Autophagy and disease

As outlined in the accompanying Minireviews, autophagy is a complicated and highly regulated process that delivers cellular material to lysosomes for degrading, recycling, and generating molecules that fuel cellular metabolism. Autophagy is important for normal cellular and organismal physiology, and both increased and decreased autophagy has been associated with disease. Importantly, these connections are already being exploited to treat patients with dozens of clinical trials that aim to manipulate autophagy to treat (or prevent) disease. This Minireview discusses some of the important issues and problems to be solved if these efforts are to be successful.

Issue 1: Autophagy does many things, so is it reasonable to try to manipulate it to treat a specific disease?
As noted in the accompanying Minireviews, we now have a quite detailed understanding of how autophagy is regulated by ATG proteins and a much better understanding of autophagy's biological roles. One major conclusion arising from all this knowledge is the breadth of those roles. Autophagy regulates many different biological processes during development, normal physiology, and in response to a wide variety of stresses. One issue that arises given the broad biological functions for this process is that attempts to manipulate autophagy therapeutically, e.g. to inhibit it, which is being done in dozens of cancer clinical trials where anti-cancer drugs are being combined with autophagy inhibitors (4), might have unintended consequences by affecting other biological functions of autophagy. Will it even be possible to inhibit autophagy without having unacceptable toxicities due to the inadvertent interference with other autophagy functions? Conversely, "healthy" interventions such as exercise are able to stimulate autophagy and alleviate disease at least in part via increased autophagy (14). Indeed, because of its wide ranging effects, autophagy enhancement has been proposed as a strategy to extend healthy aging (15). Enhancement of autophagy is also being tested in clinical trials, e.g. in neurodegenerative diseases (16). However, interventions that stimulate autophagy with the goal of preventing or treating one disease could have the unintended side effect of promoting other diseases. For example, a recent study found that dormant tumors could be reactivated based on whether the surrounding tissue can undergo autophagy (17).
To understand this issue, consider that autophagy is critical in protecting against infections by bacteria and viruses and for protecting cells, especially neurons, from toxicity caused by aggregated proteins. Autophagy is also critical for regulating responses to nutrient deprivation. The importance of these functions of macroautophagy is shown in work where the gene for an essential autophagy regulator, Atg7, was inducibly knocked out in adult mice (18). This approach could be considered an example of what we might expect with a perfect autophagy inhibitor that affects all cells in the body and directly targets the autophagy machinery. The effects were dramatic. Acute systemic ATG7 ablation resulted in all the animals dying, i.e. macroautophagy is essential for viability of adult mice. Some mice died within a few days, while the others died after a 2-3month period. The first group died due to Streptococcus infection, consistent with a known role for a specialized form of autophagy (xenophagy) in eliminating pathogenic Streptococcus from cells (19). The remaining mice all died a few weeks later due to neurodegeneration. Again, this might have been This work was supported by National Institutes of Health Grants CA150925, CA190170, and CA197887 from NCI. This is the fifth article in the Thematic Minireview series "Autophagy." The author declares that he has no conflicts of interest with the contents of this article. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. 1 To whom correspondence should be addressed.  (20,21) where neuron-specific inactivation of the essential autophagy genes Atg5 or Atg7 resulted in accumulation of polyubiquitinated and aggregated proteins and the formation of pathogenic inclusion bodies. Additionally, the Atg7-deleted mice were very sensitive to nutritional stress so that fasting for 24 h, a stress that wildtype mice can survive with no problem, was lethal for most of the animals. Again, this phenotype was what we might have expected because we know that autophagy is critical for regulating metabolic homeostasis (22,23). Consistent with this, autophagy-inhibited mice had defects in mobilizing lipids, glycogen metabolism, and muscle wasting, and the fasted mice succumbed to hypoglycemia.
