Pharmacological modulation of autophagy: therapeutic potential and persisting obstacles

Abstract

Autophagy is central to the maintenance of organismal homeostasis in both physiological and pathological situations. Accordingly, alterations in autophagy have been linked to clinically relevant conditions as diverse as cancer, neurodegeneration and cardiac disorders. Throughout the past decade, autophagy has attracted considerable attention as a target for the development of novel therapeutics. However, such efforts have not yet generated clinically viable interventions. In this Review, we discuss the therapeutic potential of autophagy modulators, analyse the obstacles that have limited their development and propose strategies that may unlock the full therapeutic potential of autophagy modulation in the clinic.

Autophagy is a highly conserved mechanism through which eukaryotic cells deliver dispensable or potentially dangerous cytoplasmic material to lysosomes for degradation¹. Thus far, three major routes for the delivery of autophagic substrates to lysosomes have been characterized: microautophagy, chaperone-mediated autophagy (CMA) and macroautophagy.

Microautophagy relies on the direct uptake of cytoplasmic material through an invagination of the lysosomal membrane². CMA involves the lysosomal-associated membrane protein 2 (LAMP2)-dependent translocation of autophagic substrates bound to cytosolic chaperones of the heat shock protein family across the lysosomal membrane³. Macroautophagy involves specialized double-membraned vesicles known as autophagosomes, which progressively sequester autophagic cargo and — upon closure — deliver the cargo to lysosomes by membrane fusion¹. The organelle that forms upon the fusion of one autophagosome and one lysosome is generally referred to as an autolysosome¹. Macroautophagy is by far the best-characterized form of autophagy. For this reason, the word autophagy is used throughout this article to refer to macroautophagy, unless otherwise specified.

Autophagy is fundamental to the preservation of organismal fitness, for multiple reasons. Constitutive autophagic responses efficiently degrade products of normal cellular metabolism that can become cytotoxic upon accumulation, such as damaged mitochondria and redox-active protein aggregates⁴. Inducible autophagic responses promote the survival of cells that respond to perturbations of intracellular or extracellular homeostasis⁵. Autophagy is indeed central to adaptation to stress, as demonstrated by the fact that pharmacological or genetic inhibition of autophagy usually precipitates the demise of cells facing infections and nutritional, metabolic, physical and chemical challenges⁶. Furthermore, autophagy is intimately connected with cell-extrinsic circuitries that operate to maintain homeostasis and support healthy ageing at the whole-body level. For instance, autophagic responses in the liver, skeletal muscle and other tissues underlie the beneficial effects of physical exercise on whole-body glucose metabolism⁷. Along similar lines, autophagy induction in malignant cells that succumb to some chemotherapies and radiotherapies results in the emission of danger signals and, ultimately, the initiation of a therapeutically relevant anticancer immune response⁸. Autophagy can also mediate cytotoxic effects, at least in specific pathophysiological settings⁹, although the term <m>autophagic cell death</m> should be used with extreme caution (BOX 1). Moreover, components of the autophagic apparatus have recently been shown to participate in processes other than the degradation of cytoplasmic material. These processes include: LC3-associated phagocytosis (LAP)¹⁰ (BOX 2), migration (mainly as a result of focal adhesion turnover)¹¹ and unconventional secretion, which is a mechanism by which cytoplasmic entities (including soluble proteins, organellar material and pathogens) are exported from the cell in a manner that does not depend on the conventional secretory route that operates between the endoplasmic reticulum and the Golgi apparatus¹².

Box 1. Autophagic cell death in development and disease.

When light microscopy was the main — if not the sole — technique to study cell biology, investigators noted that the cytoplasm of dying eukaryotic cells sometimes becomes clogged with vacuoles. Soon thereafter, the term ‘autophagic cell death’ (also known as type II cell death) was introduced to indicate instances of cellular demise that are accompanied by cytoplasmic vacuolization²⁸⁸. This expression rapidly acquired a causal implication and has subsequently been extensively used, which led to the assumption that autophagy aetiologically contributes to cell death. With the advent of modern molecular biology techniques, however, it became clear that autophagy generally mediates cytoprotective — rather than cytotoxic — effects. Indeed, pharmacological or genetic inhibition of core components of the autophagic machinery most often accelerates — rather than retards — the death of mammalian cells that experience perturbations of homeostasis²⁸⁹. Thus, autophagic responses often accompany the cellular demise (as an ultimate attempt of cells to cope with stress and to survive), but rarely cause it.

However, instances of cell death that are precipitated by the autophagic machinery have been described, both in developmental scenarios and during adaptation to stress²⁸⁹. Various autophagy-related (Atg) genes were shown to be required for the physiological demise of cells from developing Drosophila melanogaster larvae²⁹⁰. The neuron-specific knockout of Atg7 limited tissue loss in a mouse model of severe neonatal hypoxia–ischaemia⁹. Similarly, pharmacological inhibition of autophagy with 3-methyladenine (3-MA) as well as the depletion of various components of the autophagic machinery, including beclin 1 (BECN1), protected human cancer cells from a Na⁺/K⁺-ATPase-dependent form of autophagic cell death known as ‘autosis’ (REF. 94). In line with this notion, cardiac glycosides (which are potent inhibitors of the Na⁺/K⁺-ATPase) mediated robust neuroprotective effects in a rat model of neonatal hypoxia–ischaemia⁹⁴. Thus, autophagy may precipitate cell death in some circumstances. However, this possibility must be addressed experimentally with specific pharmacological and genetic tools.

Box 2. Mechanisms of autophagy.

Canonical autophagy relies on two ubiquitin-like conjugation systems. One involves autophagy-related 7 (ATG7) and ATG10, and is responsible for the formation of a supramolecular protein complex containing ATG5, ATG12 and autophagy-related 16-like 1 (ATG16L1)¹. The other ubiquitin-like conjugation system, which involves ATG3, ATG4 and ATG7, promotes the cleavage of members of the Atg8 protein family, including human LC3, and their conjugation to phosphatidylethanolamine (PE)¹. Lipidated LC3 (LC3-II) and LC3-like molecules such as GABA type A receptor-associated protein (GABARAP) are recruited to forming autophagosomes, to operate as receptors for autophagic substrates or autophagic adaptors like p62 and have largely been exploited for monitoring autophagy in vitro and in vivo²⁷⁴. ATG9, another member of the ATG protein family, has a crucial function in autophagosome nucleation, which is initiated by a supramolecular complex that contains UNC-51-like autophagy-activating kinase 1 (ULK1), RB1-inducible coiled-coil 1 (RB1CC1; also known as FIP200), ATG13 and ATG101 (REF. 1). Of note, recent data indicate that the ATG conjugation systems are less important for autophagosome formation than previously thought but are crucial for the degradation of the inner autophagosomal membrane²⁹¹ (see the figure).

Mechanistic target of rapamycin (mTOR) complex 1 (mTORC1) exerts prominent autophagy-suppressing functions by catalysing the inactivating phosphorylation of ATG13 and ULK1 (REF. 276). Such an inhibition can be relieved upon the inactivation of mTORC1 by AMP-activated protein kinase (AMPK), which is sensitive to cAMP accumulation (a consequence of ATP consumption) and also catalyses the activating phosphorylation of ULK1 and beclin 1 (BECN1)²⁹². ULK1 promotes autophagic responses by activating a multiprotein complex with phosphatidylinositol 3-kinase activity that contains BECN1, VPS34 and phosphoinositide 3-kinase regulatory subunit 4 (PIK3R4; also known as VPS15)¹. The BECN1–VPS34 complex can interact with a variety of additional regulatory factors, including UV radiation resistance-associated (UVRAG) and autophagy and beclin 1 regulator 1 (AMBRA1), which stimulate the catalytic activity of VPS34, as well as BCL-2, apoptosis regulator (BCL-2), which mediates VPS34-inhibitory effects^1,157. The expansion of autophagosomes in the course of canonical autophagic responses indeed relies on phosphatidylinositol-3-phosphate (PtdIns3P) production and PtdIns3P-binding proteins of the WD-repeat domain phosphoinositide-interacting (WIPI) family¹ (see the figure). Finally, closed autophagosomes fuse with lysosomes to generate autolysosomes, followed by luminal acidification and consequent activation of lysosomal hydrolases¹.

Several non-canonical instances of autophagy that proceed independently of specific components of the autophagic apparatus have also been described^{204,205,282,293}. This suggests the existence of functional redundancy in the molecular mechanisms that underlie autophagic responses (at least in mammals). One of these pathways, that is, LC3-associated phagocytosis (LAP), involves the recruitment of parts of the autophagic machinery to single-membraned phagosomes that form in the context of danger signalling, followed by LC3 lipidation and delivery of phagosomes to lysosomes for degradation^10,213. LAP proceeds independently of the ULK1 complex, AMBRA1 and ATG14 (which are required for canonical autophagy) but it relies on LC3, RUN and cysteine-rich domain-containing beclin 1 interacting protein (RUBCN) and NADPH oxidase 2, which are dispensable for canonical autophagy^10,213. Finally, the molecular machineries for canonical autophagy and LAP share multiple components, including BECN1, VPS34, UVRAG, ATG3, ATG5, ATG7, ATG12 and ATG16L1 (REFS 10,213). Thus, the role of LAP in various processes might have been overlooked based on the pharmacological or genetic modulation of these shared factors. AKT1S1, AKT1 substrate 1; DEPTOR, DEP domain-containing mTOR-interacting protein; MLST8, mTOR-associated protein, LST8 homologue; RPTOR, regulatory-associated protein of mTOR complex 1.

The detailed description of the molecular machinery that underlies constitutive and inducible autophagic responses is beyond the scope of this Review (BOX 2). However, it should be noted that the biochemical reactions that enable the generation of autophagosomes, the recognition of autophagic substrates, their sequestration and the delivery of autophagic cargo to lysosomes for degradation involve at least 100 different proteins¹. Thus, they provide multiple targets for the activation or inhibition of autophagy (FIG. 1). Although alterations in autophagy have been implicated in the aetiology of neurodegeneration, acute neuronal injury, ageing, cardiovascular conditions, hepatic and metabolic disorders, cancer, infectious diseases, inflammatory and autoimmune conditions, and other pathological conditions (as discussed below), no intervention aimed specifically at modulating autophagy is currently available for use in humans (TABLE 1). Indeed, although rapamycin (also known as sirolimus), chloroquine, hydroxychloroquine (HCQ) and other drugs that are approved for some indications stimulate or inhibit autophagy, they were not developed for this purpose. Thus, there is considerable, but still unrealized, potential for translating preclinical findings on autophagy modulation into therapeutic benefit for different patient populations¹³. Notably, the key role of autophagy in cell biology and its considerable therapeutic potential recently received one of the most important forms of recognition from the scientific community as the Japanese cell biologist Yoshinori Ohsumi was awarded the 2016 Nobel Prize in Physiology or Medicine for his discoveries on the mechanisms of autophagy.

Figure 1. Autophagic processes amenable to therapeutic modulation.

Several pharmacological and nutritional interventions are available to inhibit autophagy at the nucleation, elongation, fusion or degradation phase. In addition, several agents modulate autophagy through multipronged or hitherto uncharacterized molecular mechanisms. For additional details, please refer to TABLE 1. 3-MA, 3-methyladenine; AMPK, AMP-activated protein kinase; ATG4B, autophagy-related 4B cysteine peptidase; BafA1, bafilomycin A1; BECN1, beclin 1; CRM, caloric restriction mimetic; H₂S, hydrogen sulfide; HCQ, hydroxychloroquine; IFNγ, interferon-γ; Ins(1,4,5)P₃, inositol-1,4,5-trisphosphate; MAPK, mitogen-activated protein kinase; mTORC1, mechanistic target of rapamycin complex 1; ROS, reactive oxygen species; ULK1, UNC-51-like autophagy activating kinase 1.

Table 1.

