In addition to oxidizing glycolytic substrates to generate ATP, mitochondria produce metabolic intermediates that are required for the synthesis of nucleotides, lipids and proteins that are needed to increase cellular biomass (Martínez-Reyes and Chandel, 2020). As these bioactive metabolites can control cell fate, state and/or function, mitochondria also act as signaling organelles (Martínez-Reyes and Chandel, 2020). For example, cytochrome C, which normally transfers electrons from complex III to complex IV in the electron transport chain, can also trigger programmed cell death (i.e., apoptosis) when this electron carrier leaks from inside the inner membrane space into the cytosol (Garrido et al., 2006). In addition, inefficient electron transport or reverse electron transport generates reactive oxygen species (ROS) that act as signaling molecules in homeostasis but if not maintained at manageable levels can trigger inflammation and cell death (Martínez-Reyes and Chandel, 2020). Finally, mitochondria also influence chromatin structure through the production of α-ketoglutarate (α-KG) (Figure 2). α-KG activates ten-eleven translocation (TET) DNA demethylases and histone H3K27me3 demethylases (KDM6A/6B) (Carey et al., 2015), which remove repressive methyl marks on DNA and histones, respectively, allowing chromatin to unwind and transcription factor (TF) binding to target sites (Chandel, 2014).

The RPE, which is sandwiched between the photoreceptor layer and choroid, plays a critical role in maintaining metabolic homeostasis, supplying up to 60% of the overall glucose consumed by the retina, as well as removing waste products. Substance exchange is facilitated by microvilli on the apical surface of RPE cells, which increases their cellular surface area, while absorption of nutrients on the basal side is maximized by invaginations (Boulton and Dayhaw-Barker, 2001; Bonilha et al., 2006). Specialized transport proteins on the apical and basal cellular surfaces of the RPE also control ion, nutrient and metabolite transfer between the choroid and retina (Sugasawa et al., 1994). The importance of glucose transport by the RPE for retinal health was evidenced by the tissue-specific conditional deletion of Glut1 in murine RPE cells, which led to a shortening of photoreceptor outer segments, and ultimately photoreceptor cell death (Swarup et al., 2019). In response to stress, RPE cells increase glycolytic metabolism and reduce the transport of glucose to photoreceptors, which die from starvation (Pan et al., 2021). In addition, choroidal and retinal blood flow are reduced in patients with AMD, a triggering stimulus for photoreceptor degeneration that is exacerbated by obesity, smoking and other environmental triggers that compromise blood vessel health (O'Leary and Campbell, 2021). Mechanistically, these different triggers all induce oxidative stress, which leads to RPE cells stockpiling glucose and fat such that they transport less nutrients to the retina, triggering photoreceptor degeneration (Kurihara et al., 2016).

In AMD, RPE cells undergo regional or geographic atrophy in the central cone-rich macula, a degenerative process that is associated with the accumulation of lipid-rich deposits such as lipofuscin within and drusen outside the RPE (Inana et al., 2018). Aberrant lipid metabolism is included among the large number of environmental and genetic factors that increase AMD risk. Indeed, a major component of drusen in AMD patients is Apolipoprotein E (APOE), a carrier for lipids and cholesterol that is aberrantly deposited by the RPE in AMD patients (van Leeuwen et al., 2018). APOE has different allele haplotypes, with the E2 allele moderately increasing AMD risk, while the E4 allele surprisingly displays some protection against AMD, even though it is a risk factor for Alzheimer’s disease (Anderson et al., 2001). Additional AMD-associated alleles are found in other genes related to lipid metabolism, such as APOC2, ABCA1 and CETP, linking disorders of lipoprotein metabolism to disease (Deng et al., 2022). Moreover, high levels of high-density lipoprotein cholesterol confer increased risk to AMD and is associated with the accumulation of lipid peroxidation end products in lipofuscin (van Leeuwen et al., 2018). This increase in lipid oxidation is related to elevated levels of ROS in AMD, as described further below. In addition, as drusen accumulates, it is thought to form a lipid wall that reduces glucose transport to photoreceptors (Curcio, 2018). Given these linkages to altered lipid metabolism, it is not surprising that an association between metabolic syndrome and AMD has been reported (Ghaem Maralani et al., 2015). Preventative strategies based on diet have thus begun to be tested, with dietary omega-3 fatty acids and antioxidants showing some evidence of protection against AMD (van Leeuwen et al., 2018).

As highly metabolic cells, RPE cells have high energy demands that are met by producing large numbers of metabolic organelles, including mitochondria (Boulton and Dayhaw-Barker, 2001), the dysfunction of which is implicated in AMD (Terluk et al., 2015; Fisher and Ferrington, 2018; Brown et al., 2019). Indeed, in the RPE of patients with AMD, there is a reduction in mitochondrial mass (Feher et al., 2006), lower levels of electron transport chain proteins (Nordgaard et al., 2008), and mitochondrial DNA (mtDNA) damage (Karunadharma et al., 2010; Lin et al., 2011). As a consequence of mitochondrial insufficiency, ROS accumulate and respiration is defective in the RPE of patients with AMD (King et al., 2004). Supporting the importance of mitochondria for RPE function, an RPE-specific conditional knock-out (cKO) of mitochondrial transcription factor A (Tfam), which transcribes mitochondrial genes, leads to RPE hypertrophy, RPE dedifferentiation and progressive photoreceptor degeneration (Zhao et al., 2011). In RPE-specific Tfam cKOs, electron transport chain function is reduced even though mitochondrial size is increased, which in an apparent attempt to compensate for reduced energy production. In addition, RPE mTOR signaling is elevated in Tfam cKOs, and there is an associated increase in glycolysis (Zhao et al., 2011). This metabolic shift is central to the pathogenic features observed, as treating Tfam cKO mice with rapamycin, an mTOR inhibitor, suppresses RPE dedifferentiation and preserves photoreceptor health (Zhao et al., 2011). Thus, consistent with observations made in other tissues, suppressing mTOR signaling preserves retinal function during aging, an interesting feature given that suppression of this pathway is also associated with increased lifespan (Johnson et al., 2013).

