIC), and mitochondria have all been indicated as possible sources for phagophores (Lamb et al, 2013). Upon closure, autophagosomes fuse with lysosomes, forming so‐called autolysosomes, and their cargo is exposed to the catalytic activity of lysosomal hydrolases (Mizushima & Komatsu, 2011). The degradation products of the autophagosomal cargo, which includes sugars, nucleosides/nucleotides, amino acids and fatty acids, can be transported back to the cytoplasm and presumably re‐enter cellular metabolism (Fig 1) (Rabinowitz & White, 2010; Galluzzi et al, 2013). Of note, the molecular machinery that mediates autophagy is evolutionary conserved, and several components thereof have initially been characterized in yeast (He & Klionsky, 2009).

In physiological scenarios, autophagy proceeds at basal levels, ensuring the continuous removal of superfluous, ectopic or damaged (and hence potentially dangerous) entities, including organelles and/or portions thereof (Green et al, 2011). Baseline autophagy mediates a key homeostatic function, constantly operating as an intracellular quality control system (Mizushima et al, 2008; Green et al, 2011). Moreover, the autophagic flux can be upregulated in response to a wide panel of stimuli, including (but not limited to) nutritional, metabolic, oxidative, pathogenic, genotoxic and proteotoxic cues (Kroemer et al, 2010). Often, stimulus‐induced autophagy underlies and sustains an adaptive response to stress with cytoprotective functions (Kroemer et al, 2010; Mizushima & Komatsu, 2011). Indeed, the pharmacological or genetic inhibition of autophagy generally limits the ability of cells to cope with stress and restore homeostasis (Mizushima et al, 2008; Kroemer et al, 2010). This said, regulated instances of cell death that causally depend on the autophagic machinery have been described (Denton et al, 2009; Denton et al, 2012b; Liu et al, 2013b; Galluzzi et al, 2015). The detailed discussion of such forms of autophagic cell death, however, is beyond the scope of this review.

Autophagy is tightly regulated. The best characterized repressor of autophagic responses is mechanistic target of rapamycin (MTOR) complex I (MTORCI) (Laplante & Sabatini, 2012). Thus, several inducers of autophagy operate by triggering signal transduction cascades that result in the inhibition of MTORCI (Inoki et al, 2012). Among other effects, this allows for the activation of several proteins that are crucial for the initiation of autophagic responses, such as unc‐51‐like autophagy‐activating kinase 1 (ULK1, the mammalian ortholog of yeast Atg1) and autophagy‐related 13 (ATG13) (Hosokawa et al, 2009; Nazio et al, 2013). A major inhibitor of MTORCI is protein kinase, AMP‐activated (PRKA, best known as AMPK), which is sensitive to declining ATP/AMP ratios (Mihaylova & Shaw, 2011). Besides inhibiting the catalytic activity of MTORCI, AMPK directly stimulates autophagy by phosphorylating ULK1 as well as phosphatidylinositol 3‐kinase, catalytic subunit type 3 (PIK3C3, best known as VPS34) and Beclin 1 (BECN1, the mammalian ortholog of yeast Atg6), two components of a multiprotein complex that produces a lipid that is essential for the biogenesis of autophagosomes, namely phosphatidylinositol 3‐phosphate (Egan et al, 2011; Zhao & Klionsky, 2011; Kim et al, 2013). Autophagy also critically relies on two ubiquitin‐like conjugation systems, both of which involve ATG7 (Mizushima, 2007). These systems catalyze the covalent linkage of ATG5 to ATG12 and ATG16‐like 1 (ATG16L1), and that of phosphatidylethanolamine to proteins of the microtubule‐associated protein 1 light chain 3 (MAP1LC3, best known as LC3) family, including MAP1LC3B (LC3B, the mammalian ortholog of yeast Atg8) (Mizushima, 2007). A detailed discussion of additional factors that are involved in the control and execution of autophagic responses can be found in Boya et al (2013).

Importantly, autophagosomes can either take up intracellular material in a relatively non‐selective manner or deliver very specific portions of the cytoplasm to degradation, mainly depending on the initiating stimulus (Weidberg et al, 2011; Stolz et al, 2014). Thus, while non‐selective forms of autophagy normally develop in response to cell‐wide alterations, most often of a metabolic nature, highly targeted autophagic responses follow specific perturbations of intracellular homeostasis, such as the accumulation of permeabilized mitochondria (mitophagy), the formation of protein aggregates (aggrephagy), and pathogen invasion (xenophagy) (Okamoto, 2014; Randow & Youle, 2014). Several receptors participate in the selective recognition and recruitment of autophagosomal cargoes in the course of targeted autophagic responses (Rogov et al, 2014; Stolz et al, 2014). The autophagy receptor best characterized to date, that is, sequestosome 1 (SQSTM1, best known as p62), recruits ubiquitinated proteins to autophagosomes by virtue of an ubiquitin‐associated (UBA) and a LC3‐binding domain (Pankiv et al, 2007).

