Autophagy in stem cells: repair, remodelling and metabolic reprogramming

Autophagy is an intracellular catabolic mechanism by which cell components, including proteins, lipids and whole organelles, are degraded and recycled inside lysosomes (Galluzzi et al., 2017a). The degradation products are then transported back to the cytoplasm and are used to sustain cell homeostasis. Autophagy is a fundamental metabolic response to nutrient and oxygen deprivation, as well as an essential cytoplasmic quality control process (Boya et al., 2013; Galluzzi et al., 2017a). In recent years, studies have shown that autophagy plays a crucial role in various cell types, including neurons, muscle and cancer cells. However, it has also emerged that autophagy is key for the functioning and maintenance of various populations of stem cells (SCs), acting to promote their quiescence, maintain their stemness and self-renewal, and mediate their differentiation.

SCs give rise to the body's key cell lineages during embryogenesis, and also participate in tissue homeostasis and repair postnatally and throughout adult life. They divide to self-renew and to produce daughter cells that can undergo differentiation (García-Prat et al., 2017). These differentiation processes require cell remodelling, which is achieved by the autophagic elimination of structures and cell components that are no longer needed (Mizushima and Levine, 2010). Because most SCs are long-lived and need to remain functional for the life span of the organism, robust quality control mechanisms are essential for their subsistence. Autophagy, the main cellular quality control pathway, is thus thought to play a crucial role in sustaining SC homeostasis. In recent years, a number of studies have also revealed that autophagy contributes to the control of metabolism (Esteban-Martínez et al., 2017a; Kaur and Debnath, 2015) and can thus regulate the metabolism, identity and function of SCs. Indeed, the view that SCs are metabolically distinct from their differentiated counterparts, and that these metabolic features are essential for maintaining SC identity, is well-supported in the literature (Shyh-Chang et al., 2013). The fact that most SCs appear to rely more on glycolysis to generate ATP than on oxidative phosphorylation (Shyh-Chang and Ng, 2017) has been linked to the hypoxic milieu of SC niches. However, increased glycolysis is a crucial step in the conversion of differentiated cells into induced pluripotent stem cells (iPSCs; see Glossary, Box 1), suggesting that glycolytic metabolism is a requirement to maintain stemness rather than being an adaptation to the SC environment (Folmes et al., 2011; Shyh-Chang and Ng, 2017). Overall, this growing body of evidence suggests that autophagy helps to preserve SC function by simultaneously regulating cell remodelling and metabolism and serving as an important quality control mechanism.

In this Review, we discuss our current understanding of the mechanisms by which autophagy preserves SC function. We also highlight the key consequences of autophagy dysregulation in SCs on tissue development and maintenance, and discuss how SCs respond to stressful conditions. The functions of autophagy in somatic cells and disease models have been reviewed elsewhere (Schneider and Cuervo, 2014) and are therefore not discussed here.

The activation of autophagy during periods of starvation is an evolutionarily conserved response in eukaryotes (Kaur and Debnath, 2015). Under these conditions, the cell uses protein and lipid degradation to adapt its metabolism and fulfil its energy needs. Accordingly, the pharmacological or genetic downregulation of autophagy results in rapid cell death in starvation conditions (Boya et al., 2005). Other stressors, such as hypoxia, oxidative stress and infection, can also induce autophagy (Fig. 1).

The process of autophagy involves members of the autophagy-related (ATG) family of proteins. Autophagy induction is controlled by mTOR and AMP-activated protein kinase (AMPK) signalling pathways, which regulate the assembly and activation of an ATG1/ULK1 complex that in turn triggers formation of the phosphatidylinositol 3-kinase (PI3K) complex (Fig. 1). This complex regulates the incorporation of phosphatidylinositol 3-phosphate into the phagophore membrane from which autophagosomes (see Glossary, Box 1) are generated (Hurley and Young, 2017). Next, two conjugation reactions catalysed by ATG7 are necessary for autophagosome formation: one relies on ATG7 and ATG10, which induce the conjugation of ATG5 to ATG12 in the context of a multiprotein complex containing autophagy-related 16-like 1 (ATG16L1); the other results in the conjugation of phosphatidylethanolamine to LC3 (MAP1LC3) to form the autophagosome-bound form of LC3 called LC3-II. The continuous assembly of these complexes and the delivery of lipids via ATG9, the only multi-membrane-spanning ATG protein, allow the autophagosomal membrane to elongate and close to form the mature autophagosome (Hurley and Young, 2017). The later stages of autophagy are controlled by molecules that regulate autophagosome fusion with lysosomes, such as LAMPs and RAB7, and by lysosomal acidic hydrolases that regulate the degradation of the autophagy cargo (Fig. 1). The final degradation products then translocate to the cytoplasm to be recycled for use in new anabolic reactions to sustain cell homeostasis. Because autophagy is a highly dynamic process, the blockade of one of its stages, or impaired lysosomal function or biogenesis, leads to the accumulation of autophagosomes, ultimately disrupting or diminishing autophagic flux (see Glossary, Box 1) (Boya et al., 2013).

Development and differentiation are often accompanied by large-scale cellular and tissue remodelling, which is mediated by autophagy (Mizushima and Komatsu, 2011). At the cellular level, autophagy is essential for the differentiation of many cell types, including adipocytes, erythrocytes, lymphocytes and neurons (Mizushima and Levine, 2010). In addition to its role in intracellular quality control and metabolic regulation during cell differentiation, autophagy provides a rapid and efficient means of altering the composition of the cytosol, and is involved in controlling cell size through the degradation of receptors, organelles and transcription factors, all of which are processes that are crucial for cell differentiation (Mizushima and Levine, 2010). Given that autophagy is essential for the elimination of unnecessary or harmful components from cells, and is a key regulator of cell metabolism, its dysregulation has significant pathological consequences. Indeed, autophagy is implicated in a plethora of pathologies, including neurodegenerative, metabolic and immune diseases (Boya et al., 2016; Deretic et al., 2015; Galluzzi et al., 2015; Menzies et al., 2017; Stienstra et al., 2014). Defects in proteostasis and autophagy have also been described to occur in ageing, and autophagy is proposed to underlie the beneficial effects of caloric restriction (Kaushik and Cuervo, 2015; Madeo et al., 2015).

