Regulated cell death in the nervous system: a framework for execution diversity and spatially restricted degeneration

1 Introduction

Multicellular animals maintain tissue homeostasis and remove damaged cells by regulated cell death (RCD). The classical dichotomy between necrosis, the only initially recognized form of (unregulated) cell death, and apoptosis, the first discovered and most widely known form of RCD, has now proven insufficient to explain the diversity of cell death phenotypes in tissues. Advances in molecular genetics, imaging, and biochemistry have revealed multiple RCD programs that are actively executed and tightly controlled. In the nervous tissue, where cellular replacement is limited and proper functional connectivity is fundamental, even low-level RCD activation can have disproportionate consequences. Today, we recognize numerous types of RCD in addition to apoptosis. Apoptosis may follow an intrinsic pathway, in which the pro-apoptotic members of B-cell lymphoma 2 (BCL-2) family of proteins permeabilizes the mitochondrial outer membrane (MOM) to trigger death, or an extrinsic pathway, mediated by cell-surface receptors. The intrinsic and extrinsic pathways of apoptosis activate caspases, a class of proteases that mediate cell death during the execution phase, to which various forms of RCD converge. Although these mechanisms have been widely investigated, grand challenges in the field include understanding the reciprocal interactions and biological significance of the various forms of RCD, as well as their compartmentalization within neurons. This review will analyze the different types of RCD based on their molecular, structural, and morphological characteristics. Notably, the diverse forms of RCD in the nervous tissue are highly context-dependent, often incomplete, and frequently overlap at the molecular and morphological levels. Understanding the biological mechanisms of RCD is essential to cellular and molecular biology and to translating fundamental knowledge into potential therapeutic applications.

1.1 Historical notes

Following Schleiden and Schwann’s 1838 cellular theory, embryologists began studying cellular generation, development, and death. However, it took more than a century to fully understand the evolutionary and functional significance of cell death, and more than 100 publications on the subject appeared in the 19th century, attesting to its importance at the time (Clarke and Clarke, 1996). Pathologists were among the first to notice cell death in injured tissues (Virchow, 1858). However, to convince academics of the importance of cell death in normal development, growth, and aging, substantial experimental research was required. Eventually, studies of insect and amphibian metamorphosis led to the recognition of cell death as non-pathological (Clarke and Clarke, 1996).

Namely, in 1842, Vogt studied the formation of the anuran notochord, which initially developed from the chordamesoderm and was later replaced by cartilage as its cells were eliminated (Vogt, 1842). The cellular death modes of complex organisms were at that time revealed through simultaneous investigations at the organ- and tissue-levels. Various species were studied for endochondral ossification, a process in which cartilage is replaced by bone. This process involves chondrocyte loss and osteoblast replacement, leading to the formation of mature bone (Stieda, 1872). Also examined were granulosa cell degeneration in mature ovarian follicles and mammary gland cell death during breastfeeding. In 1885, Flemming began studying dying mammary gland cells and made the first detailed observations on well-preserved tissues, coining the term ‘chromatolysis’ to describe the loss of staining in dying cells (Flemming, 1885). However, as observations of physiological cell death accumulated, the conceptual framework for understanding necrosis as a distinct pathological process was being developed in parallel. Virchow was the first to recognize that more than one form of cell death could occur, applying the term necrosis to an advanced state of tissue breakdown at the gross level, while coining the term ‘necrobiosis’ to describe a separate, cell-level process in which the cell vanishes and can no longer be seen in its original form (Virchow, 1858). He explicitly defined necrobiosis as the result of “spontaneous wearing out of living parts,” distinguishing it from necrosis as a natural rather than violent form of cell death (Virchow, 1858). This early distinction between regulated and unregulated cell death would prove conceptually important, though its significance would not be fully appreciated for over a century (Buja, 2021). In 1914, Gräper proposed that some mechanism must counterbalance mitosis, and that this process likely involved the chromatolysis described by Flemming (Gräper, 1919). This paper was subsequently rediscovered by Glücksmann in 1951 (Glücksmann, 1951). In this landmark review, Glücksmann systematically compiled observations of cell death in normal vertebrate development from the literature of the preceding half century, emphasizing the frequent occurrence of cell death across a wide range of developmental contexts (Glücksmann, 1951). Glücksmann identified three phases of developmental cell death—phylogenetic, morphogenetic, and histogenetic—providing the first systematic classification of physiological cell death. This work helped to reestablish cell death as a normal developmental phenomenon rather than a purely pathological one, at a time when the field had largely turned its attention elsewhere (Fricker et al., 2018; Park et al., 2023). The direct precursor to the concept of apoptosis came from pathological studies of the liver. In 1965, Kerr ligated a branch of the hepatic portal vein in rats, inducing liver atrophy, and observed two distinct forms of cell death: classical necrosis and a separate process in which hepatocytes rounded off into small cytoplasmic bodies that were subsequently engulfed by neighboring cells or histiocytes (Kerr, 1965). Between 1967 and 1970, Kerr carried out an electron-microscopic study of this process, finding that its earliest recognizable stage involved condensation of the cytoplasm and nuclear chromatin, with compacted chromatin aggregating beneath the nuclear envelope; at the same time, the resulting membrane-bounded bodies contained well-preserved organelles (Kerr, 1969). Kerr termed this form of death “shrinkage necrosis” and, in his 1971 paper, formally described it as a distinct mode of cellular death, showing through ultrastructural analysis how it differed from classical necrosis (Kerr, 1971). In the following year, Kerr, Wyllie, and Currie (Kerr et al., 1972) were the first to coin the term apoptosis to denote this form of cell death, different from necrosis, the prototype of non-regulated cell death. The term apoptosis was derived from the Greek words “apo” and “ptosis,” which indicate “to detach” or “to fall off,” for example, of ripe fruit or decaying tree leaves. However, although at that time it was well known that many cells die during normal development and that these dying cell populations were often eliminated during the evolution of multicellular organisms, it took at least a couple of decades before apoptosis, also referred to as programmed cell death (PCD), became recognized as the prototype of several forms of RCD. Today we know that apoptosis, as an ultrastructurally well-defined form of RCD, occurs throughout normal development and tissue turnover (Migheli et al., 1994; Stewart, 1994; White, 1996; Marks and Berg, 1999). Studies of PCD in Caenorhabditis elegans, a worm in which 131 of its 1,090 somatic cells undergo apoptosis during adulthood, have substantially improved our understanding of mammalian apoptotic mechanisms by identifying the genes that regulate its progression (Horvitz, 1999). Since then, the field of RCD has expanded dramatically and revealed an unforeseen complexity. There is today accumulating evidence that other forms of RCD, distinct from apoptosis, are important under pathological conditions and have a role during normal development. For example, autophagic cell death has been shown to play a role in embryonic cavitation (Qu et al., 2007), Drosophila metamorphosis (Berry and Baehrecke, 2007), and zebrafish cardiac morphogenesis (Lee et al., 2014).

2 RCD pathways

RCD may take several forms, be mediated by distinct cellular signaling pathways, and, at least in principle, be modulated by genetic or pharmacological interventions. The pathways activated in various forms of RCD often lead to distinct morphological changes in dying cells. Classification schemes proposed by the Nomenclature Committee on Cell Death (NCCD) emphasize molecular executors rather than morphology alone (Galluzzi et al., 2018), as many ultrastructural similarities exist among the different types of RCD. Importantly, RCD pathways are not mutually exclusive, and hybrid and sequential activation states are common, particularly in neural cells.

2.1 Apoptosis

In the nervous system, apoptosis represents the archetypal RCD program. As we will discuss in the following sections, it occurs through a cellular pathway involving several molecules, the most important of which are BCL-2 and its family members, the apoptosome, and the initiator and effector caspases. Apoptosis is indispensable during development, sculpting neuronal populations and refining synaptic connectivity. Developmental apoptosis is tightly regulated by the availability of growth factors (neurotrophins and others) and by transcriptional programs primarily involving BCL-2 family proteins. In mature neurons, however, apoptotic execution is constrained by high levels of endogenous caspase inhibitors and by the spatial segregation of death signals, suggesting that neurons, at least under certain conditions, may develop apoptotic resistance (Annis et al., 2016; Hollville et al., 2019). Resistance is achieved by inhibiting the intrinsic apoptotic pathway through multiple redundant and complementary mechanisms, such as downregulating pro-apoptotic proteins (e.g., caspases, apoptosis protease-activating factor 1 [Apaf1]), elevating miR-29 levels to target Bcl-2 Homology 3 (BH3)-only proteins, and augmenting inhibitors of apoptosis proteins (IAP)-mediated suppression. Yet, despite their resistance, neurons can still undergo apoptosis in response to several pathological conditions.

Research on C. elegans and D. melanogaster has taught us much about genes involved in apoptosis (Yuan, 1995; Meier and Evan, 1998; Liu and Hengartner, 1999; Hengartner, 2001). After cloning the orthologs of C. elegans death machinery genes in mammals and D. melanogaster, it became evident that these genes exhibit analogous functions in regulating apoptosis across all species examined to date. The ‘death mechanism’ in C. elegans consists of four genes - cell death abnormality gene 3 (ced-3), 4 (ced-4), 9 (ced-9), and egg-laying defective protein 1 (egl-1) - that execute PCD (Hengartner, 1997). Subsequent genetic studies in mutant mice demonstrated that orthologs of ced-3 and ced-4 were crucial for cellular apoptosis in mammals as well (Woo et al., 2000). These genes encode proteins that act sequentially to promote apoptosis. Ced-9 inhibits ced-3 and ced-4 activity and acts as a survival factor (Hengartner et al., 1992). Cells can die due to aberrant CED-4 protein synthesis, an adaptor molecule that regulates CED-3’s lethal function. Ced-4 functions upstream of ced-3 and is regulated by ced-9. Therefore, by interacting with CED-4, CED-9 protects cells from self-destruction (Shaham and Horvitz, 1996). EGL-1 is another C. elegans death effector (Liu and Hengartner, 1999). When the egl-1 function is compromised, all apoptotic cells survive. EGL-1 can cause PCD by binding to CED-9 and inhibiting its action. Inhibition of CED-9 produces an apoptosome tetramer composed of four CED-4 molecules, which initiates apoptosis (Conradt and Horvitz, 1998).

After the discovery of the ced-9 gene in C. elegans, a comparable protein, BCL-2, was found in mammals (Vaux et al., 1992; Hengartner and Horvitz, 1994). BCL-2 was the first of a family of proteins that includes at least twenty to twenty-five CED-9-related proteins that regulate apoptosis by either promoting or inhibiting it. Numerous reports have been published on the biology of these proteins (Edlich, 2018; Kale et al., 2018; Singh et al., 2019; Vogler et al., 2025). Based on their structure and function, they can be subdivided into anti-apoptotic, multidomain pro-apoptotic, and BH3-only pro-apoptotic proteins. Each family member possesses one of the four conserved BCL2-homology (BH) domains (BH1, BH2, BH3, or BH4). B-cell lymphoma-extra-large (BCL-XL) and BCL-2-like 2 (BCL-W) are anti-apoptotic, while Bcl-2-associated X-protein (BAX), Bcl-2 homologous antagonist/killer (BAK), Bcl-XL/Bcl-2-associated death promoter (BAD), and BH3-interacting domain death agonist (BID) promote cell death. Anti-apoptotic BCL-2 family members inhibit apoptosis by preventing BAX and BAK from forming pores in the mitochondrial outer membrane (MOM), thereby preventing the passage of ions and proteins into the cytosol (Liang and Fesik, 1997). Blocking of pore formation inhibits mitochondrial cytochrome c release, apoptosome formation, and cell death (Li et al., 1998; Luo et al., 1998). In nematodes, a second mechanism that suppresses cell death involves the interaction between CED-9 and CED-4, which regulate CED-3 to prevent apoptosis. This interaction inhibits the apoptosome assembly and CED3 activation (Adams and Cory, 2002). However, it does not occur in Xenopus oocytes and mammalian cells because BCL-2 and BCL-XL bind to Apaf1 but do not block CASP9 activation (Newmeyer et al., 2000).

