Autophagy is a cellular trafficking pathway mediated by the formation of double-membraned vesicles called autophagosomes, which ultimately fuse with lysosomes where their cargo is degraded. By sequestering and clearing dysfunctional cellular components, such as protein aggregates and damaged organelles, autophagy maintains cellular homeostasis whilst also providing metabolites and energy during periods of starvation2. Studies using a range of laboratory models from yeast to mammals have established that autophagy is essential for cellular and organismal survival. For example, loss of an essential autophagy gene atg5 leads to reduced survival of Saccharomyces cerevisiae in nitrogen starvation conditions and shortened lifespan in Drosophila melanogaster10,11. Likewise, inducible knockout of Atg5 results in cell death and neurodegeneration in adult mice3,12,13. It remains unclear which of the many physiological functions of autophagy are most important for its role in maintaining cell and organismal survival. Furthermore, the role of autophagy in the quality control of cellular proteins and organelles is likely to impact a plethora of signal transduction and stress response pathways, which in turn affect metabolism, growth, and survival14,15. Untangling this complexity in vivo is challenging whilst mechanistic studies of the cellular roles of autophagy in vitro are hindered by the fact that autophagy-deficient cells are viable in cell culture12,13,16. We hypothesized that this apparent discrepancy between the requirement for functional autophagy in vivo and in vitro could be due to a metabolic shift from oxidative phosphorylation (OXPHOS) to glycolysis in tissue culture, which could mask an underlying viability defect in autophagy-deficient cells17.
Cell death underlying loss of autophagy was associated with NAD(H) depletion
A well-established strategy to reverse cellular reliance on energy generation via aerobic glycolysis and promote mitochondrial OXPHOS is to replace glucose, the major carbon source in tissue culture media, with galactose18-22. Strikingly, Atg5-/- but not wild-type mouse embryonic fibroblasts (MEFs) cultured in galactose media displayed rapid (~24 h) caspase activation and cell death (Fig. 1a, b, Extended Data Fig. 1a). This phenotype was not caused by galactose directly as it was not toxic in the presence of glucose (Extended Data Fig. 1b, c). Instead, it was driven by mitochondrial respiration since its suppression by hypoxia or mitochondrial pyruvate carrier inhibitor UK-5099 was sufficient to rescue cell death (Extended Data Fig. 1d-g). Specificity of this phenotype was validated by re-expression of Atg5, and the apoptotic nature of cell death as evidenced by caspase 3 cleavage was confirmed using Z-VAD-fmk (Extended Data Fig. 2a-c). A similar phenotype was observed in cell lines with CRISPR/Cas9 knockout of key autophagy genes such as Atg5, Atg7 or Rb1cc1 (homologue of human FIP200), as well as with the loss of lysosomal cholesterol transporter Npc1 required for efficient autophagy6 (Fig. 1c, Extended Data Fig. 2d, e).
The rapid nature of cell death suggested an underlying metabolic collapse in autophagy-deficient cells23. Loss of autophagy was previously shown to cause depletion of cellular metabolites, although the mechanisms linking these metabolic defects to the cell death phenotype are poorly understood2. To investigate the potential metabolic basis of cell death due to autophagy deficiency, we performed an unbiased metabolomics profiling of wild-type and Atg5-/- MEFs after 16 h in galactose media, i.e., prior to the onset of cell death. In agreement with a previously proposed general defect in nucleic acid recycling in autophagy-deficient cells24, a number of nucleotides were depleted in Atg5-/- MEFs (Fig. 1d, Extended Data Fig. 3a-c). By plotting the magnitude of change against the measure of significance, we identified the reduced form of nicotinamide adenine dinucleotide (NADH) as the most significantly depleted metabolite in autophagy-deficient cells (Fig. 1d, Extended Data Fig. 3b). NAD+, the oxidized form of the NAD dinucleotide, was also significantly decreased, suggesting that autophagy-deficient cells present with a depletion of the total pool of the NAD(H) (Fig. 1d, Extended Data Fig. 3b, c). We further confirmed NAD(H) deficit via a fluorescence-based assay in Atg5-/- MEFs (Fig. 1e).
