Over the last two decades, apoptotic cell death has been observed in the retinas of patients with diabetes ^{[1][2][3][4][5][6][7]}[11,12,13,17,18,21,22]. A previous human diabetic retinal study indicated that most degenerating neurons show activated caspase-3 immunopositivity; thus, most degenerating neurons appear to die by apoptosis ^[2][12]. Apoptosis is a strictly regulated cell death (RCD) event that includes chromatin condensation, DNA fragmentation, and the formation of small apoptotic bodies, which results in phagocytosis by the surrounding cells without inducing an inflammatory reaction.

Apoptotic cell death occurs in various types of retinal cells, such as pericytes ^[4][5][17,18], endothelial cells ^[6][8][20,21], and neuronal cells ^{[1][2][3][9][10][11][12]}[7,8,9,10,11,12,13], and is associated with the pathogenesis of DR ^[13][14][15][5,6,24]. Neuronal cell death is an irreversible change directly related to vision loss in patients ^[2][3][12,13]. As neuronal cell death occurs in the early stages of diabetes, early intervention, including neuroprotective therapies, is required to sustain visual function in patients with DR. The elucidation of the precise mechanism of neuronal cell death in DR is urgently required to establish neuroprotective therapies. However, the precise mechanisms underlying neuronal cell death in DR remain unclear. Apoptotic cell death pathways are broadly divided into two pathways: the intrinsic pathway, which is activated during development, DNA damage, or chemical injuries, and the extrinsic pathway, which is activated via death receptor signals ^[16][17][26,27]. In the intrinsic pathway, a sensor protein, c-Fos/c-Jun (activator protein-1 (AP-1)), transfers cell death signals to the mitochondria ^[18][19][28,29], resulting in the activation of caspase-9 and -3 in cultured retinas ^[20][21][30,31] and human diabetic retinas ^[2][3][12,13]. In the extrinsic pathway, tumor necrosis factor-α (TNF-α) and TNF receptor 1 (TNFR1) are associated with retinal neuronal cell apoptosis ^[22][32], retinal pigment epithelium apoptosis ^[23][33], and retinal endothelial cell apoptosis ^[24][34] under diabetic stress. However, in neuronal cells, the extrinsic pathway is thought to induce the activation of the intrinsic pathway by translocating truncated Bid (t-Bid) to the mitochondrial membrane after cleavage by caspase-8 ^[25][35]. Most researchers have indicated that endoplasmic reticulum (ER) stress is associated with the pathogenesis of DR ^{[19][26][27][28][29][30]}[29,36,37,38,39,40]. Briefly, ER stress sensors include the inositol-requiring ER-to-nucleus signaling protein 1 (IRE1), protein kinase-like ER eukaryotic initiation factor 2-alpha kinase (PERK), activating transcription factor-6 (ATF6), and inositol trisphosphate receptor (IP3R). Activated PERK phosphorylates eukaryotic initiation factor-2α (eIF-2α), resulting in the increased expression of activating transcription factor 4 (ATF4) ^[31][41]. The persistent activation of the PERK-ATF4 pathway facilitates apoptosis by inducing the transcription of CCAAT/enhancer-binding protein homologous protein (CHOP). CHOP induces the expression of Bcl-2 interacting mediator of cell death (BIM) and induces apoptosis by activating Bax/Bak and inhibiting Bcl-2 ^[32][42].

Activated IRE1 recruits TNFR-associated factor 2 (TRAF2) followed by activating apoptosis signal-regulating kinase 1 (ASK1) and c-Jun-N-terminal protein kinase (JNK) ^[33][34][43,44]. Previous studies, including ours, indicated that JNK is critically associated with ER stress-induced retinal cell death under diabetic stress ^{[3][19][30][35][36][37]}[13,29,40,45,46,47]. Xu et al. indicated that the anti-apoptotic effect of melatonin is associated with the suppression of the ATF6-CHOP pathway in the brain ^[38][48]. In the diabetic rat retina, ER stress markers, including ATF6 and CHOP, are upregulated ^[39][40][49,50], and vitamin B12 supplementation prevents photoreceptor cell death by suppressing ER markers ^[40][50]. Taken together, these results indicate that the ATF6-CHOP pathway is involved in retinal cell death in diabetic retinopathy. Under excessive ER stress, Ca²⁺ is released from the ER via IP3R, which induces mitochondrial Ca²⁺ accumulation ^[41][51]. Sustained Ca²⁺ accumulation in mitochondria promotes the mitochondrial permeability transition, followed by the release of cytochrome c and apoptosis-inducing factor (AIF) ^[41][51]. A previous study indicated that IP3R-related Ca²⁺ release is partly associated with capillary degeneration in DR ^[42][52]. Under normal conditions, phosphatidylserine (PS) is distributed in the intracellular phospholipid bilayer via flippases. In contrast, the scramblase exposed the PS to the extracellular layer of the bilayer. In previous studies, adenosine triphosphatase type 11C (ATP11C) and ATP11A, which belong to the type IV P-type ATPase family, were identified as ubiquitously expressed flippases in the cell membrane ^[43][44][53,54], and Xk-related protein 8 (XKR8), which belongs to the XKR family, was identified as a scramblase in the cell membrane ^[45][55]. During the late phase of apoptosis, the active form of caspase-3 cleaves flippases and scramblases, resulting in their flipping off and scrambling, respectively. As a result, PS was exposed on the surface of the cell membrane from the inside of the bilayer. The exposure of PS on the surface of the membrane bilayer is reflected as “Eat me” or “Find me” signals for phagocytosing cells, such as macrophages. In patients with diabetes, PS is more exposed on the membrane of erythrocytes than in healthy individuals via the inhibition of flippase-like activity by tubulin ^[46][56]. However, there are no reports on the use of flippases and scramblases in DR. Further studies are required to elucidate the association between flippases and scramblases and the pathogenesis of DR.