So, does this mean that inhibition of autophagy would not work for cancer therapy because you would kill patients by affecting their susceptibility to infection, causing neurodegeneration, or disrupting metabolism? Karsli-Uzunbas et al. (18) addressed this question in their inducible knockout mice. They caused formation of lung tumors using a well-established model involving loss of TP53 and activation of oncogenic KRAS. Then, after tumor formation and during the window of time when the Atg knockout mice are healthy, they blocked autophagy by knocking out Atg7. The important result was that the lung tumors regressed. This tells us that although autophagy was indeed critical for survival of the organism, this is not necessarily a barrier to getting some therapeutic benefit from autophagy inhibition. The problem, of course, is that although cured of cancer all the mice would have died shortly afterward of neurodegeneration. In a human patient, there would be little enthusiasm for a treatment that cures cancer but only at the cost of dying of neurodegeneration a month later. However, in human patients we will not be able to completely block autophagy because we would use pharmacological inhibitors to block the processes that are not going to be 100% effective. Also, this process need not be permanent as when we knock out a gene (drug treatment can be stopped). This seems to be the case in cancer patients who have been treated with drugs intended to inhibit autophagy (to date these drugs have all been using chloroquine or hydroxychloroquine to inhibit the lysosome); we have seen signs of clinical benefit, and even after extended periods of treatment on the autophagy inhibitor, no serious side effects were noted. For example, several patients where autophagy inhibition was used to reverse acquired resistance to another drug (i.e. a kinase inhibitor), all patients showed signs of clinical improvement and had limited side effects that did not include the neuronal effects or the susceptibility to infection that we might have expected based on complete autophagy inhibition in mice (24,25). Interestingly, side effects associated with this treatment have been due to the other drug working better, which is not what we would expect for autophagy inhibition itself.
The important point here rests on a core principle of pharmacology: the success or failure of a treatment rests on the therapeutic index, i.e. the amount of a drug that causes a therapeutic effect compared with the amount that causes toxicity. If you can find a dose that has a therapeutic benefit while not being unacceptably toxic, the drug can be effective. The likely reason the strategy works in the Atg7 knockout mice is that the tumor cells are more dependent on autophagy than other tissues, including the brain, and this creates a window of time when we can obtain the death of tumor cells but little death of neurons. In human patients who have been on chloroquine for extended periods of time, the likely reason the strategy works is that the tumor cells are more sensitive to autophagy inhibition so that even incomplete inhibition of autophagy can cause tumor cell death without adversely affecting normal tissues. Additionally, in treating patients with pharmacological autophagy inhibitors, we have the option of taking the treatment away to allow recovery from the toxic side effects. Therefore, even though autophagy does many things, this is not an insurmountable barrier to using autophagy modulation as a therapeutic. The goals are for beneficial effects to outweigh toxicities and to have a good understanding of the biological roles of autophagy so that we can know what to look for in regard to the unwanted effects of interference with the process.

Issue 2: Autophagy can have competing effects even in the same disease
Many commentaries and reviews about autophagy and its functions use terms like "double-edged sword" as a metaphor for the fact that autophagy often has opposing effects on the same biological response (26). This idea is well developed in cancer where it is believed that autophagy can protect against the development of cancer, but in an established tumor it can support cancer progression (8,27). Autophagy has also been reported to be necessary for replication of some infectious organisms while suppressing replication of others (7).
Even at the level of the cell, autophagy can have opposing effects. Autophagy has been widely studied in cell death (28) and is often thought of as a cell-survival mechanism. Conversely, autophagy can induce programmed cell death, e.g. a form of death called autosis (29). Autophagy has even been shown to specifically promote or inhibit apoptosis in the same cell population in response to two very similar stimuli, death receptor agonists Fas ligand/CD95 and tumor necrosis factorrelated apoptosis-inducing ligand (TRAIL) 2 (30). If autophagy can promote completely opposite effects by two stimuli that work through almost identical signaling pathways and do so even in the same population of cells, how can we ever hope to know when we should aim to increase or decrease autophagy if we care about cell death? The answer is to understand the molecular mechanisms that are at play. In the case of the competing effects of autophagy on apoptosis induction by the two death ligands, selective autophagic turnover of a negative regulator of apoptosis that affects the receptor for FasL but not the TRAIL receptor explained why FasL-induced apoptosis was stimulated by autophagy while TRAIL-induced autophagy was not (30). Conversely, autophagy-mediated down-regulation of a positive regulator of apoptosis efficiency explained why autophagy protects against TRAIL (31). Thus, by understanding mechanisms at a molecular level it is possible to make sense of even a completely opposite effect on two very similar signaling pathways working at the same time in the same cell.

Issue 3: Some biological effects when we interfere with autophagy might not be due to autophagy
The conclusion that autophagy is involved in a disease is usually based on experiments that involve inhibiting autophagy and then seeing a change in the disease process of interest. For example, knockout of an ATG gene in cells or organisms changes a disease response, and so we infer that autophagy was important in that response. The problem is that this assumes that the only role for the targeted ATG is to regulate autophagy. Unfortunately, all autophagy regulators have autophagy-independent effects, too (32). For example, ATG7 also regulates p53 (33); ATG12 regulates apoptosis (34,35); ATG5 controls mitogen-activated protein kinase (MAPK) signaling (36), apoptosis (37), and mitotic catastrophe (38), and BECN1 controls cytokinesis (39). Such autophagy-independent effects can have important implications for how we interpret and design our experiments.
