Autophagy: From Molecular Mechanisms to Disease Regulation and Therapeutic Strategies

Macroautophagy was initially described as a lysosomal transport process for stress-induced cellular homeostasis. Current research indicates it has distinct, sometimes opposing, functions in tumorigenesis and neuroprotection [9,10]. The initiation of macroautophagy involves a complex molecular cascade. Liquid–liquid phase separation (LLPS) acts as a biophysical mechanism for the rapid assembly of the ULK complex (ULK1/2/FIP200) under mTOR regulation [11]. Subsequently, the class III PI3K complex initiates nucleation. Membrane expansion and lysosomal fusion are facilitated by the ATG12–ATG5–ATG16L1 system, LC3–PE conjugation, and SNARE proteins [11]. The regulatory network of macroautophagy is multifactorial. For example, calcium signaling activates autophagy via the CaMKKbeta-AMPK pathway by inhibiting mTOR, while endoplasmic reticulum (ER)-bound Bcl-2 concurrently suppresses this process [12]. Current research often overlooks the limits of this cellular regulatory capacity. Prolonged or extreme macroautophagic stimulation in quiescent cells can exceed lysosomal degradation limits. This creates functional competition between autophagic flux and lysosomal integrity. Consequently, excessive stimulation may alter cell fate decisions and delay apoptotic clearance, complicating its therapeutic application.

Selective autophagy degrades specific intracellular targets, such as damaged mitochondria (mitophagy) or pathogens. This specificity depends on interactions between ATG8-family proteins and LC3-interacting regions (LIR) on cargo receptors [13,14]. Recent studies show this mechanism is highly dynamic. Multiple substrates, including glycogen and ribosomal proteins, can attach to lipidated ATG8 via receptors (e.g., 45, Hab1) or the Snx4-20 heterodimer. This multi-substrate integration increases degradation efficiency [15,16,17]. Selective autophagy demonstrates species-specific evolutionary conservation. In plants, the VISP1 receptor targets the viral silencing suppressor CMV2b to reduce pathogenicity [18]. However, pathogens and tumor cells frequently hijack this machinery. In glioblastoma, the RNA editing enzyme ADAR1 utilizes p62-mediated selective autophagy to promote tumor growth and temozolomide (TMZ) resistance. Similarly, the KSHV virus uses the NDP52 receptor to degrade P-bodies and facilitate viral replication [19,20,21]. The regulation of selective autophagy also involves post-translational modifications, such as TBK1-dependent phosphorylation of TAX1BP1, which enhances ubiquitin binding during autophagosome assembly. These findings indicate that while selective autophagy is a potential therapeutic target, its broad pharmacological modulation may unintentionally promote viral replication or tumor progression.

Chaperone-mediated autophagy (CMA) operates as a highly specific, non-vesicular degradation pathway. Cytosolic proteins containing a KFERQ pentapeptide motif are recognized by the heat shock cognate protein 70 (Hsc70). The foundational mechanism of CMA was delineated with the identification of the lysosomal membrane protein 2A (LAMP2A) as the essential receptor mediating substrate unfolding and direct translocation into the lysosome [22]. Recent original investigations reveal that CMA integrates key intracellular signaling and inflammatory pathways. For instance, in microglia, CMA actively degrades the NF-κB co-activator p300, which fundamentally suppresses uncontrolled NLRP3 inflammasome activation during neuroinflammation [23]. In contrast, glioma-associated CMA dysfunction stabilizes hypoxia-inducible factor (HIF)-1 alpha, thereby inducing chemo-resistance. A major limitation in CMA-targeted oncology research is the risk of systemic toxicity; inhibiting CMA to sensitize tumors to chemotherapy may inadvertently accelerate neurodegeneration by halting the clearance of neurotoxic protein aggregates [24].

Microautophagy involves the direct invagination of the lysosomal or vacuolar membrane to engulf cytosolic cargo, exhibiting distinct morphological subtypes in processes such as micropexophagy and micromitophagy. Groundbreaking mechanistic studies in mammalian models definitively demonstrated that membrane scission in microautophagy requires highly coordinated interactions between ATG proteins and the endosomal sorting complexes required for transport (ESCRT) [25]. While microautophagy fundamentally functions to repair lysosomal membranes and maintain intracellular homeostasis, its study in human diseases faces methodological bottlenecks. The lack of specific molecular markers for mammalian microautophagy limits precise in vivo tracking and complicates distinguishing its function from macroautophagy9. Advancing clinical applications requires prioritizing the discovery of specific mammalian markers and clarifying the synergistic crosstalk between these parallel autophagic pathways (Figure 1).