Autophagy can also be a highly selective process, allowing specific cytoplasmic components to be delivered to lysosomes via specific cargo-recognition proteins called autophagy receptors (Khaminets et al., 2016). The selective autophagy of mitochondria, a process termed mitophagy (see Glossary, Box 1), can also occur and allows damaged or unwanted mitochondria to be engulfed into autophagosomes for lysosomal delivery and degradation (Ashrafi and Schwarz, 2013). Under conditions of mitochondrial stress, for instance, the kinase PINK1 activates the ubiquitin ligase PRKN, which then ubiquitylates mitochondrial proteins that recruit autophagy receptors that bridge between mitochondria and autophagosomes (Lazarou et al., 2015). Mitochondria are also eliminated in developmental contexts, for example during cell differentiation, in a process named programmed mitophagy. Programmed mitophagy has been observed during erythrocyte and lens maturation and during neuronal differentiation (Ashrafi and Schwarz, 2013; Esteban-Martínez and Boya, 2017; Ney, 2015). In addition, mitophagy facilitates cell remodelling while also acting as a quality control mechanism by eliminating potential sources of oxidative stress (Takamura et al., 2011). Finally, recent evidence has demonstrated that mitophagy regulates a metabolic shift towards glycolysis in several contexts, such as during neuronal differentiation and macrophage activation (Esteban-Martínez et al., 2017b), indicating a link between mitophagy and metabolic reprogramming (Esteban-Martínez and Boya, 2017). How mitophagy controls SC homeostasis is further described in Box 2.

Given their unique properties of self-renewal, multipotency, differentiation and quiescence in adult tissues, SCs must strictly control their rates of protein and organelle turnover and of ATP production (Guan et al., 2013; Vessoni et al., 2012). Metabolic regulation is now widely believed to function as a general mechanism for controlling SC quiescence (García-Prat et al., 2017). However, a growing body of evidence indicates that autophagy is also required for SC quality control and for maintaining the cellular homeostasis of SCs (García-Prat et al., 2017; Guan et al., 2013).

Depending on their source, SCs can be classified as embryonic stem cells (ESCs) or adult SCs. ESCs are derived from the inner cell mass (ICM) of blastocysts (see Glossary, Box 1), around 3.5-5.5 days after fertilization in the case of mouse embryos and 4-9 days after fertilization for human embryos. They are grown in vitro and are pluripotent, meaning they can produce all the cells of the embryo proper (but not the placental lineages). By contrast, adult SCs are found in tissues and organs after they have completed their development. These multipotent cells have a restricted potency compared with ESCs, and only give rise to a subset of cell types to replace and repair specific tissues. Some of the best-studied types of adult SCs include haematopoietic stem cells (HSCs), neural stem cells (NSCs) and muscle stem cells (better known as satellite cells) (Rumman et al., 2015). Adult SCs are also found in tissues that exhibit high turnover, such as the intestine and skin. Such SCs are responsible not only for tissue repair after damage but also for maintaining normal tissue turnover. Conversely, HSCs are maintained in a quiescent or very low-cycling state for months, a state that they abandon to repopulate the blood in response to haematopoietic stress, and muscle satellite cells are maintained in quiescence for most of their life and only divide in response to tissue damage (Rumman et al., 2015). Maintaining a balance between stemness and differentiation is therefore of crucial importance for SCs. Excessive cell differentiation depletes the SC population (leading to SC exhaustion) and promotes ageing or decay. On the contrary, excessive SC proliferation can give rise to cancer. Thus, quality control mechanisms are essential to preserve adult SC homeostasis and the capacity of SCs to respond rapidly to environmental stressors, damage and differentiation signals to sustain tissue regeneration. Below, we review recent studies that demonstrate the essential role of autophagy in maintaining embryonic and adult SC homeostasis. We describe how autophagy functions as an intracellular quality control and repair mechanism in SCs, and how autophagy remodels cellular morphology by eliminating, for example, organelles and stemness factors that control cellular reprogramming. We also review why and how autophagy controls metabolism to sustain energy homeostasis in SCs.

After fertilization, the mammalian zygote is reprogrammed to form pluripotent cells located in the ICM of blastocysts. This reprogramming, which requires the epigenetic modification of maternal and paternal genomes, the expression of pluripotency genes, and the removal of inherited maternal proteins, involves both the ubiquitin proteasome system (UPS; see Glossary, Box 1) and autophagy (DeRenzo and Seydoux, 2004). For example, the participation of autophagy in cell reprogramming during the zygote-to-embryo transition (see Glossary, Box 1) has been documented (Hanna et al., 2010; Jopling et al., 2011). Autophagy is also essential in the very early stages of mouse embryogenesis, and is required for the embryo to reach the 4- to 8-cell stage (Tsukamoto et al., 2008; Wang et al., 2013b). The relevance of autophagy in this process was first evidenced in fertilized mouse oocytes lacking Atg5, which do not proceed beyond this stage if fertilized with Atg5-null sperm, and therefore fail to form the blastocyst and the ICM (Tsukamoto et al., 2008). More recently, it has been proposed, based on the well-documented induction of autophagy during the 4- to 8-cell stage, that markers of autophagy activity could be used in the future to determine embryonic viability in the field of assisted reproduction (Tsukamoto et al., 2014).