Mitochondria are dynamic structures, with fission, or the division of these organelles to form new mitochondria, linked to shorter lifespans in S. cerevisiae (budding yeast) (Scheckhuber et al., 2007) and C. elegans (Kim et al., 2011), as well as premature aging and neurodegenerative diseases in mammalian species (Liu et al., 2020). To assess the role of mitochondrial fission in RPE function, an RPE-specific Pgam5 cKO, was generated (Yu et al., 2020). Pgam5 encodes a mitochondrial phosphatase that dephosphorylates DRP1 to promote fission, the loss of which leads to mitochondrial hyperfusion, which in turn reduces mitochondrial turnover and subsequently leads to elevated mTOR signaling, ATP and ROS production. As a consequence, in RPE-specific Pgam5 cKOs, RPE cells enter cellular senescence (Yu et al., 2020), or permanent growth arrest, a common feature of aging (Tuttle et al., 2020). Notably, the senescent phenotype can be reverted by the expression of Drp1 in which the phospho-acceptor sites recognized by PGAM5 are mutated (Yu et al., 2020). While this study did not investigate photoreceptor degeneration, it does provide important information on the role of mitochondrial morphology in RPE cell health. Consistent with a potential role in retinal neurodegeneration, Pgam5 loss-of-function mutants display a Parkinson’s disease-like movement disorder that is associated with the appearance of dysfunctional mitochondria and cellular senescence (Lu et al., 2014). Moreover, senescent cells accumulate in many age-related diseases, and their elimination with senolytic drugs is a treatment strategy to help restore tissue homeostasis in neurodegenerative diseases (Baar et al., 2017; Childs et al., 2017). There is thus emerging support for a link between the deregulation of mitochondrial dynamics, the acquisition of a senescent cellular phenotype, and neurodegeneration.

The RPE plays a central role in protecting the retina from oxidative damage by the production of antioxidants, such as superoxide dismutase and glutathione. Treatment strategies that protect against oxidative stress have been tested in animal models of AMD. Strikingly, metformin, which stimulates AMPK signaling, mitochondrial biogenesis and ATP production via glycolysis, confers resistance to oxidative stress (e.g., sodium iodate-induced injury) and survival of RPE cells and photoreceptors (Xu et al., 2018). However, blocking oxidative stress alone, for example by delivering Sod2 using an AAV to RPE-specific Sod2 cKO mice, does not effectively prevent the appearance of AMD-like symptoms, especially once photoreceptor degeneration has progressed (Biswal et al., 2017). It has thus been proposed that metformin acts in other ways, such as its ability to reduce gluconeogenesis, to confer protection in animal models of AMD. Regardless of its mode of action, metformin is now being tested in three retinal neurodegenerative diseases in clinical trials (clinicaltrials.gov), including in patients with AMD (NCT02684578). However, metformin is not the panacea, as its’ protective effects appear to be restricted to metabolic disorders associated with increased oxidative stress. Indeed, rod photoreceptor degeneration is instead accelerated when metformin is applied in a protein-folding disorder of the retina (Athanasiou et al., 2017). Understanding disease etiology and metformin function is thus critical to the future application of this drug to treating patients with retinal degenerative diseases.

Photoreceptors (i.e., rods and cones) are highly metabolically active, consuming most of the energy supply to the retina in the form of ATP to sustain phototransduction (Narayan et al., 2017). The main energy requirement in the outer segment is the need to keep cGMP-gated, cation-selective ion channels in an open state so that cations such as Na⁺ and Ca²⁺ can enter and continuously depolarize the membrane (Okawa et al., 2008). Photoreceptors differ from most other CNS neurons in that they have a depolarized membrane potential of −40 mV when not stimulated (i.e., in darkness), and become hyperpolarized in response to progressive illumination, saturating at −65 mV (Pan et al., 2021). In the dark, high cGMP levels in the outer segments allow Na⁺ and Ca²⁺ to enter and depolarize the membrane (Okawa et al., 2008). Upon photo-stimulation, cGMP is hydrolyzed, reducing the need for ATP to open cGMP-gated channels and preventing Na⁺ and Ca²⁺ from entering the outer segments such that photoreceptor cell membranes become hyperpolarized, leading to a lower rate of neurotransmitter release into the synapse (Okawa et al., 2008). Notably, as cones are less light sensitive and more often in an inactive, depolarized state compared to rods, cones are the more energetically demanding cell type (Nikonov et al., 2006; Ingram et al., 2020).

During phototransduction, light photons provide the energy that drives a configurational change of 11-cis-retinal, an opsin-linked chromophore, to all-trans-retinal, which activates transducin, a G protein that is then activated by switching from GDP to GTP binding. The signal is amplified as one light photon activates around 400 transducin molecules (Heck and Hofmann, 2001). Transducins then activate phosphodiesterase-6, which hydrolyzes cGMP intracellularly to form GMP. The resulting reduction of cGMP levels leads to the closure of ion channels causing further hyperpolarization of the membrane.

While mature CNS neurons generally use OXPHOS to generate ATP, photoreceptors are an exception, with ∼80–90% of their energy production coming from aerobic glycolysis (Wang et al., 1997a; Petit et al., 2018). The requirement for aerobic glycolysis is especially important for rod photoreceptor function, with rod function perturbed more than cone function when hexokinase 2 (Hk2), a gene encoding an enzyme that is a gatekeeper of aerobic glycolysis, is knocked-out specifically in photoreceptors (Petit et al., 2018). Notably, earlier studies in the cat eye similarly showed a larger effect of reduced glucose levels on rod versus cone function, with cones only affected when glucose levels were sharply reduced (Macaluso et al., 1992).

Photoreceptors must balance the need to produce large amounts of ATP through catabolic pathways with a high anabolic demand that comes from the daily requirement to replace outer segments. During “housekeeping glycolysis”, glucose is broken down to pyruvate in a multi-enzymatic cascade, the last step involving pyruvate kinase (PK), which exists in two isoforms–PKM1, which is constitutively active, and PKM2, which is activated by tetramerization (Pan et al., 2021). A recent study found that PKM2’s role in generating pyruvate is essential for outer segment maintenance (Chinchore et al., 2017). Pyruvate is then reduced to lactate via lactate dehydrogenase (Ldha), the knockdown of which also inhibits outer segment biogenesis (Chinchore et al., 2017). Similar reductions in outer segment length were observed upon overexpression of Tigar (TP53-induced glycolysis and apoptosis regulator), which functionally mimics the knockdown of 6-phosphofructo-1-kinase (Pfk1), an enzyme that commits the use of glucose metabolites to glycolytic flux (Chinchore et al., 2017). Thus, glycolysis is required to meet the energy demands of phototransduction and for the anabolic process of outer segment maintenance.