The effective maintenance of cellular proteostasis is fundamentally dependent on the coordinated activation of the Coordinated Lysosomal Expression and Regulation (CLEAR) gene network, a process largely governed by the master transcription factor TFEB. The revolutionary identification of TFEB as a central regulator [26] underscores why its nuclear exclusion is a common pathological bottleneck. In AD and PD, this failure is primarily enzymatic: TFEB is excessively phosphorylated by the mTORC1-GSK3β axis, which leads to its aberrant cytoplasmic retention and a failure to clear amyloid-β and Tau aggregates [27,28]. This enzymatic suppression is further intensified in PD by PARP1-induced poly-ADP-ribosylation, which directly obstructs TFEB’s nuclear import [29]. Complementing these enzymatic defects, ALS presents a more mechanical barrier to autophagy initiation. Mutations in C9orf72 are now known to disrupt the stoichiometric interaction between the nucleoporin POM121 and the Sigma-1 receptor [30]. This defect physically compromises the nuclear pore complex, trapping TFEB in the cytosol and facilitating the pathological spread of TDP-43 [31]. Interestingly, this transcriptional “paralysis” extends even to psychiatric contexts such as major depressive disorder, where hippocampal SIRT1 deficiency leads to the repression of essential autophagy genes like ATG5 [32]. The mechanistic link between SIRT1 and the CLEAR network was recently clarified by evidence showing that SIRT1 must directly deacetylate TFEB at the Lys116 residue to permit its nuclear entry [33].

While these models offer a consistent mechanistic narrative, their clinical translatability warrants critical caution. A significant methodological limitation is that much of this data stems from acute, high-dose protein overexpression in transgenic mice, which may artificially amplify the collapse of the CLEAR network beyond what is observed in the chronic, decades-long progression of human disease. Furthermore, the temporal dynamics of TFEB activation are still debated; some clinical evidence suggests an initial compensatory upregulation of autophagy in the prodromal stages of AD, implying that TFEB exhaustion may be a consequence of late-stage systemic failure rather than an initiating driver [34]. Resolving these biological nuances is essential for developing spatiotemporal-specific modulators that avoid the risks of lysosomal over-activation.

Beyond the failure to initiate autophagy, the physical presence of protein aggregates actively sabotages the machinery, causing autophagic flux impairment or vesicular transport stasis. In HD, for instance, mature fibrillar mutant huntingtin aggregates act as competitive decoys that sequester selective receptors like p62 and essential ATG proteins [35]. This sequestration effectively depletes the functional pool of these components, arresting autophagosome maturation in distal neurites [36]. A parallel mechanism of physical obstruction exists in the non-vesicular CMA pathway during PD pathogenesis, where α-Synuclein binds to the LAMP2A receptor with abnormally high affinity, physically clogging the translocation channel [8]. These transport deficits are fundamentally compounded by downstream degradation failures, particularly in AD. Original biochemical investigations have pinpointed the primary bottleneck to a failure in lysosomal acidification, driven by defects in the v-ATPase assembly [37]. This failure in the terminal digestive step leads to a massive accumulation of undigested autophagic vacuoles (AVs), a pathology exacerbated by mutations in LC3-associated endocytosis [38].

The “aggregate-blockade” hypothesis, while widely accepted, requires a critical re-evaluation of its methodological underpinnings. For instance, much of the evidence for LAMP2A clogging remains limited to in vitro isolated lysosome assays, and quantifying this obstruction within intact, living neural networks remains a significant technical hurdle. Moreover, the role of cargo accumulation is increasingly contested by emerging perspectives on liquid–liquid phase separation (LLPS), which suggest that p62-mHTT assemblies might actually be protective scaffolds designed to isolate toxic oligomers rather than mere “garbage” causing congestion [39]. Such conflicting interpretations highlight the unresolved bottleneck of substrate-state dependence: while existing activators can clear early-stage non-fibrillar oligomers, they may paradoxically worsen the lethal accumulation of AVs when applied to late-stage pathology characterized by dense, insoluble plaques.

Failure of the autophagy-lysosome pathway to clear damaged organelles, notably mitochondria, creates a vital pathophysiological link between proteostasis collapse and chronic neuroinflammation. In healthy cells, the PINK1-Parkin pathway marks dysfunctional mitochondria and facilitates their autophagic engulfment by recruiting LC3/GABARAP [40]. This targeted mitophagy is severely impaired in PD, frequently due to LRRK2 mutations. These mutations phosphorylate Rab proteins, which in turn blocks the recruitment of essential mitophagy receptors [41]. When defective mitochondria escape degradation, they leak excessive reactive oxygen species (ROS) and mitochondrial DNA (mtDNA) into the cytosol, acting as major neuroinflammatory triggers. Evidence from major depressive disorder (MDD) models shows that ROS accumulation hyperactivates the NLRP3 inflammasome in microglia [42]. Consequently, these activated microglia release pro-inflammatory cytokines like IL-1β. This triggers NF-κB signaling, which further inhibits autophagic flux. The result is a destructive feedback loop where mitochondrial decline and escalating inflammation reinforce each other.