The induction of autophagy during embryonic reprogramming and early embryonic development seems to be controlled by different molecular mechanisms. The initial pulse of autophagy induced by fertilization is independent of mTORC1 activity; it is possible that calcium oscillations triggered by fertilization initiate autophagic responses instead (Tsukamoto et al., 2008; Yamamoto et al., 2014). However, once the 4- to 8-cell stage of embryogenesis is reached, the downregulation of mTOR expression, which is mediated by Sox2 and the nucleosome remodelling and deacetylase complex (NuRD, see Glossary, Box 1), becomes indispensable for autophagy induction, similar to the situation seen during the reprogramming of somatic cells into iPSCs (Wang et al., 2013b) (as discussed later). Interestingly, during this period, overactivation of autophagy with the mTOR inhibitor rapamycin accelerates embryonic reprogramming and the formation of the blastocyst (Wang et al., 2013b). Autophagy is also essential for the removal of maternal material that otherwise blocks the reprogramming process (DeRenzo and Seydoux, 2004), and provides recycled amino acids, nucleotides and sterols that are crucial for maintaining cellular energy homeostasis prior to pre-implantation, after which cells have access to transplacental nutrients (Tsukamoto et al., 2008).

Interestingly, autophagy appears to be dispensable for the later stages of embryogenesis. Indeed, mice null for Atg3, Atg5, Atg7, Atg9 or Atg16L1 are not embryonically lethal, although they are born with reduced body weight and generally die 1 to 2 days after birth, possibly owing to suckling defects caused by deficient neurological development (Mizushima and Levine, 2010). It is thought that the neonatal survival of these mutant embryos is due to the presence of maternally inherited ATG proteins in the oocyte cytoplasm (Mizushima and Levine, 2010; Tsukamoto et al., 2008). However, this is not the case for mice that are null for the genes encoding the phagophore-forming BECN1, AMBRA1 or FIP200 (RB1CC1) proteins, which are embryonically lethal. The origin of such phenotypic differences among ATG gene knockout mouse models is unclear. It is possible that the embryonic death affecting BECN1, AMBRA1 and FIP200-deficient mice is due to other (i.e. autophagy-independent) functions of these proteins, as has been recently demonstrated for FIP200 (Chen et al., 2016), for example. Different degrees of functional redundancy or compensatory mechanisms for the different ATG proteins have also been postulated (Mizushima and Levine, 2010). Moreover, it has also been suggested that the UPS can compensate for the absence of autophagic activity in these autophagy-deficient ESCs (Lee et al., 2017; Vilchez et al., 2012). Indeed, crosstalk between the UPS and autophagy has been observed in human ESCs, which show high levels of proteasome activity that progressively decline during differentiation, coinciding with an increase in autophagy, which possibly then degrades damaged or unnecessary proteins and organelles (Vilchez et al., 2012). FOXO transcription factors (see Glossary, Box 1) have also been shown to regulate autophagy and the UPS (Sandri, 2012; Webb and Brunet, 2014). For instance, FOXO1 has been reported to be essential for maintaining human and mouse ESC pluripotency, probably by controlling OCT4 (POU5F1) and SOX2 expression (Zhang et al., 2011), and, more recently, it has been shown that FOXO1 regulates the expression of autophagy genes and maintains a high level of autophagic flux in mouse ESCs (Liu et al., 2017).

Autophagy has also been linked to the phagocytosis of apoptotic cells by either neighbour cells or professional phagocytes, a process that is required for proper metazoan development and adult tissue homeostasis. Qu and co-workers were the first to show that mouse ESCs lacking Atg5 or Becn1 fail to cavitate during embryoid body (EB) generation (Qu et al., 2007); these EBs display defective ATP production, which results in deficient expression of the engulfment signals required for the phagocytosis of dead cells. This phenotype could be reversed by the addition of methylpyruvate, a cell-permeant intermediate of glucose metabolism, which is incorporated into the mitochondrial tricarboxylic acid cycle. Similar findings were also shown in Caenorhabditis elegans and mouse embryonic development, in which autophagy is required for the proper clearance of apoptotic cells (Cheng et al., 2013; Huang et al., 2013; Li et al., 2012; Mellén et al., 2008, 2009).

Finally, it has been shown that autophagy promotes morphological changes associated with SC differentiation (Vessoni et al., 2012). For example, autophagy contributes to the degradation of the midbody ring (MB) during ESC differentiation, a process that can be triggered by either starvation or treatment with rapamycin (Kuo et al., 2011). The MB is a circular structure that forms an intercellular bridge after cytokinesis, and is required for the separation of daughter cells (Schink and Stenmark, 2011). Studies have provided some indication of how autophagy removes MBs, via a process known as midbophagy, which involves a complex consisting of p62 (SQSTM1), ALFY (WDFY3) and TRAF6 (Isakson et al., 2013), interactions between NBR1 and CEP55 (Kuo et al., 2011), and the participation of TRIM17 (Mandell et al., 2016). A better understanding of how these structures are cleared by autophagy and how they can avoid autophagosomal degradation could greatly advance our understanding of how SC pluripotency and differentiation are controlled. Studies have also demonstrated that the primary cilium (see Glossary, Box 1) emerges from ESCs after induced lineage specification and activates autophagy. This results in the inactivation of nuclear factor erythroid-related factor 2 (Nrf2; Nfe2l2), likely by autophagy-mediated degradation of its positive regulator p62, promoting the transcriptional activation of the pluripotency factors OCT4 and NANOG and directing ESCs towards a neuroectodermal fate (Jang et al., 2016). This induction of autophagy is not observed during mesoderm differentiation, indicating that autophagy is required for the degradation of organelles and proteins only in specific differentiated cells, rather than for the removal of damaged proteins. In conclusion, autophagy is required in ESCs to fulfil the energy requirements for cell remodelling and metabolic and proteostatic changes, and for the engulfment and clearance of apoptotic cells.