Rod photoreceptors, which outnumber cone photoreceptors 20:1 in humans (Massey et al., 2006) and 35:1 in mice (Jeon et al., 1998), are also essential for cone survival, acting in part through the secretion of rod-derived cone viability factor (RdCVF). RdCVF is a thioredoxin-like factor that binds the Basigin-1 (Bsg1) receptor on cones (Byrne et al., 2015). Alternative splicing of Nucleoredoxin-like 1 (Nrxnl1) produces full-length and truncated RdCVF isoforms in rods, with the truncated form providing neurotrophic support to cones, and the longer form controlling oxidative signaling (Byrne et al., 2015). Consequently, there are defects in both rod and cone function in Nrxnl1 KOs, as well as progressive cone degeneration. Mechanistically, RdCVF forms a complex with Bsg1 and Glut1 to increase glucose uptake (Yang et al., 2009). Thus, when RdCVF is administered to Pde6b ^rd1 mice, a model of RP, aerobic glycolysis and ATP production are elevated and photoreceptor degeneration is prevented (Aït-Ali et al., 2015). The neuroprotective effects of RdCVF are thus achieved at least in part by stimulating glucose metabolism (Aït-Ali et al., 2015).

Given the high metabolic demand of photoreceptors, it is not surprising that energy failure stemming from defects in metabolism is an underlying factor in many neurodegenerative disorders of the retina (Narayan et al., 2017; Pan et al., 2021). Indeed, mutations in glycolytic genes, such as HK1 and HKDC1, are associated with RP in patients (Wang et al., 2014; Zhang et al., 2018; Pan et al., 2021). Similarly, a photoreceptor-specific Hk2 cKO in mouse, which blocks aerobic glycolysis, leads to a diminished capacity of rods and cones to respond to nutrient stress, even though they generate more mitochondria and efficiently use OXPHOS as an adaptive metabolic response (Petit et al., 2018).

The retina can be viewed as a metabolic ecosystem in which cells “share” nutrients and metabolites to sustain their function and health (Kanow et al., 2017; Jaroszynska et al., 2021). Indeed, in the outer retina, rod and cone photoreceptors form an integrated network with the RPE, Müller glia, and with each other, and there is growing evidence that disruption of this ecosystem leads to metabolic failure, photoreceptor death and vision loss (Jaroszynska et al., 2021). For instance, rods and Müller glia secrete large amounts of lactate as a glycolytic byproduct that is taken up by other cells as a metabolic fuel source, including by the RPE (Winkler et al., 2004; Kanow et al., 2017; Jaroszynska et al., 2021). By taking up lactate, which is a carbon source for OXPHOS, glycolysis is suppressed in RPE cells, such that less glucose is used up and more can be transferred to photoreceptors (Kanow et al., 2017). It is thus not surprising that defects in photoreceptor metabolism have non cell autonomous effects on other cells in the retina. For instance, as described in more detail in the next section, rod loss disrupts nutrient support from the RPE to cones, which die due to “starvation”, or reduced mTORC1 signaling (Punzo et al., 2009; Venkatesh et al., 2015). In this regard, it is interesting that in transgenic mice carrying a P23H RHO mutation, a model of RP, the reduced expression of metabolic genes in photoreceptors is mirrored by an upregulation of these same genes in Müller glia in an apparent attempt to compensate (Tomita et al., 2021). Thus, the impact of alterations in cell metabolism cannot be considered in isolation for a single cell type.

In RP, rod photoreceptors degenerate first, followed by cone photoreceptors, demonstrating the essential role that rods play in sustaining cone cell health, including in supporting the flow of nutrients from the RPE (Punzo et al., 2009; Venkatesh et al., 2015). Accordingly, in transgenic mouse models of RP, there is an association between photoreceptor cell death and reduced signaling by the metabolic regulators, AKT, PI3K, and mTOR (Jomary et al., 2006; Punzo et al., 2009; Ivanovic et al., 2011; Venkatesh et al., 2015). Highlighting the functional significance of these findings, in cone-specific Rptor;Rictor double cKOs (ablating mTORC1 and mTORC2 activity), there is an age-related decline in cone function and outer segment abnormalities, albeit without impacting cone survival (Ma et al., 2015). Conversely, insulin-induced metabolic reprogramming in a Pde6b ^–/– mouse model of RP increases mTORC1 signaling and improves cone survival (Punzo et al., 2009). Similarly, elevating mTORC1 signaling in cone photoreceptors by crossing Pten cKO or Tsc1 cKO alleles into Pde6b ^rd1 mice (Venkatesh et al., 2015), or crossing a rod specific Tsc1 cKO allele into Pde6b ^H620Q/H620Q RP mutant mice (Zhang et al., 2016), prevents cone and rod degeneration, respectively. Activation of mTORC1 signaling in these models is associated with increased expression of genes driving glucose transport (Glut1), glycolysis (Hif1a, Hk2, PK-M2) and glucose-6-phosphate dehydrogenase (G6PD), required to make NADPH from glucose in the pentose phosphate pathway (Venkatesh et al., 2015). Moreover, CASP2-dependent cell death is activated by low NADPH levels, the removal of which in Pde6b ^rd1 ;Casp2 ^−/− mice improved cone survival, demonstrating how reduced glucose metabolism may contribute to cone cell death (Venkatesh et al., 2015).

Despite these striking findings suggesting that high mTORC1 is neuroprotective in the retina, the path moving forward is not clear, as a separate study found that elevating mTORC1 in cones leads to late-stage AMD-like symptoms, including the presence of drusen-like deposits, accumulation of lipoproteins and RPE atrophy (Cheng et al., 2020). In Tsc1 cKOs, mTORC1 signaling was elevated in photoreceptors, which displayed hallmarks of senescence (p16Ink4a and p21 expression, senescence associated-β-galactosidase activity) and photoreceptor death (Rao et al., 2021). Similarly, mTORC1 signaling is elevated in a chemical model of RP, leading to photoreceptor degeneration, and rapamycin could slow down this degenerative response (Zhao et al., 2022). Additionally, metformin, which has neuroprotective effects in both chemical and genetic models of retinal neurodegeneration (Xu et al., 2018), activates AMPK to block mTORC1 signaling (Reh, 2016). Additional studies are thus required to tease apart how both the elevation and reduction of mTORC1 signaling can be neuroprotective in different models of retinal disease. A likely possibility is that any change to mTORC1 activity deregulates the capacity of cells to respond to nutrient supply through the pleiotropic effects of this signaling complex (Venkatesh et al., 2015). An alternative approach may be to target glycolysis directly, which would avoid the multiple effects of manipulating mTOR signaling. In this regard, it is important to highlight a recent study in which a mutated form of Arrestin1 was shown to increase glycolysis and lactate production by disinhibiting Eno1, leading to improved photoreceptor survival in Rho ^P23H/+ mice (Nelson et al., 2022).