The mammalian apoptosome consists of Apaf1, the homolog of CED-4; cytochrome c (also referred to as apoptotic protease activating factor 2, Apaf2); and the initiator caspase 9 (CASP9, also referred to as apoptotic protease activating factor 3, Apaf3). Caspases can degrade cellular components, but many apoptosome scaffolding proteins are required to activate them. CASP9, Apaf1, and cytochrome c are essential for the effector caspase 3 (CASP3) activation by cleaving the inactive zymogen (Liu et al., 1996; Li et al., 1997). Apaf1, which is inactive in monomeric form, binds cytochrome c released from mitochondria during apoptosis, and then associates with proCASP9 to form multimers, resulting in the formation of a heptameric wheel-like structure that forms the core of the approximately 1 MDa apoptosome (Adams and Cory, 2002). By this arrangement, the CARD (caspase recruitment domain) of Apaf1 is optimally positioned to engage with proCASP9 (Saleh et al., 1999), so that the N-terminal CARD domain of proCASP9 binds to the apoptosome and interacts directly with the Apaf1 CARD (Bratton and Salvesen, 2010). There are a few steps in the CASP9 activation process. First, the apoptosome acts as a scaffold that recruits proCASP9, which then induces a conformational change in the molecule. Then Apaf1 causes proCASP9 to oligomerize, forming active dimers. At this step, although not strictly required for initiation, the high local concentration of dimers enables rapid autocatalytic cleavage of proCASP9. Subsequently, the Apaf1 CARD can form a large, functional hetero-oligomer with active CASP9, suggesting that the CARD domain is sufficient for allosterically up-regulating protease activity. Apaf1 and CASP9 CARD interact to sequester the CASP9 CARD, which, when it is not bound, dramatically lowers the protease activity (Li et al., 2017). Activated CASP9 on the apoptosome functions as a holoenzyme to cleave and activate downstream effector caspases, namely, CASP3 and CASP7 (Bratton and Salvesen, 2010). This process is a very important step in the intrinsic apoptotic pathway. It switches CASP9 from “off” to “on” via the scaffolded heptamer. Research on human embryonic kidney 293T cells has shown that BCL-XL (Singh et al., 2019) interacts with CASP9 and Apaf1 to inhibit CASP9 maturation, a mechanism believed to be conserved from nematodes to humans (Hu et al., 1998). However, other studies have shown that members of the BCL-2 family do not interact with Apaf1 (Moriishi et al., 1999; Conus et al., 2000; Newmeyer et al., 2000).

The mammalian homologs of CED-3 are the caspases (Thornberry and Lazebnik, 1998). The first mammalian caspase identified was CASP1, also known as interleukin-1 beta-converting enzyme (ICE) (Yuan et al., 1993; Schwartz and Milligan, 1996). ICE is expressed in undifferentiated progenitors and granule cell precursors in the postnatal human cerebellar cortex, where it is associated with developmental apoptosis (Lossi et al., 1998). Further research indicated that CASP3 and CED-3 shared the greatest sequence homology (Fernandez-Alnemri et al., 1994; Nicholson et al., 1995). A total of 18 mammalian caspases have been identified across species, 12 in humans and 10 in mice. In contrast, other species have unique caspases, such as CASP13 in cattle (Koenig et al., 2001) and CASP16-18 in birds, fish, and non-placental mammals (Eckhart et al., 2008; Eckhart et al., 2025). The better characterized fourteen caspases are divided into three subfamilies based on their peptide preferences: (i) the ICE-protease subfamily of inflammatory caspases (CASP1, 4, 5, 13, and 14, along with murine CASP11 and 12); (ii) the CED-3 subfamily of apoptotic effector caspases (CASP3, 6, and 7); and (iii) the CASP2 subfamily of initiator caspases (CASP2, 8, 9, and 10) (Julien and Wells, 2017). Effector caspases with a short pro-domain degrade cell-essential proteins. However, initiator/inflammatory caspases have longer prodomains that facilitate protein interactions and may induce apoptosis only under specific conditions (perhaps not CASP1 and CASP11) (Thornberry and Lazebnik, 1998). Each caspase is synthesized as a zymogen that must be cleaved to become an active enzyme. Initiator caspases activate downstream caspases, triggering a proteolytic cascade that leads to apoptosis. Distinct pro-apoptotic signals activate different groups of caspases. CASP8, CASP10, CASP3, and CASP7 intervene in extrinsic apoptosis (see below), whereas CASP9, CASP7, and CASP3 mediate intrinsic apoptosis (Zimmermann et al., 2001). Through BID, a protein that controls mitochondrial cytochrome c release when cell-surface death receptors are activated, these two pathways are linked (Li et al., 1998; Luo et al., 1998).

CASP3 substrate selectivity is similar to that of CED-3 (Xue et al., 1996). Thus, the C. elegans CED-4/CED-3 apoptotic cascade and the D. melanogaster Dark/dApaf1/Dronc apoptosome are orthologs of the mammalian apoptosome. Remarkably, unlike mammals and D. melanogaster, C. elegans apoptotic cells display chromatin shrinkage and cytoplasmic condensation but no apoptotic bodies, highlighting distinct downstream mechanisms across phyla (Abrams et al., 1993; Kutscher and Shaham, 2017). Another notable difference with mammals is that cytochrome c is not required for apoptosome formation in flies (Yuan and Akey, 2013).

Many cell-killing events activate apoptotic pathways. The observation of comparable morphological and biochemical alterations across diverse apoptotic triggering events suggests that most apoptotic pathways converge on a few effectors (Sastry and Rao, 2000; Elmore, 2007). The order of caspase activation and mitochondrial cytochrome c release defines two main apoptotic pathways: the intrinsic and extrinsic pathways (Lossi, 2022). In intrinsic apoptosis, cytochrome c is released from the mitochondrial intermembrane space before caspase activation as a consequence of the MOM permeabilization (MOMP). Extrinsic apoptosis, on the other hand, begins with the activation of an initiator caspase (CASP8 or CASP10) independently of mitochondrial alterations. Intrinsic apoptosis is initiated by intracellular sensors that detect DNA damage, viral infection, or the absence of survival signals from surrounding cells. Pro-apoptotic signals from natural killer cells or CD8-positive cytotoxic T lymphocytes initiate the extrinsic apoptotic pathway (Yanumula and Cusick, 2021). One should not view the two pathways as separate or unrelated, as they cross-talk via BID in mitochondria. The final step of apoptosis, the execution, occurs at the end of both paths.

2.1.1 Intrinsic pathway

In neurons, the intrinsic (mitochondrial) pathway predominates. Ultrastructurally, neuronal apoptosis is characterized by chromatin margination, nuclear fragmentation, cytoplasmic condensation, and preservation of plasma membrane integrity, features that distinguish it from lytic RCD modalities (Galluzzi et al., 2007). Intrinsic apoptosis can be caused by several factors, including depletion of growth factors, DNA damage, endoplasmic reticulum (ER) stress, reactive oxygen species (ROS) buildup, replication stress, microtubule alterations, and mitotic abnormalities (Nuñez et al., 1990; Nuñez and Clarke, 1994; Brumatti et al., 2010; Czabotar et al., 2014; Roos et al., 2016; Pihán et al., 2017; Vitale et al., 2017). MOMP is a crucial step in the intrinsic apoptotic pathway (Galluzzi et al., 2018). As a consequence of MOMP, initiator caspases, particularly CASP9, are activated. Remarkably, CASP9 also promotes a feedback amplification of mitochondrial damage (Guerrero et al., 2012). The release of cytochrome c into the cytosol is also decisive. Cell-free experiments have shown that it results from a breakdown of the MOM and/or the mitochondrial permeability transition (MPT), which is governed by the voltage- and Ca²⁺⁻sensitive permeability transition pore (Blatt and Glick, 2001). Pore opening causes irreversible and extensive MOMP (Tait and Green, 2010).

The intrinsic apoptotic pathway begins with pro- and anti-apoptotic interactions among BCL-2 family members, which carefully regulate MOMP. Regulation of MOMP and its execution exhibit distinctive features in non-neuronal cells and in developing and mature neurons (Table 1). There are also differences in the regulation and implementation of MOMP between neurons and glia, particularly regarding the baseline apoptotic threshold, the expression of APAF1 and XIAP, the apoptosome competence, the primary consequences of MOMP, and the relevance of minority MOMP (Table 2).

TABLE 1. Apoptotic post-MOMP execution in mitotic cells, developing neurons, and mature neurons.

TABLE 2. Intrinsic apoptosis and post-MOMP execution in main CNS cell types.

Another mitochondrial membrane protein involved in intrinsic apoptosis is apoptosis-inducing factor (AIF). It is an ancient, bifunctional flavoprotein, similar to cytochrome c and bacterial oxidoreductases (Susin et al., 1999; Susin et al., 2000). In response to cell-killing signals, like oxidative stress or poly (ADP-Ribose) polymerase 1 (PARP1) hyperactivation, mitochondrial AIF bypasses the caspase/CAD cascade entirely. Even if caspases are chemically inhibited or genetically knocked out, AIF still translocates to the nucleus and causes chromatin condensation and DNA fragmentation. Specifically, AIF binds directly to DNA via its positively charged surface and then recruits caspase-independent partner proteins, primarily Cyclophilin A (CypA) and histone H2AX. Together, this AIF/CypA/H2AX complex acts as a functional endonuclease, causing large-scale DNA fragmentation into ∼50-kilobase fragments (Candé et al., 2002). BCL-2 overexpression reduces mitochondrial AIF release but does not block its apoptogenic activity (Susin et al., 1999; Susin et al., 2000). The mitochondrial release of AIF is independent of cytochrome c but requires PARP1 (Yu et al., 2002) and induces a distinct form of RCD called parthanatos (see below).

As mentioned, BCL-2 family members are among the most important regulators of MOMP. According to their structure and function, BCL-2 family proteins can be divided into three groups: BH3-only proteins, multidomain pro-apoptotic effectors, and anti-apoptotic members (Kale et al., 2018). BH3-only proteins (BAD, BID, BCL-2 interacting killer - BIK, BCL-2 like 11 - BIM, BCL-2-modifying factor - BMF, activator of apoptosis harakiri - HRK, phorbol-12-myristate-13-acetate-induced protein 1 - NOXA, and p53 upregulated modulator of apoptosis - PUMA) are stress sensors that respond to cell injury by activating the effectors BAX, BAK, and BCL-2 related ovarian killer (BOK). BH3-only proteins permeabilize the MOM. As a result, cytochrome c and other intermembrane-space proteins can reach the cytosol. Anti-apoptotic proteins, such as BCL-2 and BCL-XL, block both BH3-only activators and multidomain effectors, thereby preventing MOMP and apoptosis (Kluck et al., 1997; Yang et al., 1997). BCL-2 family members were initially believed to be localized solely in mitochondria; however, they are now known to be dynamically distributed across various organelles, including the ER, Golgi apparatus, nucleus, and peroxisomes, where they regulate calcium homeostasis, cell-cycle progression, cell migration, and apoptosis (Lossi et al., 2009; Lossi et al., 2010; Popgeorgiev et al., 2018). As a result, their subcellular location is quite flexible and depends on the cellular environment.