A number of studies have investigated the role of autophagy in embryonic and adult NSCs. Our group has demonstrated that the expression of Atg7, Becn1, LC3 and Ambra1 is markedly increased in cultured embryonic mouse olfactory bulb (OB)-derived NSCs during the initial period of neuronal differentiation (Vazquez et al., 2012). Furthermore, it was shown that Ambra1 and Atg5 deficiency decreases neurogenesis, a phenotype reversed by methylpyruvate supplementation, suggesting that progenitor cells activate autophagy to meet their high energy demands (Vazquez et al., 2012). Defective neurogenesis in the mouse cerebral cortex during early brain development has also been reported following Atg5 knockdown (Lv et al., 2014). A recent study using Atg16L1 hypomorph mice and primary neurons showed that NOTCH, a plasma membrane receptor and master regulator of neuronal development, is taken up into ATG16L1-positive autophagosome-precursor vesicles and modulates neurogenesis (Wu et al., 2016). Inhibition of mTOR via the autophagy-related protein Eva1a (also known as Tmem166) has also been linked to mouse NSC self-renewal and differentiation (Li et al., 2016).

In the adult mammalian brain, the best studied NSCs are those located in the subventricular zone (SVZ) of the lateral ventricles and in the subgranular zone (SGZ) of the hippocampal dentate gyrus (Doetsch et al., 1999; Palmer et al., 1997). These niches, like other SC niches in the body, are hypoxic, a condition required for stemness. NSCs, unlike terminally differentiated neurons, can expand through self-renewal and differentiate into several neural lineages. Oxidative stress, which is marked by elevated levels of reactive oxygen species (ROS), is one of the best-known factors inhibiting SC proliferation. Autophagy appears to maintain a low level of ROS in order to sustain the slow cycling of NSCs. Interestingly, the absence of FOXO1, FOXO3 and FOXO4 (i.e. in triple-null mice) leads to increased ROS production in the NSC pool, and this is accompanied by an initial increase in NSC proliferation followed by an abrupt reduction of the NSC pool and reduced neurogenesis (Paik et al., 2009; Palmer et al., 1997; Renault et al., 2009). It has been suggested that FOXO3 regulates the NSC pool by promoting quiescence, preventing premature differentiation and controlling oxygen metabolism (Renault et al., 2009). Once in the nucleus, FOXO3a might promote the expression of mitophagy genes to facilitate mitochondrial clearance, and it is therefore likely that FOXO activation decreases ROS levels via the induction of mitophagy as a protective mechanism to counterbalance mitochondrial stress (Tan and Wong, 2017). However, there is some disagreement regarding whether adult NSCs have lower or higher endogenous ROS levels than their differentiated progeny and the extent to which this could influence their self-renewal properties. A study of NSCs in the mouse SVZ reported high levels of ROS, on which the self-renewal and neurogenesis capabilities of these cells depend (Le Belle et al., 2011). The authors proposed that NSCs might maintain high ROS levels during highly proliferative stages of development, and lower levels during quiescence, suggesting a mechanism for antioxidant regulation.

Several other studies have also described the autophagy-mediated control of NSC proliferation and differentiation. For example, the downregulation of Ambra1 and Becn1 results in the decreased proliferation and increased apoptosis of mouse NSCs (Yazdankhah et al., 2014), and downregulation of miR-34a gives rise to increased expression of synaptic proteins and Atg9a, which appear to also regulate mouse NSC differentiation in vitro (Morgado et al., 2015). Together, these observations suggest that autophagy participates in the regulation of ROS levels through mitophagy, thereby controlling adult NSC proliferation, although the involvement of other molecular players or signalling pathways cannot be ruled out. Further studies will be required to identify whether crosstalk exists between autophagy and other catabolic systems in controlling NSC stemness, as well as the mechanisms that regulate the relative quiescence of NSCs during adult life.

The regulation of adult neurogenesis by autophagy has received surprisingly little research attention, with only a handful of studies supporting a role for autophagy in the maintenance and differentiation of adult SCs into different neuronal lineages. Wang and colleagues reported that ablation of FIP200, but not of Atg5, Atg16L1 or Atg7, results in increased ROS that leads to a progressive loss of NSCs caused by p53 (Trp53)-dependent apoptotic and cell cycle arrest (Wang et al., 2016a, 2013a). The authors showed that FIP200-null NSCs, but not NSCs deficient for other autophagy genes, display p62 aggregates and increased SOD1 retained in the cytoplasm, leading to increased levels of superoxide (Wang et al., 2016a). Loss of FIP200 causes defects in neuronal differentiation and both failures can be rescued by treatment with the antioxidant N-acetylcysteine. Interestingly, FIP200 deletion causes NSC depletion in the postnatal mouse brain but does not affect embryonic NSCs (Wang et al., 2013a). The same group recently showed that autophagy regulates NSCs through cell-extrinsic mechanisms. They demonstrated that FIP200 regulates the differentiation of NSCs in the mouse postnatal SVZ by restricting microglia infiltration and activation (Wang et al., 2017). However, other studies, using different genetic approaches to modulate gene expression, found that Atg5 deletion impairs adult neurogenesis in the SGZ of the mouse hippocampus (Xi et al., 2016) and that Atg5 downregulation leads to increased proliferation and decreased differentiation of mouse embryonic cortical neural precursor cells (Lv et al., 2014). It is therefore possible that autophagy is required for the maintenance and differentiation of NSCs at different stages of postnatal life.