Mitochondria are abundant in the inner segments of both rods and cones (Figure 6), raising the question of their main function(s). Photoreceptors convert 80% of glucose to lactate via glycolysis which produces ATP and metabolic intermediates for outer segment biogenesis (Wang et al., 1997a; Wang et al., 1997b). However, OXPHOS also contributes to energy production in the inner segments, generating ATP to fuel cGMP-gated channels to repolarize photoreceptor membranes in the dark (Okawa et al., 2008). Accordingly, mutations in TCA cycle genes, such as IDH3B have been identified in patients with nonsyndromic RP (Fattal-Valevski et al., 2017; Pierrache et al., 2017), highlighting the importance of mitochondria in photoreceptor survival. Notably, given that 80% of glucose is metabolized by glycolysis, other carbon sources, such as lipids, including the long-chain fatty acid palmitate, are thought to fuel the TCA cycle in photoreceptor inner segments (Joyal et al., 2016). Finally, photoreceptor mitochondria in the inner segment also regulate Ca²⁺ ion homeostasis, regulating levels of this important cation to regulate membrane depolarization and neurotransmission (Jaroszynska et al., 2021).

Lysosomes play a critical role in sustaining photoreceptor health and function, beginning with their critical role in phototransduction through their degradation of photoreceptor opsins, including rhodopsin, which must be continually replaced (Santo and Conte, 2021). In addition, circadian cycles of lysosome-mediated autophagy are observed in the rod inner segment following the peak of disc shedding, which is triggered by the constant exposure of photoreceptors to light and subsequent oxidative stress (Santo and Conte, 2021; Reme´ et al., 1986). Given the important role that lysosomes play in protein and organelle clearing and metabolic sensing, it is not surprising that several lysosomal storage disorders are associated with retinal dystrophies (Intartaglia et al., 2020) and other neurodegenerative diseases (Zhu et al., 2020). Indeed, autophagy flux is elevated in RHO ^P23H mice, a model of RP, and the genetic or pharmacological inhibition of autophagy promotes photoreceptor function and survival in these mice (Yao J. et al., 2018). Similarly, autophagy is increased in response to light damage in the retina, and autophagy inhibitors were shown to reduce light-induced photoreceptor cell death (Kunchithapautham and Rohrer, 2007). Conversely, even though autophagy is reduced in Rd10 mice, activating autophagy with rapamycin, an mTOR inhibitor, accelerates photoreceptor cell death (Rodríguez-Muela et al., 2015). However, the field is not without controversy, as other studies have reported that experimental induction of autophagy improves photoreceptor cell survival. Indeed, metformin activates AMPK and its protective effects on photoreceptor survival in Rd10 mice, a model of RP, have at least in part been attributed to the activation of lysosome-mediated autophagy (Xu et al., 2018). In addition, an increase in autophagy is seen in a rat model of retinal detachment, and inhibiting autophagy accelerates photoreceptor cell death (Besirli et al., 2011; Santo and Conte, 2021). Despite these controversies, which may reflect differences in model systems and assays used, there is a clear link between lysosomal dysfunction and neurodegenerative disorders of the retina. While outside the scope of this review, notable examples include Batten Disease, which is a lysosomal storage disorder associated with mutations in CLN3 that result in juvenile onset vision loss due to the death of retinal interneurons (Johnson et al., 2019). Strikingly, it was recently demonstrated that AAV-mediated rescue of CLN3 expression improves bipolar cell survival and sustains vision in a mouse model, providing hope for new therapies in the future (Kleine Holthaus et al., 2020).

So far, many of the studies linking reduced energy metabolism to retinal degeneration have either been correlative (i.e., mutations in glycolytic genes) or examined the effects of altering mTORC1 signaling in animal models. A question that remains is whether metabolic rate reductions are indeed observed in the retina during neurodegenerative disease. Interestingly, Fluoro-Deoxy-Glucose Positron Emission Tomography (FDG-PET) has been applied to individuals with Alzheimer’s disease, revealing reduced rate of glucose metabolism that can distinguish these patients from those with other types of dementia (e.g., frontotemporal dementia, Lewy body disease) (Marcus et al., 2014). FDG-PET has also been used to examine glucose metabolism in the retina and FDG uptake is observed in Müller glia (Poitry-Yamate et al., 2013). In zebrafish models, live imaging of 2-NBDG uptake, a fluorescent glucose analog, has been used in screens for drugs that alter glucose flux in a model of diabetes (Jaroszynska et al., 2021). Future experiments could investigate whether similar metabolic abnormalities can be used for early diagnosis and to distinguish patterns of degeneration in AMD and RP.

Neurons have preset genetic programs in place that ensure they are maintained for the lifetime of an animal (Lin et al., 2020). A recent study revealed that this neuronal survival program occurs at the level of the mitochondria with post-transcriptional regulation of the pro-apoptotic protein BAK1 attenuating apoptosis in the brain (Lin et al., 2020). Whether a similar survival mechanism occurs in retinal neurons, and furthermore, whether mitochondrial-dependent cell death pathways are perturbed in retinal neurodegenerative disease remains to be determined. However, like the brain, BAK1 expression is normally excluded from the retina, likely through the same nonsense medicated decay (NMD) pathway (Lin et al., 2020). Moreover, as highlighted above, CASP2-dependent cell death is activated by low NADPH levels, which is found in Pde6b ^rd1 mice and associated with retinal degeneration (Venkatesh et al., 2015). In this regard, it is of interest that deleting Nmnat1 (NAD⁺ synthase nicotinamide mononucleotide adenylyltransferase-1), which regulates the formation of NAD⁺ to shuttle electrons, in retinal progenitor cells created with a Six3-Cre driver leads to defects in photoreceptor maturation and survival, with multiple cell death pathways activated (i.e., apoptosis, pyroptosis, necroptosis) (Sokolov et al., 2021). Interestingly, there is a striking disruption of multiple retinal metabolites (39/112 metabolites de-regulated, as determined by LC-MS/MS), which taken together reveal defects in central carbon metabolism, including glycolytic flux (Sokolov et al., 2021). The link between these metabolic disturbances and each of the cell death pathways remains to be determined. This question is of importance as multiple mutations in NMNAT1 have been identified in patients with Leber congenital Amaurosis (LCA), another neurodegenerative disease targeting photoreceptors (Chiang et al., 2012; Falk et al., 2012; Koenekoop et al., 2012; Perrault et al., 2012).