In non-apoptotic cells, BAX constantly moves between the cytosol and the MOM. BAK, on the other hand, is primarily localized to mitochondria but can also return to the cytosol, a process known as retrotranslocation (Cheng et al., 2003; Edlich et al., 2011; Todt et al., 2015). Retrotranslocation, mostly mediated by BCL-XL, is an essential anti-apoptotic mechanism that maintains reduced mitochondrial levels of active BAX and BAK (Edlich et al., 2011; Garner et al., 2016). For this to be effective, the C-terminal helix of BCL-XL and the BAX BH3 domain must interact. This interaction removes BAX from the MOM. BH3-only activators such as BIM and BID prevent BCL-XL from acting during apoptosis. As a consequence, it prevents retrotranslocation, allowing BAX and BAK to accumulate, join together, and form pores in the mitochondria. BH3-mimetic drugs (such as Bcl-2 Inhibitor VI - ABT-737) facilitate apoptosis by inhibiting BCL-XL activity and enhancing mitochondrial BAX retention (Renault et al., 2015; Murad and Garcia-Saez, 2021).

BH3-only proteins are activated either transcriptionally (e.g., PUMA, BIM, NOXA) or post-translationally (e.g., BID) in response to compartment-specific stress signals (Bouillet et al., 1999; Villunger et al., 2003; Elmore, 2007). When they come into contact with BAX or BAK, they alter their shape, expose the BH3 domain, favoring dimerization, and leading to the formation of higher-order oligomers (Brouwer et al., 2001; Alsop et al., 2015). Super-resolution imaging has shown that BAX/BAK oligomers make toroidal lipidic pores in MOM. In these pores, membrane lipids help shape the pore rather than just forming proteinaceous channels (Große et al., 2016; Cowan et al., 2020).

Not all BH3-only proteins directly activate BAX/BAK. Sensitizer BH3-only proteins (e.g., BAD, BMF, HRK) improve MOMP by selectively binding anti-apoptotic BCL-2 members, therefore freeing activator BH3-only proteins and effectors (Giam et al., 2008). BCL-2, BCL-XL, BCL-W, MCL-1, and BCL-2-related protein A1 (BCL-2A1) are pro-survival proteins that prevent MOMP by either directly sequestering BAX/BAK or enabling them to return to the cytosol (Chen et al., 2005; Llambi et al., 2011; Czabotar et al., 2014; Lee et al., 2016). The differences in binding affinity between these interactions constitute a selective network that determines whether apoptosis occurs.

Pro-survival BCL-2 family members inhibit MOMP through two distinct ways that align with the apoptotic primed state of the cell (Certo et al., 2006). In healthy, unprimed cells, BAX and BAK are largely inactive, and mitochondrial integrity does not require continuous anti-apoptotic restraint. In contrast, primed cells accumulate BH3-only activator proteins at anti-apoptotic BCL-2 homologs and become dependent on them for survival. Dependency on BCL-2 homologs may occur in two modes.

In mode 1, anti-apoptotic proteins, including BCL-2, BCL-XL, and MCL-1, prevent BH3-only proteins such as BID, BIM, and PUMA from exerting their pro-apoptotic functions. This association is quite unstable and can be disrupted by BH3-only sensitizer proteins (such as BAD and NOXA), which displace the activators, allowing BAX/BAK activation. In mode 2, which occurs when stress is high or persists for a long time, pro-survival proteins bind directly to active BAX and BAK, reducing the likelihood of MOMP. If unsuccessful, BAX/BAK can oligomerize, producing pores in the MOM. Inactive BAX and BAK help maintain mitochondrial homeostasis by promoting mitochondrial fusion in healthy cells. When activated, they alter mitochondrial function, leading to mitochondrial fragmentation (Kalkavan and Green, 2018). MCL-1 is a pro-survival member with a short half-life and is regulated at the transcriptional, translational, and post-translational levels, allowing apoptotic thresholds to be quickly adjusted. Caspases cleave MCL-1, generating a pro-apoptotic fragment and thereby rendering the execution phase irreversible (Michels et al., 2005). Genetic studies demonstrate that the intrinsic apoptotic machinery is broadly conserved across cell types, yet individual BCL-2 homologs fulfill non-redundant, tissue-specific survival functions. For example, BCL-2 is essential for lymphocyte survival, BCL-XL is required for erythropoiesis and neurogenesis, and BCL-W supports spermatogenesis (Madden Felicella and Jen, 2013). Remarkably, combined deletion of Bax, Bak1, and Bok makes cells strongly resistant to multiple apoptotic stimuli but causes perinatal lethality (Campbell et al., 2012; Alsop et al., 2015; Ke et al., 2015).

In the nervous system, perturbation of the BCL-2 control has marked developmental and postnatal consequences, as reduced developmental neuronal apoptosis occurs in BCL-XL/BAX-deficient mice (Shindler et al., 1997), while reduced neuronal vulnerability to apoptosis is described in BCL-XL-overexpressing models (Parsadanian et al., 1998); compromised neuronal survival occurs in BCL-2-null mice (Michaelidis et al., 1996; Tanabe et al., 1997; Llambi et al., 2016), and sufficient MCL-1 levels are essential for cortical neurogenesis (Arbour et al., 2008). These results underscore that apoptotic regulation in vivo is highly complex and sustained by combinatorial interactions that are difficult to reproduce in conventional in vitro systems.

The conventional model asserts that BH3-only proteins are essential for effector activation; however, transformed cells deficient in BID, BIM, PUMA, and NOXA can still undergo apoptosis when anti-apoptotic BCL-2 proteins are counterbalanced, suggesting that BAX/BAK can self-activate under specific conditions (Chen et al., 2015). Moreover, BOK can trigger MOMP without the involvement of BAX/BAK or other BCL-2 family members (Einsele-Scholz et al., 2016; Llambi et al., 2016), but in vivo its physiological roles are primarily fulfilled by BAX and BAK (Ke et al., 2012).

Various molecular hypotheses have been put forward to elucidate the interactions among BH3-only proteins, anti-apoptotic proteins, and BAX/BAK (Shamas-Din et al., 2013). In the direct activation model, BH3 activators (truncated BID [tBID], BIM, PUMA) directly interact with BAX/BAK, while sensitizers shift them from anti-apoptotic proteins (Certo et al., 2006; Kim et al., 2009). In the displacement model, BAX/BAK are always active but controlled by anti-apoptotic proteins. When BH3-only proteins dislocate them, apoptosis occurs (Chen et al., 2005). In the embedded-together model, all interactions arise at the mitochondrial membrane, and the membrane environment regulates binding equilibria (Llambi et al., 2011). The integrated model combines the two previous ideas into Mode 1 (BH3 activator sequestration) and Mode 2 (BAX/BAK sequestration). The prevailing mode depends on the kind of cell and the level of stress. In the membrane (lipid)-mediated spontaneous activation model, clustered regularly interspaced short palindromic repeats (CRISPR)-engineered systems demonstrate that upon neutralization of anti-apoptotic proteins, BAX can spontaneously integrate into the MOM through helix 9. Additionally, the lipid composition of the membrane directly facilitates effector activation, while BH3-only proteins primarily inhibit pro-survival members.

After MOMP, the cytoplasmic release of cytochrome c and the second mitochondria-derived activator of caspase/direct inhibitor of apoptosis-binding protein with low pI (SMAC/DIABLO) from the mitochondrial intermembrane space allows the assembly of the apoptosome. Cytochrome c is typically engaged within intracristal compartments, and its mobilization is facilitated by cristae remodeling (Scorrano et al., 2002). This process entails the breakdown of optic atrophy 1 protein (OPA1) oligomers, which can be facilitated by zinc metallopeptidase protein (OMA1)-dependent OPA1 processing and dynamin-related protein 1 (DRP1)-mediated mitochondrial dynamics following BAX/BAK activation. In the cytosol, cytochrome c binds to APAF1, which then oligomerizes into the heptameric apoptosome. This recruits and activates proCASP9 (Li et al., 1997). After that, active CASP9 processes executioner caspases CASP3 and CASP7. Simultaneously, SMAC/DIABLO undergoes proteolytic maturation, exposing its IAP-binding motif and enabling it to neutralize IAPs, especially X-linked inhibitor of apoptosis protein (XIAP), thereby alleviating caspase inhibition (Verhagen et al., 2000; Eckelman and Salvesen, 2006). Although caspases are the principal executioners of apoptosis downstream of MOMP, they are not strictly required for cell death to occur. When post-mitochondrial caspase activity is inhibited, cellular demise is delayed rather than prevented, and caspase-independent cell death pathways are frequently engaged instead (Galluzzi et al., 2015).

It is also worth noting that MOMP need not be an all-or-nothing event. When permeabilization is confined to only a subset of mitochondria—a phenomenon termed minority MOMP—the resulting sublethal caspase activation is insufficient to kill the cell but can cause DNA damage and genomic instability, with potentially oncogenic consequences (Ichim et al., 2015). Furthermore, if the pro-apoptotic stimulus is transient, cells that have undergone partial caspase activation may recover through a process called anastasis, in which executioner caspase activity is reversibly engaged and subsequently extinguished (Zaitceva et al., 2021).

Beyond caspase activation, MOMP has direct effects on bioenergetics. The release of cytochrome c that accompanies MOMP progressively dissipates the mitochondrial transmembrane potential (Δψm), ultimately leading to the collapse of cellular energy metabolism (Nunnari and Suomalainen, 2012). This bioenergetic failure represents a key mechanism by which cells can die following MOMP in a caspase-independent manner.

In neurons, as in other cell types, the onset of MOMP represents a critical commitment step in intrinsic apoptosis (Tait and Green, 2010). However, the downstream execution phase is characterized by considerably higher activation thresholds and greater buffering capacity than in rapidly dividing non-transformed cells, such as proliferating progenitors or fibroblasts, reflecting the neuron’s pronounced resistance to caspase activation following cytochrome c release (Deshmukh and Johnson, 1998; Martinou and Youle, 2011) (Table 1). Multiple lines of evidence demonstrate that post-mitotic neurons exhibit limited apoptosome competence (Table 1), a phenomenon largely attributable to low APAF1 expression and high levels of endogenous caspase inhibitors (e.g., XIAP). Inhibitors of caspases, therefore, require that the proteases be activated only if a very high threshold is surpassed, thereby providing an important safeguard against unwarranted loss of non-dividing irreplaceable cells (Potts et al., 2003; Wright et al., 2004; Wright et al., 2007). Genetic deletion of XIAP reinstates cytochrome c–dependent caspase activation in mature neurons, indicating that the limitation of the apoptosome is not attributable to an inherent incapacity to form the complex, but rather to tonic caspase suppression (Potts et al., 2003). The release of SMAC/DIABLO after MOMP is crucial in neurons because it neutralizes XIAP, thereby allowing caspase activation (Table 1). Even with cytosolic cytochrome c, many differentiated neurons do not activate executioner caspases unless XIAP is also suppressed. MOMP, therefore, is required but not sufficient to trigger intrinsic apoptosis in neurons (Hollville et al., 2019). After MOMP, neuronal death frequently proceeds via caspase-independent mechanisms due to the post-mitochondrial brake. Loss of cytochrome c causes respiratory chain dysfunction, progressive dissipation of Δψm, impaired ATP production, defective mitochondrial protein import, and increased ROS generation. Given neurons’ extreme dependence on oxidative phosphorylation, these bioenergetic defects alone can drive irreversible degeneration even when caspase activity is minimal. As we will discuss in a subsequent section of this paper, this metabolic collapse is particularly relevant in axons and synaptic compartments.