During Mycobacterium tuberculosis infection, bacterial replication occurs in macrophages in association with autophagosome-associated proteins (40). Moreover, mice with macrophage deletion of the essential autophagy gene Atg5 are particularly susceptible to tuberculosis infection (41), which leads to the hypothesis that induction of autophagy would protect against tuberculosis infection (42). However, more recent work where several other autophagy regulators (ATG16L1, ATG12, ATG3, ATG7, and ATG14) were deleted in macrophages all failed to show the same sensitization effect as ATG5 (43). This indicates that an autophagy-independent function of ATG5 is critical for suppressing bacterial replication. This study highlights the extensive studies (in this case using multiple strains of knockout mice) that might need to be done to determine whether an effect is due to autophagy or to an autophagy-independent function of a targeted gene; it is risky to make conclusions based on targeting just one or two autophagy regulators.
Other examples where core components of the autophagy machinery but not autophagy itself are important in disease include autoimmune diseases such as lupus, which involves an autophagy-related process called LC3-associated phagocytosis (LAP) (44). Interestingly, although LAP requires many components of the core autophagy machinery, it can be discriminated from macroautophagy through a distinct function for the autophagy regulator RUBICON (45). More restricted cases where several but not all autophagy regulators are necessary for an effect are also found. For example, certain autophagy regulators, including ULK1, Beclin 1, ATG14L, and VPS34, are required for the intracellular bacterium Brucella abortus to replicate in infected cells, but ATG5, ATG7, ATG16L1, and LC3B are not required (46).
By designing experiments based on a detailed understanding of the molecular mechanisms by which autophagy regulators work, it is feasible to discriminate between autophagy-dependent and autophagy-independent effects. A good example is FIP200, which is required for early steps in autophagy induction as part of the ULK1 complex (47) where it binds to ATG13. FIP200 knockout mice are embryonic lethal, but this is due to autophagy-independent effects as shown by the generation of a mouse with point mutations in FIP200 that ablate only its autophagy-dependent functions by disrupting ATG13 binding without affecting its other roles; these mice have a similar phenotype to Atg5 or Atg7 knockout mice (48). In contrast, FIP200's ability to regulate autophagy is critical for tumor growth (48). The moral of the story is that we need to have a very complete molecular understanding of different functions that any given regulator has if we hope to really understand how the disease we are interested in is controlled by the molecules we are studying.
Similar problems arise with pharmacological inhibitors of autophagy. A large number of clinical trials are based on the idea that autophagy inhibition with lysosomal inhibitors like chloroquine can sensitize tumor cells to other anti-cancer drugs (4). However, this chemosensitization can occur by autophagy-independent means that are not mimicked by genetic inhibition of ATG genes (49). In other cases chemosensitization by chloroquine is mimicked by ATG knockdown and is thought to be due to the autophagy inhibition (24). The important point is that to know if autophagy itself is critical in a disease, one must test the requirement of multiple components of the autophagy machinery.
Strategies to enhance autophagy have a similar problem. We have few, if any, methods to enhance autophagy without also having other effects. As noted in the Minireview by Delorme-Axford and Klionsky (72), the regulation of autophagy is complicated, and we have plenty of drugs and other interventions that can be used to enhance autophagy. Thus, if one inhibits mTOR, one will usually enhance autophagy, and mTOR inhibitors such as rapamycin are widely used for this purpose. Other agents such as the nutraceutical disaccharide trehalose can enhance autophagy through mTOR independent means (50), e.g. by inhibiting glucose transporters and thus treating liver disease (51). Caloric restriction mimetics, including many relatively non-toxic agents, can induce autophagy (52), and in some cases they have been reported to have beneficial effects in disease settings, e.g. by enhancing anti-tumor immunosurveillance (53). Other presumably even more benign interventions can also be used such as exercise (54). However, as with autophagy inhibitors, it is often difficult to determine whether biological effects are in fact due to autophagy because all these approaches also have autophagy-independent effects. Experiments can be designed to try to address this issue. Most commonly, this is done by stimulating with your autophagy inducer while simultaneously inhibiting autophagy by knocking out an essential ATG and hoping that the effect caused by your inducer goes away. Although such experiments can show that autophagy was needed for the biological effect being studied, they usually cannot completely exclude the possibility that another effect of the intervention was critical, too.
By understanding the molecular mechanisms that control autophagy, it may be possible to design more specific inducers. A model was provided by Levine and co-workers (55), where a peptide from the autophagy regulator Beclin 1 was used to induce autophagy. Importantly, these authors were able to show that a cell-permeable version of their peptide could THEMATIC MINIREVIEW: Autophagy and disease enhance autophagy both in vitro and in vivo and successfully treat mice following viral infection by enhancing autophagy. This peptide was identified by understanding how a viral protein activates autophagy and thus provides another good example of how understanding molecular mechanisms can lead to practically useful ways to approach disease treatment.