Mitochondrial fission, or fragmentation, is associated with photoreceptor degeneration in a model of retinal detachment (She et al., 2018). This degeneration occurs in response to DRP1 activation, which induces mitochondrial fission and apoptosis (She et al., 2018). In the brain, DRP1 was found to be necessary for engulfment of fragmented mitochondria in autophagic vesicles (Goldsmith et al., 2022). Interestingly, mutations of OPA1, which result in a loss of inner mitochondrial membrane fusion, lead to an autosomal dominant optic atrophy (from which OPA1 derives its name) and childhood blindness. Moreover, when an OPA1-like homolog was deleted in Drosophila, light-induced neurodegeneration of photoreceptors was observed (McQuibban et al., 2006). Finally, in a stroke model, succinate can also induce mitochondrial fission, leading to mitochondrial dysfunction and contributing to neurodegeneration. Thus, mitochondrial dynamics must be tightly regulated to sustain neuronal health, but a role in photoreceptor survival has yet to be addressed (Wu et al., 2017).

While links between mitochondrial morphology and metabolic fuel selection is now well established, how individual nutrients influence photoreceptor health through changes in mitochondrial morphology is an important future area of consideration. For instance, fuel selection influences T lymphocyte fate selection—transitioning from glucose to fatty acid usage correlates with increased mitochondrial fusion and the generation of memory T cells (Buck et al., 2016; Raud et al., 2018). In addition, a high fat diet triggers mitochondrial fission in microglia in the hypothalamus, leading to microglial activation and an inflammatory response that precedes weight gain (Kim et al., 2019). Notably, recent studies have demonstrated that the equivalent of a Western diet (i.e., high fat, fructose and glucose) induced metabolic syndrome, with characteristic liver inflammation and fibrosis, as well as hallmark features of AMD (Roddy et al., 2020). How dietary changes impact retinal health at the mechanistic level should be further explored as this study did not directly investigate mitochondrial or other metabolic effects of the dietary changes in the retina (Roddy et al., 2020).

In general, neural progenitor cells in the embryonic brain favor glycolysis to produce ATP, while differentiated neurons use OXPHOS (Khacho and Slack, 2018). A similar scenario plays out during retinal development, as revealed by the more pronounced drop in ATP production that is observed in differentiated retinal cells versus RPCs when NaN₃ is used to block complex IV of the electron transport chain (Agathocleous et al., 2012). Notably, glycolysis in RPCs is preferentially fueled by glucose generated from glycogen stores (and not extracellular glucose), as demonstrated by the combined treatment of frog retinas with NaN₃ and glycogen phosphorylase inhibitor (GPI) resulting in a greater decline in ATP production compared to NaN₃ and 2-deoxy-glucose (2DG) treatments (Agathocleous et al., 2012). As a consequence of a glycolytic block from the co-administration of 2-DG and GPI, Xenopus RPCs reduce their proliferation rates and undergo apoptosis, even though ATP levels remain steady due to compensatory OXPHOS (Agathocleous et al., 2012). Glycolytic metabolism thus has additional roles in supporting RPC mitotic capacity and survival that are unrelated to energy production.

Glycolytic metabolism is directly regulated by mTORC1 signaling, including in the retina. In RPC-specific Tsc1 cKOs, glycolytic gene expression is transcriptionally elevated by the TF hypoxia-induced factor 1-alpha (Hif1a) (Lim et al., 2021). In Tsc1 cKOs, RPC cell cycle length is shortened, and more RPCs actively divide (Choi et al., 2018), an increased proliferative response that is triggered by increased glycolysis as it can be suppressed by 2DG (Lim et al., 2021). Strikingly, the early proliferative expansion of Tsc1 cKO RPCs is followed by accelerated mitotic aging (i.e., telomere shortening), and senescence-associated cell death, indicating that RPCs have a limited mitotic life-span or division limit (Lim et al., 2021). However, the hyper-proliferation of Tsc1 cKO RPCs was only observed when Tsc1 was removed from a small subset of RPCs in the ciliary margin (Tyrp1-Cre), and not when Tsc1 was deleted more broadly in RPCs using Chx10-Cre or Rax-Cre (Lim et al., 2021). These differences suggest that Tsc1 cKO RPCs undergo competitive clonal expansion when surrounded by wild-type RPCs, ultimately leading to Tsc1 cKO RPCs reaching their division limit more quickly. This competitive scenario is not observed when RPCs are all equally capable of dividing. Similarly, Pten cKO RPCs generated with a Pax6-Cre driver that is only active in peripheral RPCs upregulate mTORC1 signaling and outcompete their wild-type neighbors, leading to the formation of a shrunken central hamartoma comprised of wild-type cells (Tachibana et al., 2018). Additional defects in the timing of retinal cell differentiation are also observed in RPC-specific Pten cKO retinas (Cantrup et al., 2012; Jo et al., 2012; Sakagami et al., 2012; Tachibana et al., 2016; Tachibana et al., 2018).

In addition to mTORC1 signaling being sufficient to promote RPC proliferation, there is also a requirement for mTORC1 signaling to sustain RPC division and prevent precocious differentiation (Choi et al., 2018; Jones et al., 2019). Raptor (Rptor) is an adaptor protein that along with mTOR and other subunits makes up the mTORC1 complex. In RPC-specific Rptor cKOs generated with Lhx2-Cre (Jones et al., 2019) or Chx10-Cre (Choi et al., 2018), mTORC1 signaling is reduced and RPC proliferation rates decline, with a change in the timing of ganglion cell differentiation also observed in the Lhx2-Cre study. Thus, signaling pathways that regulate metabolism are central determinants of cell fate decisions during retinal development.