Some authors have proposed very different roles for autophagy in the maintenance of adult rat hippocampal NSCs. Chung and co-workers found that rat hippocampal NSCs undergo autophagic cell death in response to insulin withdrawal despite the presence of intact apoptotic machinery, and that this effect is suppressed by the knockdown of Atg7 (Chung et al., 2015). Conversely, Yu and colleagues reported that the tight regulation of calpains – a family of calcium-dependent cytosolic proteases – by the proteasome and by Ca²⁺ levels switches the mode of hippocampal NSC cell death from being autophagic, when calpain levels are low, to apoptotic, when calpains levels are high (Yu et al., 2008). These apparently contradictory findings demonstrate that there is still much to be discovered in order to fully understand the role of autophagy in adult neurogenesis. In recent years, other populations of NSCs have been detected in other regions of the mouse adult nervous system, such as Müller glia in the retina (Jorstad et al., 2017) and nestin-expressing progenitors in the cerebellum (Wojcinski et al., 2017), and it has been shown that these cells are able to repopulate their respective tissues upon induced damage. Mouse adult spinal cord glial cells can also be reprogrammed in vivo to generate neurons upon injury (Wang et al., 2016b). These studies constitute a compelling incentive to elucidate the molecular mechanisms through which autophagy mediates neuronal differentiation, as that knowledge could be applied to stimulate these resident SCs for adult brain repair.

HSCs sustain the production of all blood cells (Fig. 2). The first HSCs appear in mid-gestation in mouse embryos and progressively colonize the foetal liver, the main haematopoietic organ during these embryonic stages. Just before birth, the bone marrow replaces the liver as the main reservoir of HSCs. In adulthood, haematopoiesis is maintained by multipotent bone marrow-resident HSCs (Hirschi, 2012). These quiescent HSCs self-renew but can enter the cell cycle and differentiate into multipotent progenitors during physiological haematopoiesis (Crisan and Dzierzak, 2016) to balance the massive destruction of blood cells that occurs daily, and also as a consequence of haematological stress (e.g. bone marrow transplantation, cytotoxic drugs, radiation). Dysregulation of the fine balance between quiescence, self-renewal and differentiation can result in the development of blood disorders, such as anaemia and leukaemia (Ansó et al., 2017). Understanding the mechanisms that protect HSCs from damage is therefore essential to treat haematopoietic malignancies.

Single-cell RNA sequencing has recently demonstrated high levels of transcription activity of autophagy-related genes during foetal HSC formation in mouse embryos (Hu et al., 2017; Zhou et al., 2016). Unlike adult SCs, which are mainly quiescent, foetal HSCs undergo rapid cycling. The cytofluorimetric analysis of cells isolated from transgenic GFP-LC3 mice, which express the autophagosomal marker LC3 tagged to GFP (Mizushima et al., 2004), has revealed increased autophagic flux activity in HSCs, compared with their differentiated progeny (Watson et al., 2015). Autophagy is also increased in HSCs from GFP-LC3 mice after cytokine withdrawal and is induced in a FOXO3a-dependent manner in response to metabolic stress (Warr et al., 2013). These data indicate that HSCs exhibit a high degree of basal and induced autophagy, supporting the view that autophagy is crucial for preserving HSC function.

Autophagy also plays an important role in maintaining foetal and adult HSCs. The conditional deletion of FIP200 in mouse HSCs (i.e. in FIP200^flox/flox; Tie2-Cre mice) results in a massive reduction in the number of liver embryonic HSCs, resulting in foetal/perinatal lethality (Liu et al., 2010). Competitive reconstitution experiments in lethally irradiated recipients show that FIP200-null foetal liver HSCs also fail to provide long-term multilineage reconstitution. At the cellular level, these cells display increased mitochondrial mass and elevated ROS, and a slight increase in bromodeoxyuridine (BrdU) incorporation, suggesting that the observed exhaustion of HSCs might be due to increased proliferation following autophagy downregulation (Liu et al., 2010). Autophagy is also essential for adult HSC function. Indeed, foetal and adult HSCs isolated from Atg7^flox/flox; Vav-Cre mice are unable to form secondary colonies in colony-forming assays, and in transplantation experiments Atg7-deficient HSCs are unable to rescue lethally irradiated hosts (Mortensen et al., 2011). Interestingly, foetal liver Atg7-deficient HSCs can rescue lethally irradiated recipients, suggesting that Atg7 is less crucial for foetal than for adult HSC function. Mice with Atg7-deficient HSCs also have increased numbers of mitochondria as well as elevated oxidative stress, DNA damage and cell proliferation, which might promote the onset of blood malignancies observed in these mice (Mortensen et al., 2011). A similar phenotype has also been documented in Atg5^flox/flox; Vav-Cre mice (Watson et al., 2015), and heterozygous loss of Atg5 in a model of acute myeloid leukaemia leads to increased HSC proliferation in vitro and to the development of more aggressive leukaemias in vivo (Watson et al., 2015). Deleting Atg12 in 4-week-old mice using the pan-haematopoietic promoter Mx, which responds to the synthetic analogue of double-stranded RNA polyIpolyC, results in defective HSC self-renewal and myeloid-biased differentiation (Ho et al., 2017). Interestingly, this myeloid lineage expansion is also observed in other autophagy-deficient HSCs (Liu et al., 2010). In conclusion, these data support the view that autophagy preserves HSC stemness by reducing oxidative damage and limiting HSC proliferation.