Evolutionarily conserved role of autophagy in maintaining NAD(H) levels
Autophagy is required for the survival of eukaryotic organisms from yeast to man25. We investigated whether the role of autophagy in the maintenance of intracellular NAD(H) pools is evolutionarily conserved. Knockdown of Atg5 in the fruit fly, Drosophila melanogaster, reduced autophagic flux as was evident from accumulation of autophagy substrate Ref(2)P26, and resulted in significant depletion of NAD(H) (Fig. 1f, Extended Data Fig. 3d). We further analyzed nitrogen-deprived Saccharomyces cerevisiae yeast that are dependent on mitochondrial respiration, wherein autophagy deficiency causes loss of respiratory capacity and cell death11. We found that nitrogen-deprived atg5Δ yeast exhibited both the loss of autophagy flux as monitored by Atg8-GFP cleavage26 and a striking depletion of NAD(H) levels (Fig. 1g, Extended Data Fig. 3e). Together with our observations in MEFs, these data imply an evolutionary-conserved role of autophagy in the preservation of NAD(H) levels.
Increased NADase activity mediated NAD(H) depletion in autophagy-deficient cells
We further investigated the role of NAD(H) in mediating cytotoxicity underlying autophagy deficiency. Inhibition of nicotinamide phosphoribosyltransferase (NAMPT) involved in NAD biosynthesis via a salvage pathway27 (Fig. 2a) by FK866 compromised the viability of wild-type MEFs in galactose, but not glucose, medium, indicating that NAD(H) is a limiting factor for the survival of OXPHOS-dependent cells (Extended Data Fig. 4a, b). We therefore investigated the mechanism of NAD(H) depletion in autophagy-deficient cells. Activities of two main classes of NAD-consuming enzymes, poly-ADP-ribose polymerases (PARPs) and deacetylases of sirtuin family (SIRTs)28, were increased in Atg5-/- MEFs after 14 h culture in galactose media. This was evident from elevated levels of poly-ADP-ribosylation (PARylation) and reduced protein acetylation, respectively (Fig. 2b). PARylation, but not deacetylation, activity remained elevated after 20 h of culture (Extended Data Fig. 4c), consistent with the reliance of SIRTs but not PARPs on high cellular NAD+ levels15. SIRTs and PARPs are activated by reactive oxygen species (ROS) and DNA damage that were found to be significantly elevated in respiring Atg5-/- MEFs (Extended Data Fig. 4d-h), likely resulting from mitochondrial dysfunction due to loss of autophagic quality control14,29. Indeed, OXPHOS-dependent Atg5-/- MEFs displayed disrupted mitochondrial morphology, altered levels of electron transport chain supercomplexes, and reduced ATP generation via OXPHOS (and increased reliance on glycolysis in glucose media) (Extended Data Fig. 4i-l).
We did not find evidence for PARP hyperactivation as a direct cause of cell death which is mediated by the mitochondria-to-nucleus translocation of apoptosis-inducing factor (AIF) (Extended Data Fig. 5a). Therefore, we hypothesized that the loss of cell viability is triggered by NAD(H) exhaustion due to uncontrolled NAD+ cleavage. Indeed, pharmacological inhibition of SIRTs with sirtinol or PARPs with olaparib partially rescued both, intracellular NAD(H) levels and cell viability of Atg5-/- MEFs (Fig. 2a, c, d, Extended Data Fig. 5b-d). The role of NADases was further validated by siRNA-mediated knockdown of Sirt1 or Parp1, indicating involvement of these enzymes in NAD(H) depletion and cell death (Extended Data Fig. 5e-h).
Boosting intracellular NAD(H) rescued viability of autophagy-deficient cells
To test whether boosting intracellular NAD(H) levels is sufficient to rescue viability of autophagy-deficient cells, we utilized the native cellular capacity for NAD+ synthesis (Fig. 2a). Supplementation of bioavailable NAD+ precursors, nicotinamide (NAM) or nicotinamide riboside (NR), led to the recovery of intracellular NAD+ and NADH levels, and rescued viability of Atg5-/- MEFs (Fig. 2e, f, Extended Data Fig. 6a). The effect of NAM (but not NR) was abrogated by FK866 (Fig. 2e, f, Extended Data Fig. 6a), consistent with the requirement for conversion of these precursors to NAD(H) via the salvage pathway (Fig. 2a). Both precursors also inhibited PARP and/or SIRT activity28 (Extended Data Fig. 6b). However, rescue of NAD(H) and cell viability did not correlate with PARPs/SIRTs activity because FK866, which also suppressed PARylation/deacetylation by reducing NAD(H) levels, did not rescue cell viability (Fig. 2e, f, Extended Data Fig. 6a, b). Furthermore, in the presence of FK866, NR (but not NAM which requires NAMPT for conversion to NAD+) improved NAD(H) levels and cell viability whilst PARylation/deacetylation activity was instead partially rescued compared to FK866 alone (Fig. 2e, f, Extended Data Fig. 6a, b). We conclude that boosting NAD(H) levels rescued cell death downstream of PARPs/SIRTs.