Programmed mitophagy, or the directed autophagy of mitochondria, occurs in several developing tissues to drive metabolic reprogramming from OXPHOS to glycolysis (Ney, 2015). In the retina, there is a striking reduction in mitochondrial mass in the retina during the period when the first ganglion cells are beginning to differentiate (Esteban-Martínez et al., 2017). During embryonic development, when the first wave of RPCs is differentiating, glycolytic metabolites were detected at higher levels in the retina than at postnatal stages, and extracellular acidification rate, a proxy measure of glycolysis, is also elevated (Esteban-Martínez et al., 2017). Strikingly, drugs that inhibit mitophagy prevent the developmental loss of mitochondria, indicating that these organelles are actively targeted for mitophagy in the embryonic retina (Esteban-Martínez et al., 2017). Moreover, by preventing mitophagy, an increase in glycolytic gene expression is no longer observed, indicating that metabolic reprogramming to a glycolytic phenotype requires the programmed destruction of fragmented mitochondria (Esteban-Martínez et al., 2017). This block in mitophagy furthermore reduces ganglion cell differentiation, an effect that is phenocopied by blocking glycolysis with glycolytic inhibitors. Thus, mitochondria must be cleared during early embryonic development to allow ganglion cells to differentiate properly (Esteban-Martínez et al., 2017).

A future avenue that remains to be explored is how glycolytic pathways intersect with the transcriptional cascades that are known to drive cell differentiation and cell fate specification in the retina. For instance, a critical determinant of an RGC fate is the TF Atoh7, which has a conserved role across species. In chick retina, Atoh7 transactivates Hes5.3 in RPCs that become committed to a ganglion cell fate (Brodier et al., 2021). Strikingly, mitochondrial number is reduced in Hes5.3 expressing RPCs when RGCs begin to differentiate, correlating with reduced PGC1-α expression, a TF that drives mitochondrial biogenesis (Brodier et al., 2021). Accordingly, the authors propose that reduced mitochondrial activity may slow down the cell cycle long enough for Atoh7 to induce a ganglion cell fate (Brodier et al., 2021). This model is in accordance with the cell cycle length hypothesis, which states that progenitor cells poised to differentiate have longer cell cycles so that exposure times to lineage-specifying TFs are increased (Dalton, 2015). To test this model, Brodier et al. (2021) treated chick retinas with FCCP, a mitochondrial uncoupler that binds to protons and equilibrates them across the IMM, thereby dissipating the proton gradient (Brodier et al., 2021). The resulting mitochondrial uncoupling not only increases RPC cell cycle length, but also increases the number of RPCs expressing high levels of Atoh7, a readout of ganglion cell fate commitment (Brodier et al., 2021). Notably, in this chick study, mitophagy is not detected (Esteban-Martínez et al., 2017). Thus, even though fewer mitochondria are required for ganglion cell differentiation in both chick and mouse, the two species use different methods to achieve the same result (i.e., reduced biogenesis in chick versus increased mitophagy in mouse).

Metabolic dynamics may also more directly regulate retinal gene expression by altering chromatin accessibility. Indeed, several major metabolites are essential co-factors for chromatin modifying enzymes. Notable examples include a-ketoglutarate, a co-factor for histone demethylases, which play a key role in maintaining embryonic stem cell pluripotency (Carey et al., 2015). In addition, S-adenosyl-methionine is a methyl donor for histone and DNA methyltransferases, and NAD + activates SIRT1 histone deacetylase activity in skeletal muscle stem cell to activate myogenic gene expression (Ryall et al., 2015). Metabolic enzymes also can directly regulate chromatin structure–for instance, acetyl-CoA synthetase translocates to the nucleus and regulates histone acetylation in the hippocampus to consolidate spatial memory (Mews et al., 2017). A comprehensive analysis of chromatin accessibility and modifications during RPC differentiation will allow investigators to elucidate the link between metabolic dynamics, gene expression and chromatin structure during retinal development.

During glycolysis, accumulated pyruvate is converted to lactate in the cytosol and exported out of the cell. Lactate has long been thought to be a simple waste product, but there is growing support for the idea that lactate participates in cell-to-cell communication (Nasu et al., 2021). For instance, in the nervous system, lactate released by astrocytes is taken up by neurons to fuel the TCA cycle in what has been termed an astrocyte-to-neuron lactate shuttle. New genetic tools, such as the eLACCO1.1 biosensor (Nasu et al., 2021), have been developed that would allow to trace lactate transfer in the developing retina, which may provide novel insights into how lactate production by RPCs may fuel energy demands of differentiating retinal cells, including photoreceptors. Another TCA cycle intermediate is succinate, which accumulates in low O₂. Interestingly, by binding an orphan G-coupled receptor, succinate can induce retinal angiogenesis in response to hypoxia (Sapieha et al., 2008). Future work to delineate the role of metabolites like lactate and succinate and their transfer between cells in retinal development will be of value to the field.

When lactate is exported out of the cell by lactate/H⁺ symporters (Slc16a1 family genes), it is transported along with one H⁺. The net effect is lower extra-cellular pH (pHe) and higher intra-cellular pH (pHi) (Oginuma et al., 2020). Consequently, when glycolysis increases, so too does pHi. Tissues naturally have pHi and differentiation gradients–e.g., there is an anterior (low pHi, more differentiated cells) to posterior (high pHi, more progenitor cells) gradients in the embryonic chick tailbud (Oginuma et al., 2020). Future experiments could determine whether similar pH gradients exist in the developing retina using ratiometric pH sensors, such as pHluorin, which has a bimodal activation spectrum—higher 488/405-nm ratio indicates lower pHi. Hence, the consequences of altered pH on retinal development could be investigated using these molecular probes. For instance, lowering pHi biases neuromesodermal progenitor cells to a neural fate in the embryonic chick tailbud, acting in part by promoting enzyme-independent β-catenin acetylation to activate Wnt signaling (Oginuma et al., 2020). Notably, Wnt signaling inhibits RPC differentiation and sustains a proliferative RPC phenotype by inhibiting the expression of the proneural bHLH TFs (Kubo et al., 2005). Whether Wnt or other signaling pathways are similarly influenced by elevated glycolysis in the retina, which sustains RPC proliferation, is an exciting open question that once answered will uncover a new level of signaling pathways regulation. A future focus may be to investigate glycolysis modulation as a novel way to direct cell fate in retinal regenerative strategies as an alternative to targeting the signaling pathways directly.