Main modulators of autophagy available to date and their limitations

Agent	Mode of action	Blood–brain barrier permeant	Status	Major limitations	Refs
Activators
A-769662	AMPK activation?	Unknown	In preclinical development	Unclear MOA (requires upstream AMPK-activating kinase)	130
Antimycobacterial antibiotics	Altered ROS production	Yes	Approved for treatment of mycobacterial infections	Potentially interferes with many ROS-sensitive processes	178
BECN1-derived peptide	BECN1 activator	Yes	In preclinical development	Limitations associated with chemical nature (shelf stability and potential immunogenicity)	256
BRD5631	Unknown	Unknown	In preclinical development	Unclear MOA (independent of mTORC1)	179
Caloric restriction	Multiple	NA	NA	Potentially dangerous for subjects with weight problems (for example, patients with cachectic cancer); compliance issues	10,54,118,128,13 1,135,148,150,16 4,180,188,231,237,240
Carbamazepine	Reduction in Ins(1,4,5)P3 and inositol levels	Yes	Approved for treatment of seizures and bipolar disorders	Psychoactive agent, inhibits various neuronal functions	134,137
Carbon monoxide	Altered ROS production	Yes	Experimental agent	Potentially interferes with many ROS-sensitive processes	182
Chloramphenicol	Unknown	Yes	Approved for second-line treatment of bacterial infections	Unclear MOA and potentially mitochondriotoxic	116
Everolimus (also known as RAD-001)	mTORC1 inhibition	Yes	Approved for cancer therapy	Inhibits multiple mTORC1-dependent processes and has robust immunosuppressive effects	219,220
Hydrogen sulfide	AMPK activation?	Yes	Experimental agent	Unclear MOA and potentially toxic for the respiratory trait	133
Hydroxycitrate	CRM	Unknown	Clinically tested for treatment of diabetes, now discontinued	May cause weight loss upon chronic administration	163
IFNγ	MAPK activation?	No	In clinical trials, mainly for cancer immunotherapy	Unclear MOA (involves MAPK signalling)	169
Lithium	Reduction in Ins(1,4,5)P3 and inositol levels	Yes	Approved for treatment of bipolar disorders	Psychoactive agent, inhibits various neuronal functions	42,50,55
Melatonin*	Altered ROS production	Yes	In clinical trials for the treatment of a wide range of conditions	Potentially interferes with ROS-sensitive processes and has been associated with autophagy inhibition in some models	97
Metformin	AMPK activation?	Yes	Approved for treatment of type 2 diabetes	Mediates multiple AMPK-unrelated effects, including inhibition of respiratory complex I	136
Physical exercise	Multiple	NA	NA	Not appropriate for patients affected by cardiovascular or skeletal disorders but safe in most other cases	7,119,129
Rapamycin (also known as sirolimus)	mTORC1 inhibition	Yes	Approved for use in coronary stents (to prevent transplant rejection) and to treat a rare pulmonary disease	Inhibits multiple mTORC1-dependent processes, has robust immunosuppressive effects and may cause mTORC2 inhibition on chronic administration	10,22,31,32,38,40, 41,43–47,50,51,56, 60,68–70,77,79,80, 90,95,96,99, 101–103,132,134, 146,180,183,184, 196,198,219,220, 223,225–228,234, 241,244,245, 255,258–260
Resveratrol	CRM	Yes	Nutritional supplement that is available over the counter; in clinical trials for treatment of several disorders	Potentially causes nephrotoxicity at high concentrations	33,132,147,229,235
Retinoic acid	Unknown	Yes	Approved for cancer therapy (ATRA)	Reported to specifically inhibit CMA in some settings	100
Simvastatin	AMPK activation?	Yes	Approved for treatment of obesity	Unclear MOA (associated with AMPK activation) and potentially mitochondriotoxic	99
Spermidine	CRM	Yes	Nutritional supplement that is available over the counter	Degradation products include ROS and potentially cytotoxic aldehydes	163,230
Temsirolimus (also known as CCI-779)	mTORC1 inhibition	Yes	Approved for cancer therapy	Inhibits multiple mTORC1-dependent processes and has robust immunosuppressive effects	51
Torins	mTORC1 inhibition	Unknown	Experimental agents	Inhibit multiple mTORC1-dependent processes	246
Trehalose	Unknown	Yes	In clinical trials for treatment of bipolar disorders, dry eye syndrome and vascular ageing	Unclear MOA (independent of mTORC1) but safe	39,53,65
Trichostatin A	Unknown	No	Discontinued from clinical tests	Unclear MOA, potentially linked to transcriptional effects	98
Vorinostat	Unknown	Yes	Approved for cancer therapy	Unclear MOA, potentially linked to transcriptional effects	115
Inhibitors
3-MA	VPS34 inhibition	Yes	Experimental agent	Inhibits various class III PI3Ks	10,36,63,76,78–80, 82,84–88,93,96,98, 99,105,243,255
Azithromycin	Unknown	Yes	Approved for treatment of multiple bacterial infections	Unclear MOA, blocks autophagic flux	187
Bafilomycin A1	Lysosomal inhibition	Yes	Experimental agent	Inhibitor of lysosomal functions	36,58,79,86–88,105,180,255
Cardiac glycosides	Na+/K+-ATPase inhibition	Yes	Extensively used in the past for treatment of cardiac disorders	Narrow therapeutic window but specific for autosis	94
Chloroquine	Lysosomal inhibition	Yes	Extensively used in the past as an antimalarial agent	Inhibitor of lysosomal functions	100,104,134, 206,215,224
Compound C (also known as dorsomorphin)	AMPK inhibition	Unknown	In preclinical development	Potentially interferes with AMPK-dependent processes	82
Edavarone	ROS scavenger	Unknown	Experimental agent	Potentially interferes with many ROS-sensitive processes	89
HCQ	Lysosomal inhibition	Yes	Extensively used in the past as an antimalarial agent	Inhibitor of lysosomal functions	190,206,215
LY294002	VPS34 inhibition	Yes	In clinical trials for the treatment of refractory neuroblastoma	Exhibits improved selectivity compared with 3-MA and wortmannin but commonly considered nonspecific	84
Lys05	Lysosomal inhibition	Yes	In preclinical development	Exhibits increased potency compared with HCQ	294
Mdivi-1	Mitophagy inhibition	Yes	In preclinical development	Inhibitor of mitochondrial fragmentation	250
Melatonin*	Altered ROS production	Yes	In clinical trials for treatment of a wide panel of conditions	Potentially interferes with ROS-sensitive processes and has been associated with autophagy activation in some models	73
MRT67307	ULK1 inhibition	Unknown	In preclinical development	Also inhibits ULK2, IKK and TBK1	299
MRT68921	ULK1 inhibition	Unknown	In preclinical development	Also inhibits ULK2	299
NSC185058	ATG4B inhibition	Unknown	In preclinical development	Exhibits improved selectivity for the autophagic pathway	297
SAR405	VPS34 inhibition	Unknown	In preclinical development	Exhibits improved selectivity compared with 3-MA and wortmannin	295
SBI-0206965	ULK1 inhibition	Unknown	In preclinical development	Exhibits improved selectivity for the autophagic pathway	298
VPS34-IN1	VPS34 inhibition	Unknown	In preclinical development	Exhibits improved selectivity compared with 3-MA and wortmannin	296
Wortmannin	VPS34 inhibition	No	Experimental agent	Inhibits various class III PI3Ks	99

3-MA, 3-methyladenine; AMPK, AMP-activated protein kinase; ATG4B, autophagy-related 4B cysteine peptidase; ATRA, all-trans-retinoic acid; BECN1, beclin 1; CMA, chaperone-mediated autophagy; CRM, caloric restriction mimetic; HCQ, hydroxychloroquine; IFNγ, interferon-γ; IKK, inhibitor of nuclear factor-κB kinase; Ins(1,4,5)P₃, inositol-1,4,5-trisphosphate; MAPK, mitogen-activated protein kinase; MOA, mode of action; mTORC, mechanistic target of rapamycin complex; NA, not applicable; PI3K, phosphoinositide 3-kinase; ROS, reactive oxygen species; TBK1, TANK-binding kinase 1; ULK, UNC-51-like autophagy activating kinase.

^*Conflicting data exist on the ability of melatonin to modulate autophagy. Adapted with permission from REF. 9, Macmillan Publishers Limited.

Here, we discuss recent progress on the therapeutic potential of pharmacological and nutritional modulators of autophagy, dissect the obstacles that have limited the development of these interventions thus far, and propose strategies by which such hurdles may be circumvented in the near future to obtain clinically relevant interventions for a variety of human disorders.

Autophagy as a therapeutic target

Whole-body knockout studies in mice have demonstrated that specific components of the autophagic machinery are required for embryonic development or are critical for animals to survive birth and reach adulthood^14–16. Three main approaches have been pursued as alternatives to the use of whole-body knockout mice to study the role of autophagy in physiology and disease: the generation of animals with partial autophagic defects at the whole-body level (such as Becn1^+/− mice); the engineering of tissue-specific or inducible knockout models; and the restoration of autophagic activity in key organs (such as the central nervous system (CNS)) in animals with whole-body autophagic defects¹⁷. Results obtained with these models have implicated alterations of autophagy or autophagy-associated processes in a wide range of clinically relevant disorders (as discussed below), which supports the possibility that pharmacological modulators (activators or inhibitors) of autophagy may be beneficial for large patient populations.

Neurodegeneration

The deletion of autophagy-related 5 (Atg5) or Atg7 in the mouse CNS during embryonic development results in the delivery of viable pups that survive birth, but develop progressive motor and behavioural deficits starting at 3 weeks of age^18,19. The cortex and cerebellum of these animals exhibit swelling, markers of <m>regulated cell death</m> (RCD) and ubiquitin-containing inclusions^18,19, which are pathological hallmarks of various neurodegenerative disorders, including Alzheimer disease (AD), Parkinson disease (PD), dementia with Lewy bodies (DLB), Huntington disease (HD), amyotrophic lateral sclerosis (ALS) and Lafora disease²⁰. Autophagy robustly protects neurons from RCD by preventing the accumulation of cytotoxic protein aggregates and by preserving metabolic homeostasis²¹. In line with this idea, markers of impaired autophagy — such as activation of mechanistic target of rapamycin (mTOR), autophagosome accumulation and limited degradation of sequestosome 1 (SQSTM1; also known as p62) — have been detected in samples from patients with multiple forms of neurodegeneration²². Moreover, many of the genes that are mutated in familiar variants of PD — including parkin RBR E3 ubiquitin protein ligase (PARK2), Parkinsonism-associated deglycase (PARK7), PTEN-induced putative kinase 1 (PINK1) and leucine-rich repeat kinase 2 (LRRK2) — are involved in <m>mitophagy or aggrephagy</m>^23,24. Furthermore, AD-associated variants of <m>presenilin 1</m> (PSEN1) block autophagy as a result of impaired lysosomal acidification²⁵; mutations in SQSTM1, optineurin (OPTN, which encodes another <m>autophagic adaptor</m>) and TANK-binding kinase 1 (TBK1, which encodes a regulator of both p62 and OPTN) are common among individuals with familial and sporadic ALS^26–28; and both laforin glucan phosphatase (EPM2A) and NHL repeat-containing E3 ubiquitin protein ligase 1 (NHLRC1), which are mutated in individuals with Lafora disease, also seem to promote autophagy²⁹. Finally, mutations in WD repeat domain 45 (WDR45) — which encodes an ATG9 interactor of the WD repeat domain phosphoinositide-interacting (WIPI) family — have been shown to be involved in the pathogenesis of static encephalopathy of childhood with neurodegeneration in adulthood, which is a rare neurological disorder³⁰. Interventions that promote autophagy or autophagy-associated processes have been shown to mediate beneficial effects in animal models of neurodegeneration.

Alzheimer disease

Administration of the mTOR inhibitor rapamycin — which potently activates autophagy — ameliorates cognitive deficits and alleviates the accumulation of β-amyloid in mice expressing mutant <m>myloid-β precursor protein</m> (APP)³¹ as well as in 3xTg-AD mice (which bear three distinct genetic alterations that are associated with AD in humans)³². Along similar lines, resveratrol, which is a natural polyphenol that promotes autophagy by operating as a <m>caloric restriction mimetic</m> (CRM), decreased β-amyloid overload in mice expressing a chimeric variant of mutant APP and AD-associated human PSEN1. This effect was ascribed to the AMP-activated protein kinase (AMPK)-dependent inhibition of mTOR complex 1 (mTORC1)³³ (BOX 2). Of note, behavioural alterations that develop in mice engineered to express one or several APP mutations that are linked to AD in humans could also be ameliorated by the concomitant deletion of genes that encode endogenous inhibitors of lysosomal proteases, such as cystatin B (Cstb) or cystatin C (Cst3)^34,35.