The importance of mitophagy in the haematopoietic system has also been recognized. Mice deficient for the mitophagy receptor BNIP3L (also known as NIX) develop anaemia as a result of the defective elimination of mitochondria in red blood cells (Diwan et al., 2007; Sandoval et al., 2008). Recent findings also indicate that mitophagy is important not only for the elimination of mitochondria from mature erythrocytes but also to preserve stemness in HSCs (Ito et al., 2016). Ito and co-workers demonstrated that in a newly described HSC population [Tie2 (Tek)-positive cells; hereafter Tie2⁺] the upregulation of two essential mitophagy regulators, PINK1 and PRKN, results in high levels of mitophagy that maintain HSC division potential. Silencing Pink1 and Prkn results in a failure to reconstitute the haematopoietic system in irradiated recipient mice (Ito et al., 2016). Interestingly, in single-cell gene expression assays this Tie2⁺ population displays increased expression of fatty acid oxidation and peroxisome proliferator-activated receptor-delta (PPAR) genes. PPAR agonist treatment results in increased mitophagy through enhanced transcriptional expression of Pink1 via FOXO3 signalling, in agreement with earlier data showing that FOXO3-dependent autophagy preserves HSC function (Warr et al., 2013). These findings contrast with those of another study in which HSCs from Prkn-deficient animals were found to reconstitute the blood in transplantation experiments (Ho et al., 2017). Further studies are therefore required to understand better the putative role of mitophagy in HSC function (Joshi and Kundu, 2013).

The ability of HSCs to differentiate depends on their ability to activate mitochondrial oxidative phosphorylation (Kohli and Passegué, 2014). Accordingly, deletion of the mitochondrial phosphatase PTPMT1, which regulates the transition from anaerobic glycolysis to oxidative phosphorylation, results in accumulation of HSCs that are unable to differentiate (Yu et al., 2013). PTPMT1-Vav-Cre mice display important alterations in the HSC pool, as well as cell cycle delay and differentiation defects due to an inability to upregulate mitochondrial fatty acid oxidative metabolism. Moreover, deletion of the same gene in myeloid or lymphoid lineage progenitors does not impair normal development, indicating that PTPMT1 plays a pivotal role in HSCs but is not essential in late lineage progenitors (Yu et al., 2013). Other data support the existence of a close relationship between mitochondrial metabolism and HSC function. A recent report demonstrated that loss of the mitochondrial complex III subunit, Rieske iron sulfur protein (RISP), in mouse foetal HSCs has no effect on cell proliferation but does alter differentiation, leading to severe anaemia and embryonic death. Furthermore, deletion of the same gene in adult HSCs leads to the loss of HSC quiescence and to increased BrdU incorporation, indicative of cell cycle entry, and results in severe deficiencies in red and white cells as well as platelets (Ansó et al., 2017).

As mentioned earlier, autophagy is crucial for the later stages of the differentiation of many blood cell types, as evidenced by the dramatic alterations in cell function and differentiation observed following cell-specific autophagy blockade. Interestingly, GATA1, a master regulator of haematopoiesis, regulates several autophagy genes (Kang et al., 2012). Phenotypically, dysregulated autophagy is characterized by alterations in cellular quality control and by marked metabolic changes, which affect, for example, the glycolytic shift that occurs during proinflammatory macrophage activation (Esteban-Martínez and Boya, 2017; Riffelmacher et al., 2017). Autophagy activity is therefore not only necessary to sustain the self-renewal of HSCs but also to control the terminal differentiation of different blood cell types to maintain haemostasis (Fig. 2).

Skeletal muscle is composed of muscle fibres (myofibres), which consist of multinucleated syncytial cells. These postmitotic cells are unable to sustain muscle growth and repair, which instead relies on a unique population of muscle stem cells, also known as satellite cells. Named for their location beneath the basal lamina of muscle fibres, these are somite-derived myoblasts that have not fused to other myoblasts and remain as stem cells throughout adult life (Wang and Rudnicki, 2011).

During muscle development, embryonic myoblasts differentiate to generate primary myofibres that will serve as template fibres for subsequent myogenesis before birth. Later, rapid and extensive proliferation of postnatal myoblasts is responsible for further muscle growth and maturation during neonatal myogenesis (Wang and Rudnicki, 2011). By the third week of life in mice, the muscle is mature, and the number of satellite cells reaches a steady state as they enter quiescence (Fig. 3). Quiescent satellite cells are then activated in response to muscle damage and enter the cell cycle to give rise to committed proliferating muscle precursors, which differentiate and fuse to repair the damaged muscle. As in the case of HSCs, transplantation experiments are often used to evaluate the capacity of self-renewal to replenish the SC pool and to generate committed descendants that will proliferate and differentiate to orchestrate tissue repair.

Studies have only recently begun to investigate the role of autophagy in satellite cells. In Atg7-deficient mouse satellite cells (generated by crossing Pax7-Cre with Atg7^flox/flox mice), satellite cell numbers are severely reduced (García-Prat et al., 2016). At the cellular level, this autophagy deficiency is very similar to the phenotype observed in aged satellite cells, which are characterized by increased oxidative stress, DNA damage, accumulation of p62 and ubiquitin aggregates, damaged mitochondria and presence of senescence markers, including p16INK4a (CDKN2A), p21CIP1 (CDKN1A) and P15INK4b (CDKN2B) (Fig. 3). Interestingly, satellite cells isolated from aged mice show a reduction in autophagic flux, which is completely restored in vivo in response to mTOR inhibition with rapamycin or administration of spermidine, a natural polyamine that has been shown to extend the lifespan of mice in an autophagy-dependent manner (Eisenberg et al., 2016). These data demonstrate the potential role of autophagy in preserving muscle homeostasis and preventing age-dependent senescence (García-Prat et al., 2016). Interestingly, a recent study found that adult mouse satellite cells express genes involved in autophagy in a circadian manner, and aged satellite cells exhibit a markedly reduced capacity to cyclically recycle damaged components that are generated on a daily basis (Solanas et al., 2017).

In addition to the role of basal autophagy in maintaining SC stemness, induced autophagy is also essential for supporting the bioenergetic demands generated during satellite cell activation (Tang and Rando, 2014). When autophagy is acutely downregulated by pharmacological or genetic approaches in isolated ex vivo mouse myofibres or in sorted satellite cells, reduced BrdU incorporation during spontaneous activation in culture is observed, suggesting a delay in satellite cell activation (Fig. 3). Moreover, autophagy downregulation in mouse satellite cells reduces ATP levels, which can be restored by supplementing the cultures with sodium pyruvate, which also partially restores levels of cell proliferation (Tang and Rando, 2014).