In neurons, the bioenergetic aspect of MOMP is frequently more deadly than the proteolytic caspase cascade itself (Galluzzi et al., 2009). Thus, neurons are particularly vulnerable to minority MOMP because they live for long periods and have numerous mitochondria, given their very high energy consumption (Table 1). Sublethal caspase activation, resulting in DNA damage, signals for cell-cycle re-entry and synaptic dysfunction without causing rapid cell death, and can be a consequence of minority MOMP. This condition has been associated with neurodegeneration, wherein persistent subthreshold apoptotic signaling leads to neuronal dysfunction rather than immediate loss (Ichim et al., 2015).

In neurons, apoptosome competence is developmentally regulated. During embryonic and early postnatal development, neurons exhibit higher APAF1 levels and lower XIAP levels, enabling efficient cytochrome c–driven caspase activation for physiological developmental apoptosis. In mature neurons, APAF1 is transcriptionally inhibited, directing the system towards MOMP-mediated, caspase-limited degeneration. This arrangement enhances survival following transitory mitochondrial damage but predisposes to gradual, caspase-independent cell death under pathological conditions (Wright et al., 2004).

Anoikis (from the Greek for “homelessness”) is a particular form of intrinsic apoptosis triggered by loss of integrin-mediated adhesion to the extracellular matrix (ECM) or glial substrates. ECM detachment activates an intracellular path involving BIM and BMF, which trigger BAX/BAK-dependent MOMP and subsequent CASP3 activation (Puthalakath et al., 2001; Mailleux et al., 2007; Paoli et al., 2013; Buchheit et al., 2014). Anoikis is crucial during neuronal migration, axonal pathfinding, synaptic remodeling, neurodegeneration, and ECM degradation (László et al., 2020). In this case, caspase activation is often restrained, and degeneration can proceed via mitochondrial and calpain-dependent pathways.

Table 3 summarizes the main findings regarding the role of intrinsic apoptosis in central nervous system (CNS) pathologies, along with a few key mechanistic notes.

TABLE 3. Involvement of intrinsic apoptosis in main CNS diseases.

2.1.2 Extrinsic pathway

Extrinsic apoptosis is a type of RCD that is initiated by signals from outside the cell (Sastry and Rao, 2000). It is carried out by some members of the tumor necrosis factor receptor superfamily (TNFRSF) characterized by an intracellular death domain (DD) and commonly referred to as death receptors. In the nervous system, it is particularly relevant in glial cells, which can promote neuroinflammation, immune-mediated neurotoxicity, and specific neurodegenerative contexts. In addition to death receptors, dependency receptors also can trigger extrinsic apoptosis when their specific ligands fall below a threshold (Wajant, 2002; Gibert and Mehlen, 2015).

TNFRSF comprises FS-7-associated surface antigen receptor (FasR), tumor necrosis factor receptor 1 (TNFR1), TNF-related apoptosis-inducing ligand receptor 1 (TRAILR1, or DR4) and 2 (TRAILR2, or DR5), tumor necrosis factor receptor superfamily 25 (TNFRSF25, or DR3) and 21 (TNFRSF21, or DR6), ectodysplasin A receptor (EDAR), and p75 neurotrophin receptor (p75^NTR) (Guicciardi and Gores, 2009). These canonical death receptors have an intracellular DD of 80 amino acids that recruits adaptor proteins after ligand binding, receptor trimerization, and a conformational change in the cytoplasmic tail. FasR and TRAILs directly recruit FADD, whereas TNFR1 first recruits TRADD and then FADD (Wajant, 2002). FADD engagement leads to the assembly of the death-inducing signaling complex (DISC) and the activation of CASP8 and CASP10. CASP8 forms death-effector domain (DED)-mediated filaments, undergoes homodimerization and autoproteolytic maturation, and then directly activates CASP3, or cleaves BID and engages MOMP. Cellular FLICE inhibitory protein (c-FLIP) isoforms finely adjust the composition and activity of DISC. The stoichiometry of c-FLIP isoforms with CASP8 determines whether signaling is pro-apoptotic or pro-survival (Yeh et al., 2000). The transcription of CASP8 and FADD-like apoptosis regulator (CFLAR), which encodes c-FLIP, depends on NF-κB. Expression of c-FLIP is therefore a crucial checkpoint that connects inflammatory signals to resistance to apoptosis (Chang et al., 2002).

Efficient apoptotic signaling necessitates higher-order receptor organization after S-palmitoylation (for FasR and TRAILRs), UDP-N-acetyl-alpha-d-galactosamine polypeptide N-acetylgalactosaminyltransferase 14 (GALNT14)-dependent (Seyrek et al., 2025) O-glycosylation (Matarrese et al., 2005; Toscano et al., 2007; Lichtenstein and Rabinovich, 2013), particularly during neurodegeneration (Lee et al., 2007; Lichtenstein and Rabinovich, 2013), localization in lipid rafts, and the formation of signaling protein oligomerization transduction structures (SPOTS) (Verbrugge et al., 2010). After the FasR/FS-7-associated surface antigen ligand (FasL) interaction, FADD initiates CASP8, which, in turn, initiates CASP3, which finally executes cell death (Blatt and Glick, 2001). Full activation of CASP8 requires that FasR is engulfed in receptosomes after internalization via clathrin/dynamin, followed by coupling to ezrin and actin (Marion et al., 2011).

In contrast, TRAIL receptors can signal apoptosis from the plasma membrane (Wajant, 2003). Ligand presentation is critical to the outcome of ligand-receptor binding, as membrane-bound FasL is pro-apoptotic, whereas soluble FasL is often pro-survival/immunomodulatory. In standard trophic deprivation experiments, blocking FasR and FasL reduces cell death in cultured pheochromocytoma cells 12 (PC12), cerebellar granule cells, and spinal motoneurons (Brunet et al., 1999; Le Niculescu et al., 1999; Raoul et al., 1999). FasL upregulation in neurons is a common early event in the apoptotic response to damage or stress (Raoul et al., 2000). Jun amino-terminal kinase (JNK) increases FasL expression in neurons, thereby inducing apoptotic death; in parallel, death can be the result of the inhibition of protein kinase B (PKB) that increases JNK (Brunet et al., 1999; Francois and Grimes, 1999; Le Niculescu et al., 1999; Martin-Villalba et al., 1999; Brunet et al., 2001). However, FasR stimulation can lead to neuronal death even in the absence of FasL overexpression (Raoul et al., 1999).

TNFR1 has a death domain that can promote survival and inflammation by activating NF-κB and mitogen-activated protein kinases (MAPKs), or trigger apoptosis by interfering with DISC assembly (Wajant et al., 2003). The final result depends on several factors, including the ubiquitination status of RIPK1, the recruitment of linear ubiquitination assembly complex (LUBAC), and the activity of CASP8 (Douglas and Saleh, 2020). Tumor necrosis factor receptor 2 (TNFR2), in contrast, lacks a death domain. It enhances cell survival after binding to TRAF proteins and promoting NF-κB activation (Patel and Brewer, 2008). Specifically in the nervous system, the two receptors exhibit antagonistic functions during neurodegenerative processes in vivo, with TNFR1 aggravating and TNFR2 reducing neuronal cell loss (Fontaine et al., 2002; Ortí-Casañ et al., 2023).

In neurodegenerative conditions, TRAIL receptors exhibit different behaviors: TRAILR1 and TRAILR2 are proapoptotic; TNF-related apoptosis-inducing ligand receptor 3 (TRAILR3) and 4 (TRAILR4) are decoy receptors, i.e., they bind to specific ligands with high affinity but lack the internal domain required to transmit signals; and the soluble TRAIL receptor osteoprotegerin (OPG) can inhibit TRAIL-induced apoptosis (Emery et al., 1998; Aktas et al., 2007). It is noteworthy that receptor heterodimerization further amplifies signaling events. TRAILR deficiency protects from amyloid β (Aβ) toxicity (Di Benedetto et al., 2022), and astrocyte TRAIL contributes to motor neuron death in amyotrophic lateral sclerosis (ALS) (Chai et al., 2016).

TNFRSF25 (DR3) is another member of the TNFR superfamily that contains a death domain. It triggers apoptosis or NF-κB activation (Heyninck and Beyaert, 2001). In the CNS, it is upregulated under several different conditions, including binge alcohol exposure (Liu et al., 2021), facilitates Aβ-induced neuronal apoptosis (Guan et al., 2023), and is augmented by aging-related exosomes (Liu et al., 2025). These observations are consistent with the neuroprotective effect of TNFRSF25 knockdown (Guan et al., 2023).

Apoptosis during embryonic development, tissue modeling, and neuronal pruning may be promoted by developmental death receptors (TNFRSF21, EDAR, p75^NTR). These receptors primarily transmit signals independently from the DISC pathways. TNFRSF21 (DR6) plays a role in axonal pruning and developmental connectivity through several mechanisms (Bittner et al., 2009). These comprise amyloid precursor protein (APP) binding and cleavage, leading to caspase-dependent axon self-destruction (Kamal et al., 2001). The receptor is also highly active in differentiated neurons, and its activity increases after injury (Xu et al., 2015). It has unique traits, including the ability to induce apoptosis independently of FADD/CASP8 and its interaction with BAX (Zeng et al., 2012). EDAR initiates a cellular pathway comprising ectodysplasin-A receptor-associated adapter protein (EDARADD), tumor necrosis factor receptor-associated factors (TRAFs), transforming growth factor β-activated kinase 1 (TAK1, or MAP3K7), IκB kinase (IKK), and NF-κB (Morlon et al., 2005). EDAR regulates ectodermal differentiation rather than apoptosis (Thomas et al., 2004). p75^NTR shows context-dependent signaling: when it activates JNK, it promotes apoptosis (Harrington et al., 2002), whereas when it activates TRADD and NF-κB, it promotes survival (Yazidi-Belkoura et al., 2003). Proteolytic cleavage and the presence of its Trk coreceptors control the activity of p75^NTR. In vivo effects of p75^NTR include neuronal apoptosis (Barker, 1998; Barrett, 2000; Raoul et al., 2000; Miller and Kaplan, 2001), suppression of axonal growth, and synaptic regulation (Haase et al., 2008). Differential proteolytic cleavage affects p75^NTR signaling in hippocampus and cerebellar granule cells (Vicario et al., 2015).

Dependence receptors (deleted in colorectal carcinoma - DCC, patched 1 protein - PTCH1, unc-5 netrin receptor B/D - UNC5B/UNC5D, neurotrophic receptor tyrosine kinase 3/high-affinity tyrosine kinase neurotrophin receptor C-NTRK3/TrkC) are named for their different activity in the presence or absence of ligand. When a ligand is present, these receptors promote survival and/or differentiation; in its absence, they trigger caspase-dependent apoptosis (Gibert and Mehlen, 2015). The mechanisms of action include CASP9 activation (Fombonne et al., 2012), engagement of p53-dependent pathways (Raveh et al., 2001), and MOMP (Ichim et al., 2013; Zhu et al., 2013). Broadly speaking, they function as tumor suppressors (Mehlen and Tauszig-Delamasure, 2014) and developmental sculpting molecules (Mehlen and Thibert, 2004).

2.2 Non-apoptotic RCD pathways and the concept of non-canonical RCD

There is some confusion in the literature regarding the use of the terms “non-apoptotic RCD” and “non-canonical RCD.” As apoptosis was the first form of RCD to be described, all subsequent modalities of RCD were generally defined as “non-apoptotic RCD” (Galluzzi et al., 2018). This definition encompasses, among others, necroptosis, ferroptosis, parthanatos, autophagy-dependent cell death (ADCD), pyroptosis, axon degeneration (Wallerian-like), and mitoptosis (Table 4). These forms of RCD are defined positively by their own distinct molecular machinery, not by their relationship to the apoptotic pathway. Therefore, non-apoptotic RCD is a broader, modality-based classification, simply meaning a form of RCD that is not apoptosis, regardless of whether the mechanism involved is canonical or non-canonical within its own category.