Owing to its key role in the preservation of intracellular homeostasis, autophagy constitutes a barrier against various degenerative processes that may affect healthy cells, including malignant transformation. Thus, autophagy mediates oncosuppressive effects. Accordingly, proteins with bona fide oncogenic potential inhibit autophagy, while many proteins that prevent malignant transformation stimulate autophagic responses (Morselli et al, 2011). Moreover, autophagy is involved in several aspects of anticancer immunosurveillance, that is, the process whereby the immune system constantly eliminates potentially tumorigenic cells before they establish malignant lesions (Ma et al, 2013). However, autophagy also sustains the survival and proliferation of neoplastic cells exposed to intracellular and environmental stress, hence supporting tumor growth, invasion and metastatic dissemination, at least in some settings (Kroemer et al, 2010; Guo et al, 2013b). Here, we discuss the molecular and cellular mechanisms accounting for the differential impact of autophagy on malignant transformation and tumor progression.

In various murine models, defects in the autophagic machinery caused by the whole‐body or tissue‐specific, heterozygous or homozygous knockout of essential autophagy genes accelerate oncogenesis. For instance, Becn1^+/− mice (Becn1^−/− animals are not viable) spontaneously develop various malignancies, including lymphomas as well as lung and liver carcinomas (Liang et al, 1999; Qu et al, 2003; Yue et al, 2003; Mortensen et al, 2011), and are more susceptible to parity‐associated and Wnt1‐driven mammary carcinogenesis than their wild‐type counterparts (Cicchini et al, 2014). Similarly, mice lacking one copy of the gene coding for the BECN1 interactor autophagy/beclin‐1 regulator 1 (AMBRA1) also exhibit a higher rate of spontaneous tumorigenesis than their wild‐type littermates (Cianfanelli et al, 2015). Mice bearing a systemic mosaic deletion of Atg5 or a liver‐specific knockout of Atg7 spontaneously develop benign hepatic neoplasms more frequently than their wild‐type counterparts (Takamura et al, 2011). Moreover, carcinogen‐induced fibrosarcomas appear at an accelerated pace in autophagy‐deficient Atg4c^−/− mice (Marino et al, 2007), as do KRAS^G12D‐driven and BRAF^V600E‐driven lung carcinomas in mice bearing lung‐restricted Atg5 or Atg7 deletions, respectively (Strohecker et al, 2013; Rao et al, 2014). The pancreas‐specific knockout of Atg5 or Atg7 also precipitates the emergence of KRAS^G12D‐driven pre‐malignant pancreatic lesions (Rosenfeldt et al, 2013; Yang et al, 2014).

Several mechanisms can explain, at least in part, the oncosuppressive functions of autophagy. Proficient autophagic responses may suppress the accumulation of genetic and genomic defects that accompanies malignant transformation, through a variety of mechanisms. Reactive oxygen species (ROS) are highly genotoxic, and autophagy prevents their overproduction by removing dysfunctional mitochondria (Green et al, 2011; Takahashi et al, 2013) as well as redox‐active aggregates of ubiquitinated proteins (Komatsu et al, 2007; Mathew et al, 2009). In addition, autophagic responses have been involved in the disposal of micronuclei arising upon perturbation of the cell cycle (Rello‐Varona et al, 2012), in the degradation of retrotransposing RNAs (Guo et al, 2014), as well as in the control of the levels of ras homolog family member A (RHOA), a small GTPase involved in cytokinesis (Belaid et al, 2013). Finally, various components of the autophagic machinery appear to be required for cells to mount adequate responses to genotoxic stress (Karantza‐Wadsworth et al, 2007; Mathew et al, 2007; Park et al, 2014). This said, the precise mechanisms underlying such genome‐stabilizing effects remain elusive, implying that the impact of autophagy on DNA‐damage responses may be indirect. Further investigation is required to shed light on this possibility.

Autophagy is intimately implicated in the maintenance of physiological metabolic homeostasis (Galluzzi et al, 2014; Kenific & Debnath, 2015). Malignant transformation generally occurs along with a shift from a predominantly catabolic consumption of glycolysis‐derived pyruvate by oxidative phosphorylation to a metabolic pattern in which: (1) glucose uptake is significantly augmented to sustain anabolic reactions and antioxidant defenses, (2) mitochondrial respiration remains high to satisfy increased energy demands; and (3) several amino acids, including glutamine and serine, become essential as a means to cope with exacerbated metabolic functions (Hanahan & Weinberg, 2011; Galluzzi et al, 2013). Autophagy preserves optimal bioenergetic functions by ensuring the removal of dysfunctional mitochondria (Green et al, 2011), de facto counteracting the metabolic rewiring that accompanies malignant transformation. Moreover, the autophagic degradation of p62 participates in a feedback circuitry that regulates MTORCI activation in response to nutrient availability (Linares et al, 2013; Valencia et al, 2014).