Autophagy is also essential in the later stages of muscle cell differentiation, with autophagy blockade resulting in altered myocyte fusion and myotube formation during muscle differentiation (Fortini et al., 2016; Sin et al., 2016). Interestingly, autophagy is activated during the early, compensatory regenerative stages in the mdx mouse model of Duchenne muscular dystrophy, and impaired autophagy activation in late stages of disease progression correlates with fibrotic tissue deposition and with diminished regeneration in dystrophic muscles (Fiacco et al., 2016). These findings indicate that autophagy is essential for preserving muscle homeostasis, serving as a quality control mechanism by preventing satellite cell senescence and meeting the bioenergetic demands of satellite cells during activation (Fig. 3).

The advent of cell reprogramming methodologies has enabled the generation of iPSCs. These are pluripotent SCs generated from differentiated somatic cells that are reset to a pluripotent state (Yamanaka and Blau, 2010) by the pluripotency transcription factors, which include Oct4, Sox2, Klf4 and Myc (together called OSKM) (Takahashi and Yamanaka, 2006; Yamanaka, 2012). In recent years, a number of studies have revealed that autophagy is implicated in this reprogramming process, particularly during the early stages (Buckley et al., 2012; Hansson et al., 2012; Tsukamoto et al., 2008; Wang et al., 2013b) (Fig. 4). For example, the deletion of Atg5, Atg3 or Atg7 in mouse embryonic fibroblasts impairs reprogramming efficiency (Tsukamoto et al., 2008; Wang et al., 2013b). Furthermore, autophagy degrades nuclear pluripotency-associated proteins that are normally only transiently expressed (Cho et al., 2014). Transducing mouse fibroblasts with OSKM triggers a transient pulse of increased autophagy from day 1 that peaks the following day and subsequently declines to basal levels by day 3, correlating with mTOR downregulation at both the mRNA and protein levels (Menendez et al., 2011; Wang et al., 2013b). This transient mTOR inhibition is essential, as demonstrated by experiments in which cells treated with rapamycin during the first 3 days after OSKM transduction show increased efficiency of cell reprogramming; by contrast, rapamycin treatment at later stages, or very high concentrations of rapamycin, abolish iPSC generation, suggesting that autophagy needs to be downregulated soon after it peaks (Chen et al., 2011; He et al., 2012). Accordingly, iPSC reprogramming is also inhibited by increasing mTOR activity by knocking down the mTOR negative regulator Tsc2 or by using Tsc2^−/− mouse embryonic fibroblasts in which mTOR activity is hyperactivated (He et al., 2012). At the molecular level, it has been shown that Sox2, which is one of the OSKM factors, controls mTOR expression via the NuRD complex (Hu and Wade, 2012; Wang et al., 2013b). During the early stages of cell reprogramming, the NuRD complex is recruited by Sox2 to a repressive region of the mTOR promoter to mediate its transcriptional repression, and it dissociates from it 2 days after the induction of reprogramming. However, little is known about the feedback mechanisms that shut down the activity of the NuRD complex to enable the transcription and translation of mTOR to resume in the later stages of reprogramming (Rais et al., 2013). Such mechanisms might be essential for establishing the pluripotency signalling network required for cell reprogramming.

A role for mitochondria and mitophagy has also been implicated in iPSCs. As in ESCs, the mass, maturation status and number of mitochondria are significantly reduced in iPSCs compared with somatic cells (Armstrong et al., 2010; Facucho-Oliveira and St John, 2009; Prigione et al., 2010; Sena and Chandel, 2012; St John et al., 2006), suggesting that pluripotent SCs rely more heavily on glycolysis for energy production (Fig. 4). Although it is commonly accepted that mitophagy plays an important role in creating the conditions necessary to establish pluripotency (Vessoni et al., 2012; Wang et al., 2013b) and mediates mitochondrial rejuvenation to prevent iPSC differentiation, the form of autophagy involved and the molecular mechanisms that govern these processes remain a matter of debate (Vessoni et al., 2012; Wang et al., 2013b). A recent study reported that mitophagy is essential for iPSC reprogramming and is regulated by Atg5-independent, AMPK-dependent autophagy (Ma et al., 2015). However, how mitophagy is controlled and how it functions in pluripotency reprogramming remains to be elucidated.

The role of autophagy in cancer stem cells (CSCs) has been studied in detail (Auberger and Puissant, 2017; Hamaï et al., 2014). Below, we briefly summarize how the autophagic-lysosomal pathway contributes to the unique characteristics of CSCs. We also discuss the potential value of targeting autophagy as a means of eradicating CSCs.

Recent in vivo lineage-tracing approaches support the involvement of CSCs in many cancers (Beck and Blanpain, 2013; Pattabiraman and Weinberg, 2014; Singh et al., 2015). Similar to other SCs, CSCs can self-renew (Beck and Blanpain, 2013) but they also have a potential for malignancy (Beck and Blanpain, 2013; Pattabiraman and Weinberg, 2014; Tam and Weinberg, 2013). Of note, they are highly resistant to cancer therapy and recent evidence suggests they can initiate tumour metastasis (de Sousa e Melo et al., 2017; Lawson et al., 2015; Massagué and Obenauf, 2016; Pascual et al., 2017). Moreover, it has been shown that CSCs harbour the ability to convert into non-cancer SCs and vice versa, a phenomenon known as CSC plasticity (Beck and Blanpain, 2013; Pattabiraman and Weinberg, 2014; Singh et al., 2015). This phenomenon of CSC plasticity increases the complexity of the relationship between autophagy and CSCs. Thus, although active autophagy is a recognized hallmark of tumours (Galluzzi et al., 2015; White, 2015), it can serve as a tumour-suppressing mechanism or can promote tumour formation, depending on the type of cancer and the stage of development (Galluzzi et al., 2015). Autophagy is also implicated in the crosstalk between cancer cells and the microenvironment, host tissues and the immune system (Galluzzi et al., 2017b, 2015; Zhong et al., 2016). Compelling evidence indicates that autophagy is a major cellular pathway involved in the origin, maintenance and differentiation of CSCs (Auberger and Puissant, 2017; Guan et al., 2013; Hamaï et al., 2014; Pan et al., 2013); CSCs reside in hypoxic, nutrient-poor and acidic environments, conditions known to induce autophagy, and many studies have shown that CSCs are highly responsive to these stimuli (Fig. 5) and, hence, that the basal rate of autophagy is frequently higher in CSCs than in non-cancer SCs.