Another recent study takes this concept an exciting step further (56). As noted above, one confusing and problematic issue with autophagy manipulation is that we might not want to just generally activate all possible types of autophagy. Thus, to protect against neurodegenerative disease caused by aggregated proteins, it would be optimal to enhance selective autophagy capable of removing aggregated proteins, but we might not want to also enhance other forms of autophagy (e.g. mitophagy, which degrades mitochondria). The paper by Sakamaki et al. (56) focused on understanding epigenetic regulation of autophagy and made the interesting finding that the bromodomain protein BRD4 is a negative regulator of autophagy. Importantly, this effect was selective such that BRD4 depletion enhanced autophagy in some situations but not in others and could, for example, reduce aggregated proteins with no effect on mitophagy or xenophagy. This implies that BRD4 inhibitors, which have been developed for cancer therapy, could be repurposed to treat diseases associated with protein aggregation, although they may not be useful to treat bacterial infections. This establishes an important precedent-it may be feasible to design interventions that would not just generally enhance autophagy but instead would enhance different kinds of selective autophagy. If feasible to do this pharmacologically, as suggested by working with the BRD4 inhibitor, this should allow widening of the therapeutic index for autophagy-inducing drugs.
Issue 4: Will we be able to tell whether we changed autophagy when we are trying to manipulate the process for therapeutic purposes?
The practical application of the ideas discussed above is limited by another major problem. It is difficult enough to reliably measure autophagy in tissue culture, and it is much harder to tell whether autophagy has been altered in vivo. It is especially difficult to do so in a clinical setting where technical tricks that we might use in the lab, e.g. following the behavior of a fluorescent autophagosome-associated protein like LC3B (57), are impossible. Some of the problem is inherent in the dynamic nature of the process. As explained in the accompanying Minireviews, the full process of autophagy involves formation of autophagosomes, fusion with lysosomes, and then degradation of the cargo along with components of the autophagosome like LC3. This sometimes causes confusion among inexperienced autophagy investigators who do not realize that accumulation of processed LC3 II could either mean that autophagosome generation was increased (i.e. enhancement of autophagy) or that autophagosome fusion and degradation was inhibited (i.e. inhibition of autophagy). In tissue culture experiments we control for these effects by comparing LC3 II levels with and without lysosome inhibitors (58). Unfortunately, in tissue samples it is much harder to discriminate between increased, decreased, or unaltered autophagy.
In the clinical studies with chloroquine-based drugs, different methods have been attempted to assess whether autophagy was indeed blocked (59 -62), also see Levy et al. (4) for discussion of potential markers that are used in current trials. Several studies have attempted to measure the number of autophagosomes using immunohistochemistry of proteins such as LC3. However, although one can measure LC3 foci this way in clinical specimens (63,64), as noted above it is hard to discriminate between induced or blocked autophagy with this assay.
Another problem is that levels of autophagy-inhibiting drugs may be different in the target tissue (e.g. the tumor) and the blood (65); this means that even if one did reliably measure a change in autophagy in blood cells, it might not mean that autophagy was changed in the tissue you care about. Potential solutions to this problem include the use of surrogate markers of other activities that are controlled by autophagy and can be measured in blood. For example, secretion of cytokines and other molecules can be controlled by autophagy (66 -69), and the autophagy-regulated secretome has been suggested as a surrogate that would allow assessment of whether altering autophagy in a patient has been successful (70).

Conclusions
Although there is plenty of room for improvement, it is encouraging to see that we have gone from the identification of the first autophagy-regulating genes (71) to clinical benefits in patients (25) in just 20 years. Successfully addressing the issues that I raised will certainly allow us to improve these efforts, but these can only be answered by better understanding at a molecular level how autophagy works and what it does. This represents a nice example where basic science that is intended to develop a detailed mechanistic understanding of a biological process is critical for the practical application of our knowledge to treat disease; we will not be able to do worthwhile translational and clinical studies without a solid underpinning of basic science. As we develop more detailed understanding of these processes, it will hopefully become possible to widen therapeutic indices when we are using autophagy-modulating therapies and to refine our manipulations to target the kind of autophagy we want to affect and not just all aspects of the autophagy response. Achieving those goals would be a fitting validation of the insight that the Nobel Committee demonstrated when they awarded the Prize for Medicine to Professor Ohsumi for his basic research discoveries.