Autophagy appears to ensure the maintenance of normal stem cells. This is particularly relevant for hematological malignancies, which are normally characterized by changes in proliferation or differentiation potential that alter the delicate equilibrium between toti‐, pluri‐ and unipotent precursors in the bone marrow (Greim et al, 2014). The ablation of Atg7 in murine hematopoietic stem cells (HSCs) has been shown to disrupt tissue architecture, eventually resulting in the expansion of a population of bone marrow progenitor cells with neoplastic features (Mortensen et al, 2011). Along similar lines, the tissue‐specific deletion of the gene coding for the ULK1 interactor RB1‐inducible coiled‐coil 1 (RB1CC1, best known as FIP200) alters the fetal HSC compartment in mice, resulting in severe anemia and perinatal lethality (Liu et al, 2010). Interestingly, murine Rb1cc1^−/− HSCs do not exhibit increased rates of apoptosis, but an accrued proliferative capacity (Liu et al, 2010). The deletion of Rb1cc1 in murine neuronal stem cells (NSCs) also causes a functional impairment that compromises postnatal neuronal differentiation (Wang et al, 2013). However, this effect appears to stem from the failure of murine Rb1cc1^−/− HNCs to control redox homeostasis, resulting in the activation of a tumor protein p53 (TP53)‐dependent apoptotic response (Wang et al, 2013). Finally, Becn1^+/− mice display an expansion of progenitor‐like mammary epithelial cells (Cicchini et al, 2014). Of note, autophagy also appears to be required for the preservation of normal stem cell compartments in the human system. Indeed, human hematopoietic, dermal, and epidermal stem cells transfected with a short‐hairpin RNA (shRNA) specific for ATG5 lose their ability to self‐renew while differentiating into neutrophils, fibroblasts, and keratinocytes, respectively (Salemi et al, 2012).

It has been proposed that autophagy contributes to oncogene‐induced cell death or oncogene‐induced senescence, two fundamental oncosuppressive mechanisms. The activation of various oncogenes imposes indeed a significant stress on healthy cells, a situation that is normally aborted through the execution of a cell death program (Elgendy et al, 2011), or upon the establishment of permanent proliferative arrest (cell senescence) that engages the innate arm of the immune system (Iannello et al, 2013). The partial depletion of ATG5, ATG7 or BECN1 limited the demise of human ovarian cancer cells pharmacologically stimulated to express HRAS^G12V from an inducible construct (Elgendy et al, 2011). Similarly, shRNAs specific for ATG5 or ATG7 prevented oncogene‐induced senescence in primary human melanocytes or human diploid fibroblasts (HDFs) expressing BRAF^V600E or HRAS^G12V (Young et al, 2009; Liu et al, 2013a). Accordingly, the overexpression of the ULK1 homolog ULK3 was sufficient to limit the proliferative potential of HDFs while promoting autophagy (Young et al, 2009). Moreover, both pharmacological inhibitors of autophagy and small‐interfering RNAs targeting ATG5, ATG7 or BECN1 prevented spontaneous senescence in HDFs while preventing the degradation of an endogenous, dominant‐negative TP53 variant (Horikawa et al, 2014). Finally, ectopic ATG5 expression reduced the colony‐forming ability of melanoma cell lines normally characterized by low ATG5 levels, an effect that could be reproduced by the administration of autophagy inducers (Liu et al, 2013a). Apparently at odds with these results, HRAS^G12V fails to induce senescence in mouse embryonic fibroblasts (MEFs) lacking transformation‐related protein 53 binding protein 2 (Trp53bp2), correlating with the stabilization of Atg5/Atg12 complexes and consequent upregulation of the autophagic flux. In line with this notion, ectopic expression of Atg5 prevented Trp53bp2‐sufficient MEFs from entering senescence upon overexpression of HRAS^G12V (Wang et al, 2012b). Thus, while in some cells autophagy appears to inhibit malignant transformation by favoring oncogene‐induced senescence, this may not be a general mechanism of autophagy‐mediated oncosuppression.

It has been suggested that autophagy is involved in the degradation of oncogenic proteins, including mutant (but not wild‐type) TP53 (Rodriguez et al, 2012; Choudhury et al, 2013; Garufi et al, 2014), p62 (Duran et al, 2008; Mathew et al, 2009; Ling et al, 2012), PML‐RARA (Isakson et al, 2010; Wang et al, 2011), and BCR‐ABL1 (Goussetis et al, 2012). Mutant TP53 often accumulates in neoplastic cells and operates as a dominant‐negative factor, thereby interfering with the oncosuppressive function of the wild‐type protein (de Vries et al, 2002). Cancer cells depleted of ULK1, BECN1 or ATG5 tend to accumulate increased amounts of mutant TP53, whereas the transgene‐driven overexpression of BECN1 or ATG5 results in mutant TP53 depletion (Choudhury et al, 2013). Such an autophagy‐dependent degradation of mutant TP53 would therefore restore the ability of wild‐type TP53 to inhibit malignant transformation, at least in some settings. It is worth noting that both ATG5 and ATG7 have been involved in the regulation of TP53‐dependent adaptive responses to stress (Lee et al, 2012; Salemi et al, 2012). However, this activity appears to be independent of autophagy, at least in the case of ATG7 (Lee et al, 2012). Interestingly, p62 itself has been ascribed with potentially oncogenic functions, including a key role in the transduction of RAS‐elicited signals as well as in the activation of a feedforward loop involving the cytoprotective transcription factor NF‐κB driven by oncogenic stress (Duran et al, 2008; Mathew et al, 2009; Takamura et al, 2011; Ling et al, 2012). Autophagy may therefore inhibit oncogenesis by limiting p62 availability (Mathew et al, 2009), at least in some settings.