We performed an unbiased metabolomics profiling of Atg5-/- MEF in galactose medium, treated with or without NAM, to assess any correlation between the rescue of cell death and recovery of intracellular metabolites. By plotting the magnitude of rescue against the measure of significance, we found NADH to be the only metabolite that primarily correlated with Atg5-/- MEF viability, i.e., it was first found to be significantly depleted in Atg5-/- MEF (Fig. 1d, Extended Data Fig. 3b, c) and then significantly restored by NAM supplementation (Fig. 2g, Extended Data Fig. 6c, d). NAD+ followed a similar trend (Fig. 1d, 2g, Extended Data Fig. 3b, c, 6c, d). Therefore, the rescue of cell death is mediated by increased NAD(H), but not, for example, ATP (Fig. 2g, Extended Data Fig. 6c, d). The role of NAD(H), but not PARP/SIRT directly, was further supported by NAD(H) supplementation via de novo pathway using L-tryptophan (Fig. 2a), which rescued NAD(H) levels and cell death whilst having no inhibitory effect on PARP/SIRT activities (Fig. 2h, i, Extended Data Fig. 6e, f). Together, the salvage and de novo pathways remain active in autophagy-deficient cells and NAD(H) depletion is primarily mediated by its increased consumption.
Since Atg5-/- MEFs manifested with general nucleotide depletion (Fig. 1d, Extended Data Fig. 3b, c), we supplemented cells in galactose medium with five nucleosides and found that all were able to restore cell viability whilst NAD(H) levels were not rescued (Extended Data Fig. 7a-c). Therefore, NAD(H) decline is independent of the previously proposed purine/pyrimidine depletion mechanism of cell death in autophagy-deficient tumor-derived cells24. Inhibition of Nrf2, previously shown to mediate cell death in autophagy deficiency30, also rescued viability without affecting NAD(H) levels (Extended Data Fig. 7d-f). Therefore, NAD(H) decline is an additional and previously unidentified cause of cell death due to autophagy deficiency.
NAD(H) depletion mediated cell death via mitochondrial depolarization
NADH was predominantly detected in a mitochondria-enriched cell fraction where it was depleted in Atg5-/- MEFs cultured in galactose medium (Fig. 3a). Oxidation of NADH generates mitochondrial membrane potential (ΔΨm) across the inner mitochondrial membrane and we hypothesized that depletion of mitochondrial NADH is the cause of mitochondrial depolarization and apoptosis14,31. Indeed, boosting NAD(H) levels with NAM rescued membrane depolarization in Atg5-/- MEFs (Fig. 3b, c). Furthermore, preventing dissipation of ΔΨm by suppressing ATP synthase activity (using a low dose of oligomycin)32, partially prevented depletion of NADH and rescued cell death (Fig. 3b-e, Extended Data Fig. 8a). Consistent with no effect of oligomycin on NAD+ levels (Fig. 3d), SIRT and PARP activities remained unaffected (Extended Data Fig. 8b), indicating that oligomycin acted downstream. To further test the role of mitochondrial NADH in the cell death phenotype, we overexpressed a non-proton pumping alternative NADH oxidase, NDI133. In galactose medium, overexpression of NDI1 led to an increased oxidation state of the NAD(H) pool which was sufficient to enhance the apoptotic phenotype of Atg5-/- MEFs (Fig. 3f, g, Extended Data Fig. 8c, d). We conclude that NADH is the limiting factor in the survival of autophagy-deficient cells.