Parkinsonism

The intracerebral injection of a lentiviral vector encoding ATG7 or BECN1 decreases neuronal inclusions in synuclein-α (SNCA)-expressing mice (a model of PD and DLB), and this has been associated with reduced neurodegeneration^22,36. Along similar lines, the intracerebral administration of an adenoviral vector encoding BECN1 or transcription factor EB (TFEB) — which is a master transcriptional regulator of autophagy and lysosomal functions — to rats expressing SNCA in the brain limited the accumulation of SNCA aggregates within dopaminergic neurons and prevented behavioural impairment³⁷. Comparable results have been obtained by activating autophagy with systemic or intracerebral administration of rapamycin, <m>trehalose</m> or <m>valproate</m> in several mouse models of PD and DLB, including SNCA-expressing mice²², mice expressing mutant Park2 (REF. 38), Park2^−/− mice expressing human microtubule-associated protein tau (MAPT)³⁹ as well as mice or rats that develop parkinsonism upon administration of 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP)^40–42, 6-hydroxydopamine (6-OHDA)^43,44 or lactacystin⁴⁵. Rapamycin also alleviated neurotoxicity in Drosophila melanogaster with mutations in parkin (park; the fly orthologue of human PARK2) or Pink1 (REF. 46) as well as in flies treated with the dopaminergic toxin paraquat⁴⁷. Of note, defects in late-stage autophagy leading to PD-like and DLB-like disorders in mice have also been ascribed to loss of type I interferon signalling⁴⁸. Accordingly, intracerebral interferon β1 (Ifnb1) delivery by a lentiviral vector boosted autophagy and limited the loss of dopaminergic neurons in rats expressing human SNCA in the brain⁴⁸. These findings identify cytokine signalling as a potential target for the treatment of PD and DLB through the induction of autophagy. Interestingly, defects in CMA may also be implicated in the development of PD and DLB, as was demonstrated recently in rats⁴⁹. Whether boosting CMA ameliorates the manifestations of disease in animal models of parkinsonism remains to be elucidated.

Huntington disease

In flies and mice expressing HD-associated variants of human huntingtin, rapamycin (and other mTOR inhibitors, including temsirolimus and everolimus) alone or combined with <m>lithium</m> exerts considerable neuroprotective effects, as has been determined histologically and in behavioural tests^50,51. However, the adenovirus-mediated intracerebral delivery of either of two mTORC1 activators — namely, RASD family member 2 (RASD2; also known as RHES) and RAS homologue enriched in brain (RHEB) — in a constitutively active form alleviated metabolic, histological and behavioural manifestations of the disease in HD-prone mice⁵². mTORC1 activation by RHEB was paradoxically associated with an increase in multiple biomarkers of autophagy⁵². Although this was not formally investigated, RHEB and/or RHES may be involved in the regulation of autophagy or autophagy-associated processes that are independent or downstream of mTORC1 (TABLE 2).

Table 2.

Pathologies potentially responding to autophagy-modulatory interventions

Disease	Approach	Observations	Refs
Acute brain intoxication	Debated	• Role of autophagy in neuronal responses to toxicants has not been studied in appropriate models	9,58–60
Ageing	Activation	• Most lifespan-extending interventions activate autophagy, and their beneficial effects depend on an intact autophagic machinery in multiple model organisms	47,229–240
Atherosclerosis	Activation	• Efficient autophagic responses in macrophages from arterial lesions limit disease progression • Coronary angioplasty generally involves rapamycin-eluting stents	111
Autoimmune disorders	Debated	• LAP defects cause an SLE-like disease in mice • Autophagy activators mediate therapeutic effects in multiple autoimmune diseases • Autophagy supports cell populations or processes that underlie disease	166,167,210–228
Bacterial infections	Activation	• Elimination of intracellular bacteria relies on xenophagic responses, which couple danger signalling to autophagosome formation	10,168–190
Cancer	Debated	• Autophagy inhibition may exacerbate effect of cytotoxic therapies • Autophagy underlies the activation of therapeutically relevant immune responses	8,15,16,152–167
Cardiac stroke	Debated	• Both autophagy activators and the Becn1+/− genotype are associated with cardioprotection in models of stroke	110,115–117,22,125
Cardiomyopathy	Debated	• Autophagic defects in cardiomyocytes provoke cardiomyopathies • Autophagy activators mediate beneficial effects in multiple disease models	107–109,112–114, 118–121,124,126
Cerebral stroke	Debated	• There is conflicting literature on the impact of autophagy on disease outcome in models of 4VO, pMCAO and tMCAO	9,73–89
COPD	Inhibition	• Mitophagy may participate in the pathogenesis of cigarette smoke-associated COPD	248–251
Diabetes	Debated	• Autophagy activation improves insulin sensitivity in type 2 diabetes models • Insulin release by β-cells is negatively regulated by autophagy during fasting	130,141,145–150
Eukaryotic parasites	Debated	• LAP activation may control fungal and eukaryotic parasites	10,200–208
Hepatic disorders	Activation	• Autophagy activators limit hepatic fat accumulation and damage in models of steatosis and α1-antitrypsin deficiency-associated liver disease	127,129,133–137, 140,142,143
Lysosomal storage disorders	Debated	• The inhibition of autophagy upstream of autophagosome formation may ameliorate lysosomal overload. Similar effects may be achieved with autophagy activators that overcome lysosomal blockage	262–265
Metabolic syndrome	Activation	• Exercise links intact autophagic responses in multiple organs with leanness and improved systemic metabolism	7,128–133,135,140, 141,144,145,151
Neonatal asphyxia	Inhibition	• Pharmacological and genetic inhibition of autosis or autophagy mediates neuroprotective effects	9,72,90–94
Neurodegenerative disorders	Activation	• Autophagy is inhibited in the ageing brain, which contributes to the accumulation of pathogenic and pathognomonic neuronal inclusions	18–20, 22–57
Neurotrauma	Debated	• Autophagy activation limits functional impairment and promotes recovery in models of SCI and SAH, whereas data are conflicting in models of TBI	95–105
Pulmonary disorders other than COPD	Activation	• Autophagic defects have been documented in patients with certain pulmonary disorders. Preclinical data support the benefit of autophagy activation in mouse models of cystic fibrosis, IPF and PAH	241–247
Renal conditions	Activation	• Deletion of Atg5 or Atg7 in podocytes induces glomerulosclerosis and aggravates renal degeneration that is caused by PAN administration or kidney overload	252–255
Seizures	Activation	• Preclinical data and epidemiological studies link excitotoxicity to defective autophagic responses	61–73
Skeletal conditions	Debated	• Autophagy is crucial for bone growth • Administration of rapamycin has been associated with both beneficial and detrimental effects in mouse models of bone fracture	256–260
Viral infections	Debated	• Preclinical data link autophagic responses to improved cellular and organismal control of viruses • Some viruses exploit autophagic responses in the host	191–199
Vision disorders	Activation	• Efficient autophagic responses and LAP are required for the survival and function of retinal photoreceptors	266–269,271–273

4VO, 4-vessel occlusion; Atg, autophagy-related; Becn1, beclin 1; COPD, chronic obstructive pulmonary disease; IPF, idiopathic pulmonary fibrosis; LAP, LC3-associated phagocytosis; PAH, pulmonary arterial hypertension; PAN, puromycin aminonucleoside; pMCAO, permanent middle carotid artery occlusion; SAH, subarachnoid haemorrhage; SCI, spinal cord injury; SLE, systemic lupus erythematosus; TBI, traumatic brain injury; tMCAO, transient middle carotid artery occlusion.

Amyotrophic lateral sclerosis

Activation of autophagy with caloric restriction, trehalose, valproate or lithium delays disease onset, reduces neurological deficits and prolongs survival in mice expressing an ALS-associated variant of superoxide dismutase 1 (SOD1), namely SOD1-G93A^53–55. However, rapamycin had detrimental effects on motor neuron degeneration and overall survival in mice expressing SOD1-G93A⁵⁶. Moreover, crossing these mice with Becn1^+/− mice resulted in a paradoxical increase in lifespan that was accompanied by p62 accumulation and an unexpected interaction between SOD1-G93A and BECN1 (REF. 57). The reasons that underlie these contrasting observations remain to be elucidated. At least theoretically, when lysosomal degradation is congested, limiting the initiation of autophagy may be more beneficial than attempting to boost autophagic responses with interventions that fail to overcome the block in autophagosomal processing (see below). It remains to be determined whether this applies to ALS (TABLE 2).

Acute neuronal injury

Acute challenges to the CNS, including drug and ethanol intoxication, seizures, adult stroke, neonatal asphyxia and traumatic brain injury (TBI), have been associated with biomarkers of ongoing or blocked autophagic responses (see below), which suggests that autophagy modulators might provide therapeutic benefits⁹. However, a clear aetiological link between failing autophagic responses and acute neuronal injury has not yet been confirmed, not only owing to methodological issues but also owing to an inherent heterogeneity in models (see below).

Acute intoxication

Autophagic responses have been documented in the cortex of mice that were exposed to cocaine⁵⁸, but other neuronal populations, including the nucleus accumbens, respond to systemic cocaine with mTORC1 activation (and hence autophagy inhibition)⁵⁹. Depletion of mTOR or regulatory-associated protein of mTOR complex 1 (RPTOR; a key component of mTORC1) from specific neuronal populations reduced <m>locomotor sensitization</m> (one of the symptoms of cocaine administration) in mice⁵⁹, as did rapamycin administration in rats⁶⁰. By contrast, inhibition of autophagy with small interfering RNAs (siRNAs) targeting Atg5 or Becn1 limited the capacity of cocaine to trigger RCD in mouse primary cortical neurons⁵⁸. Similar results were obtained in vitro with chemical inhibitors of autophagy, including 3-methyladenine (3-MA) and wortmannin, which target phosphatidylinositol 3-kinase catalytic subunit type 3 (PIK3C3; also known as VPS34), as well as the lysosomal inhibitor bafilomycin A1 (BafA1)⁵⁸. Thus, some aspects of cocaine neurotoxicity (for example, cocaine-driven RCD) may be aggravated, whereas others (for example, locomotor sensitization) may be alleviated, by efficient autophagic responses. As the pathogenesis of acute cocaine intoxication involves multiple cell-extrinsic components (for example, neuroinflammation), great caution should be taken before extrapolating in vitro data to human disease. Likewise, numerous studies suggest that autophagic responses are beneficial in the course of acute brain intoxication with methamphetamine, ethanol or olanzapine (an antipsychotic drug)⁹. However, studies to elucidate the relationship between autophagy or autophagy-associated processes and acute neuronal responses to toxicants in vivo are urgently awaited to clarify the potential of modulating autophagy for therapeutic purposes in this setting (TABLE 2).

Seizures

One of the mechanisms by which seizures mediate neurotoxic effects involves the sustained depolarization of postsynaptic terminals and consequent influx of Ca²⁺ ions from the synaptic cleft, which is a process that is commonly known as excitotoxicity⁶¹. Cytosolic Ca²⁺ is a potent activator of autophagy, and several excitotoxic neurotransmitters — including glutamate, N-methyl-D-aspartate (NMDA) and kainic acid — induce the expression of autophagy biomarkers in the rodent brain, possibly as a compensatory response to damage^62,63. The depletion of BECN1 by siRNAs as well as pharmacological inhibitors of autophagy aggravated (whereas rapamycin and trehalose limited) the demise of primary rodent neurons and multiple neuronal cell lines of human origin that were exposed to glutamate, NMDA or kainic acid in vitro^64,65, which suggests that autophagy supports neuronal viability in the course of excitotoxic challenges. Further corroborating this idea, mutations in three different genes encoding endogenous inhibitors of mTORC1 — namely, tuberous sclerosis 1 (TSC1), TSC2 and phosphatase and tensin homologue (PTEN) — are associated with an increased predisposition to epilepsy in humans⁶⁶. Accordingly, mice lacking Tsc1, Tsc2, Pten or Atg7 in neuronal or glial cell populations experienced spontaneous epileptic episodes that were associated with premature death, which is an outcome that could be limited by systemic rapamycin administration^67–70. However, mice receiving an siRNA targeting Atg7 in the hippo campus were more resistant to the neurotoxic effects of kainic acid than mice receiving a control siRNA⁷¹. Moreover, the stable depletion of Becn1 or Atg7 expression as well as the administration of 3-MA, LY294002 (another VPS34-targeting agent) or lysosomal inhibitors limited the death of cultured rodent neurons from the striatum, cerebellum and cortex that were challenged with NMDA or kainic acid^72,73. Further experiments are required to understand whether autophagy modulators may indeed benefit patients with seizures and to what extent this may involve autophagy-associated processes (TABLE 2).