The t(9;22)(q34;q11) translocation is found in about 90% of chronic myeloid leukemia patients, resulting in the synthesis of a fusion protein that involves breakpoint cluster region (BCR) and ABL proto‐oncogene 1 (ABL1) (Ben‐Neriah et al, 1986). BCR‐ABL1 is a constitutively active kinase and is etiologically involved in leukemogenesis, as demonstrated by the outstanding clinical success of imatinib mesylate, a BCR‐ABL1‐targeting kinase inhibitor (Druker et al, 2001). Arsenic trioxide, a chemotherapeutic agent commonly employed against various forms of leukemia, appears to trigger the p62‐dependent and cathepsin B‐dependent degradation of BCR‐ABL1 in leukemic progenitors (Goussetis et al, 2012). In line with this notion, the pharmacological or genetic inhibition of autophagy or cathepsin B reportedly limits the antileukemic potential of arsenic trioxide (Goussetis et al, 2012). The t(15;17)(q22;q21) translocation can be documented in 95% of promyelocytic leukemia cases, resulting in the expression of a chimera that involves promyelocytic leukemia (PML) and retinoic acid receptor, alpha (RARA) (Goddard et al, 1991). PML‐RARA blocks normal retinoic acid‐dependent myeloid differentiation, de facto driving leukemogenesis (Rousselot et al, 1994). Patients expressing PML‐RARA generally benefit from the administration of all‐trans retinoic acid (ATRA), resulting in PML‐RARA degradation and restored myeloid differentiation (Wang et al, 2011). Pharmacological and genetic evidence suggests that autophagy is implicated in both ATRA‐ and arsenic trioxide‐driven PML‐RARA degradation (Isakson et al, 2010; Wang et al, 2011). Further experimentation is required to understand whether autophagy degrades potentially oncogenic proteins in cells not exposed to chemotherapeutic agents.

Autophagy is implicated in immune responses that prevent the establishment and proliferation of malignant cells (Ma et al, 2013). At least in some circumstances, dying malignant cells are capable of recruiting antigen‐presenting cells (APCs) and other cellular components of the immune system, resulting in the elicitation of innate and/or adaptive antitumor immune responses (Deretic et al, 2013; Kroemer et al, 2013). On the one hand, autophagic responses are required for dying neoplastic cells to release ATP in optimal amounts, which not only recruits APCs through purinergic receptor P2Y, G‐protein coupled, 2 (P2RY2), but also activates them to release immunostimulatory chemokines through purinergic receptor P2X, ligand‐gated ion channel, 7 (P2RX7) (Michaud et al, 2011). On the other hand, autophagy in immune cells is implicated in several steps of both adaptive and innate immune responses (Ma et al, 2013). Thus, both cancer cell‐intrinsic and systemic defects in autophagy may prevent the host immune system to properly recognize and eliminate pre‐malignant and malignant cells.

Autophagy mediates potent anti‐inflammatory effects (Deretic et al, 2013). At least in some cases, malignant transformation is stimulated by an inflammatory microenvironment, which contains high amounts of potentially genotoxic ROS as well as various mitogenic cytokines (Coussens et al, 2013). Proficient autophagic responses limit inflammation as: (1) they efficiently dispose of the so‐called inflammasomes (the supramolecular platforms that are responsible for the maturation and secretion of pro‐inflammatory interleukin‐1β and interleukin‐18), as well as damaged mitochondria, which would otherwise release endogenous inflammasome activators (Nakahira et al, 2011; Zitvogel et al, 2012); (2) they are linked to the inhibition of pro‐inflammatory signals delivered by some pattern recognition receptors, such as RIG‐I‐like receptors (Jounai et al, 2007); (3) they limit the abundance of B‐cell CLL/lymphoma 10 (BCL10), a protein involved in pro‐inflammatory NF‐κB signaling (Paul et al, 2012); (4) they are connected to the inhibition of transmembrane protein 173 (TM173, best known as STING), a pattern recognition receptor involved in the delivery of pro‐inflammatory cues in response to cytosolic nucleic acids (Saitoh et al, 2009).