The gene regulatory networks (GRNs) that underlie cell fate decisions in the developing retina are beginning to be deciphered with the use of single cell transcriptomic and open chromatin analyses (Lyu et al., 2021). However, many transcripts expressed in the developing retina are not translated or translated with a delay, especially before eye opening (Chen et al., 2021). Interestingly, mRNA translation is regionally localized, with late endosomes serving as translation hubs in growing retinal ganglion cell axons (Cioni et al., 2019). Strikingly, as these translation hubs move towards the growth cone, they often pause on mitochondria they encounter, which serve as focal hotspots of translation (Cioni et al., 2019). Given that mTOR signaling regulates protein translation by inducing mRNA interactions with eukaryotic translation initiation factor 4F (eIF4F), rRNA synthesis, and the subsequent recruitment of ribosome subunits (Nandagopal and Roux, 2015), it seems plausible that regulation of mTOR might control when and where mRNA is translated in the developing retina. Establishing where mTOR is localized during axonal development, and how this localization guides mTOR function and possibly mRNA target selection will be important questions to address.

In Drosophila, mTOR signaling was shown to facilitate photoreceptor differentiation by downregulating Unkempt (Unk), a zinc finger/RING domain protein that acts as negative regulator of photoreceptor differentiation (Avet-Rochex et al., 2014). One of the critical ways in which mTORC1 mediates downstream effects is by activating ribosomal S6 kinases (P70S6K), which phosphorylates 4E-BP1, thereby removing inhibitory controls on eukaryotic initiation factor 4E (eIF-4E), a central regulator of cap-dependent translation (Averous and Proud, 2006). Accordingly, elevated Hif1a levels in Tsc1 cKO RPCs have been attributed in part to increased translation (Lim et al., 2021). Conversely, the increase in cell cycle proteins that are observed in Tsc1 cKO RPCs persists even after rapamycin treatment, an mTORC1 inhibitor. Instead, mTORC1 activation promotes more rapid cyclin degradation, which likely accounts for increased oscillations of these proteins to facilitate the enhanced rate of cell cycle progression in Tsc1 cKO RPCs (Choi et al., 2018). These studies highlight the need to further dissect apart how metabolic signaling pathways regulate retinal development. Understanding how mTORC1 regulates neuronal differentiation, either via either translation or protein degradation will be an exciting future area of investigation.

In dormant stem cells, mitochondria fuse and form elongated networks, whereas they undergo fission (fragmentation) following stem cell activation (van Velthoven and Rando, 2019). Such mitochondrial shape changes, or dynamics, drive stem cell transitions–e.g., increased fission induces muscle stem cell proliferation and differentiation (Khacho et al., 2014; Khacho et al., 2016; Khacho et al., 2019). In brain neural stem cells, mitochondrial fission directs cells towards a committed fate, while fusion directs stem cell self-renewal (Khacho et al., 2016). Investigating the role of mitochondrial fission and fusion in balancing RPC proliferation and differentiation is an important future area of investigation. Indeed, as mitochondria are important hubs for integration of numerous signaling pathways, they have roles in cell physiology outside of production of ATP, central among which is the production of metabolic intermediates to generate biomass. In this regard it is interesting that mitochondrial mass is reduced in RPCs during early embryonic development, when ganglion, horizontal, cone and amacrine cells are forming, but not during the postnatal period, when rod photoreceptors, bipolar cells and Müller glia are being generated (Esteban-Martínez et al., 2017). This finding raises the question of whether mitochondria play an important role in guiding later stage differentiation decisions, and also whether late RPCs like early RPCs use glycolysis for energy production. Interestingly, mitochondrial ROS has been proposed to act as a rheostat, with small dynamic changes in ROS levels regulating hematopoietic stem cell proliferation and differentiation decisions, while larger increases in ROS can trigger pathogenic reactions such as apoptosis (Maryanovich and Gross, 2013).The biology of ROS signaling–including how it is generated, detected, and acted upon at the molecular level—is an untapped area of investigation in the field of retinal development.

As introduced above, lysosomes degrade unwanted proteins and organelles in a regulated manner to maintain cellular homeostasis, but also play an underappreciated role as metabolic sensors (Todkar et al., 2017). In this capacity, lysosomes detect starvation cues that promote stem cell quiescence, as observed in the hematopoietic system (García-Prat et al., 2021a). The switch between hematopoietic stem cell quiescence and activation is governed by a balance between lysosomal and mitochondrial biogenesis, mutually exclusive events: when one is up, the other is down (Todkar et al., 2017). These two events are regulated by two opposing transcription factors, Myc and Tfeb (García-Prat et al., 2021a), which compete for the same binding sites in the genome with opposing effects (Annunziata et al., 2019). mTORC1 promotes mitochondrial biogenesis by activating Myc (de la Cruz López et al., 2019), and suppresses lysosomal biogenesis by phosphorylating and sequestering Tfeb in the cytosol (Martina et al., 2012; Settembre et al., 2012). The importance of the lysosome in stem cell maintenance was further supported by a CRISPR-screen that identified several lysosomal genes that regulate in embryonic stem cell differentiation (Villegas et al., 2019). Lysosomal activation also correlates with the differentiation of neural stem cells into specific cell fates (Delaney et al., 2020). Thus, further investigation of how lysosomal biogenesis and function regulates the choice by multipotent retinal progenitor cells to differentiate into specific cell fates is worth investigating.

To understand the inability of mammalian Müller glia to regenerate, clues may be found in comparisons to stem cell regulation in other cell lineages. Stem cells support tissue homeostasis in regenerative species such as fish and in the few mammalian tissues that can regenerate (e.g., skeletal muscle, hematopoietic system). Studies across organisms and species have revealed that somatic stem cells that support regeneration exist in a continuum of states, transiting from deep dormancy to primed quiescence, or G_ALERT, before they are activated to proliferate and differentiate (Rodgers et al., 2014; Llorens-Bobadilla et al., 2015; Urbán et al., 2016; Laurenti and Göttgens, 2018; Urbán et al., 2019). G_ALERT stem cells are poised to respond to injury, and increase their metabolism (mitochondrial biogenesis and fission, glycolysis) to support a proliferative response (Khacho et al., 2019; Rodgers et al., 2014; Liang et al., 2020; Dell'Orso et al., 2019).