Adult stroke

Tissues that are served by an occluded artery experience nutrient deprivation and hypoxia, which are two potent activators of autophagy, and this is potentially followed by excessive reactive oxygen species (ROS) production (at reperfusion), which has also been associated with autophagy activation⁹. Consistent with this, biomarkers of autophagy activation have been detected in the brain of adult rodents experiencing stroke⁹. The intracerebral delivery of a lentiviral vector encoding a Tsc1-targeting short hairpin RNA (shRNA) aggravated neuronal loss in rats experiencing permanent middle carotid artery occlusion (pMCAO), which supports the notion that autophagy mediates neuroprotective effects in the course of a stroke⁷⁴. Similarly, inhibition of autophagy through deletion of Prkaa2 (which encodes AMP-activated, α2 catalytic subunit) or sirtuin 1 (Sirt1), or by downregulation of ATG7 or TSC1, aggravated the cytotoxicity of oxygen glucose deprivation (OGD) in primary mouse⁷⁵ or rat^74,76 cortical neurons. In addition, autophagy activation (with rapamycin) or inhibition (with 3-MA, BafA1 or AMPK inhibitors) improved or worsened, respectively, disease outcome in rodents that were experiencing transient middle carotid artery occlusion (tMCAO) or pMCAO^76,77, as well as neuroprotection in multiple models of <m>ischaemic preconditioning</m> in vivo^78–82. Stable downregulation of Becn1 in the rat brain by stereotactic injection of a lentiviral vector limited the neurotoxic effects of tMCAO⁸³. Moreover, both 3-MA and BafA1 limited infarct size in various rodent models of tMCAO^84,85, pMCAO^86,87 or four-vessel occlusion⁸⁸, and multiple molecules that protect neurons in adult rodents experiencing stroke (for example, melatonin and edavarone) also seem to inhibit autophagy^73,89, possibly through their antioxidant effects. Thus, although autophagy seemingly participates in the pathophysiology of stroke in adults, whether therapeutic interventions should aim to activate or inhibit autophagy remains to be clarified. Although not yet demonstrated, the time at which autophagy modulators are administered to a patient experiencing stroke (for example, before versus after reperfusion) may considerably influence their therapeutic effects (TABLE 2).

Neonatal asphyxia

Although it has been suggested that autophagy would be activated as an adaptive response to neonatal asphyxia⁹⁰, accumulating evidence indicates that autophagy contributes mechanistically to neuronal loss in the course of neonatal asphyxia⁹. Newborn rats that received a Becn1-specific shRNA intracerebrally displayed twice the amount of intact striatal tissue 24 hours after ischaemia compared with their control counterparts (that received a non-targeting shRNA)⁷². In addition, the neuron-specific knockout of Atg7 in mice provided a high degree of long-term protection from neonatal asphyxia to hippocampal cornu ammonis regions⁹¹. Similarly, the hippocampus, thalamus, cortex and striatum of newborn mice lacking Atg7 in neurons were considerably protected from RCD following hypoxia–ischaemia, which corresponds to a global decrease in tissue loss of approximately 40%⁹². Finally, newborn rats treated with intracerebroventricular 3-MA up to 4 hours after ischaemia exhibited considerably smaller lesions than did control animals⁹³, as did newborn rats receiving the <m>cardiac glycoside</m> neriifolin, which is an inhibitor of autosis (BOX 1), immediately after carotid artery occlusion⁹⁴. In summary, inhibiting autophagy is expected to limit the severity of hypoxic–ischaemic encephalopathy in newborn rodents (TABLE 2).

Neurotrauma

Consistent with the hypothesis that autophagy is beneficial, but usually impaired, in the context of neurotrauma, rapamycin and other molecules with autophagy-inducing potential (for example, melatonin and retinoic acid) limit CNS damage, support regeneration and improve the restoration of neuromuscular functions in rodents experiencing spinal cord injury (SCI)⁹⁵ or subarachnoid haemorrhage (SAH)^96–98. Chemical inhibitors of autophagy, including 3-MA, wortmannin and the antimalarial drug chloroquine (which blocks lysosomal degradation), aggravated neurological damage imposed by SCI or SAH^96,99, and abolished the neuroprotective effects of autophagy inducers^98,100. Activating autophagy with rapamycin or melatonin also had beneficial effects in rodents experiencing several forms of TBI, including closed-head injury¹⁰¹, weight-drop damage¹⁰² and hemicerebellectomy¹⁰³, even when the drug was administered after traumatic injury. In this latter model, neuroprotection by rapamycin was lost in Becn1^+/− mice, which were a priori more sensitive to hemicerebellectomy-induced damage¹⁰³. Of note, melatonin has been suggested to inhibit autophagy in other settings⁷³. Moreover, post-injury chloroquine administration limited neuronal damage and improved neurological recovery in rats that were subjected to controlled cortical impact¹⁰⁴, as did 3-MA and BafA1 administered as a prophylactic intervention in a mouse weight-drop model¹⁰⁵. Thus, autophagy activators may be beneficial for patients experiencing SCI or SAH, even if administered in a therapeutic (as opposed to prophylactic) setting. It remains to be clarified whether the same holds true for other forms of neurotrauma, including TBI (TABLE 2).

Cardiovascular conditions

The efficient disposal of various autophagic substrates seems to be crucial for the maintenance of cardiovascular homeostasis, in both physiological and pathological conditions¹⁰⁶. For example, ageing Lamp2^−/− mice, which display defective CMA, develop a disorder that is characterized by cardioskeletal myopathy similar to that associated with Danon disease, and patients with this condition exhibit LAMP2 defects^107,108. Similarly, the temporally controlled deletion of Atg5 from mouse cardiomyocytes promoted cardiac hypertrophy¹⁰⁹, and the hearts of Park2^−/− mice, which are mitophagy-incompetent, failed to benefit from cardiac ischaemic preconditioning ex vivo¹¹⁰. In addition, the conditional knockout of Atg5 in macrophages of Ldlr^−/− mice fed a high-fat diet (HFD) — which are prone to accumulate atherosclerotic plaques — aggravated arterial lesions (by facilitating apoptotic and necrotic RCD, and by worsening <m>efferocytosis</m>¹¹¹). Furthermore, the cardiomyocyte-specific deletion of Dnase2a (which encodes a lysosomal nuclease that is involved in the autophagic degradation of mitochondrial DNA) sensitized mice to pressure overload-driven hypertrophy that was accompanied by a robust inflammatory response¹¹². Finally, the Becn1^+/− genotype aggravated the disease progression in mice overexpressing mutant crystallin αB (CRYAB) in the heart (which is a model of desmin-related cardiomyopathy)¹¹³. Conversely, the cardiomyocyte-specific overexpression of ATG7 reduced biochemical and functional biomarkers of the disease in this model, as did physical exercise (which is an established activator of autophagy)¹¹⁴. Consistent with this, autophagy activators — including (but not limited to) caloric restriction, physical exercise, rapamycin, <m>spermidine</m> and a peptide derived from the BECN1 region that interacts with the HIV-1 protein Nef (whose mechanism of action has not been characterized yet) — had beneficial effects in models of myocardial ischaemia reperfusion^115–117, pressure overload-driven hypertrophy or heart failure^118–120, and cardiac senescence¹²¹. Notably, the devices that are currently used in the clinic for <m>coronary angioplasty</m> generally deliver rapamycin¹²², although the underlying rationale resides in the antiproliferative activity of this drug¹²³. Moreover, endurance exercise may preserve cardiovascular and metabolic fitness by activating autophagy in multiple organs⁷. Interestingly, the Becn1^+/− genotype was associated with some extent of cardioprotection in the setting of myocardial ischaemia reperfusion, pressure overload-driven hypertrophy and diabetic cardiomyopathy, which the authors ascribed to reduced maladaptive autophagy^124–126. Data obtained in mice subjected to the inducible cardiomyocyte-specific deletion of Atg5, however, argue against these findings¹⁰⁹, which highlights the potential bias introduced by the organismal adaptation to life-long gene knockout (see below).

In summary, activation of autophagy seems to be a main goal not only for the preservation of cardiovascular fitness but also for the management of multiple cardiovascular disorders (TABLE 2).

Hepatic and metabolic disorders

Several reports suggest that the cell-intrinsic and cell-extrinsic functions of autophagy in the liver and pancreas are crucial not only for the maintenance of organ homeo stasis but also for the regulation of neuroendocrine circuitries that control systemic metabolism¹²⁷. Thus, acutely activating autophagy with caloric restriction, physical exercise, rapamycin, CRMs (that is, <m>metformin</m> or resveratrol), AMPK-targeting agents or hydrogen sulfide limits non-alcoholic steatohepatitis (NASH) and/or improves various metabolic parameters (including body weight, circulating glucose or triglyceride levels, and insulin sensitivity) in 24-month-old rats¹²⁸, rodents that had been fed a HFD^129–133, mice acutely or chronically exposed to ethanol¹³⁴ as well as in <m>db/db mice</m> or <m>ob/ob mice</m> (which are two genetic models of hyperphagia and thus metabolic syndrome)^130,135,136. Moreover, activation of autophagy with carbamazepine considerably reduced hepatic fibrosis in a model of α1<m>-antitrypsin deficiency</m>-associated liver disease¹³⁷. However, rapamycin administered according to specific schedules also causes insulin resistance as it inhibits mTORC2 (REFS 138,139). This suggests that adequate administration schedules or alternative autophagy activators are preferable for the treatment of metabolic disorders that are linked to type 2 diabetes.

An Atg7-targeting shRNA delivered by an adenoviral vector, an Atg7-targeting siRNA that was injected intravenously, the whole-body knockout of forkhead box O3 (Foxo3, which encodes a transcription factor that supports autophagic responses), or the conditional co-deletion of Prkab1 and Prkab2 (which encode the AMPK subunit β1 and the AMPK subunit β2, respectively) in adipocytes aggravated hepatic damage and metabolic dysregulation in HFD-fed mice¹⁴⁰, ob/ob mice¹⁴¹ and mouse models of ethanol intoxication^142,143, which further corroborates the beneficial role of autophagic responses in hepatic and metabolic conditions. The hepatocyte-specific deletion of serine/threonine kinase 11 (Stk11, which encodes the main AMPK activator and is also known as Lkb1) compromised glucose homeostasis in mice and rendered them insensitive to metformin¹⁴⁴. Furthermore, mice expressing a non-phosphorylatable mutant BCL-2, apoptosis regulator (BCL-2) — a model in which inducible autophagy is selectively blocked — were unable to obtain metabolic benefits from physical exercise⁷, whereas ob/ob mice specifically lacking Atg7 in the myeloid compartment were more susceptible to obesity-induced diabetes than their autophagy-proficient littermates¹⁴⁵. These findings are intriguing, as they suggest that systemic metabolic homeostasis may (at least partially) rely on efficient autophagic responses in compartments other than the liver, pancreas and adipose tissue. Interestingly, acute caloric restriction, rapamycin and resveratrol also attenuated multiple manifestations of the diabetic syndrome induced by <m>streptozotocin</m> — including nephropathy — in mice and rats^146–148. In addition, Atg7 was required for pancreatic β-cells to develop normally and to ensure physiological glucose control¹⁴⁹. However, autophagic responses within pancreatic β-cells seem to contribute to the physiological inhibition of insulin release by fasting¹⁵⁰. It remains to be elucidated whether autophagy inhibitors may support insulin secretion and systemic glucose control in patients with type 1 diabetes (which is characterized by a primary defect in insulin secretion). Finally, the deletion of Atg5 or Atg12 in UCP1⁺ adipocytes prevented the beige-to-white fat transition in mice, hence limiting HFD-driven obesity and glucose intolerance¹⁵¹. This inhibition of beige-to-white fat transition suggests that — although autophagy seems to globally support metabolic fitness — the development, survival or functions of some detrimental cells (such as white adipocytes) may also rely on efficient autophagic responses.