Finally, autophagy may suppress carcinogenesis owing to its key role in the first line of defense against viral and bacterial infection (Deretic et al, 2013). Indeed, several potentially carcinogenic pathogens potently activate autophagy upon infection. These pathogens include hepatitis B virus (which promotes hepatocellular carcinoma), human herpesvirus 8 (which causes Kaposi's sarcoma and contributes to the pathogenesis of primary effusion lymphoma and multicentric Castleman's disease), human papillomavirus type 16 and 18 (HPV‐16 and HPV‐18, which cause cervical carcinoma), Epstein–Barr virus and Helicobacter pylori (both of which are associated with gastric carcinoma), Streptococcus bovis (which causes colorectal carcinoma), Salmonella enterica (which is associated with an increased incidence of Crohn's disease, hence sustaining colorectal carcinogenesis, and gallbladder carcinoma), as well as Chlamydia pneumoniae (an etiological determinant in some forms of lung cancer) (Nakagawa et al, 2004; Travassos et al, 2010; Yasir et al, 2011; Conway et al, 2013; Griffin et al, 2013; Zhang et al, 2014). Such a xenophagic response is required for the rapid clearance of intracellular pathogens as well as for the stimulation of pathogen‐specific immune responses (Deretic et al, 2013; Ma et al, 2013). Accordingly, epithelial cells bearing molecular defects in the autophagic machinery, such as those provoked by Crohn's disease‐associated point mutations in ATG16L1 and nucleotide‐binding oligomerization domain containing 2 (NOD2) (Lassen et al, 2014), are more susceptible to infection by intracellular pathogens than their wild‐type counterparts. In line with this notion, reduced levels of autophagic markers including BECN1 have recently been correlated with HPV‐16 and HPV‐18 infection in a cohort of cervical carcinoma patients (Wang et al, 2014). Thus, autophagy may exert oncosuppressive effects also by virtue of its antiviral and antibacterial activity.

Taken together, these observations suggest that autophagy prevents malignant transformation by preserving both cellular and organismal homeostasis in conditions that pose a risk for oncogenesis (Fig 2).

In agreement with the oncosuppressive activity of autophagy, several oncoproteins, that is, proteins that drive malignant transformation upon overexpression‐ or mutation‐dependent hyperactivation, inhibit autophagic responses (Maiuri et al, 2009). Along similar lines, many bona fide oncosuppressor proteins, that is, proteins that are inactivated or lost in the course of oncogenesis, stimulate autophagy (Morselli et al, 2011) (Table 1).

Autophagic responses generally support the growth and progression of established tumors by reducing their sensitivity to cell‐intrinsic as well as microenvironmental stimuli that would normally promote their demise, in particular upon the so‐called epithelial‐to‐mesenchymal transition (Kroemer et al, 2010; Avivar‐Valderas et al, 2013; Cai et al, 2014). This notion is supported by a growing amount of data indicating that defects in the autophagic machinery often restrain the proliferation, dissemination and metastatic potential of malignant cells, as discussed below. Moreover, advanced human tumors generally exhibit an increased autophagic flux, correlating with an invasive/metastatic phenotype, high nuclear grade, and poor disease outcome (Lazova et al, 2012; Mikhaylova et al, 2012).

Although mice with a systemic mosaic deletion of Atg5 or a liver‐specific knockout of Atg7 develop spontaneous hepatic neoplasms more frequently than their wild‐type counterparts (see above), these malignancies are mostly benign (pointing to defects in tumor progression) and their size can be further decreased by the simultaneous deletion of Sqstm1 (Takamura et al, 2011). In accord with these data, p62 has been shown to support the progression of both endogenous (ERBB2‐driven) and xenografted mammary tumors through a variety of mechanisms, including the activation of NRF2 (see above) (Chen et al, 2013; Cai‐McRae et al, 2014). As opposed to their autophagy‐competent counterparts, highly metastatic hepatocellular carcinoma cell lines infected with lentiviruses that stably downregulate BECN1 or ATG5 are virtually unable to survive within the metastatic niche, although they normally proliferate, invade surrounding tissues and undergo the epithelial‐to‐mesenchymal transition (Peng et al, 2013). Along similar lines, the shRNA‐mediated downregulation of Atg5 or p62 abolishes the ability of Tsc2^−/−Trp53^−/− MEFs to develop macroscopic tumors upon inoculation into nude mice (Parkhitko et al, 2011). Moreover, the robust antimetastatic effects of N‐myc downstream regulated 1 (NDRG1) have recently been ascribed to its ability to suppress stress‐induced autophagic responses (Sahni et al, 2014).