The Nomenclature Committee on Cell Death (NCCD) defines pyroptosis as RCD accompanied by the formation of plasma membrane pores by the gasdermin protein family, which is often induced by inflammatory caspase activation ^[47][57]. Diabetes mellitus is a chronic inflammatory disease; thus, pyroptosis, an inflammation-related RCD, is associated with the pathogenesis of diabetes mellitus and DR ^[48][49][58,59].

The first priming signals of the classical pathway of pyroptosis are to bind pathogens or cytokines, including TNF-α and IL-1β, to Toll-like receptor (TLR) followed by activating NF-κB ^[50][61]. NF-κB transcripts include pro-IL-1β, pro-IL-18, and a component of the inflammasome, NLRP3. The induction of TLR4 expression in retinal endothelial cells has been observed under high-glucose conditions ^[51][62]. In addition to TLR4, TLR2, NF-κB, TNF-α, and IL-8 are increased in RGCs under high glucose exposure ^[52][63]. The second signal is exposure to NLRP3 agonists, which include damage-associated molecular patterns (DAMPs) and pathogen-associated molecular patterns (PAMPs). DAMPs and PAMPs induce mitochondrial damage, followed by increased reactive oxygen species (ROS) production and NLRP3 activation ^[53][54][64,65]. The activated NLRP3 undergoes oligomerization, resulting in the recruitment of ASC, MEK7, and pro-caspase-1, followed by the formation of an active NLRP3 inflammasome ^[55][66]. The activated NLRP3 inflammasome activates caspase-1 by cleaving pro-caspase-1 and produces mature IL-1β and IL-18. Furthermore, caspase-1’s cleavage of gasdermin D (GSDMD) and the 33-mer N-terminus of GSDMD results in GSDMD pores approximately 22 nm in diameter in the plasma membrane ^[56][57][67,68]. The GSDMD pores release low-molecular-weight DAMPs, IL-1β, and IL-18 to extracellular spaces. Furthermore, the passive plasma membrane rupture mediated by NINJ1 exacerbates inflammatory reactions by releasing high-molecular-weight DAMPs ^[58][60]. In the non-classical pathway, endotoxins such as lipopolysaccharide activate caspase-4/5/11’s cleavage of GSDMD in a caspase-1-independent manner ^[59][69]. Several studies have indicated that NLRP3 activation, caspase-1 activation, and the upregulation of IL-1β and IL-18 are found in retinal endothelial cells in vitro and in vivo ^[60][61][62][70,71,72]. Several previous studies using human retinal pericytes have indicated that GSDMD activation and pore formation followed by releasing IL-1β and IL-18 were induced by high glucose exposure in a dose- and time-dependent manner ^[63][73] and that in human retinal pericytes exposed to advanced glycation end-products, caspase-1 and GSDMD were activated followed by increases in IL-1β, IL-18, and lactate dehydrogenase (LDH) ^[64][74]. These results indicate that pyroptosis is partly associated with pericyte loss in DR. In Müller cells, angiotensin-converting enzyme, the active form of caspase-1, and IL-1β were increased under diabetic stress in vitro and in vivo, and the NLRP3 inhibitor MCC950 reduced their expression ^[65][75]. These results indicate that the NLRP3 inflammasome pathway is activated in Müller cells in DR. A recent study indicated that the knockdown of transient receptor potential channel 6 reduced pyroptosis in rat retinal Müller cells by inhibiting ROS and NLRP3 ^[66][76]. A previous study indicated that LDH release, the upregulation of IL-1β and NLRP3, and the activation of caspase-1 and GSDMD were observed in microglia under high glucose exposure ^[67][77]. Because caspase-1 and NLRP3 inhibitors prevent microglial cell death, pyroptosis is associated with microglial cell death in DR ^[67][77]. A recent study indicated that scutellarin protected RGC pyroptosis in DR via the inhibition of caspase-1, GSDMD, NLRP3, IL-1β, and IL-18 ^[68][78]. Collectively, these results suggest that pyroptosis is associated with neurovascular cell death in DR. However, it is not known why these two steps are involved in the plasma membrane rupture during pyroptosis. One possible reason is that the first step (i.e., GSDMD pore formation) may be still a reversible change, and NINJ1-mediated plasma membrane rupture may be a “point of no return”. Thus, pyroptosis may stop before the NINJ1-mediated plasma membrane rupture. Further studies are required to elucidate the association between pyroptosis and the pathogenesis of DR and to establish therapeutic strategies to protect against pyroptosis before the point of no return.