Although this loop is evident in vitro, translating these findings to human pathology requires careful assessment. Current research relies heavily on acute in vivo stimulation, such as high-dose lipopolysaccharide (LPS) injections. Such models rarely reflect the chronic, low-grade “inflammaging” seen in the aging human brain. Additionally, the causal order of autophagic decline and neuroinflammation remains debated. While most theories label autophagic failure as the primary cause, longitudinal data suggests that early, brief inflammatory bursts might actually trigger compensatory autophagy as a defense. Pathway exhaustion may only occur after long-term exposure. Addressing this causal uncertainty is essential for creating treatments that decouple mitophagy defects from inflammasome activation without causing broad immunosuppression.

Beyond cell-autonomous defects, the microbiota–gut–brain axis (MGBA) functions as a key systemic regulator of neuronal proteostasis. Initial evidence for this connection came from studies demonstrating that gut microbiota are essential for the motor deficits and neuroinflammation seen in α-Synuclein overexpressing mice [43]. Under normal conditions, symbiotic gut bacteria generate metabolites like short-chain fatty acids (SCFAs). These SCFAs cross the blood–brain barrier and stimulate neuronal autophagic flux through the AMPK-mTOR axis [44]. In contrast, gut dysbiosis impairs the intestinal epithelial barrier. This “leaky gut” state permits bacterial endotoxins, such as LPS, to enter systemic circulation. Once circulating, these toxins activate TLR4 signaling to trigger neuroinflammation, which subsequently inhibits neuronal autophagy. This process accelerates the buildup of Aβand α-Synuclein. These mechanisms support the “gut-first” hypothesis in PD. This theory suggests that pathogenic aggregation starts in the enteric nervous system and moves to the brain through the vagus nerve.

Despite the novel systems-biology perspective offered by the MGBA, clinical translation faces significant hurdles. Most human studies establish merely a correlation, rather than causation, between gut dysbiosis and neurodegeneration. While fecal microbiota transplantation (FMT) has reactivated autophagy in rodent models [45], large-scale clinical validation in humans is still lacking. Furthermore, profound interspecies differences in microbiome composition mean that results from highly controlled laboratory settings often fail to replicate in heterogeneous human populations. A key scientific challenge remains identifying specific microbial metabolites—such as certain SCFAs or mitophagy-inducers like Urolithin A—that can selectively modulate brain autophagy without causing systemic disturbances [46]. Whether the “gut-first” hypothesis applies universally or only to specific clinical subtypes remains to be validated by prospective longitudinal studies (Figure 3).

Given the profound mechanistic convergence in the collapse of the autophagy-lysosome pathway across diverse neurological disorders, contemporary therapeutic innovation is shifting from strictly disease-specific symptomatic treatments toward unified strategies centered on restoring neuronal autophagic flux. The most classical approach in this domain involves broad-spectrum mTOR inhibitors, such as rapamycin. By disinhibiting autophagy, these agents have demonstrated significant potential in promoting the clearance of toxic aggregates in preclinical models of AD, PD, and HD [5,47]. Beyond general modulators, current research is shifting toward high-precision interventions. In Alzheimer’s disease (AD) studies, for example, researchers are developing proteolysis-targeting chimera (PROTAC) molecules. These are engineered to identify and eliminate Tau aggregates through the autophagic system [48]. Progress in Amyotrophic Lateral Sclerosis (ALS) treatments has likewise moved toward genetic and receptor-level repair. One approach uses SIGMAR1 agonists, such as pridopidine, to restore the TFEB-mediated transcriptional network [30]. Another strategy involves AAV9-based gene therapy to replace missing optineurin (OPTN), which restarts the specific mitochondrial clearance process [49]. Additionally, evidence shows that standard antidepressants can trigger autophagy via protein kinase A-dependent pathways. This discovery suggests a new method for breaking the neuroinflammatory feedback loops associated with Major Depressive Disorder (MDD) [42,50] (Figure 4).