TABLE 4. Features of the main non-apoptotic forms of RCD in the nervous tissue.

The term “non-canonical” should be used in relation to a defined reference pathway. So non-canonical RCD is a relative term—it means regulated, but mechanistically divergent from whatever the canonical route is for that cell death modality. Non-canonical RCD, therefore, refers to a cell death modality that is still regulated and programmed, but deviates from its standard cellular pathway at one or more steps. The key conceptual difference between the two terms is that non-canonical RCD is a mechanistic descriptor—indicating that the death pathway deviates from an established sequence, but not indicating which RCD modality is involved. Non-apoptotic RCD is a modality descriptor—it indicates that the death is not due to apoptosis. Still, it does not indicate whether the mechanism within that modality is canonical. The two definitions, therefore, overlap incompletely. Necroptosis, for example, is non-apoptotic yet canonical within its own framework (Yu et al., 2021). A further complication is that some non-apoptotic RCD pathways share molecular components with apoptosis—for example, CASP3 cleaves gasdermin (GSDM) to initiate pyroptosis (Rogers et al., 2017), and CASP8 plays roles in both apoptosis and in regulating necroptosis (Fritsch et al., 2019)—blurring clean categorical boundaries.

We have very recently reviewed the main types of RCD and their epigenetic mechanisms in the nervous system (Merighi et al., 2026). Readers can consult this paper for a detailed synthesis of past and current research on this topic. Here, for clarity, we provide a concise integrative synthesis of this matter, highlighting the importance of epigenetic control as a unifying principle across the different types of RCD observed in neurons and glia (Tables 4, 5).

TABLE 5. Summary of the occurrence and main features of different types of non-apoptotic RCD in CNS cells. For references, see the main text.

3 Comparative analysis of the neuronal execution phase in apoptotic and non-apoptotic RCD pathways

The execution phase is the point of no return in RCD, when molecular events that disrupt cellular integrity take over from upstream signaling and stress-sensing pathways. The initiation and licensing phases are highly context-dependent and epigenetically modulated. At the same time, execution depends on the activation of terminal effectors that directly damage cell membranes and organelles, primarily mitochondria (Misgeld and Schwarz, 2017), as well as the cytoskeletal architecture and nuclear integrity (Figure 1). In neurons, execution phases are often diminished, postponed, or spatially confined, indicative of energy limitations, post-mitotic conditions, and pronounced compartmentalization (Figure 1). Neuronal RCD execution does not map cleanly onto the canonical modalities defined in dividing cells (Hollville et al., 2019). Instead, it can be characterized along several independent axes that conventional classifications conflate: whether execution is proteolytic or non-proteolytic; whether it is membrane-centric or nuclear-centric; whether it unfolds rapidly or over extended timescales; and whether it affects the cell globally or in a spatially compartmentalized manner (e.g., restricted to the axon while the soma survives). In neurons, these dimensions combine in ways that cut across canonical boundaries—a death process may, for instance, be non-proteolytic and compartmentalized, or membrane-centric and delayed—reflecting the distinct structural and metabolic constraints of post-mitotic cells.

3.1 Apoptotic execution: caspase-centric but spatially modulated

The execution phase of apoptosis comprises a sequence of actions regulated by executioner caspases (Grinshpon et al., 2019), primarily CASP3 and 7 (Table 1). Activated effector caspases cleave numerous substrates, eventually leading to chromatin condensation, DNA fragmentation (Bortner et al., 1995), phosphatidylserine (PS) exposure (Segawa and Nagata, 2015), and the formation of apoptotic bodies (Arur et al., 2003). CASP3 leads to DNA fragmentation by accelerating the proteolytic inactivation of caspase-activated DNase inhibitor (ICAD), thereby enabling caspase-activated DNase (CAD) to resume catalytic activity (Enari et al., 1998). The protease also regulates PS internalization and externalization. PS exposure at the cell membrane is achieved by activating phospholipid scramblases, which promote externalization, or by inactivating phospholipid flippases, which favor internalization. At least two distinct mechanisms exist by which cleaved (active) CASP3 induces PS exposure in response to apoptotic stimuli. Initially, CASP3 cleaves XK-related protein 8 (XKR8), which associates with basigin (BSG) or neuroplastin protein (NPTN) to form a phospholipid-scrambling complex that facilitates PS externalization. In the second instance, CASP3 inhibits the ATPase phospholipid-transporting activity of ATPase phospholipid transporter 11A (ATP11A) and 11C (ATP11C) (Segawa et al., 2014). Notably, PS exposure is not consistently correlated with intrinsic or extrinsic apoptosis, as it is a cell-type-specific phenomenon that may be associated with non-apoptotic cell death in necroptosis and efferocytosis (Fadeel et al., 1999; Shlomovitz et al., 2019).

Extrinsic apoptosis can occur in two modes. Certain cells (type I) undergo apoptosis independently of mitochondrial involvement, whilst others (type II) depend on mitochondria to initiate apoptosis (Ozören and El-Deiry, 2002; Barnhart et al., 2003). In neurons that are classified as type II cells, and where CASP3 and CASP7 are blocked by XIAP, extrinsic apoptosis necessitates the proteolytic cleavage of BID by CASP8 (Jost et al., 2009; Urresti et al., 2016). Notably, another IAP family member, survivin, has been shown to modulate the basal activity of CASP3 in cerebellar granule neurons cultivated ex vivo (Lossi et al., 2016). In neurons, complete apoptotic execution is relatively rare outside development (Lossi and Merighi, 2003). Several adaptations constrain caspase execution, including high IAP expression (Lossi et al., 2016), compartmental restriction of caspase activation (e.g., synapses and dendrites), and inefficient apoptosome assembly due to low Apaf1 levels in mature neurons. As a result, neurons frequently exhibit partial or sublethal caspase activity, sufficient for synaptic pruning or structural remodeling but insufficient for nuclear collapse (Li and Sheng, 2012; Györffy et al., 2018). Therefore, the execution phase of neuronal apoptosis is a graded process rather than an all-or-none event.

3.2 Necroptotic execution: MLKL-driven membrane catastrophe

Necroptosis is induced by ligand binding to TNF-family death-domain receptors, toll-like receptors 3 and 4 (TLR3/4), pattern-recognition receptors, or virus sensors, all of which converge on receptor interacting serine/threonine kinase 3 (RIPK3) activation via a shared receptor homotypic interaction motif (RHIM) domain (Grootjans et al., 2017). For TNF family receptors, the initiator is RIPK1; for toll-like receptors 3 and 4 (TLR3/4), the initiator is the RHIM-containing adaptor known as TIR-domain-containing adapter-inducing interferon-β (TRIF, or TICAM-1). In the case of the Z-DNA-binding protein ZBP1/DAI, the cytosolic viral sensor itself possesses an RHIM domain.

RIPK1 exists in two primary forms: a “closed conformation,” which is kinase-deficient and stimulates NF-κB and MAPK pathways, and an “open conformation,” which is kinase-active and triggers apoptosis and necroptosis. Via homotypic contact mediated by the RHIM domain, the active kinase form of RIPK1 recruits and activates RIPK3. Inhibition of RIPK3 initiates apoptosis through the ripoptosome, a signaling platform that typically cleaves RIPK3 (Gupta and Liu, 2021). The ripoptosome induces cell death via necroptosis or RIPK3-dependent apoptosis following phosphorylation of RIPK3 at S369 (Gupta and Liu, 2021). The ripoptosome is assembled during the G2 and M phases of the cell cycle, a significant finding given the correlation between apoptosis and neuronal cell proliferation (Barral et al., 2014; Gionchiglia et al., 2021; Merighi et al., 2021). The mixed lineage kinase domain-like pseudokinase (MLKL) serves as the final executioner in necroptosis. MLKL is activated (phosphorylated) by RIPK3 and forms oligomers that move to the cell plasma membrane, producing holes that cause the cell to burst and release its contents (Grootjans et al., 2017), a process referred to as an internal cellular danger signal (DAMP) (Yu et al., 2021). Transgenic models (knockout, kinase-dead knock-in) and pharmacological inhibition suggest that RIPK1, RIPK3, or MLKL are involved in numerous inflammatory, degenerative, and infectious disorders. Necroptotic execution, in contrast to apoptosis, operates independently of proteolytic cascades and results in lytic cell death accompanied by the loss of intracellular contents (Vanden Berghe et al., 2010; Vandenabeele et al., 2010). The epigenetic inhibition of RIPK3 and MLKL is common in neurons, making full necroptotic execution infrequent. When initiated, neuronal necroptosis often remains incomplete due to energetic constraints, leading to dendritic swelling, mitochondrial dysfunction, and neuroinflammation without rapid cell lysis (Dawson and Dawson, 2017). Necroptotic execution in neurons is context-dependent and, in certain neurodegenerative conditions, may be delayed or manifest as protracted structural degeneration rather than rapid membrane rupture. Pathological examinations of Alzheimer’s disease (AD)-affected brains reveal the accumulation of phosphorylated RIPK1, RIPK3, and, predominantly, phosphorylated MLKL (pMLKL) within granulovacuolar degeneration (GVD) in neurons (Koper et al., 2022). In standard necroptosis, the assembly of the necrosome and the translocation of pMLKL to the plasma membrane result in cellular swelling and membrane rupture within hours. In AD neurons with GVD granules, necrosome components concentrate intracellularly, suggesting that membrane rupture and cell death occur with delayed or impaired execution relative to classical necroptosis. These observations have been construed as a variant or prolonged form of necroptosis in neurons. In cases of acute damage (such as ischemia/reperfusion), pRIPK1, pRIPK3, and pMLKL in brain tissue usually peak hours to days following the injury. This timeframe reflects the kinetics of induction rather than a mechanistically prolonged execution, highlighting that neuronal necroptosis occurs over extended durations compared to fast necroptosis in certain non-neural cells (Liao et al., 2020).

3.3 Ferroptotic execution: lipid peroxidation as the lethal event

Uncontrolled iron-dependent lipid peroxidation triggers ferroptosis, leading to severe membrane damage, particularly of polyunsaturated phospholipids. The critical executional event is the dysfunction of antioxidant mechanisms, particularly glutathione peroxidase 4 (GPX4), leading to the buildup of deleterious lipid hydroperoxides (Yang and Stockwell, 2016). The execution pathway proceeds through three basic steps: the iron-dependent Fenton reaction, lipid peroxidation of polyunsaturated fatty acids (PUFAs), and inhibition of the glutathione (GSH)/GPX4 interaction. The Fenton reaction is a chemical process in which ferrous iron and hydrogen peroxide react to generate hydroxyl radicals and ferric iron, both of which are highly reactive and lethal to cells (Dixon and Stockwell, 2014). GPX4 is the main defense against toxic hydroxyl radicals and ferrous iron. GSH acts as a cofactor in the reduction of toxic lipid peroxides into non-toxic lipid alcohols (Costa et al., 2023). If this mechanism does not work, either because the cystine/glutamate antiporter (system Xc⁻) is blocked or GPX4 is directly inhibited, ROS accumulate quickly. The system Xc⁻ actually provides the cystine needed to synthesize GSH. Intracellular labile iron, often increased through transferrin receptor 1 (TfR1) uptake or reduced storage (ferritinophagy), promotes the conversion of hydrogen peroxide into harmful radicals through the Fenton reaction, hence triggering lipid peroxidation. The enzymes acyl-CoA synthetase long-chain family member 4 (ACSL4) and lysophosphatidylcholine acyltransferase 3 (LPCAT3) augment the membrane content of PUFAs, which are then oxidized by iron-dependent lipoxygenases (LOXs), causing direct damage to membrane integrity. Alongside GPX4, the ferroptosis suppressor protein 1 (FSP1) reduces oxidized coenzyme Q10 (ubiquinone) to its reduced form (ubiquinol), thereby providing a complementary antioxidant mechanism at the plasma membrane that inhibits lipid peroxidation.