In regenerative species like zebrafish, metabolism lies at the centre of regenerative pathways, inhibiting glucose metabolism with 2DG resulted in the loss of fin regeneration capability (Sinclair et al., 2021). In mammals, mTORC1 expression and activity is highly up-regulated in G_ALERT stem cells in the skeletal muscle of the contralateral limb, compared to quiescent stem cells in an unperturbed animal. Using the muscle stem cell-specific Pax7 ^CREER driver to ablate TSC1 (tuberous sclerosis protein 1), a negative regulator of mTORC1, the authors were able to stimulate stem cells to enter G_ALERT stem cells from a totally quiescent stem cell populations without any contralateral injury (Rodgers et al., 2014). In contrast, the ablation of Raptor, a key component of mTORC1, suppressed mTORC1 signaling and resulted in a complete loss of G_ALERT stem cells in the contralateral stem cell niche, even with injury (Rodgers et al., 2014). Along with other experiments, mTORC1 was firmly demonstrated to be necessary and sufficient for G₀-G_ALERT transition, and in the absence of regenerative cues mTORC1 signaling maintains G_ALERT stem cells in a primed state ready to respond (Rodgers et al., 2014). mTORC1 drives regenerative processes through the animal kingdom and across tissues (Lund-Ricard et al., 2020), including in chick Müller glia (Zelinka et al., 2016), via unknown mechanisms. Given its interplay with metabolic dynamics, understanding how mTORC1 signaling and downstream glycolytic pathways regulates Müller glia activation is critical to the design of future regenerative strategies.

The failure of mammalian Müller glia to regenerate can be attributed to the lack of inductive/permissive signals or/and the presence of inhibitory cues. Most groups focus their effort on studying potential inductive signals that can drive Müller glia to proliferate, which includes the triggering event of the injury itself and the growth factor signaling response. However, what molecular machinery constitutes the brake that transitions mammalian Müller glia into quiescence and thus prevents retinal repair, remains an open question (Hoang et al., 2020). A ground-breaking study identified gene regulatory networks (GRNs) that are associated with a return-to-quiescence program by mammalian Müller glia post-injury (Hoang et al., 2020). However, the importance of GRNs that control metabolic reprogramming have yet to be investigated. Interestingly, in hematopoietic stem cells, lysosomal and mitochondrial biogenesis are mutually exclusive events (i.e., when one is up, the other is down) (Todkar et al., 2017). As detailed above, this switch is governed by a Myc-centered mitochondrial GRN and Tfeb-centered lysosomal GRN (Figure 12). Understanding the contribution of Myc and Tfeb to Müller glia regeneration could provide important new clues to the role of metabolic events in regulating glial cell regenerative capacity.

The quest is on to find the TFs and other inducing signals that may trigger Müller glia to act like stem cells and repair the damaged retina, but the importance of metabolic pathways may be under appreciated (Blackshaw and Sanes, 2021). If mitochondrial activation facilitated Müller glia proliferation, conversion of Müller glia to neurons may also be under the control of metabolic programming. Indeed, there is growing evidence that neuronal reprogramming in the brain is influenced by metabolic pathways. For example, in vivo reprogramming of glia to neurons is improved when Bcl12 is expressed, but not for the expected anti-apoptotic effects (Gascón et al., 2016). Instead, Bcl2 reduces neuronal reprogramming-associated lipid peroxidation, the blockage of which prevents induced neurons from undergoing ferroptosis. Additionally, a proteomic study revealed striking differences in mitochondrial proteins in neurons versus astrocytes, and the specific expression of neuronal genes improved neuronal reprogramming efficacy (Russo et al., 2021). Given the central role of metabolism, the potential impact of diet and dietary supplements on Müller glia regenerative response has, to the best of our knowledge, not been reported. Interestingly, in zebrafish, in which Müller glia are normally regenerative, Müller glia display a reduced capacity to proliferate and generate new neurons in a genetic model of chronic diabetes (Schmitner et al., 2021). Moreover, altering metabolism in this diabetic model inactivates Notch signaling, an essential component of the regenerative response in the zebrafish retina. It is important to note, however, that metabolic changes associated with chronic disease are not necessarily directly comparable to changes observed in acute injury models (Schmitner et al., 2021). In the rat, a recent study found that high glucose exposure induces oxidative stress in retinal Müller glia in vitro, and identified Nrf2 as a critical transcription factor in this response (Albert-Garay et al., 2022). More studies are required to understand how metabolic transitions influence Müller glial cell activation and the repair response.

Once new neurons have been generated, an important remaining question remains is how required functional connections be established? Indeed, CNS neurons do not regenerate axons after injury. Much of the injury response is dependent on glial cells, which guide axonal growth. Interestingly, in the Drosophila CNS, artificially increasing glycolysis in glial cells led to the accumulation of excess metabolites (l-lactate, L-2HG), which acted on GABA_B receptors to promote neuronal outgrowth (Li et al., 2020). Moreover, l-lactate could stimulate regrowth of corticospinal axons in a mouse model of spinal cord injury, leading to improved functional outcomes (Li et al., 2020). Thus, metabolism is a central regulator of axonal growth.

In the murine retina, Müller glia can be induced to differentiate into bipolar cell-like neurons by expressing neurogenic bHLH TFs, together with a stimulating injury and HDAC inhibitor (Todd et al., 2021). In another study, Müller glia were successfully converted into functional photoreceptor-like cells in a degenerative model by activating Wnt signaling and overexpressing 3 TFs that promote neurogenesis and photoreceptor fate specification (Yao K. et al., 2018). However, although these photoreceptors were efficient in capturing light, the response recorded from RCGs cells was weak and resembled an unstimulated RCG response in a healthy retina. The authors hypothesized that this weak response could be due to the low number of reprogrammed cells, but other possibilities include incomplete differentiation, leading to aberrant connectivity and synapse formation. Not investigated in these studies is the potential significance of metabolites in guiding axon outgrowth, as highlighted above. Thus, a full understanding of how metabolic pathways are regulated during regenerative events is essential for the future design of cell therapeutics.