The clinical transition of these promising therapies, however, is not without its perils. Despite the success of autophagy activators in transgenic models, their human application is obstructed by formidable pharmacokinetic and safety hurdles. Primarily, blood–brain barrier (BBB) permeability remains a decisive bottleneck; many potent modulators, particularly large macromolecular degraders and viral vectors, fail to achieve therapeutically effective concentrations within the central nervous system upon systemic administration. Secondly, the fundamental nature of autophagy as a universal homeostatic mechanism poses a risk of systemic toxicity. Blindly upregulating autophagic flux across all tissues may inadvertently disrupt the metabolism of healthy peripheral organs or even trigger autophagic cell death upon hyperactivation. Perhaps the most critical obstacle is the current lack of reliable, non-invasive in vivo biomarkers. Without the ability to dynamically monitor autophagic flux within the living human brain, it remains exceptionally difficult to ascertain target engagement or to define a safe “optimal therapeutic window.” Addressing these translational gaps—through advanced nanodelivery systems and the identification of cerebrospinal fluid or blood exosome biomarkers—represents the next essential frontier for autophagy-targeted medicine.

Obesity is a systemic metabolic disease that significantly increases the risk of T2D, metabolic-associated fatty liver illness, and cardiovascular diseases [55]. A shared mechanism across these metabolic states is the disruption of the nutrient-sensing AMPK/mTOR axis. Under physiological conditions, nutrient starvation recruits AMPK to initiate autophagy via ULK1 [56]. Conversely, chronic nutrient excess hyperactivates mTORC1 and suppresses AMPK signaling [56]. This signaling imbalance leads to lipotoxicity and inflammation, exhibiting distinct organ-specific bottlenecks [57]. The downstream consequences share a common autophagic impairment. In adipose tissue, an increase in CD36 transporters causes lysosomal calcium overload. This process stops the breakdown of lipid droplets [58]. Within the liver, the primary lysosomal regulator, TFEB, fails to move into the nucleus, which prevents lysosomal genesis [58]. Similarly, in Type 1 Diabetes (T1D), faulty autophagosome-lysosome connections lead to p62 accumulation. This lack of LC3-LAMP1 colocalization triggers β-cell apoptosis [59]. The AMPK/mTOR axis serves as a general framework, yet local microenvironments significantly shift these pathways. For example, metabolites like lactate can trigger autophagy through the ERK/p90RSK pathway, bypassing mTOR. Additionally, TGF-β/Smad signaling mediators help determine β-cell survival [60]. Such metabolic diversity indicates that using broad-spectrum autophagic modulators might result in inconsistent clinical outcomes.

The crosstalk between autophagy and the NLRP3 inflammasome is a critical mechanistic hub in metabolic diseases. Autophagy functions to clear damaged mitochondria that produce reactive oxygen species [61]; this process fails during diabetic complications, leading to NLRP3 hyperactivation [62]. In T2D, selective mitophagy is initially activated to maintain homeostasis [63]. However, chronic metabolic stress eventually overwhelms this adaptation, leading to damaged organelles and islet disintegration [59]. The causal relationship between autophagic decline and metabolic inflammation requires strict evaluation. Recent evidence indicates that natural substances like kaempferol normalize metabolic imbalance in diet-induced obese (DIO) mice by targeting the mitochondrial protein TUFM to revive autophagy [64]. Yet, the evolutionary conservation of these pathways, such as the Stb5 transcription factor controlling autophagy through NADPH synthesis and redox status in the yeast Saccharomyces cerevisiae [65], does not fully recapitulate the multi-organ complexity of human metabolic syndromes.

Therapies capable of restoring this fragile signaling state are gaining emphasis [66]. Pharmacologic interventions, including Metformin and GLP-1 receptor agonists, reconnect the AMPK axis to shield cells against glucotoxicity. Concurrently, non-pharmacologic interventions like exercise and Vitamin D3 supplementation aim to resolve dysregulated autophagic flux [67], providing an opportunity to protect functional β-cell mass and stop disease onset [68]. Furthermore, emerging treatments such as marine-derived plasmalogens (PE-P), rapamycin, and ω-3 polyunsaturated fatty acids utilize various pathways (e.g., PPAR-γ) to increase autophagy, ultimately diminishing systemic lipotoxicity and metabolic hazard [69,70,71] (Figure 5). The fundamental challenge in metabolic autophagy therapeutics is achieving tissue-specific targeting. Since autophagy serves critical homeostatic functions, systemic pharmacological activation risks unintended consequences, necessitating the development of precise delivery systems. To synthesize the diverse pharmacological landscape discussed above, Table 1 summarizes the key autophagy-targeted modulators, their distinct mechanisms of action, and their current clinical stages across different pathological contexts.

No new data were created or analysed in this study. Data sharing is not applicable to this article.