The above-described pathway is universal, but neurons are uniquely susceptible to ferroptosis due to their iron-handling mechanisms, high PUFA content, and glutamatergic signaling. Neurons cannot access blood-transferrin-bound iron directly because of the blood-brain barrier (BBB), and thus have a distinctive iron uptake and storage mechanism (Kawabata, 2019). Consequently, neurons must synthesize their own transferrin (Tf) and use TfR1 to take up iron. Neurons also have a poorer ability to store iron in ferritin than glial cells do. Therefore, in nervous tissue, there is a higher percentage of free, redox-active iron (the labile iron pool), which directly catalyzes the Fenton reaction to generate ROS (Ou et al., 2022). Dopaminergic neurons in the substantia nigra (and others) have a specific mechanism of iron storage. These neurons synthesize neuromelanin, a neuron-specific kind of melanin that contains considerable amounts of iron (Bianchi and Merighi, 1986). Neuromelanin is protective against ferroptosis. However, in Parkinson’s disease (PD), iron-storing neuromelanin may degrade, releasing iron and inducing substantial oxidative stress and ferroptosis (Kaftan Öcal and Armagan, 2025). Neurons are also particularly vulnerable to glutamate-induced ferroptosis. Elevated extracellular glutamate levels, observed, e.g., in AD and PD, inhibit the function of the system Xc⁻ transporter. As a consequence, cystine uptake is blocked, and GSH synthesis is halted, inactivating GPX4 (Benarroch, 2018; Benarroch, 2023).

Oxytosis is a specialized form of neuronal ferroptosis, previously described as a calcium-dependent, glutamate-induced form of cell death, in which inhibition of system Xc⁻ causes GSH depletion. Still, subsequent damage is primarily driven by ROS accumulation rather than by iron-mediated lipid peroxidation (Tan et al., 2001). The relationship between oxytosis and ferroptosis has been extensively investigated over the past decades (Tang and Kang, 2023). Oxytosis was originally described as a glutamate-induced, calcium-dependent form of neuronal death characterized by inhibition of the Xc⁻ system, depletion of intracellular GSH, accumulation of ROS, and lipid oxidative damage (Tan et al., 2001). Following the identification of ferroptosis in 2012 (Dixon et al., 2012), it became evident that the two processes share key molecular features, including cystine deprivation, GSH depletion, impaired antioxidant defenses, and lethal lipid peroxidation. Subsequent studies demonstrated that oxytosis can be considered a neuronal manifestation of ferroptosis, although additional mechanisms, such as calcium influx and specific neuronal signaling pathways, may also contribute to the oxytotic phenotype. As a result, the concept of ferroptosis has largely subsumed the earlier notion of oxytosis and provided a broader framework for understanding oxidative cell death in both neurological and non-neurological disorders (Tan et al., 2023).

Unlike apoptosis and necroptosis, no caspases or pore-forming proteins are involved in ferroptosis; nuclear morphology is initially preserved; death progresses from membrane dysfunction outward, rather than from the nucleus. In neurons, ferroptotic execution is often compartmentalized, preferentially affecting synapses and axons, consistent with their high lipid content and local iron accumulation. Therefore, ferroptosis is considered a locally executed, metabolically driven death modality (Mattson and Arumugam, 2018).

3.4 Parthanatos: metabolic collapse leads to nuclear execution

Due to severe DNA damage, e.g., from oxidative stress, excitotoxicity, or ischemia, the threshold for PARP1 activation may be exceeded, triggering parthanatos. As a consequence, massive poly (ADP-ribose) (PAR) synthesis, NAD⁺ and ATP depletion, and the release and nuclear translocation of AIF occur (Wang et al., 2009; Fatokun et al., 2014). PAR polymer accumulation is the main death signal in parthanatos. PAR polymers are toxic to the cell; they initially accumulate in the nucleus, then translocate to the cytosol, and interact with mitochondrial proteins. AIF is the irreversible step in the execution commitment. It induces large-scale DNA fragmentation independently of caspases (Wang et al., 2009). As described in Section 2.1, AIF release engages CypA to drive large-scale DNA fragmentation independently of classical nuclease activity (Candé et al., 2004). This produces fragments of ∼50 kb, unlike oligosomal nuclear laddering in apoptosis (Zhang et al., 2002). In parallel with AIF-mediated nuclear damage, the cell undergoes a bioenergetic collapse characterized by NAD⁺ and ATP depletion, glycolytic failure, and mitochondrial dysfunction. This metabolic catastrophe accelerates structural disintegration but is not sufficient on its own to execute parthanatos, as AIF nuclear translocation is required. Execution in parthanatos is transcription-independent, rapid once PARP1 activity exceeds a threshold, tightly coupled to chromatin architecture and DNA damage burden. In neurons, parthanatos is a metabolically catastrophic mode of execution, particularly relevant in excitotoxicity, ischemia, and aging, where DNA damage and oxidative stress converge. Neurons undergo parthanatos via a mechanism that, while sharing the core PARP1-dependent pathway with other cell types, is heavily influenced by the unique, high-energy, and excitotoxic environment of the nervous system. The neuronal process is characterized by specific triggers, intense glutamate excitotoxicity, and oxidative stress that drive rapid accumulation of PAR polymers and, often, a combination of parthanatos with other cell death forms, such as apoptosis or necrosis. Overstimulation of N-methyl-D-aspartate (NMDA) receptors, which permit calcium entry into the cell, is a common cause of neuronal parthanatos. The calcium influx stimulates nitric oxide synthase (nNOS), which produces nitric oxide (NO) and ROS, causing severe DNA damage. Parthanatos entails the translocation of AIF to the nucleus across all cell types; however, neurons frequently demonstrate a particular reliance on the mitochondrial-associated proteins BID and BCL-2 interacting protein 3 (BNIP3), or on calpain activation (Cheng et al., 2018), to facilitate AIF release, thereby distinguishing the mitochondrial damage process. Parthanatos is involved in PD, AD, ALS, and stroke (Yang et al., 2024). However, a very recent study demonstrated that PARP1 activity is essential for retinal photoreceptor survival in the human homologous RhoI255del mouse model of autosomal dominant retinitis pigmentosa, underscoring the importance of PARP1 for rod photoreceptor viability rather than death (Zhu et al., 2025). This observation remarks that parthanatos in neurons may be strongly context-dependent (Lee et al., 2024).

3.5 Pyroptotic execution: GSDM-mediated lysis

Pyroptosis execution relies on GSDM pore formation in the plasma membrane following inflammasome-driven CASP1 (canonical execution) or CASP4/5/11 (non-canonical execution) activation. This results in rapid cell swelling, lysis, and cytokine release.

The first step in canonical pyroptotic execution is the assembly of the inflammasome. Danger signals (pathogen-associated molecular patterns - PAMPs, and damage-associated molecular patterns - DAMPs) activate cytosolic sensors (e.g., NLRP3, NLR family CARD domain containing 4 - NLRC4, absent in melanoma 2 - AIM2), leading to apoptosis-associated speck-like protein containing a CARD (ASC) oligomerization, recruitment, and activation of CASP1 (Broz and Dixit, 2016). This initial step commits the cell to pyroptotic execution. Gasdermin D (GSDMD) cleavage represents the irreversible execution step. Active CASP1 cleaves GSDMD and releases its N-terminal fragment from autoinhibition (Shi et al., 2015). The N-terminal segment of GSDMD is then directly responsible for pore formation (Ding et al., 2016). It associates with acidic phospholipids (phosphatidylinositol phosphates, cardiolipin), oligomerizes within the plasma membrane, and generates 10–20 nm holes. Consequently, K⁺ efflux, Na⁺ influx, and water entry occur, resulting in cellular swelling and disruption of osmotic balance. Pores also allow the release of interleukin 1β (IL-1β) and interleukin-18 (IL-18) (Flores-Romero et al., 2020). In parallel, CASP1 cleaves pro-IL-1β and pro-IL-18, and mature cytokines exit via GSDMD pores. Cytokine release can precede the complete cell lysis. Membrane rupture amplifies inflammatory signaling. As pore density increases, there is a massive osmotic imbalance, plasma membrane rupture, and release of DAMPs, leading to strong inflammatory propagation.

In non-canonical pyroptosis, in response to cytosolic lipopolysaccharide (LPS), CASP4/5 (human) or CASP11 (mouse) directly binds LPS. Caspases then cleave GSDMD as described above. K⁺ efflux stimulates NLRP3 as a secondary consequence, thereby activating CASP1 and amplifying cytokine maturation (Downs et al., 2020).

Particular features of neuronal pyroptosis encompass the activation of distinct sensors, namely, AIM2 and NLR family pyrin domain containing 1 (NLRP1), as well as the participation of specific chemokine receptors, C-C chemokine receptor type 5 (CCR5) (Yan et al., 2021b) and C-X-C chemokine receptor type 4 (CXCR4) (Liu et al., 2023), which have been demonstrated to facilitate inflammasome assembly following stroke or hemorrhagic brain injury. Neurons infrequently experience pyroptosis owing to diminished GSDM expression and restricted inflammasome activity (Hao and Feng, 2023). The pyroptotic machinery in the CNS primarily operates in microglia, indirectly facilitating neuronal damage by releasing cytokines rather than directly lysing neurons (Friker et al., 2020). AD is linked to neuroinflammation, characterized by microglial activation driven by Aβ accumulation, and ASCs bind to and cross-seed extracellular Aβ. In microglia, the NLRP3 inflammasome can incorporate ASC released by a pyroptotic cell, and exposure to ASC-Aβ complexes enhances the proinflammatory response, resulting in pyroptotic cell death, the release of functional ASC, and the initiation of a self-perpetuating, detrimental cycle (Friker et al., 2020). The aggregation of ASC-Aβ fibrils also hinders microglial clearance of Aβ. Collectively, these data suggest a shift from acute to chronic Aβ-associated neuroinflammation driven by the formation of ASC-Aβ complexes.

3.6 ADCD: execution depending upon delayed transcription

In contrast to other RCDs, ADCD requires continuous autophagic flux, leading to excessive organelle destruction and metabolic collapse. Slow execution is dependent on chromatin accessibility and ongoing gene expression (Galluzzi et al., 2018). ADCD starts with a commitment phase of sustained autophagic activation. Several persistent cell stressors, e.g., growth factor deprivation, mechanistic target of rapamycin (mTOR) inhibition, ER stress, developmental remodeling signals, and certain neurodegenerative disorders, may lead to mechanistic target of rapamycin complex 1 (mTORC1) inhibition and Unc-51-like autophagy activating kinase 1 (ULK1) activation, thereby initiating phagophore formation (Mizushima and Komatsu, 2011). The phagophore then undergoes nucleation and expansion through the activation of the class III phosphatidylinositol-3 kinase (PI3KC3) complex and the autophagy-related (ATG) conjugation systems, which constitute the core execution machinery. The activation of the PI3KC3 complex (beclin 1, vacuolar protein sorting 34 - VPS34 - and 15 - VPS15, ATG14L) generates phosphatidylinositol 3-phosphate (PI3P) with recruitment of WD-repeat protein interacting with phosphoinositides (WIPI) proteins and nucleation of the isolation membrane (Mizushima and Komatsu, 2011). Microtubule-associated protein 1 light chain 3 (LC3) lipidation is essential for ADCD execution. In ADCD, the lethal step involves hyperactivation of autophagic flux, massive formation of autophagosomes, accelerated fusion with lysosomes, and bulk degradation of cytoplasmic components. Several mechanisms underlie the selective degradation of essential cellular components, including mitochondria (mitophagy beyond the viability threshold), anti-apoptotic proteins, and excessive turnover of the ER and ribosomes. The loss of organelles then leads to metabolic collapse, bioenergetic failure, and structural dismantling. Sustained autophagic flux can result in lysosomal overload and lysosomal membrane permeabilization (LMP), leading to the release of cathepsin and proteolytic damage.