Taken together, these observations suggest that systemic autophagy activators may mediate therapeutic activity in patients with a variety of metabolic disorders, although the inhibition of autophagic responses in specific cell compartments might amplify such a beneficial effect (TABLE 2).

Cancer

Autophagic responses contribute to preservation of homeostasis and adaptation to stress in both normal and malignant cells¹⁵². Thus, autophagy has been shown to inhibit <m>malignant transformation</m> in a variety of models and by a multitude of mechanisms¹⁵². Accordingly, Becn1^+/− mice were more prone to developing spontaneous malignancies as they aged than were their wild-type littermates^15,16. Mice with a systemic mosaic deletion of Atg5 or the liver-specific knockout of Atg7 spontaneously accumulated benign liver adenomas¹⁵³, and the local deletion of Atg5 or Atg7 markedly accelerated the onset of KRAS-G12D-driven or BRAF-V600E-driven pancreatic or pulmonary adenomas in mice^154–156. In addition, multiple <m>oncosuppressor genes</m> — including tumour protein 53 (TP53) and PTEN — support autophagic responses, whereas several <m>proto-oncogenes</m> — such as BCL2, AKT serine/threonine kinase 1 (AKT1) and epidermal growth factor receptor (EGFR) — inhibit them^157–160. However, autophagy also promotes <m>tumour progression</m> and resistance to treatment, at least at the cancer cell-intrinsic level, through a multitude of mechanisms¹⁵². Thus, the conversion of early KRAS-G12D-driven or BRAF-V600E-driven pancreatic or pulmonary adenomas into advanced, invasive adenocarcinomas was attenuated in the context of local Atg5 or Atg7 deletion^154–156. Along similar lines, the stable depletion of ATG5, ATG7 or BECN1 with shRNA-coding constructs limited the growth of multiple human cancer cells that were subcutaneously or orthotopically xenografted into athymic immunodeficient mice¹⁶¹. Furthermore, human cancer cells that were implanted in immunodeficient hosts were more sensitive to chemotherapy or radiotherapy in the presence of pharmacological inhibitors of autophagy, including 3-MA, wortmannin, chloroquine and HCQ (which is another antimalarial agent)¹⁶¹. Taken together, these findings suggested that autophagy inhibitors would be useful agents for the clinical management of cancer, either as a standalone intervention or as a means to sensitize malignant cells to therapy¹⁶².

Recent clinical trials testing chloroquine or HCQ (alone or combined with chemotherapy or radiotherapy) in patients with lymphoma, melanoma, glioblastoma and other solid neoplasms established the safety of this approach¹⁶¹, which fostered the initiation of additional phase II and phase III clinical studies in Europe and the United States (ClinicalTrials.gov). However, none of these clinical trials has formally confirmed the hypothesis that inhibiting autophagy in cancer cells provides therapeutic benefits to patients with cancer¹⁶¹. Moreover, preclinical findings indicate that intact autophagic responses in malignant cells are required for appropriate danger signalling (and hence for the initiation of therapeutically relevant anticancer immune responses) in tumours established in syngeneic immunocompetent hosts and treated with immunogenic chemotherapy or radiotherapy^8,163. Consistent with this, caloric restriction and various CRMs enhanced (rather than limited) the therapeutic efficacy of mitoxantrone, oxaliplatin and radiotherapy in the same tumour models^161,163,164. Furthermore, biomarkers of autophagic responses in malignant cells were associated with intensified <m>immunosurveillance</m> and improved disease outcome in cohorts of patients with breast carcinoma who were treated with anthracyclines¹⁶⁵. Finally, local as well as systemic inhibition of autophagy could have short-term and long-term detrimental effects in patients with cancer for two reasons. First, autophagy is important for the survival, proliferation and effector functions of immune cell subtypes that are involved in tumour control^166,167. Second, at least hypothetically, inhibiting autophagy may increase the risk of healthy tissues to undergo malignant transformation or experience other toxic effects. Thus, the activation of autophagy with safe nutritional measures stands out as a promising approach to improve the clinical efficacy of anticancer agents that operate (at least in part) by promoting tumour-specific immune responses (TABLE 2).

Infectious diseases

Autophagy is required for the cellular and organismal control of multiple pathogens, including bacterial, viral and eukaryotic parasites (such as fungi).

Bacterial infections

Autophagic responses that are specific for cytoplasmic bacteria (referred to as xenophagy) are a crucial component of the innate immune system and have been shown to restrict the growth of bacterial pathogens, including Salmonella enterica subsp. enterica serovar Typhimurium¹⁶⁸, Mycobacterium tuberculosis^169,170, Listeria monocytogenes¹⁷¹ and group A Streptococcus spp.¹⁷². Accordingly, many bacteria have evolved strategies to inhibit autophagic responses in the host. These include (but are not limited to) the production of cAMP-elevating toxins (Vibrio cholera and Bacillus anthracis)¹⁷³, the normalization of otherwise dwindling amino acid levels at the surface of bacterium-containing vacuoles (S. Typhimurium)¹⁷⁴, the deconjugation of microtubule-associated protein 1 light chain 3β (MAP1LC3B; also known as LC3) (Legionella pneumophila)¹⁷⁵, the inactivation of GTPases that are required for normal vesicular trafficking (Shigella flexneri and pathogenic Escherichia coli)¹⁷⁶ and the escape from autophagic recognition (L. monocytogenes)¹⁷⁷. Activation of autophagy through starvation or treatment with rapamycin, a BECN1-derived peptide or other agents restricted bacterial growth and improved cellular or organismal resistance to infection caused by M. tuberculosis (in D. melanogaster and mouse macrophages)^169,178, S. enterica (in human cancer cell lines)¹⁷⁹, or pathogenic E. coli (in mice and human cancer cell lines)^180,181. Moreover, autophagic responses to carbon monoxide protected mice from sepsis induced by cecal ligation and puncture¹⁸². Of note, LAP (BOX 2) is also involved in the control of intracellular bacteria, including Burkholderia pseudomallei and L. monocytogenes, by monocytes^10,183,184.

Interestingly, the ability of mice to control M. tuberculosis infection was partially abrogated by the Park2^−/− genotype (which imposes a selective defect on mitophagy)¹⁸⁵ as well as by the conditional deletion of Atg5 (but not other autophagy-related genes) from monocytes and neutrophils, possibly as a consequence of exacerbated lung inflammation¹⁸⁶. Thus, caution should be taken when extrapolating data obtained from single-gene knockouts to entire cellular processes (see below). Nevertheless, activators of autophagy, mitophagy and LAP may constitute promising tools for the clinical management of some bacterial infections, whereas molecules with unsuspected autophagy-inhibitory functions may be detrimental (such as azithromycin for patients with cystic fibrosis)¹⁸⁷. That said, other pathogenic bacteria, including Anaplasma phagocytophilum (which causes a tick-borne disease with relatively mild symptoms) and Coxiella burnetii (which causes a severe endocarditis), stimulate autophagy in the host to support their own metabolic needs^188,189. In this situation, autophagy inhibitors (including HCQ) have been shown to provide some clinical benefits¹⁹⁰ (TABLE 2).

Viral infections

Several viruses are efficiently controlled by autophagic responses in host cells¹⁹¹. The autophagic degradation of viruses, which is commonly referred to as ‘virophagy’, relies on core components of the autophagic machinery, including BECN1 (REF. 192), as well as on proteins that also participate in mitophagy, such as SMAD-specific E3 ubiquitin protein ligase 1 (SMURF1) and Fanconi anaemia complementation group C (FANCC)^191,193. Consistent with an important role of virophagy in the control of viral infections, some viruses evolved virulence factors that actively inhibit autophagy, such as the BECN1 inhibitor ICP34.5 from herpes simplex virus 1 (HSV-1)¹⁹⁴. Interestingly, Becn^+/− and Fancc^−/− mice were more susceptible to Sindbis virus infection than their wild-type littermates, and D. melanogaster that had received an Atg18-targeting shRNA exhibited increased sensitivity to vesicular stomatitis virus infection^192,193,195. Futher findings corroborate the potential therapeutic value of autophagy activators for the control of viral infections: rapamycin and a BECN1-derived peptide efficiently limited HIV-1 replication in a human lymphoblastoid cell line¹⁹⁶, in primary human monocyte-derived macrophages¹⁹⁷ and in peripheral blood lymphocytes (PBLs) from healthy donors^196,198 as well as in severe combined immunodeficient (SCID) mice reconstituted with human PBLs (which is a model for the study of HIV-1 infection in vivo)¹⁹⁸. The same BECN1-derived peptide restrained viral replication and improved overall survival in mouse models of West Nile virus and chikungunya virus¹⁹⁷. Specific viruses, such as coxsackievirus B3, however, may have a replication advantage linked to autophagy activation, as was demonstrated in mice with a conditional deletion of Atg5 in pancreatic acinar cells¹⁹⁹ (TABLE 2).

Eukaryotic parasites

Mice lacking Atg5 or Atg7 specifically in the myeloid cell compartment are more susceptible to an intravenous challenge with Candida albicans (but not with Cryptococcus neoformans) than their wild-type counterparts, possibly as a result of defective neutrophil recruitment^200,201. Similarly, mice bearing Atg7-deficient or Becn1-deficient macrophages, as well as Rubcn^−/− mice, were less resistant to intra-nasal Aspergillus fumigatus infection than their control litter mates¹⁰. Moreover, some fungal pathogens, such as A. fumigatus, seem to have evolved strategies to block LAP, including the production of a cell wall component that specifically inhibits the activation of NADPH oxidases at the phagosome²⁰². Atg5, Atg7 and autophagy-related 16-like 1 (Atg16l1) in macrophages were also required for mice to control infection caused by the eukaryotic parasite Toxoplasma gondii, although such a process was not accompanied by autophagic degradation^203–205. These data point to an essential contribution of autophagy and, to a greater degree, LAP in the control of fungal infections and eukaryotic parasites. Interestingly, two commonly used inhibitors of autophagy (chloroquine and HCQ) have long been used for the treatment of malaria, which is caused by the eukaryote Plasmodium falciparum²⁰⁶. Chloroquine and HCQ, however, inhibit both the autophagic and non-autophagic functions of lysosomes, and have been proposed to exhibit antimalarial properties through inhibition of haem polymerase, which is toxic to the parasite²⁰⁷. Taken together, these observations suggest that autophagy activators may be beneficial for the treatment of fungi and other eukaryotic parasites (TABLE 2). However, it should be borne in mind that autophagy is conserved across the eukaryotic kingdom, which implies that pathogenic eukaryotes may also take advantage of enhanced autophagic responses²⁰⁸. In this setting, specific LAP activators may represent superior therapeutic tools, and efforts should be dedicated to the development of such molecules.

Of note, autophagy engages in extensive crosstalk with Toll-like receptor signalling and plays a crucial part in antigen cross-presentation²⁰⁹, which together underlie optimal innate and adaptive immunity against bacteria, viruses and eukaryotic parasites. Thus, autophagy activators may also support the eradication of invading pathogens by promoting innate and adaptive immune responses.

Inflammatory and autoimmune conditions

Autophagy mediates prominent anti-inflammatory effects, which reflects its ability to degrade inflammasomes as well as to limit the availability of endogenous inflammasome activators, including ROS and mitochondrial DNA^210,211. However, intact autophagic responses may also support the survival, proliferation and activity of multiple cells that contribute to the aetiology of autoimmune disorders^166,167. Thus, although autophagy represents a promising target for the treatment of multiple inflammatory and autoimmune disorders, including (but not limited to) systemic lupus erythematosus (SLE), Crohn’s disease, rheumatoid arthritis and multiple sclerosis, the implementation of autophagy-modulatory interventions for the management of these pathologies may be less straightforward than was initially envisioned.