The lung‐specific deletion of Atg7 during the late stages of BRAF^V600E‐driven carcinogenesis favors the development of small oncocytomas (which are relatively benign tumors) rather than adenocarcinomas, a shift that is accompanied by the accumulation of dysfunctional mitochondria and an increased dependency on exogenous glutamine (Strohecker et al, 2013). Similar results have been obtained in models of KRAS^G12D‐driven lung and pancreatic carcinogenesis, upon the tissue‐specific deletion of Atg5 or Atg7 (Guo et al, 2013a; Rosenfeldt et al, 2013; Rao et al, 2014; Yang et al, 2014), as well as in models of breast carcinoma driven by the mammary‐gland specific knockout of partner and localizer of BRCA2 (Palb2), upon the monoallelic deletion of Becn1 (Huo et al, 2013). Intriguingly, the downregulation of Atg1 (the fly ortholog of ULK1) by RNA interference also suppressed a hyperproliferative eye phenotype caused in Drosophila melanogaster by the ectopic expression of Ras1^12V (a constitutively active variant of the fly ortholog of mammalian RAS proteins) but had an opposite effect on the overgrowth of the eye epithelium provoked by mutations in the oncosuppressor gene scribbled (scrib) (Perez et al, 2014). Similarly, the individual depletion of 12 distinct components of the autophagic machinery (i.e., Atg1, Atg6, Atg12, Atg5, Atg7, Atg4a, Atg4b, Atg8a, Atg8b, Atg3, Atg9 and Atg18) promoted, rather than limited, the overgrowth of adult Drosophila eyes and their larval precursor tissues in so‐called eyeful flies (a model of Notch‐driven carcino‐genesis) (Perez et al, 2014).

In some models of endogenous mammalian carcinogenesis, the Trp53^−/− genotype prevents genetic interventions that target the autophagic machinery from provoking metabolic and bioenergetic alterations that limit tumor progression (Huo et al, 2013; Rosenfeldt et al, 2013; Rao et al, 2014). Conversely, the response of pancreatic cancer cells, xenografts and KRAS^G12D‐driven autochthonous adenocarcinomas to genetic or pharmacological autophagy inhibition persists in the context of TP53 loss‐of‐heterozygosity (Yang et al, 2011; Yang et al, 2014).

KRAS^G12D‐driven pancreatic adenocarcinoma cells entering a state of dormancy (rather than succumbing) in response to oncogene ablation (i.e., the shutdown of oncogenic KRAS signaling) have recently been show to activate autophagy to efficiently counteract metabolic stress (Viale et al, 2014). Of note, whereas primary KRAS^G12D‐expressing cells generally exhibit increased glucose and glutamine uptake, as well as an elevated anabolic flux via the pentose phosphate pathway (Ying et al, 2012; Rosenfeldt et al, 2013), the survival of oncogene‐depleted pancreatic adenocarcinoma cells critically relies on oxidative phosphorylation (Viale et al, 2014). Thus, exposing such dormant pancreatic adenocarcinoma cells to the inhibitor of oxidative phosphorylation oligomycin reportedly abolishes their ability to form tumors upon KRAS^G12D re‐expression (Viale et al, 2014).

Interestingly, KRAS^G12D‐expressing pancreatic adenocarcinoma cells driven into dormancy upon oncogene ablation also display functional and phenotypic features of cancer stem cells (CSCs) (Viale et al, 2014). Moreover, mammary CSCs (which propagate in culture as mammospheres) are often characterized by an elevated autophagic flux, and their ability to efficiently form tumors in vivo appears to rely on autophagy, as tumor formation can be abolished by the genetic inhibition of BECN1 or ATG4A (Gong et al, 2013; Wolf et al, 2013). Thus, autophagy may also sustain tumor progression by preserving the viability of the CSC compartment and/or by promoting the persistence of dormant cancer cells (Viale et al, 2014).

Cancer cells isolated from established tumors and subjected to the genetic or pharmacological inhibition of autophagy are less resistant to exogenous stimuli than their wild‐type counterparts (Boya et al, 2005; Amaravadi et al, 2007; Kroemer et al, 2010). In line with this notion, autophagy‐deficient tumors are often more sensitive to several chemotherapeutic agents as well as to radiation therapy than their autophagy‐proficient counterparts (Janku et al, 2011; Ko et al, 2014; Levy et al, 2014). This does not necessarily hold true for immunocompetent mice. Indeed, autophagic responses preceding the demise of cancer cells exposed to a selected panel of agents are required (though not sufficient) for cell death to be perceived as immunogenic and hence to elicit a therapeutically relevant immune response (Kroemer et al, 2013; Ko et al, 2014). Cancer cells exposed to therapeutic interventions can also undergo senescence (Lopez‐Otin et al, 2013). Although senescent cells do not proliferate, they may support disease relapse by releasing a wide panel of pro‐inflammatory and mitogenic cytokines into the microenvironment (underlying the so‐called senescence‐associated secretory phenotype, SASP) (Lopez‐Otin et al, 2013). Interestingly, these cells are highly dependent on autophagic responses for survival, and pharmacological inhibitors of autophagy have been shown to synergize with various chemotherapeutics in experimental models of lymphoma that are susceptible to acquire the SASP in response to treatment (Young et al, 2009; Dorr et al, 2013).

Taken together, these observations suggest that autophagy supports the progression of established neoplasms through several mechanisms (Fig 3) and that pharmacological inhibitors of autophagy may exert robust antineoplastic effects, at least in some settings.