In neurons, ADCD execution is rare and typically reflects a failure of adaptive autophagy control rather than the activation of a dedicated death machinery. In these cells, ADCD occurs through mechanisms that differ from those of other cell types, principally because of their post-mitotic characteristics, specific polarization (axons/dendrites), and dependence on autophagy for homeostasis. ADCD mechanisms in neurons are often geared toward eliminating damaged neurons when repair mechanisms (like mitophagy) fail, rather than the rapid, cleanup-oriented death seen in proliferating cells (Yang et al., 2008; Nixon and Yang, 2012). In neurodegenerative disorders such as AD, neurons display a distinctive pattern of increased autophagy failure. Unlike other cells that might die from simple starvation, neurons often suffer from lysosomal failure. Failure causes a massive buildup of undegraded autophagic vacuoles laden with Aβ, leading to neuronal cell-autonomous death, a form of death referred to as autophagy-lysosomal associated death (ALAD) (Nixon, 2024). In AD, neuronal ADCD is also often associated with a unique abortive attempt to re-enter the cell cycle, which initiates apoptosis but often proceeds via autophagy-dependent pathways (Verdaguer et al., 2002a; Verdaguer et al., 2002b; Guan et al., 2022). A specific subtype of ADCD known as autosis occurs in the neonatal rat brain after hypoxia-ischemia. This form is unique in that it depends not only on autophagy (ATG genes) but also on the Na⁺/K⁺-ATPase pump (Depierre et al., 2024).

3.7 Paraptosis: a cell death program independent from caspases requiring de novo protein synthesis

The NCCD has not officially categorized paraptosis as a separate RCD entity with specific molecular markers (Galluzzi et al., 2018); nonetheless, it is mechanistically replicable across several systems. The initiation phase of paraptosis encompasses sustained MAPK/extracellular signal-regulated kinase (ERK) signaling following activation of a signaling pathway involving growth factor receptors, rat sarcoma virus (Ras), rapidly accelerated fibrosarcoma (Raf), and MAPK/ERK kinase (MEK) (Sperandio et al., 2000; Sperandio et al., 2004). Unlike proliferative ERK signaling, persistent ERK activation leads to cell vacuolization and death (Sperandio et al., 2004). The possibility of blocking paraptosis with cycloheximide indicates that it requires de novo protein synthesis (Wang et al., 2012). The central execution feature of paraptosis is massive ER and or mitochondrial swelling. In ER-dominant paraptosis, which can be observed after protein kinase R-like ER kinase (PERK) activation, eukaryotic translation initiation factor 2A (eIF2α) phosphorylation, activating transcription factor 4 (ATF4) induction, or C/EBP homologous protein (CHOP) upregulation (Kunst et al., 2024), ER stress and UPR activation (Merighi and Lossi, 2022) occur. The disruption of proteostasis results from proteasome inhibition, accumulation of misfolded proteins, disturbed Ca²⁺ homeostasis, and ER dilation. Distended ER cisternae eventually give rise to large vacuoles (Sperandio et al., 2000). Mitochondrial swelling, cristae disorganization, and Δψm dissipation, as a late event, characterize mitochondria-dominant paraptosis. Swelling may result from ion dysregulation, ROS accumulation, or bioenergetic imbalance. Osmotic swelling of the cell is central to its execution (Kunst et al., 2024). Paraptosis in neurons exhibits a morphological outcome similar to that in other cell types, but the triggers and regulatory pathways can be distinctive. Paraptosis is the primary pathway of neuronal death following glutamate excitotoxicity or ischemic injury; it is primarily mediated by Ca²⁺ overload and ROS accumulation, and mature neurons have highly restricted apoptotic pathways compared to those of other cell types (Hollville et al., 2019).

3.8 Axonal and organelle-restricted execution programs

Diverse cellular and molecular modes of axon degeneration have been described. Axon retraction, axosome discharge, and local degeneration of axon branches are the three primary mechanisms by which developmental axon elimination occurs. In contrast, widespread degeneration (Wallerian degeneration) of the severed portion of the axon is the consequence of axotomy (Neukomm and Freeman, 2014). Axonal execution is mediated by sterile alpha and TIR motif-containing protein 1 (SARM1)-dependent NAD⁺ depletion, resulting in ATP collapse, calcium influx, and cytoskeletal disintegration. Axonal execution is independent of nuclear or somatic death pathways and totally autonomous (Gerdts et al., 2015; Coleman and Höke, 2020). Organelle execution programs comprise mitoptosis, ER stress-induced apoptosis, and lysosomal/autophagic cell death (Skulachev, 2002). Mitoptosis consists of selective mitochondrial fragmentation and elimination (Tinari et al., 2007). While protective at low levels, excessive mitoptosis can precipitate energetic failure without immediate plasma membrane rupture (Youle and Narendra, 2011).

4 Morphological and ultrastructural correlates

Morphological analysis remains a keystone for identifying RCD in neurons (Lossi and Merighi, 1998; Lossi et al., 2002a; Lossi et al., 2004; Lossi et al., 2015), particularly given the frequent coexistence of multiple RCD pathways within neurons. Transmission electron microscopy (TEM) provides unparalleled resolution of subcellular alterations, but these must be interpreted within a molecular and temporal framework. Key parameters for a sound ultrastructural analysis include nuclear morphology (chromatin condensation, fragmentation, or dispersion), mitochondrial size and electron density, plasma membrane integrity, status of intracellular organelles (ER, lysosomes), and the presence or absence of cytoplasmic vacuolization. Importantly, no single ultrastructural feature is pathognomonic; rather, the identification of RCD modalities depends on constellations of features. The detection of initial chromatin margination, nuclear fragmentation, mitochondrial condensation, and preserved plasma membranes signifies apoptosis (Lossi et al., 2002b). Necroptosis and MPT-driven necrosis should be suspected when extensive organelle swelling, loss of membrane integrity, and cytoplasmic translucency are prevalent (Naganska and Matyja, 2001). Ferroptosis should be surmised when mitochondria exhibit shrinkage, increased membrane density, and diminished cristae in the absence of pronounced nuclear fragmentation (Qin et al., 2021). Paraptosis exhibits extensive cytoplasmic vacuolization originating from the ER and mitochondria, while the cell maintains a relatively intact nuclear structure (Schneider et al., 2004). Neural cells often exhibit hybrid or sequential morphologies, indicative of cross-talk between routes or temporal progression. For example, apoptotic signaling may transition into secondary necrosis, while ferroptotic lipid peroxidation can coexist with necroptotic membrane rupture. Fixation artifacts, post-mortem delay, and regional metabolic differences further complicate interpretation. Consequently, molecular markers and functional assays must complement ultrastructural analysis (Lossi et al., 2002a).

5 Compartmentalization of neuronal cell death modalities

Neurons are uniquely vulnerable yet remarkably resilient cells. Their post-mitotic state, extreme longevity, and polarized architecture impose biological constraints that fundamentally distinguish neuronal death from that of proliferating somatic cells. Classical models of RCD were derived mainly from non-neuronal systems and assume a relatively uniform, cell-wide execution of death programs. In neurons, however, death is frequently spatially restricted, temporally staged, and compartment-specific, with axons, dendrites, synapses, mitochondria, and even subnuclear domains capable of engaging autonomous degenerative programs. This compartmentalization is not merely a morphological curiosity; it represents a core organizing principle of neuronal vulnerability in development, injury, aging, and neurodegenerative disease (Figure 1). Neuronal death, therefore, emerges as a hierarchically organized process, in which local degeneration may precede, substitute for, or never culminate in somatic cell death (Figure 2).

5.1 Structural determinants of compartmentalized death in neurons

Both physical and molecular barriers enforce neuronal compartmentalization. Axons can be separated from the soma by distances exceeding a meter in humans and large animals, necessitating local control of metabolism, calcium buffering, redox balance, and proteostasis. Although dendrites travel much shorter distances, they have similar metabolic requirements, specifically in type I Golgi neurons. Axonal and synaptic compartments depend on localized protein synthesis and degradation mechanisms, region-specific calcium microdomains, mitochondrial transport and anchoring, and compartment-specific signaling scaffolds. These characteristics enable neurons to activate localized stress responses and apoptotic pathways without immediate nuclear participation or transcriptional reprogramming (Bading, 2013; Misgeld and Schwarz, 2017). As a consequence, RCD effectors such as caspases, calpains, PARP1, and lipid peroxidation cascades can be activated in a sublethal, spatially restricted manner.

5.2 Axonal death as an autonomous compartmentalized program

As detailed in Section 3.8, axons possess a genetically encoded, NMNAT2/SARM1-dependent self-destruction program that is molecularly and functionally distinct from apoptosis (Gerdts et al., 2015; Coleman and Höke, 2020). Significantly, axonal degeneration occurs independently of the soma, which may survive for extended periods. This autonomy is evident not only following physical trauma but also in chronic neurodegenerative disorders, such as ALS, peripheral neuropathies, and PD, where axonal pathology occurs before neuronal loss, a condition referred to as “dying-back” degeneration (Coleman and Perry, 2002; Salvadores et al., 2017; Cao and Yang, 2024). Axonal degeneration is the prototype of compartment-restricted necrotic-like death, driven by metabolic processes and independent of transcription, so reframing neuronal death as a modular rather than a unitary phenomenon.

5.3 Synaptic degeneration and sublethal caspase activation

Synapses are the most labile neuronal compartments and the earliest sites of pathology in many brain diseases. During development, synaptic pruning is mediated by complement signaling (C1q, C3) and microglial phagocytosis, as well as by localized activation of the apoptotic machinery, notably CASP3 (Mattson and Duan, 1999; Györffy et al., 2018), within dendritic spines and presynaptic terminals (Figure 3). Importantly, this caspase activity is spatially confined, low-level, and transient, and excluded from the nucleus. Consequently, canonical executioner caspases are reallocated toward structural remodeling instead of cellular destruction. In AD and its comorbidities, synaptic loss exhibits a stronger correlation with cognitive impairment than neuronal death. Aβ oligomers and tau pathology cause localized calcium dysregulation, mitochondrial impairment, and caspase activation at synapses, resulting in a synapse-specific degenerative death, known as synaptosis, which may not progress to somatic apoptosis (Selkoe, 2002; Mattson and Arumugam, 2018).

5.4 Dendritic compartmentalization of death pathways

Dendrites in general receive synaptic input and exhibit sharp sensitivity to calcium-dependent stress. Excessive glutamatergic stimulation can trigger local excitotoxic cascades, encompassing MPT, calpain activation, ROS production, and lipid peroxidation. These mechanisms may result in dendritic beading, retraction, and fragmentation without causing rapid neuronal death (Hasbani et al., 2001). Localized activation of CASP3 in dendrites has been detected in cerebellar slices using a FRET probe, demonstrating its responsiveness to diverse pro-apoptotic or pro-survival stimuli (Lossi et al., 2016). Other data indicate that ferroptosis-like pathways may contribute to dendritic degeneration associated with aging and disease, particularly in the context of compromised antioxidant defenses (Skouta et al., 2014; Stockwell, 2022).