Ferroptosis was first reported by Dixon et al. in 2012 as an iron-dependent form of RCD ^[69][79]. During ferroptosis, excessive peroxidation of polyunsaturated fatty acids (PUFAs) occurs in the plasma membrane, resulting in the disruption of plasma membrane integrity and cell swelling, such as necrotic cell death ^[69][79]. The NCCD defines ferroptosis as RCD initiated by oxidative perturbations of the intracellular microenvironment, constitutively controlled by glutathione peroxidase 4 (GPX4). Ferroptosis is inhibited by iron chelators and lipophilic antioxidants ^[47][57]. Ferroptosis does not require caspase activation; therefore, it is thought to be an evolutionarily more classical form of RCD than apoptosis ^[70][80]. Although the precise mechanism of ferroptosis remains unclear, two transcription factors, nuclear factor-erythroid 2-related factor 2 (NRF2) and BTB and CNC homology 1 (BACH1), competitively regulate ferroptosis ^[71][81]. In addition, three ferroptosis regulatory systems inhibit lipid peroxidation: the glutathione (GSH)–glutathione peroxidase 4 (GPX4), ferroptosis suppressor protein 1 (FSP1), coenzyme Q₁₀ (CoQ₁₀), and GTP cyclohydrolase 1 (GCH1)–tetrahydrobiopterin (BH₄) pathways ^[72][82]. NRF2 and BACH1 regulate gene expression involving regulatory systems, such as the subunit of system Xc, SLC7A11, FSP1, GCH1, ferritin, and GPX4 ^[71][81]. The Fenton reaction is a chemical reaction which forms toxic hydroxyl radicals (HO•) by reducing H₂O₂ in the presence of Fe²⁺ (H₂O₂ + Fe²⁺→HO•+ OH⁻ + Fe³⁺). Because Fenton reactions induce lipid peroxidation, they play a key role in ferroptosis. The GCH1–BH₄ pathway inhibits phospholipid hydroperoxide (PLOOH), while the GSH–GPX4 pathway catalyzes the reduction of PLOOH.

Growing evidence indicates that ferroptosis is associated with the pathogenesis of diabetes mellitus and its complications, including DR ^[73][85]. Ferrostatin-1 is a synthetic compound that acts as a classical hydroperoxyl radical scavenger; however, Miotto et al. indicated that ferrostatin-1 eliminates lipid hydroperoxides and produces the same anti-ferroptotic effect as GPX4 in the presence of reduced iron ^[74][86]. Shao et al. indicated that ferrostatin-1 reduces ferroptosis by improving the antioxidant capacity of the Xc-GPX4 pathway in retinal epithelial cell line cultures exposed to high-glucose media and in animal models of DR ^[75][87]. Fatty acid binding protein 4 (FABP4) is an independent prognostic marker of DR ^[76][77][88,89]. Fan et al. indicated that FABP4 inhibition alleviates lipid metabolism and oxidative stress by regulating peroxisome proliferator-activated receptor γ (PPARγ)-mediated ferroptosis and reduces ferroptosis by upregulating PPARγ activity in ARPE-19 cells cultured in high-glucose media ^[78][90]. In addition, the study suggests that FABP4 inhibition reduces ferroptosis in retinal tissues in a diabetic animal model ^[78][90]. Liu et al. indicated that glial maturation factor Β, a neurodegenerative factor that is upregulated in the vitreous in the early stage of DR, is involved in the lysosomal degradation process in autophagy, resulting in ASCL4 accumulation and ferroptosis in RPE cells cultured in high-glucose media ^[79][91]. In addition, the study suggests that the ferroptosis inhibitor liproxstatin-1 is effective in protecting retinal tissues in early DR and maintaining visual function in a diabetic rat model in vivo ^[79][91]. Liu et al. demonstrated that in human retinal endothelial cells cultured under high-glucose conditions, long non-coding RNA zinc finger antisense 1 (ZFAS1) is upregulated and activates ferroptosis by modulating the expression of ACSL4 ^[80][92]. A recent clinical study indicated that compared to those of the normal group, the serum levels of GPX4 and GSH were significantly lower and lipid peroxide, iron, and ROS were significantly higher in patients with DR ^[81][93]. Thus, ferroptosis-related biomarkers may be involved in the pathological processes of DR ^[81][93]. Natural compounds may effectively inhibit ferroptosis in patients with DR. A recent study indicated that amygdalin, an effective component of bitter almonds, inhibits ferroptosis in human retinal endothelial cells exposed to high glucose levels by activating the NRF2/antioxidant response element signaling pathway ^[82][94]. Another study indicates that 1,8-cineole, the main component of volatile oils in aromatic plants, inhibits the ferroptosis of the retinal pigment epithelium under diabetic conditions via the PPARγ/thioredoxin-interacting protein pathways ^[83][95]. Although the point of no return of ferroptosis remains unclear, ferroptosis may be a therapeutic target for preventing the progression of DR. Ferroptosis is likely related to vascular cell death in DR. Further studies are required to elucidate the precise mechanisms underlying ferroptosis in the neurovascular impairment in DR.