Based on the data presented above, it is tempting to speculate that the multistep process leading from a healthy tissue to a metastatic and therapy‐resistant, and hence life‐threatening, neoplasm involves a temporary loss of autophagic competence (or the gain of molecular functions that antagonize, at least transitorily, autophagy‐dependent oncosuppression). Initially, defects in the autophagic process might facilitate the acquisition of malignant features by healthy cells. Later on, once malignancy is established, the restoration of proficient autophagic responses may be essential to support the survival, proliferation and growth of cancer cells in the presence of adverse microenvironmental conditions (Fig 4A). How proficient autophagic responses are reconstituted after an initial phase of autophagy inhibition, however, has not yet been established. As a possibility, the genetic or epigenetic instability that characterizes progressing tumors may restore autophagy in specific cells, rendering them able to overcome their neighboring autophagy‐incompetent counterparts. Formal experimental evidence in support of this model is lacking. At least in some settings, oncogenesis and tumor progression may indeed rely on a stable loss or gain of autophagic proficiency (Fig 4B and C).

Importantly, genetic data indicating that the inhibition of autophagy exerts bona fide antineoplastic effects against established tumors have been obtained mainly in RAS‐driven or RAS‐related (BRAF^V600E‐driven) models of oncogenesis (Guo et al, 2013a; Rosenfeldt et al, 2013; Strohecker et al, 2013; Perez et al, 2014; Rao et al, 2014; Yang et al, 2014). In other scenarios, including the loss of scrib and the eyeful genotype in Drosophila, disabling autophagy by genetic means de facto accelerates tumor progression (Perez et al, 2014). Moreover, the antineoplastic effects of genetic and pharmacological interventions that inhibit autophagy vary in models in which the TP53 system is lost by different modalities (i.e., homozygous knockout versus loss‐of‐heterozygosity) (Yang et al, 2011; Huo et al, 2013; Rosenfeldt et al, 2013; Rao et al, 2014; Yang et al, 2014). Thus, the impact of autophagy on tumor progression may exhibit a significant degree of context dependency. Accordingly, recent data indicate that only tumors that are addicted to autophagy even in nutrient‐rich conditions and in the absence of stressful stimuli respond to autophagy inhibitors in vivo (Maycotte et al, 2014). This suggests that only a fraction of cancer patients may benefit from the administration of autophagy inhibitors. Along similar lines, autophagy has been shown to underlie, at least in part, the therapeutic activity of some anticancer regimens (Salazar et al, 2009; Torres et al, 2011; Vara et al, 2011). Moreover, autophagy is required not only for the emission of immunostimulatory signals by malignant cells succumbing to specific anticancer agents (Kroemer et al, 2013), but also for the activation of tumor‐targeting innate and adaptive immune responses (Ma et al, 2013). Efforts should therefore be focused on the identification of precise clinical scenarios in which autophagy supports, rather than counteracts, disease progression and resistance to therapy. This is particularly important not only because autophagy inhibitors may one day become part of the clinical routine, but also because most (if not all) anticancer agents that are currently employed in the clinic modulate autophagy (Kroemer et al, 2010).

Of note, the genetic inhibition of autophagy in models of mammalian carcinogenesis has near‐to‐invariably been achieved with the whole‐body or conditional (heterozygous or homozygous) knockout of Atg5, Atg7 or Becn1 (Guo et al, 2013a; Rosenfeldt et al, 2013; Strohecker et al, 2013; Rao et al, 2014; Yang et al, 2014). As an increasing number of activities is being ascribed to these and other autophagy mediators (Cosse et al, 2010; Lee et al, 2012; Liu et al, 2012a; Moscat & Diaz‐Meco, 2012; Zhao et al, 2012; Maskey et al, 2013; Elgendy et al, 2014), it remains possible that other, autophagy‐independent functions of ATG5, ATG7 and BECN1 support the progression of KRAS^G12D‐ and BRAF^V600E‐driven tumors.