Systemic lupus erythematosus

Genetic polymorphisms in ATG5 and possibly ATG7 have been associated with SLE in multiple studies²¹², which suggests that autophagic defects may contribute to the pathogenesis of disease. Accordingly, mice lacking Atg5, Atg7 or Becn1 in LysM⁺ cells — which comprise macrophages, monocytes, some neutrophils and some dendritic cells — as well as Rubcn^−/− and Cybb^−/− mice (which are characterized by a specific defect in LAP) spontaneously developed an SLE-like autoimmune disorder linked to deficient phagocytic uptake of dying cells and consequent production of pro-inflammatory cytokines²¹³. Importantly, a similar phenotype is not observed in Ulk1^−/− mice and mice lacking RB1-inducible coiled-coil 1 (Rb1cc1) in LysM⁺ cells (which are two genetic alterations that provoke an autophagic defect but spare LAP)²¹³. Thus, similarly to wild-type mice, autophagy-deficient but LAP-competent mice do not develop an SLE-like disease and exhibit a normal phagocytic response to dying cells coupled to the production of anti-inflammatory mediators such as interleukin-10 (IL-10)²¹³. These findings suggest that pharmacological LAP activators may be beneficial for patients with SLE. It remains to be elucidated whether general autophagy activators may compensate for LAP defects. Of note, autophagy has been involved in the production of <m>neutrophil extracellular traps</m> (NETs), which contributes to SLE pathogenesis²¹⁴. Thus, the inhibition of autophagy in specific cell populations may also provide therapeutic benefits to patients with SLE. Indeed, chloroquine and HCQ have been used for the treatment of SLE with some success²¹⁵. However, this clinical activity probably reflects the immunosuppressive effects of these molecules rather than their capacity to block autophagy. Indeed, neither the cell populations nor the cellular processes that are targeted by HCQ have been formally investigated in the context of SLE (TABLE 2).

Crohn’s disease

Various non-synonymous polymorphisms that are associated with an increased susceptibility to Crohn’s disease (data on prevalence are not available for all polymorphisms and vary considerably in different studies) negatively affect the activity of proteins that participate in autophagic or xenophagic responses, such as ATG16L1, UNC-51-like autophagy-activating kinase 1 (ULK1) and nucleotide-binding oligomerization domain-containing 2 (NOD2)^216,217. This reflects the multifactorial aetiology of this disorder, which involves both a microbial or epithelial and an autoimmune component²¹⁸, and suggests that autophagy activators may be beneficial for patients with Crohn’s disease. Indeed, everolimus ameliorated disease severity in Il10^−/− mice, which spontaneously develop a Crohn’s disease-like syndrome²¹⁹. Moreover, everolimus has been used with some success in the management of specific SLE cases²²⁰. However, it remains to be determined whether the activity of everolimus in this context truly stems from the activation of autophagy or from its autophagy-independent immunosuppressive effects (TABLE 2).

Rheumatoid arthritis

There are contrasting observations about autophagy in CD4⁺ T cells in the context of rheumatoid arthritis. Naive CD4⁺ T lymphocytes from patients with rheumatoid arthritis have been reported to exhibit autophagic defects that are secondary to a metabolic reprogramming that affects glycolysis²²¹. By contrast, CD4⁺ T cells from patients with rheumatoid arthritis reportedly display increased autophagic responses, hence resembling CD4⁺ T lymphocytes that are activated in vitro²²². Whether this apparent discrepancy reflects the CD4⁺ T cell subset under consideration (naive versus total or activated CD4⁺ T cells) or a methodological bias remains unclear. Irrespectively, HCQ limits the resistance to apoptosis displayed by CD4⁺ T cells from patients with rheumatoid arthritis ex vivo and reduces disease severity in a mouse model of collagen-induced arthritis²²². In addition, the specific deletion of Atg5 from mouse CD4⁺ T cells limits their proliferation and activation ex vivo²²². Taken together, these observations suggest that the inhibition of autophagy in specific immune cell populations, notably CD4⁺ T cells, may limit disease progression in patients with rheumatoid arthritis (TABLE 2).

Multiple sclerosis

Rapamycin treatment limits skin and lung fibrosis in mouse models of multiple sclerosis induced by bleomycin administration or a hetero zygous mutation in fibrillin 1 (Fbn1), which is accompanied by reduced production of fibrogenic cytokines and decreased levels of hypergammaglobulinaemia and anti-DNA topoisomerase 1 antibodies (two circulating markers of disease)²²³. However, these findings were not mechanistically linked to the activation of autophagy, which implies that they could reflect the established antiproliferative and immunosuppressive activity of rapamycin. Conversely, the specific deletion of Atg7 from dendritic cells in mice ameliorates experimental autoimmune encephalomyelitis — which is a model of CD4⁺ T cell-dependent multiple sclerosis — by reducing the priming of autoimmune responses²²⁴. Consistent with this, chloroquine delays disease progression if administered before disease onset and reduces disease severity if administered after onset²²⁴. These findings suggest that the specific inhibition of autophagy in dendritic cells may constitute a desirable therapeutic objective for the management of some forms of multiple sclerosis (TABLE 2).

Other conditions

Autophagy may also be involved in the pathogenesis of other inflammatory and autoimmune conditions. Rapamycin administration provided beneficial effects in mouse models of autoimmune myositis²²⁵, autoimmune encephalomyelitis²²⁶ and autoimmune uveitis (only when used at high doses)²²⁷, as well as in patients with autoimmune lymphoproliferative syndrome²²⁸. Conversely, low-dose rapamycin administration aggravated experimental autoimmune uveitis in mice²²⁷. It remains to be clarified to what extent these findings relate to autophagy activation versus immuno-suppression by rapamycin (TABLE 2).

Ageing

The healthy lifespan of multiple model organisms, including Saccharomyces cerevisiae (yeast)^229,230, Caenorhabditis elegans (worm)^231–233, D. melanogaster (fly)^47,230, Mus musculus (mouse)^234–236 and Macaca mulatta (monkey)²³⁷, can be experimentally extended by autophagy-activating measures, and this lifespan extension almost invariably depends on an intact autophagic machinery²³⁸. Indeed, autophagy mediates robust homeostatic functions at both the cell-intrinsic and cell-extrinsic levels, which directly counteract several processes that are associated with ageing. These processes include (but may not be limited to): the accumulation of macromolecular damage that drives cellular senescence and RCD (which are particularly detrimental in the context of an aged stem cell compartment); systemic metabolic deregulation; chronic, mild inflammation (so-called inflammaging); and accrued oncogenesis²³⁹. It is common knowledge that a balanced dietary regimen coupled with regular exercise preserves organismal fitness and can postpone several, if not all, of the manifestations of ageing. Accumulating evidence suggests that various beneficial effects of a healthy lifestyle depend (at least in part) on the activation of autophagy²⁴⁰. It will therefore be interesting to see whether CRMs or other pharmacological activators of autophagy can be used to extend lifespan (TABLE 2).

Other disorders

Autophagy modulators may also be beneficial for patients that are affected by pathologies such as pulmonary, renal and skeletal diseases.

Pulmonary disorders

On the one hand, autophagic defects have been documented in the lungs of patients with idiopathic pulmonary fibrosis (IPF)²⁴¹, pulmonary arterial hypertension (PAH)²⁴² and cystic fibrosis²⁴³. Accordingly, rapamycin treatment partially protects mice from IPF caused by bleomycin²⁴¹ or radiation²⁴⁴, as well as from hypoxia-induced PAH²⁴⁵. Similarly, the intranasal delivery of a Becn1-coding lentivirus successfully restores autophagy and limits inflammation in Cftr^F508Δ mice (a model of cystic fibrosis)²⁴³, as does the genetic inactivation of Rptor or Mtor in mouse models of hyperoxia-induced or lipopolysaccharide-induced acute lung injury^246,247. Moreover, Map1lc3b^−/− mice exhibit aggravated PAH upon chronic hypoxia²⁴². Thus, pharmacological activators of autophagy may be beneficial in patients with some pulmonary disorders (TABLE 2).

On the other hand, autophagy contributes to (rather than counteracts) the aetiology of chronic obstructive pulmonary disease (COPD). Indeed, Map1lc3b^−/− and Becn1^+/− mice develop limited emphysema upon chronic cigarette smoke exposure compared with their wild-type littermates^248,249. Similar observations were made in Pink1^−/− mice as well as in mice receiving the mitophagy inhibitor Mdivi-1 (REF. 250). Interestingly, the transcription factor early growth response 1 (EGR1) seems to be mechanistically involved in detrimental autophagic responses that underlie COPD²⁵¹. However, no direct links between EGR1 signalling and mitophagy have been established yet. Similarly, whether pharmacological inhibitors of EGR1 signalling may be beneficial for individuals with COPD remains to be determined (TABLE 2).

Renal conditions

Autophagy may mediate important homeostatic functions in the kidney. The podocyte-specific deletion of Atg5 or Atg7 in mice results in spontaneous glomerulosclerosis that is preceded by mitochondrial alterations (which are also documented in patients with idiopathic <m>focal segmental glomerulosclerosis</m>), excessive ROS production and podocyte loss^252,253. Similarly, mice bearing Atg5^−/− podocytes exhibit increased glomerular degeneration that is caused by puromycin aminonucleoside (PAN) or doxorubicin compared with their wild-type littermates²⁵². In addition, the podocyte-specific knockout of Atg7 considerably sensitizes mice to kidney overload imposed by unilateral nephrectomy²⁵⁴. Comparable results were obtained in PAN-treated rats receiving chemical autophagy inhibitors (3-MA or chloroquine), whereas autophagy activation with rapamycin limited glomerular degeneration in this model²⁵⁵. Thus, autophagy activators may not only benefit diabetic patients who are at risk of nephropathy (see above) but also individuals with other renal conditions (TABLE 2).

Skeletal disorders

Autophagy has recently been identified as an important mediator of bone growth in response to fibroblast growth factor (FGF) signalling²⁵⁶. Thus, the bone growth defects that are imposed by the Fgf18^+/− or Fgfr4^−/− phenotype can be rescued by intraperitoneal administration of a BECN1-derived peptide²⁵⁶. However, the deletion of Mtor or Rptor from PRX1⁺ cells (which are found in the limb, cranial and interlimb mesenchymal tissues) significantly impairs skeletal growth in mouse embryos, which results in a severe phenotype that is associated with death shortly after birth²⁵⁷. Similarly, rapamycin administration has been found to mediate beneficial²⁵⁸ and detrimental^259,260 effects in rodent models of bone fracture. It remains to be determined whether these apparent discrepancies reflect the pleiotropic effects of mTORC1 (which, among various functions, controls cell proliferation). Additional investigation is required to elucidate the potential benefits that are associated with the use of autophagy modulators in skeletal disorders (TABLE 2).

Lysosomal storage disorders

Lysosomal storage disorders (LSDs) are a heterogeneous group of rare inheritable conditions that originate from defects in lysosomal activity, which result in the cytotoxic accumulation of specific lysosomal substrates²⁶¹. The conditional deletion of Atg5 or Atg7 from myocytes alleviates lysosomal overload and enables enzyme replacement therapy in a mouse model of Pompe disease (which is characterized by glycogen accumulation)^262,263. Similarly, rapamycin is toxic for inducible pluripotent stem cells (iPSCs) from patients with Gaucher disease (which is characterized by <m>glucosylceramide</m> accumulation)²⁶⁴. Conversely, rapamycin, carbamazepine, trehalose and other autophagy inducers have cytoprotective effects on iPSCs from patients with Niemann–Pick type C disease (which is characterized by the accumulation of cholesterol and glycolipids)²⁶⁵. These observations suggest that the inhibition of autophagy upstream of autophagosome formation may be beneficial for patients with certain LSDs. By contrast, autophagy activators may have detrimental effects, unless they successfully overcome the lysosomal blockage that characterizes these disorders.

Vision disorders

Autophagy and related processes sustain healthy vision by a number of mechanisms. Retinal photoreceptor cells, especially cones, exhibit high mitochondrial turnover by mitophagy and rely on AMPK-dependent autophagic responses to cope with glucose deprivation²⁶⁶. Consistent with this, the cone-restricted deletion of Atg5 results in a functional decline that is accompanied by the accumulation of damaged mitochondria, an increased sensitivity to light toxicity and glucose deprivation²⁶⁶. Moreover, the inhibition of autophagy reportedly contributes to photoreceptor cell loss in models of <m>retinitis pigmentosa</m>²⁶⁷. ATG5-dependent and BECN1-dependent LAP has been involved in a process known as the ‘visual cycle’, in which shedding photoreceptor outer segments are engulfed by the retinal pigment epithelium (RPE), and all-trans-retinoic acid is converted to 11-cis-retinal for photoreceptor regeneration. Thus, mice lacking Atg5 (but not Ulk1) in the RPE display defective 11-cis-retinal generation and reduced vision, which can be transiently corrected by the administration of all-trans-retinol²⁶⁸. Mice with a macrophage-specific deletion of Atg5 as well as mice carrying the Atg16l1^T300A mutation exhibit increased susceptibility to lipopolysaccharide-induced uveitis that is associated with increased inflammasome activation²⁶⁹. In humans, the ATG16L1^T300A mutation is associated with Crohn’s disease (see above), which is also linked to an increased predisposition for uveitis²⁷⁰. Finally, mutations in OPTN as well as duplications of TBK1 are associated with normal-tension glaucoma^271,272. One of these mutants (namely, OPTN-E50K) displays an increased interaction with TBK1, and mice that are engineered to express the OPTN-E50K variant develop a disease that phenocopies the human disorder²⁷³. At this point, it is unclear whether these mutations promote normal-tension glaucoma as a consequence of changes in autophagy or other TBK1-related processes. However, activators of autophagy or LAP may counteract the pathogenesis of multiple vision disorders.