Given the multiple functions of autophagy in CSCs (Fig. 5), a number of studies have investigated if and how autophagy can influence tumorigenesis. It has been shown that autophagy supports the survival of human breast malignant precursor cells but that treatment with chloroquine, a lysosomotropic inhibitor of autophagy, blocks the generation of breast ductal carcinoma in situ spheroids in vitro and abrogates xenograft tumour formation (Espina et al., 2010). The silencing of BECN1 and ATG7 impairs the in vitro self-renewal of ALDH1-positive breast cancer cell lines or CSCs isolated from human breast cancer specimens and inhibits their growth in xenografts in mice (Gong et al., 2013; Yue et al., 2013). Similarly, the silencing of ATG7, ATG12 or ATG8/LC3 impairs the in vitro growth of CD44⁺CD24^−/low breast cancer stem cells (Cufi et al., 2011; Maycotte et al., 2015). Interestingly, the inhibition of autophagy in CD44⁺CD24⁻ breast CSCs decreases the secretion of IL6 (Maycotte et al., 2015), a cytokine important for CSC maintenance (Iliopoulos et al., 2011).

The role of autophagy in the survival of CSCs and the maintenance of stemness has also been described in other tumour types (Auberger and Puissant, 2017; Hamaï et al., 2014; Lin et al., 2015; Marcucci et al., 2017; Ojha et al., 2015). However, the role of autophagy in CSCs is probably more complex, as demonstrated by the fact that autophagy inhibition decreases the viability of chronic myeloid leukaemia CD34⁺ progenitor cells, whereas its inhibition in HSCs favours the expansion of acute myeloid leukaemia progenitor cells (Auberger and Puissant, 2017). In summary, these data underscore the importance of autophagy in CSC function but highlight the need for further studies to investigate the potential of cancer therapies that target autophagy in CSCs.

In this Review, we have discussed studies that have provided important insights into the pivotal roles that autophagy plays in embryonic and adult SCs, including in the maintenance of stemness, the promotion of cellular reprogramming and the differentiation of SCs (summarized in Fig. 6). Together, these findings indicate that: (1) autophagy is used for cell remodelling to degrade organelles and stemness factors during SC reprogramming, activation or differentiation; (2) autophagy-mediated cell repair and quality control mechanisms are essential to preserve homeostasis in most if not all SCs, and this is usually associated with eliminating damaged mitochondria, the most usual source of cellular ROS; and (3) autophagy and mitophagy are essential to preserve the energy homeostasis and metabolic reprogramming that allow different SC types to maintain quiescence, self-renewal, activation and differentiation. Accordingly, autophagy deficiency results in significant alterations in SC function including SC exhaustion, senescence, aging and cell death (Fig. 6).

Although pharmacological approaches to modulate SC fate have shown some promise (Angelos et al., 2017; Bouchez et al., 2011; Fares et al., 2014; Rentas et al., 2016), few studies have successfully modulated SC function by targeting autophagy. SMER28, a small molecule capable of inducing autophagy (Sarkar et al., 2007), has been used to reverse erythropoiesis abnormalities in patients with Diamond–Blackfan anaemia (DBA), a congenital disorder characterized by severely diminished red blood cell production due to defective erythroid progenitor differentiation (Doulatov et al., 2017). Classical autophagy-modulating approaches, such as rapamycin or spermidine treatment, caloric restriction, and low protein diets, have also been shown to preserve HSC and muscle satellite cell function (Cerletti et al., 2012; Fiacco et al., 2016; García-Prat et al., 2016; Kohli and Passegué, 2014). These studies suggest that SC function could indeed be modulated by targeting autophagy, a finding that has relevant therapeutic implications.

A better understanding of the molecular mechanisms through which autophagy regulates the function of SCs and their differentiation into specific cell types could hold significant promise for the development of new therapies for haematological, muscular and neurological diseases, as well as for some cancers. Other questions on the role of autophagy in SC maintenance and differentiation remain unanswered. For example, it is still unclear whether autophagy failure during ESC expansion or differentiation has deleterious consequences in the adult organism. How autophagy is involved in maintaining adult stem cell quiescence, exit and re-entry into this cellular stage is an emerging field that has only recently started to be explored in HSCs and muscle satellite cells and remains dark for other SC types, such as NSCs. Furthermore, we also emphasize the need to finely decipher the distinct autophagy requirements for healthy adult SCs and ESCs versus CSCs in order to find autophagy-based targets for potential anti-cancer therapies and to prevent malignant reprogramming into CSCs. Our knowledge on the understanding of autophagy function in SC biology has expanded dramatically in the last decade. However, although it is clear that autophagy is a key determinant of SC self-renewal, stemness and differentiation, many aspects of this relationship remain unclear. Answering these questions will increase our insight into SC biology and human development, improve the efficiency of iPSC reprogramming and differentiation protocols, and facilitate the design of strategies to delay the onset of degenerative and age-associated diseases.

We thank O. Howard for English language editing. The authors apologize for work not being mentioned owing to space limitations.