In experimental settings, the pharmacological inhibition of autophagy is most often realized with the administration of chloroquine (CQ) and hydroxychloroquine (HCQ), two lysosomotropic drugs approved by the US Food and Drug Administration for the prophylactic treatment of malaria and for the management of (chronic, discoid or systemic) lupus erythematosus as well as acute or chronic rheumatoid arthritis (Rubinsztein et al, 2012). Similar to other lysosomotropic agents, both CQ and HCQ block autophagy by inhibiting the fusion between autophagosomes with lysosomes and their degradation (Rote & Rechsteiner, 1983). Results from several Phase I–II clinical data indicate that HCQ can be safely employed at relatively high doses to improve the clinical activity of radiation therapy as well as of distinct anticancer chemotherapeutics, including temozolomide (an alkylating agent), vorinostat (a histone deacetylase inhibitor) and bortezomib (a proteasome inhibitor) (Barnard et al, 2014; Mahalingam et al, 2014; Rangwala et al, 2014a; Rangwala et al, 2014b; Rosenfeld et al, 2014; Vogl et al, 2014; Wolpin et al, 2014). However, the potency and specificity of HCQ and CQ are poor, and both these compounds have been shown to mediate antineoplastic effects via multiple autophagy‐independent pathways, including lethal lysosomal destabilization (Boya et al, 2003; Maycotte et al, 2012) and the normalization of the tumor vasculature (Maes et al, 2014). Thus, the abundant scientific literature concluding that pharmacological inhibitors of autophagy constitute a convenient means to arrest tumor progression or sensitize malignant cells to therapy based on results obtained with CQ and HCQ only should be taken with caution. Lys05, a potent dimeric variant of CQ, is currently being characterized in preclinical tumor models (McAfee et al, 2012). Lys05, however, seems to share the limited specificity of CQ and HCQ. Small molecules that block autophagy in a highly specific manner are therefore urgently awaited. Recently, a specific VPS34 inhibitor has been developed and shown to efficiently inhibit autophagy. However, its putative antineoplastic effects may reflect the pleiotropic activity of VPS34, which is also involved in non‐autophagic vesicle trafficking (Ronan et al, 2014). Finally, it will be interesting to develop molecules that inhibit autophagy in malignant cells but not in their normal counterparts, perhaps by targeting upstream signal transducers rather than downstream effectors. Proof‐of‐principle data in support of the therapeutic activity of such an approach in preclinical models have already been generated (Wilkinson et al, 2009). This is particularly relevant given the key contribution of autophagy to the maintenance of homeostasis in healthy cells. Indeed, at least on theoretical grounds, efficiently inhibiting autophagy in non‐transformed cells may have deleterious consequences ranging from an accrued propensity to malignant transformation to overt cytotoxicity.

Intriguingly, several experimental maneuvers that increase the lifespan of model organisms as evolutionary distant as nematodes, flies and mice, including caloric restriction as well as the administration of the MTOR inhibitor rapamycin, activate autophagy (Harrison et al, 2009; Morselli et al, 2010; Madeo et al, 2014). Moreover, these interventions generally lose their lifespan‐extending activity in autophagy‐deficient hosts (Morselli et al, 2010; Madeo et al, 2014). However, to which extent the lifespan‐prolonging effects of autophagy directly relate to its major oncosuppressive functions remains to be determined.

Recently, several studies have demonstrated that autophagy is regulated by epigenetic alterations, including histone methylation and acetylation (Artal‐Martinez de Narvajas et al, 2013; Lam et al, 2013; Eisenberg et al, 2014). In addition, transcription factors other than TP53 and HIF‐1, such as transcription factor EB (TFEB) and cAMP‐responsive element binding protein 1 (CREB1), are intimately involved in autophagic responses (Settembre et al, 2011; Seok et al, 2014). The precise mechanisms through which cancer‐associated epigenetic alterations (and/or the consequent transcriptional reprogramming) modulate autophagy have not yet been elucidated. Obtaining profound insights into this issue may pave the way to the development of novel, cancer‐specific inhibitors of autophagy with therapeutic potential.

Irrespective of these incognita, autophagy stands out as key system for the maintenance of homeostasis, hence exerting a differential impact on malignant transformation and tumor progression (Box 1).

LG, KMR and GK conceived the review. LG wrote the review, centralized and integrated comments from co‐authors, conceived the figures, and revised the review upon editorial feedback. JMBSP designed the figures. FP constructed the table. All authors corrected the review and provided valuable input to obtain a consensus view.

Vassiliki Karantza is an employee of Merck Research Laboratories. All other authors declare that they have no conflict of interest.

GK is supported by the Ligue contre le Cancer (équipe labelisée); Agence National de la Recherche (ANR); Association pour la recherche sur le cancer (ARC); Cancéropôle Ile‐de‐France; Institut National du Cancer (INCa); Fondation Bettencourt‐Schueller; Fondation de France; Fondation pour la Recherche Médicale (FRM); the European Commission (ArtForce); the European Research Council (ERC); the LabEx Immuno‐Oncology; the SIRIC Stratified Oncology Cell DNA Repair and Tumor Immune Elimination (SOCRATE); the SIRIC Cancer Research and Personalized Medicine (CARPEM); and the Paris Alliance of Cancer Research Institutes (PACRI). DG is supported by the American Cancer Society (Institutional Research Grant IRG‐14‐192‐40). SK is supported by the National Health and Medical Research Council of Australia Project Grant (1041807) and a Senior Principal Research Fellowship (1002863). MP is supported by grants from the Ministry of Health of Italy (‘Ricerca Corrente’ and ‘Ricerca Finalizzata’) and Associazione Italiana per la Ricerca sul Cancro (AIRC). HUS is supported by the Swiss National Science Foundation, Swiss Cancer League, and the European Commission (MEL‐PLEX). GV is supported by grants from the Spanish Ministry of Economy and Competitiveness (MINECO), Fondo Europeo de Desarrollo Regional (FEDER) (PI12/02248, FR2009‐0052 and IT2009‐0053), Fundación Mutua Madrileña (AP101042012) and Fundació La Marató de TV3 (20134031).