Challenges in developing autophagy modulators

In spite of great potential, no interventions that are specifically aimed at modulating autophagy are currently available for use in humans. Indeed, although rapamycin, chloroquine, HCQ and several drugs that are licensed for the use in humans activate or inhibit autophagy, they were not developed for this purpose. What are the obstacles that, until now, have prevented the development of autophagy modulators for clinical use? How can we circumvent them and finally harness the entire therapeutic potential of autophagy activators and inhibitors?

Specificity

Besides being used with limited rigour (that is, at non-standardized concentrations or time points), multiple chemical agents that are currently available to activate or inhibit autophagy have limited specificity for the autophagic process, for either of two reasons. First, some of these molecules have an intrinsically low pharmacological specificity for their target. This is the case for 3-MA and wortmannin, both of which are non-selective phosphoinositide 3-kinase (PI3K) inhibitors and hence block the catalytic activity of several PI3Ks beyond VPS34 (REF. 274). Along similar lines, although acute rapamycin administration results in the relatively specific inhibition of mTORC1 through FK506-binding protein 1A (FKBP1A), chronic exposure to rapamycin promotes mTORC2 disassembly¹³⁹. Second, several components of the autophagic machinery operate at the interface of multiple cellular processes, that is, they mediate autophagy-independent functions^203,275. Thus, the inhibition of mTORC1 using rapamycin not only activates autophagy but also inhibits translation, cellular growth and proliferation²⁷⁶. Similarly, the blockage of lysosomal functions using chloroquine, HCQ or BafA1 inhibits not only the disposal of autophagosomes but also the degradation of endosomes and impairs vesicular trafficking²⁷⁷. In line with this notion, rapamycin and multiple <m>rapalogues</m> mediate robust immunosuppressive effects, as they block T cell proliferation²⁷⁸, whereas both chloroquine and HCQ exert immunomodulatory as well as antineoplastic effects that are independent of autophagy^279–281. Several approaches are being explored to circumvent these specificity issues, including the development of novel autophagy modulators (BOX 3).

Box 3. Emerging autophagy modulators with increased specificity.

Considerable efforts are being dedicated to the development of agents with increased pharmacological specificity (TABLE 1), including new lysosomal inhibitors (such as Lys05)²⁹⁴ as well as new mechanistic target of rapamycin complex 1 (mTORC1)-targeting and VPS34-targeting agents (including SAR405 and VPS34-IN1)^295,296 (see also Navitor Pharmaceuticals). In parallel, increasing attention has been dedicated to components of the autophagic machinery that seem to have limited roles in other processes, such as autophagy-related 4B cysteine peptidase (ATG4B; which can be targeted by NSC185058)²⁹⁷ and UNC-51-like autophagy-activating kinase 1 (ULK1; which can be targeted by SBI-0206965, MRT67307 and MRT68921)^298,299, as well as to specific autophagic pathways, including mitophagy and LC3-associated phagocytocis (LAP)^193,213. Lys05, SAR405, NSC185058, and SBI-0206965 have been shown to mediate anticancer effects in vitro and in vivo^294,295,297. Lys05 had single-agent antineoplastic activity in xenograft models of human metastatic melanoma (1205Lu cells) and colorectal carcinoma (HT29 cells)²⁹⁴, as did NSC185058 in xenograft models of human osteosarcoma (Saos-2 cells)²⁹⁷. SAR405 and SBI-0206965 synergized with mTORC1 inhibitors in arresting the proliferation of human lung (H1299 and A549) and kidney (ACHN and 786-O) cells in vitro^295,298. Conversely, the potential therapeutic activity of VPS34-IN1 (which has been characterized biochemically in human osteosarcoma U2OS cells), MRT67307 and MRT68921 (both of which have been studied for their biochemical properties in mouse embryonic fibroblasts) has not yet been tested in relevant disease models^296,299. Thus, the actual therapeutic potential of these approaches remains largely unexplored.

An additional specificity-related issue derives from the complex architecture of all tissues, which contain several different cell types that are engaged in extensive homologous and heterologous interactions. Most autophagy modulators available at present are also poorly specific because they do not preferentially target one cell type. Thus, evaluating the actual impact of autophagy activators or inhibitors in disease scenarios in which autophagic responses in different tissue compartments may have opposite effects is challenging. Addressing this complexity by studying the effects of highly targeted autophagy modulators in disease models with cell-specific autophagic defects (see below), is key to the development of clinically viable strategies to activate or inhibit autophagy. Of note, this issue may predominantly concern the development of autophagy inhibitors. Indeed, there is not necessarily a downside to activating autophagy at the whole-body level for the treatment of diseases in which autophagy activation may be useful.

Biomarkers and assays

Multiple biomarkers that are routinely used to monitor autophagy in vitro and in vivo are intrinsically unsuitable to monitor autophagic flux, that is, the actual degradation of autophagosomes and their content within lysosomes²⁷⁴. Several of these indicators, such as the levels of lipidated LC3 (measured by immunoblotting) or the amount of GFP–LC3⁺ puncta (measured by fluorescence microscopy or flow cytometry upon cell permeabilization), statically monitor the size of the autophagosomal compartment, which expands not only in the course of productive autophagic responses (increased on-rate) but also when the formation or activity of autolysosomes is blocked (decreased off-rate). Such a technical concern de facto invalidates the conclusions of multiple studies in which autophagy was monitored only with static autophagosomal biomarkers⁹.

Besides assessing the actual degradation of autophagic substrates (which may be cumbersome and cannot be carried out in all experimental scenarios), several strategies have been developed to circumvent this issue²⁷⁴. One common approach relies on comparing LC3 lipidation (or the accumulation of GFP–LC3⁺ puncta) in the presence or the absence of lysosomal inhibitors, such as BafA1 (REF. 282). This approach can differentiate between situations of increased on-rate and decreased off-rate and can be conveniently applied to human peripheral blood mononuclear cells ex vivo²⁷⁴. As an alternative, tandem fluorescence-tagged versions of LC3 have been successfully used to identify cells in which autophagic flux is operational in vitro and in vivo^274,283. In paraffin-fixed formalin-embedded patient samples, encouraging results have been obtained by simultaneously quantifying LC3⁺ puncta and p62 abundance¹⁶⁵. We are convinced that accurately measuring autophagic flux is key to the development of clinically viable modulators of autophagy.

An additional problem relates to the lack of reliable methods for discriminating between distinct forms of autophagy (for example, general autophagy versus mitophagy), especially in vivo and in patient material²⁷⁴. This issue is particularly relevant, as it prevents the establishment of proof of principle (that is, the mechanistic correlation between an autophagy-modulating intervention and actual autophagy modulation) and proof of concept (that is, the mechanistic correlation between an autophagy-modulating intervention and disease outcome) for most pathologies. Moreover, despite considerable efforts towards standardization²⁷⁴, many assays to measure autophagy and autophagy-related processes are still implemented with consistent variability across the scientific community. Thus, new highly standardized approaches that allow for monitoring specific forms of autophagy and autophagy-related processes in vivo and in patient material are urgently awaited.

Genetic models

Many of the genetic models that have been used so far for studying the impact of autophagy on the pathogenesis of disease and developing autophagy-targeting agents suffer from limitations that are often overlooked. One of the initial obstacles was the embryonic or perinatal lethality imposed by the whole-body knockout of several components of the autophagic machinery^14–16. As mentioned above, this issue has been partially circumvented by the generation of mice with partial autophagic defects at the whole-body level, such as Becn1^+/− or Atg4b^−/− mice^15,16,284. These animals have been highly instrumental for studying the role of autophagy in multiple disorders. However, they preserve some proficiency in <m>canonical autophagy</m> and mount normal non-canonical autophagic responses in several tissues, including the liver²⁸². This implies that observing normal disease progression or the response to treatment in Becn1^+/− or Atg4b^−/− mice does not formally exclude a role for autophagy.

As an alternative approach, multiple models of tissue-specific autophagic incompetence have been generated. Most of these models rely on the deletion of ‘floxed’ Atg5, Atg7 or Becn1 in selected cell types, which is based on the expression of Cre recombinase under the control of a promoter of choice²⁸⁵. In this scenario, gene knockout occurs upon the physiological activation of the Cre-controlling promoter, which introduces considerable variability. Thus, whereas some promoters — such as the collagen, type II, alpha 1 (Col2a1) promoter²⁸⁶ — impose an autophagic defect early during differentiation or embryonic development, others — such as the cyclin-dependent kinase inhibitor 2A (Cdkn2a) promoter²⁸⁷ — abolish autophagic proficiency in terminally differentiated cells (in this case, senescent) cells. One of the major issues in this setting is the occurrence of compensatory processes, that is, a general reorganization of cellular functions that partially (if not totally) compensate for the lack of a specific protein. This is generally favoured when gene knockout occurs early during cellular or organismal lifespan as well as when the degree of functional redundancy is high (as is the case for the autophagic machinery)¹.

Two main strategies have been devised to minimize this potential source of bias: the tissue-specific administration of viral vectors coding for Cre recombinase¹⁵⁴ or the use of Cre variants that can be activated pharmacologically (for example, by the systemic administration of tamoxifen)¹⁰⁹. One potential problem with these models relates to the efficiency of Cre activation. Indeed, viral infection may not affect all target cells, or tamoxifen concentrations may be suboptimal in some tissue areas, resulting in partial autophagic defects. However, both these approaches generally display high efficiency and result in widespread gene knockout within a specific tissue.

All types of conditional knockout models (including those that rely on viral Cre delivery) also suffer from some degree of nonspecificity, which may impose autophagic defects in tissues that are unintentionally affected. For instance, genes that are under the control of the Col2a1 promoter are expressed not only by chondrogenic tissues but also — transiently — in non-chondrogenic tissues, including the notochord, eye, heart, epidermis and discrete areas of the brain²⁸⁶. Although this issue is generally marginal, it may have affected — at least to some degree — data interpretation in specific cases. Finally, it should be kept in mind that most, if not all, components of the autophagic machinery mediate known or hitherto undiscovered autophagy-unrelated functions²⁰³. This implies that great caution should be taken before attributing the phenotype that results from whole-body or conditional knockout of a single component of the autophagic machinery to autophagy as a process.

Developing new models that circumvent these persisting obstacles may be less straightforward than desirable. Nonetheless, careful consideration of the limitations that are associated with each model may enable the correct interpretation of data and hence support drug discovery and development in this field.

Future development of autophagy modulators

There seems to be considerable confusion as to how autophagy modulators should be developed and used for the treatment of several pathologies (BOX 4), which has limited the development of novel therapeutic agents. Here, we summarize some principles that must be carefully considered in the future development and use of autophagy-targeting interventions in the context of three specific scenarios.

Box 4. Improving the pharmacological audit trail for autophagy modulators.

The so-called pharmacological audit trail (PhAT) is a conceptual guide to the preclinical and clinical development of novel therapeutic agents that sequentially assesses the risk of failure at key steps of the entire process as it guides decision-making³⁰⁰. It is clear that therapeutic interventions aimed at modulating autophagy or autophagy-related processes cannot be developed according to a generic PhAT (see the figure, left). On the one hand, it will be imperative to identify with precision: the pathologies that are mechanistically determined or aggravated by alterations in autophagy (patient selection); the cell population or populations in which such alterations actually underlie disease, as opposed to bystander cell populations that may exhibit secondary autophagic defects (cell type identification); and how autophagy is specifically altered in such cells (for example, hyperactivation, inhibition before autophagosome formation or inhibition after autophagosome formation) (autophagic pathway characterization). This will enable the development of a therapeutic strategy that is aimed at activating or inhibiting autophagy in specific cell types (including diseased and bystander cells) with one (or more) targeted drug candidate or candidates. On the other hand, besides carrying out conventional pharmacokinetic and pharmacodynamic studies, it will be important to determine whether and how the autophagic flux changes in response to the drug candidates in both target and bystander cells, followed by a step of hypothesis validation with intermediate end points of clinical efficacy. In case of disease progression, compensatory mechanisms (related to autophagy) that are possibly operating in diseased or bystander cells will have to be assessed, which may lead to a complete re-evaluation of the initial therapeutic strategy (see the figure, right).