The NCCD defines necroptosis as a type of RCD triggered by perturbations of intracellular or extracellular homeostasis which critically depends on the kinase activities of mixed-lineage kinase ligand (MLKL), receptor-interacting protein kinase 3 (RIPK3), and RIPK1 ^[47][57]. However, studies on the association between RIPK1 expression and necroptosis are relatively limited ^[47][57]. Necroptosis is characterized by a necrosis-like appearance, including cell swelling, mitochondrial membrane permeabilization, and membrane rupture, resulting in an inflammatory reaction in a caspase-independent manner ^[84][96]. There are three necroptosis inducers: (1) death ligands which bind with death receptors including TNF-α, Fas, or TNF-related apoptosis-inducing ligand (TRAIL); (2) pathogens which are recognized by TLR family members, such as TLR3 or TLR4; and (3) Z-DNA which is recognized by Z-DNA binding protein 1 (ZBP1) ^[85][97]. All intracellular signals from these inducers aggregate into RIPK3. Toll/IL-1R domain-containing adaptor-inducing interferon β (TRIF)-mediated necroptosis and ZBP1-mediated necroptosis are RIPK1 independent ^[47][57]. RIPK1 was first identified as a regulatory factor in necroptosis ^[86][98], and RIPK1 is thought to bind to RIPK3 via self-phosphorylation ^[87][99]. However, TRIF and ZBP1 directly bind to RIPK3 and induce necroptosis in an RIPK1 independent manner, and RIPK1 inhibits necroptosis mediated by TRIF and ZBP1 ^[88][89][100,101]. The precise mechanisms of MLKL pores in the plasma membrane are debatable. However, the four-helix bundle (4HB) domain exists at the N-terminus of MLKL, and the 4HB domain is integrated with the plasma membrane and thought to form the MLKL pore ^[90][102]. Unlike pyroptosis, NINJ1 does not require membrane rupturing during necroptosis ^[58][60]. Therefore, the MLKL pores are completely different from the GSDMD pores. In addition, it is unclear how necroptosis is induced under pathological conditions in vivo. Further studies are required to elucidate the association between necroptosis and pathological events.

Very few studies have demonstrated an association between necroptosis and DR because the mechanisms underlying the induction of necroptosis in vivo remain unclear. A recent in vitro study indicated that in RGCs cultured in high-glucose conditions, the expression of RIPK1 and RIPK3 was significantly increased, and necrostatin-1 protected against retinal ganglion cell necroptosis ^[91][103]. Xu et al. indicated that an intravitreal injection of Dickkopf-1 protected streptozotocin-induced diabetic rats against retinal tissue necroptosis in vivo ^[92][104]. A recent study indicated that in the diabetic retina, the expression of RIPK1, RIPK3, and MLKL is increased in activated microglia, and that the necroptosis inhibitor GSK-872 reduces neuroinflammation and neurodegeneration, followed by an improvement of visual function in diabetic mice ^[93][105]. They concluded that microglial necroptosis was a therapeutic target in early DR ^[93][105].

Researchers should be aware that MLKL pore formation is not always a point of no return in the process of necroptosis. Due to the repair mechanisms of the cell membrane, some necroptotic cells with MLKL pores remain alive ^[94][106]. Living cells release inflammatory cytokines and induce inflammation ^[94][106]. Further studies are required to elucidate the role of necroptosis in the pathogenesis of DR.