Generation of autophagy-deficient human neurons from ATG5-/- hESCs
To test the relevance of our findings in a physiologically relevant human system, we employed autophagy-deficient (ATG5-/-) human neurons by harnessing the pluripotency property of human embryonic stem cells (hESCs). Towards this, we first generated ATG5-/- hESCs by knockout of ATG5 exon 3 via genome editing with transcription activator-like effector nucleases (TALENs) (Extended Data Fig. 2d, 9a, b). ATG5-/- hESCs expressed pluripotency markers, such as NANOG, SOX2, OCT4, SSEA4 and TRA-1-60, comparable to wild-type (ATG5+/+) hESCs (Extended Data Fig. 9c-e). Loss of autophagy was confirmed by the absence of ATG5-ATG12 conjugate and LC3-II (autophagosomes), and accumulation of p62 (autophagy substrate)26 in ATG5-/- hESCs (Extended Data Fig. 10a-d). Multiple clones of ATG5-/- hESCs were differentiated into neural precursors (NPs) and then to neurons, which were confirmed for their cellular identity with cell-specific markers (Fig. 4a, Extended Data Fig. 11a-e) and loss of autophagy (Fig. 4b, c, Extended Data Fig. 11f-h). Further validation of our hESC-based human cellular platforms with autophagy deficiency was made by complementing ATG5-/- hESCs with human ATG5 mRNA via lipid nanoparticles (LNPs), formulated with the ionizable lipid C12-20034 (Extended Data Fig. 12a). C12-200 LNP-mediated delivery of human ATG5 mRNA, but not GFP mRNA, generated the ATG5-ATG12 conjugate and restored functional autophagic flux in ATG5-/- hESCs and neurons (Extended Data Fig. 12b-e).
Loss of NAD(H) underlies cytotoxicity in human neurons with autophagy deficiency
We next investigated the mechanistic link between NAD(H) and cell survival in human neurons. While hESCs rely on glycolysis for energy production and pluripotency35, differentiated cells like neurons with high energy demands are dependent on mitochondrial OXPHOS for ATP generation36. This allows analysis in neurons to be performed at basal state by eliminating the need for medium switch to galactose. We found that NAD(H) levels were significantly depleted in ATG5-/- hESC-derived neurons 3 weeks after differentiation (Fig. 4d), further supporting the evolutionarily conserved role of autophagy in maintaining NAD(H) pool (Fig. 1e-g). Since oxidation of NADH generates ΔΨm 14, depletion of NADH was associated with mitochondrial depolarization (Extended Data Fig. 13a). Autophagy-deficient neurons also exhibited elevation in ROS (Fig. 4e), mitochondrial fragmentation and increased mitochondrial load (Extended Data Fig. 13b-f), presumably due to accumulation of damaged mitochondria arising from abrogation of autophagic clearance. In agreement with a role of NAD-consuming enzymes in depleting NAD(H) levels, PARylation and deacetylation activities of PARPs and SIRTs, respectively, were increased in ATG5-/- hESC-derived neurons (Fig. 4f). Notably, increased cell death was observed at a basal level in ATG5-/- neurons between 2 and 4 weeks after differentiation (Fig. 4g-i, Extended Data Fig. 13g, h). Collectively, these data in autophagy-deficient human neurons were consistent with our finding in Atg5-/- MEFs.
Pharmacological elevation of NAD(H) is cytoprotective in human neurons with autophagy deficiency or dysfunction
Next, we evaluated the effects of pharmacologically modulating NAD(H) levels (Fig. 2a) on cell viability of ATG5-/- neurons. NAM increased NAD(H) levels, and consequently restored ΔΨm in ATG5-/- hESC-derived neurons (Fig. 5a, b). These changes were associated with improvement in axonal length and rescue of cell death in ATG5-/- neurons after NAM treatment, which did not affect ATP or ADP levels (Fig. 5c-f, Extended Data Fig. 14a-c). NR also rescued the viability of ATG5-/- neurons (Extended Data Fig. 14d, e). Inhibition of NAD(H) production by FK866 further augmented cell death in ATG5-/- neurons (Extended Data Fig. 15a-d) and abolished the cytoprotective effect of NAM (Fig. 5f), thus implying a role of NAD(H) in governing neuronal survival.
Finally, we utilized disease-affected human neurons differentiated from patient-derived human induced pluripotent stem cells (hiPSCs) of a neurodegenerative lysosomal storage disorder, Niemann-Pick type C1 (NPC1) disease. As previously reported37, NPC1 neurons presented with a severe autophagic defect (Extended Data Fig. 16a-c) and cell death between 3 and 4 weeks after differentiation (Fig. 5g). Similar to our findings in autophagy-deficient models, autophagy dysfunction in NPC1 neurons correlated with NAD(H) depletion (Fig. 5h). Treatment with NAM was able to rescue both, NAD(H) levels and cell death phenotype in NPC1 neurons (Fig. 5g, h), thus showing therapeutic benefits.