Autophagy activation underlies disease

For cases in which the activation of autophagy underlies disease, pharmacological inhibition in specific cell populations constitutes an obvious therapeutic objective. However, the functional outcome and actual therapeutic value of such an intervention may vary considerably depending on the stage of the autophagic cascade that is targeted. For instance, whereas inhibiting autophagy by targeting upstream modulators — such as AMPK (with compound C), ULK1 (with SBI-0206965) or Na⁺/K⁺-ATPase (with cardiac glycosides) — seems to be a potentially safe intervention, using lysosomal inhibitors may increase (rather than decrease) the detrimental effects of autophagy activation by causing the accumulation of non-functional autophagosomes and autolysosomes, and hence a general blockage in vesicular trafficking (FIG. 2a).

Figure 2. Principles of autophagy modulation.

a | When autophagy mechanistically contributes to the aetiology of the disease, inhibiting autophagy initiation in diseased cells is expected to restore normal autophagic degradation and mediate therapeutic effects, whereas blocking lysosomal degradation may favour a detrimental accumulation of non-functional autophagosomes and autolysosomes. b | In the presence of initiation defects, stimulating autophagy upstream of autophagosome formation (in diseased cells) is expected to normalize autophagic flux (at least in part) and hence mediate beneficial effects, whereas boosting lysosomal degradation may exert limited (if any) therapeutic activity. c | In the presence of degradation defects, activating autophagy upstream of autophagosome formation (in diseased cells) may aggravate disease severity by exacerbating (thick arrows) the accumulation of non-functional autophagosomes and autolysosomes. Conversely, accelerating lysosomal degradation or inhibiting initiation (in diseased cells) may exert beneficial effects, at least to some degree (dashed arrow). d | Boosting autophagic flux in cells that survived a pathological insult or in non‐diseased cells may favour functional recovery and/or mediate beneficial effects linked to improved inflammatory tissue homeostasis. Beneficial and detrimental interventions are indicated by green and red arrows, respectively.

Autophagy inhibition underlies disease

Autophagic defects can contribute to disease by two general, non-mutually exclusive mechanisms: the accumulation of potentially dangerous autophagic substrates or non-functional autophagosomes and autolysosomes, and the lack of potentially beneficial autophagic products or functions. In this situation, the activation of autophagy with nutritional or pharmacological interventions may ameliorate disease outcome. However, the exact nature of the autophagic defect determines which approach should be undertaken to obtain beneficial (rather than detrimental) effects. Autophagy can be blocked upstream of autophagosome formation, which results in a limited number of autophagosomes and reduced autophagic activity. In this case, nutritional interventions (that are mainly detected by the AMPK–mTORC1 signalling node) or molecules that promote the formation of phosphatidylinositol-3-phosphate (PtdIns3P) by the BECN1–VPS34 complex are expected to mediate positive effects, as they boost autophagy initiation (FIG. 2b). Conversely, interventions that are aimed at the acceleration of lysosomal degradation, such as the inhibition of CST3 or CSTB, may not mediate therapeutic effects. Autophagy can also be blocked downstream of autophagosome formation, for instance, as a consequence of a lysosomal defect. In this case, further boosting the generation of autophagosomes is not only expected to have little therapeutic activity but may also aggravate the homeostatic perturbations imposed by the lysosomal blockage. On the contrary, the acceleration of lysosomal degradation may relieve the blockage in the disposal of autophagosomes and autolysosomes, and mediate beneficial effects. Of note, the inhibition of autophagy upstream of autophagosome formation (by targeting AMPK or ULK1, for instance) may also mediate therapeutic activity in this setting, which reflects a relative normalization of the autophagosomal and autolysosomal compartments (FIG. 2c). Thus, clinically useful autophagy activators should be designed based on careful characterization of the autophagic defect that accompanies each specific disease. At least theoretically, the combination of upstream autophagy activators with molecules that accelerate lysosomal degradation might overcome multiple forms of autophagic blockade. However, it has not yet been formally demonstrated whether this approach mediates superior benefits in models of disease that are caused by autophagic defects.

Autophagy activation compensates for disease

In some cases, autophagy alterations have little (if any) relationship to primary disease aetiology, but autophagy activation may still support compensatory mechanisms, including the degeneration of cells and tissues. Although many preclinical models that have been used thus far do not allow discrimination between these two aspects of pathogenesis (aetiology versus recovery; see below), interventions that boost autophagic activity in cells that are not directly affected by disease may ameliorate the long-term outcome in multiple disorders, even when administered after the primary pathological insult. This is particularly true for conditions with multifactorial aetiology, such as cerebral stroke or neurotrauma. In this scenario, efficient autophagic responses in glial cells that survive stroke considerably limit RCD-driven inflammation and the consequent loss of additional neurons (which is known to participate in the pathogenesis of disease), hence mitigating long-term functional impairment⁹. Thus, the activation of autophagy in cells that survive the primary pathological insult or in non-diseased cells (which exhibit normal autophagic capacity) may constitute a promising therapeutic intervention for multiple disorders with complex aetiologies (FIG. 2d). It is possible that patients with some of these pathologies may benefit from a sequential approach in which autophagy is inhibited first (or selectively in diseased cells) and activated subsequently (or selectively in non-diseased cells). Future studies are required to elucidate this possibility in specific pathologies.

Conclusions and perspectives

As discussed throughout this Review, the therapeutic potential of interventions that activate or inhibit autophagy is enormous. Nonetheless, multiple obstacles of pharmacological, technical or experimental nature have hampered the straightforward implementation of autophagy modulators in the clinic. We are confident that many of these hurdles can be circumvented upon the development of more selective autophagy modulators, more precise biomarkers of the autophagic flux and more physiological models of autophagy deficiency in vivo. The very nature of autophagy imposes obstacles that must also be overcome for the development of clinically viable autophagy activators and inhibitors. Additional work is required to understand how to inhibit or activate autophagy independently of cellular processes including proliferation and RCD, and hence harness the full therapeutic potential of autophagy modulators.

Glossary

Autophagic cell death

A variant of regulated cell death (RCD) that is precipitated by the autophagic machinery and hence can be retarded with pharmacological or genetic inhibitors of autophagy

Regulated cell death (RCD)

A form of cell death that relies on the activation of genetically encoded machinery and hence can be retarded or accelerated with specific pharmacological or genetic interventions

Mitophagy

Autophagic response that is specific for depolarized or otherwise damaged mitochondria.

Aggrephagy

Autophagic response that is specific for intracellular protein aggregates, which often are highly ubiquitylated

Presenilin 1 (PSEN1)

Component of the γ-secretase complex that contributes to the accumulation of amyloid plaques in the brain of patients with Alzheimer disease (AD).

Autophagic adaptor

A protein that directs autophagic substrates to forming autophagosomes through its capacity to bind ubiquitylated structures and lipidated Atg8 family members

Amyloid-β precursor protein

(APP). A protein that — upon cleavage — accounts for the majority of amyloid plaques in the brain of patients with Alzheimer disease (AD).

Caloric restriction mimetic (CRM)

A molecule that mimics the biochemical and cellular effects of fasting, including autophagy activation and cytosolic acetyl-CoA depletion, but does not provoke a sizeable weight loss

Trehalose

A natural α-linked disaccharide that potently activates autophagy through poorly characterized mechanisms

Valproate

A widely used antiepileptic drug that induces autophagy by affecting myo-inositol-1,4,5-trisphosphate levels

Lithium

An antidepressant that promotes autophagic responses by altering myo-inositol-1,4,5-trisphosphate levels

Locomotor sensitization

Long-lasting exacerbation of a psychostimulant-induced locomotor response, which is brought about by repeated intermittent administration of the same psychoactive agent

Ischaemic preconditioning

An experimental technique for increasing the resistance of neurons or cardiomyocytes to prolonged, severe ischaemia based on the repeated administration of short, mild ischaemic episodes

Cardiac glycoside

A natural compound that exerts positive inotropic effects and retards some forms of autophagic cell death as it inhibits the plasma membrane Na+/K+-ATPase

Efferocytosis

The removal of dying or dead cells by professional phagocytes

Spermidine

A natural polyamine that potently activates autophagy by operating as a caloric restriction mimetic (CRM)

Coronary angioplasty

A minimally invasive surgical procedure for the treatment of narrowed or weakened arteries, which consists of the insertion of a small mesh tube (stent) through the femoral artery

Metformin

An antidiabetic agent with pleiotropic effects, including the capacity to trigger autophagy by acting as a caloric restriction mimetic (CRM)

db/db mice

Mice homozygous for the spontaneous db (for diabetes) mutation in leptin receptor (Lepr), which causes limited leptin signalling. These animals are commonly used as models for type 2 diabetes and metabolic syndrome

ob/ob mice

Mice homozygous for the spontaneous ob (for obesity) mutation in leptin receptor (Lepr), which causes absent leptin signalling. These animals are commonly used as models for obesity and metabolic syndrome

Carbamazepine

A widely used antiepileptic drug that induces autophagy by altering myo-inositol-1,4,5-triphosphate levels

α1-antitrypsin deficiency

A genetic disease that causes the defective production of serpin family A member 1 (SERPINA1; also known as α1-antitrypsin) in the lungs and liver, which results in pulmonary disorders that are often associated with hepatic symptoms.

Streptozotocin

A naturally occurring toxin that is commonly used to generate rodent models of type 1 diabetes owing to its pronounced selectivity for pancreatic β-cells

Malignant transformation

The conversion of a healthy, normal cell into a neoplastic cell precursor. Malignant transformation is insufficient to drive tumorigenesis

Oncosuppressor genes

Genes mutated or silenced in familial or sporadic forms of cancer. Many of these genes encode proteins that are involved in the maintenance of cellular homeostasis or in the activation of regulated cell death (RCD)

Proto-oncogenes

Genes overexpressed or hyperactivated in familial or sporadic forms of cancer. Many of these genes encode positive regulators of cellular proliferation or proteins that inhibit regulated cell death (RCD)

Tumour progression

A process through which a neoplastic cell precursor acquires additional genetic or epigenetic alterations that allow it to escape cell-intrinsic and cell-extrinsic control mechanisms and form aggressive tumours

Immunosurveillance

A process in which the immune system recognizes and eliminates a potentially dangerous entity, including invading pathogens as well as pre-malignant and malignant cells

Inflammasomes

Supramolecular platforms that support caspase 1 activation, hence allowing for the proteolytic maturation and secretion of pro-inflammatory interleukin-1β (IL-1β) and IL-18

Neutrophil extracellular traps (NETs)

Chromatin-based and granule protein-containing fibres that are released by neutrophils to immobilize and kill invading microorganisms.

Cellular senescence

A permanent proliferative arrest that is generally associated with specific morphological and biochemical features, including the secretion of multiple cytokines and other biologically active factors

Chronic obstructive pulmonary disease (COPD)

A progressive lung disease that is characterized by long-term limited airflow, which is often caused or aggravated by tobacco smoke

Focal segmental glomerulosclerosis

A leading cause of kidney failure in adults that is characterized by the degeneration of sections of the glomerulus with a focal (as opposed to diffuse) intrarenal distribution

Glucosylceramide

A sphingolipid that accumulates in patients with Gaucher disease (mostly in the macrophages) as a result of mutations in glucosylceramidase beta (GBA)

Retinitis pigmentosa

An inherited, degenerative eye disease that causes severe vision impairment owing to the progressive degeneration of the rod photoreceptor cells

Rapalogues

Rapamycin derivatives with improved pharmacodynamic and pharmacokinetic properties

Canonical autophagy

A term commonly used to refer to an autophagic response that is dependent on autophagy-related 5 (ATG5), ATG7, beclin 1 (BECN1) and phosphatidylinositol-3-phosphate (PtdIns3P) production