5.5 Somatic and nuclear execution of neuronal death

When damage signals surpass compartmental containment, they converge on the soma and nucleus, activating transcription-dependent cell death pathways. These encompass intrinsic apoptosis, parthanatos, and PANoptosis (Malireddi et al., 2019; Li and Qu, 2025), an inflammatory form of RCD that integrates molecular features of pyroptosis, apoptosis, and necroptosis through the assembly of a PANoptosome complex, thereby coordinating multiple cell death pathways in response to infection, tissue injury, and inflammatory stimuli. The somatic compartment is particularly responsive to epigenetic status, chromatin accessibility, and DNA repair capability, which together establish the threshold for irreversible mortality (Dawson and Dawson, 2017; Mattson and Arumugam, 2018).

5.6 Mitochondrial and organelle-level compartmentalization

Mitochondria function as semi-autonomous centers of apoptosis within neurons. Localized mitochondrial impairment can initiate mitophagy, which facilitates survival; mitoptosis, characterized by selective mitochondrial fragmentation and removal; and localized ATP deficiency, resulting in axonal or synaptic degeneration. Mitochondrial malfunction activates global apoptotic or necrotic pathways only when it becomes pervasive. This hierarchical escalation highlights the significance of organelle-level quality control in maintaining neuronal survival (Youle and Narendra, 2011).

5.7 Aging, epigenetic drift, and compartmental integrity degradation

The aging process gradually undermines the systems that regulate cell death in a segregated manner. Decreased axonal transport, compromised mitophagy, altered calcium buffering, and epigenetic alterations lower the threshold for inter-compartmental signal transmission. Consequently, localized degenerative occurrences progressively engage global RCD pathways, transforming adaptive compartmental damage into permanent neuronal loss (López-Otín et al., 2013; Mattson and Arumugam, 2018). In neurodegenerative conditions, neuronal integrity deteriorates in a sequential process that commences with synapse failure, progresses to axonal degeneration, and culminates in somatic death. Aging is frequently linked to epigenetic drift (Teschendorff et al., 2013), a progressive, stochastic modification of DNA methylation patterns and other epigenetic markers that accumulates over an individual’s life. We recently examined the epigenetic processes governing the main types of RCD in the nervous system (Merighi et al., 2026).

6 Note on RCD in neurological diseases and therapeutic targeting

The role of RCD in neurological diseases has been the subject of numerous authoritative reviews over the last 10 years (Table 6). Here, we report some fundamental information about this issue and discuss only the key aspects. It is worth considering that distinct neurological conditions engage characteristic but overlapping RCD repertoires, shaped by injury kinetics, metabolic stress, and inflammatory context (Table 7).

TABLE 6. Reviews on RCD and neurodegeneration in the last 10 years (source PubMed).

TABLE 7. Regulated cell death (RCD) in neurological diseases.

Pathway redundancy, temporal dynamics, and cell-type selectivity further complicate therapeutic manipulation of RCD in the nervous system (Table 8). Remarkably, effective intervention relies more on aligning the treatment target with the predominant mode of death at a specific disease stage than on absolute pathway blocking.

TABLE 8. Therapeutic targeting of RCD pathways.

7 Conclusion

Reliable identification and targeting of RCD in neural tissues require multiparametric approaches integrating biochemical assays, imaging, ultrastructural analysis, and genetic manipulation. Significant barriers to developing effective strategies to combat RCD in the nervous system include compensatory activation of parallel death pathways, systemic toxicity of RCD inhibitors, and difficulty achieving cell-type–specific delivery. These challenges argue for combinatorial, temporally staged therapies rather than single-pathway blockade. Single-marker methods sometimes inaccurately categorize death modalities, especially in chronic disease models characterized by hybrid phenotypes. Principal future problems include delineating predominant RCD routes in vivo, determining whether partial activation of death programs assists adaptive roles, and incorporating cell death into systems-level models of neural network impairment.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Glossary

Δψm mitochondrial transmembrane potential ACSL4 acyl-CoA synthetase long-chain family member 4 AD Alzheimer’s disease ADCD autophagy-dependent cell death AIF apoptosis-inducing factor AIM2 absent in melanoma 2 ALS amyotrophic lateral sclerosis Apaf1 apoptosis protease-activated factor 1 APP β-amyloid precursor protein ASC apoptosis-associated speck-like protein containing a CARD ATF4 activating transcription factor 4 ATG autophagy-related ATP11A ATPase phospholipid transporting 11A ATP11C ATPase phospholipid transporting 11C Aβ amyloid β BAD Bcl-XL/Bcl-2-associated death promoter BAK Bcl-2 homologous antagonist/killer BAX Bcl-2-associated X-protein BBB blood-brain barrier BCL-2 B-cell lymphoma 2 BCL-W BCL-2 like 2 (or BCL-2L2, KIAA0271, protein phosphatase 1 regulatory subnit 51, PPP1R51) BH BCL2-homology BID BH3-interacting domain death agonist BIK BCL-2 interacting killer or NBK BIM BCL-2L11 (or BOD, BCL-2 like 11, Bcl-2 like protein 11) BLC-XL B-cell lymphoma-extra large (or BCL-2 like 1, BCL-2L1, Bcl-X, protein phosphatase 1 regulatory subunit 52, PPP1R52) BMF BCL-2-modifying factor BNIP3 BCL-2 interacting protein 3 BOK BCL-2 related ovarian killer BSG basigin CAD caspase-activated DNase CARD caspase recruitment domain CARD8 caspase recruitment domain family member 8 CASP1 cysteinyl aspartate-specific protease 1 (or caspase 1, interleukin converting enzyme 1, ICE) CASP10 cysteinyl aspartate-specific protease 10 (or caspase 10) CASP3 cysteinyl aspartate-specific protease 3 (or caspase 3) CASP7 cysteinyl aspartate-specific protease 7 (or caspase 7) CASP8 cysteinyl aspartate-specific protease 8 (or caspase 8, FLICE, MASH, Mch5) ced-3 cell death abnormality gene 3 ced-4 cell death abnormality gene 4 ced-9 cell death abnormality gene 9 CFLAR CASP8 and FADD-like apoptosis regulator (or casper, C-FLIP) c-FLIP cellular FLICE inhibitory protein CHOP C/EBP homologous protein CRISPR clustered regularly interspaced short palindromic repeats DAMPs damage-associated molecular patterns DCC deleted in colorectal carcinoma DISC death-inducing signaling complex DRP1 dynamin-related protein 1 ECM extracellular matrix EDAR ectodysplasin A receptor EDARADD ectodysplasin-A receptor-associated adapter protein egl-1 egg-laying defective protein 1 ER endoplasmic reticulum ERK extracellular signal-regulated kinase EZH2 zeste 2 polycomb repressive complex 2 subunit FADD FAS-associated protein with death domain FAK focal adhesion kinase FAS FS-7-associated surface antigen FasL FS-7-associated surface antigen ligand FasR FS-7-associated surface antigen receptor (or CD95 or APO-1) FasR/TNf FS-7-associated surface antigen receptor/tumor necrosis factor FDX1 ferredoxin 1 FSP1 ferroptosis suppressor protein 1 FTH1 ferritin heavy chain 1 GALNT14 UDP-N-acetyl-alpha-d-galactosamine:polypeptide N-acetylgalactosaminyltransferase 14 GCLC glutamate-cysteine ligase catalytic subunit GDF15 growth/differentiation factor 15 GIT1 G-protein-coupled receptor kinase-interacting protein 1 GPX4 glutathione peroxidase 4 GSDM gasdermin GSDMD gasdermin D GSDME gasdermin E or DFNA5 GSH glutathione HMGB1 high mobility group box 1 HMGCL 3-hydroxy-3-methylglutaryl-CoA lyase HMT histone methyltransferase HRD hereditary retinal degeneration HRK activator of apoptosis harakiri (or HRK, DP5, harakiri, BCL2 interacting protein) HSPs heat shock proteins IAP inhibitors of apoptosis proteins ICAD caspase-activated DNase inhibitor ICE interleukin-1-converting enzyme IGF-1 insulin-like growth factor 1 IGF-1R insulin-like growth factor 1 receptor IL-18 interleukin-18 IL-1β interleukin 1β JNK Jun amino-terminal kinase (or JNK-46 or JNK1or JNK1A2) LC3 microtubule-associated protein 1 light chain 3 LPS lipopolysaccharide LUBAC linear ubiquitination assembly complex MAPK mitogen-activated protein kinase MCL-1 myeloid cell leukaemia-1 MLKL mixed lineage kinase domain-like protein MND motor neuron disease MOM mitochondrial outer membrane MOMP mitochondrial outer membrane permeabilization MPT mitochondrial permeability transition mPTP mitochondrial permeability transition pore MS multiple sclerosis MTP mitochondrial transition pore NF-Κb nuclear factor kappa-light-chain-enhancer of activated B cells NLRP3 nucleotide-binding oligomerization domain, leucine-rich repeat, and pyrin domain 3 NOXA phorbol-12-myristate-13-acetate-induced protein 1 NPTN neuroplastin protein NTRK3 neurotrophic receptor tyrosine kinase 3 OMA1 zinc metallopeptidase protein (or 2010001O09Rik, DAB1, MPRP-1, YKR087C, or ZMPpeptidase) OPA1 optic atrophy 1 protein p75^NTR p75 neurotrophin receptor PAR poly (ADP-ribose) PARP1 Poly(ADP-Ribose) Polymerase 1 (or PARP) PC12 pheochromocytoma cell 12 PCD programmed cell death PD Parkinson’s disease PI3K phosphatidylinositol 3-kinase PKB protein kinase B (or Akt) PS phosphatidylserine PT permeability transition PTCH1 patched 1 protein PUFA polyunsaturated fatty acids PUMA p53 upregulated modulator of apoptosis (or BC3, JFY-1, BCL-2 binding component 3) Ras rat sarcoma virus RCD regulated cell death RER rough endoplasmic reticulum RIPK1 receptor-interacting protein kinase 1 RIPK3 receptor-interacting protein kinase 3 ROS reactive oxygen species SMAC/DIABLO second mitochondria-derived activator of caspase/direct inhibitor of apoptosis-binding protein with low PI SPOTS signaling protein oligomerization transduction structures TAB2 transforming growth factor β-activated protein kinase 1 (TAK1) -binding protein (or KIAA0733, MAP3K7IP2) TAK1 mitogen-activated protein kinase kinase kinase 7 (or MAP3K7) TBI traumatic brain injury tBID truncated BID TEM transmission electron microscopy TLR toll-like receptor TNFR1 tumor necrosis factor receptor 1 (or TNFRSF1A) TNFRSF tumor necrosis factor receptor superfamily TNFRSF21 tumor necrosis factor receptor superfamily21 (or DR6) TNFRSF25 tumor necrosis factor receptor superfamily 25 (or DR3) TRADD TNFRSF1A associated via death domain TRAFs tumor necrosis factor receptor-associated factors (or EBI6, MGC: 10353, TNF receptor) TRAIL TNF-related apoptosis-inducing ligand (or APO2 ligand) TRAILR1 TNF-related apoptosis-inducing ligand receptor 1 (or TNFRSF10A, DR4) TRAILR2 TNF-related apoptosis-inducing ligand receptor 2 (or TNFRSF10B, DR5) TRAILR4 TNF-related apoptosis-inducing ligand receptor 4 (or TNFRSF10D) Trk high-affinity tyrosine kinase neurotrophin receptor ULK1 Unc-51-like autophagy activating kinase 1 UNC5B unc-5 netrin receptor B (or p53RDL1, UNC5H2) UNC5D unc-5 netrin receptor D (or KIAA1777, Unc5h4) XIAP X-linked inhibitor of apoptosis protein (or API3, BIRC4, IAP-3, ILP1, MIHA, XLP2) XKR8 XK-related protein 8 ZBP1 Z-DNA binding protein 1

Footnotes

Edited by: Adrian Ting, Mayo Clinic, United States

Reviewed by: Annette Vivi Jacobsen, Masaryk University, Czechia

Andrew Davidson, University of Glasgow, United Kingdom