Oxidative stress and inflammation are two critical pathological factors in the development and progression of diabetic retinopathy (DR). In the type 2 diabetes mellitus (T2DM) environment, persistent hyperglycemia promotes increased oxidative stress within the retina, leading to significantly elevated reactive oxygen species (ROS) (25).

Glutamate, as the principal excitatory neurotransmitter in the retina, plays a crucial role in visual signal transmission and retinal neuronal function under physiological conditions. However, the hyperglycemic state of type 2 diabetes mellitus (T2DM) profoundly disrupts glutamate metabolic pathways, leading to its excessive accumulation and exacerbating retinal neuronal toxicity (65). During diabetic retinopathy (DR) progression, excessive glutamate accumulation combined with impaired uptake may lead to significantly elevated extracellular levels, causing excessive activation of N-methyl-D-aspartate (NMDA) receptors on retinal neurons.

NMDA receptor overactivation triggers a cascade of neurotoxic events: (1) Calcium overload — excessive Ca²⁺ influx exceeds cellular buffering capacity, disrupts mitochondrial function, and triggers apoptotic pathways; (2) Increased oxidative stress — elevated intracellular calcium stimulates reactive oxygen species (ROS) production, further damaging cellular components; (3) Inflammatory cascade activation — glutamate excitotoxicity promotes the release of pro-inflammatory cytokines (including TNF-α and IL-6), forming a self-perpetuating cycle of inflammation and neuronal damage (73).

The role of retinal Müller cells (RMCs) in glutamate homeostasis is particularly critical. Under diabetic conditions, RMC proliferation and activation are regulated by the PPP1CA/YAP/GS/Gln/mTORC1 pathway (74). Glutamine synthetase (GS) converts neurotoxic glutamate (Glu) to non-toxic glutamine (Gln), thereby activating mechanistic target of rapamycin complex 1 (mTORC1) and promoting RMC activation. However, in DR, this protective mechanism becomes dysregulated. Blockade of the PPP1CA/YAP/GS/Gln/mTORC1 pathway inhibits RMC function, leading to impaired glutamate clearance and accumulation of neurotoxic levels (74).

Retinal microglial activation further amplifies the toxic effects of glutamate (75). Activated microglia release additional pro-inflammatory mediators and promote chronic neuroinflammation. Furthermore, glutamate metabolism is closely linked to other amino acid metabolic pathways (particularly aspartate and glutamine metabolism), indicating that metabolic dysregulation may synergistically affect retinal health through multiple pathways (66). [Preclinical evidence] These findings collectively suggest that glutamate metabolism may represent a central node in the pathological network driving retinal neurodegeneration in DR. However, direct human retinal evidence remains limited, and the translational relevance of these preclinical observations requires further validation.

Although the role of glutamate excitotoxicity in DR neurodegeneration is conceptually attractive, the quality and translational potential of existing evidence requires careful assessment. While animal studies have provided important mechanistic insights, they have significant methodological limitations. Current preclinical evidence primarily derives from STZ-induced acute hyperglycemic models and genetic models such as db/db mice, which although capable of reproducing certain DR features, cannot fully simulate the chronic, multifactorial pathological processes of human type 2 diabetes (65, 76). More critically, most animal studies have relatively short observation periods (4–12 weeks), whereas human DR development typically spans years or even decades. This temporal compression may lead us to overestimate the effects of acute metabolic interventions while underestimating the importance of chronic adaptive mechanisms.

From the perspective of evidence consistency, animal studies do show a strong correlation between elevated glutamate and neuronal damage, but the causal direction of this correlation is not entirely clear. The studies by Dionysopoulou et al. (76) and Shivashankar et al. (75) present contradictions in the temporal sequence of microglial activation and glutamate elevation — the former observed microglial activation preceding glutamate elevation, while the latter considered glutamate accumulation to be the triggering factor. This contradiction suggests that glutamate dysregulation may be both a cause and consequence of DR pathology, forming a self-reinforcing vicious cycle. Disentangling the “primary driving factor” in this cycle is crucial for determining optimal intervention timing.

The lack of human evidence represents the greatest shortcoming in the current research field. The few existing human studies (66, 71, 77) have small sample sizes (n = 25–48) and mostly employ cross-sectional designs, which can only establish correlations rather than causality. More importantly, glutamate levels measured in these studies are mostly derived from plasma or vitreous fluid, and whether they accurately reflect glutamate concentrations in the retinal neuronal microenvironment remains questionable. The presence of the blood-retinal barrier means that the correlation between peripheral blood glutamate levels and local retinal glutamate homeostasis may be weak, a precedent that has been observed in neurological disease research (73). Additionally, the glutamate quantification methods employed by different studies (HPLC, LC-MS, enzymatic methods) lack standardization and cross-validation, making quantitative comparisons between studies difficult.

From a clinical translation perspective, glutamate modulation strategies face multiple challenges. First is the issue of specificity: glutamate, as the principal excitatory neurotransmitter of the retina, is indispensable for normal visual signal transmission. Any intervention aimed at reducing pathological glutamate levels must be precisely controlled to avoid interfering with physiological glutamatergic transmission. NMDA receptor antagonists, while showing neuroprotective effects in animal models, may impair visual function with long-term use (76). Second is the route of administration: many natural compounds with glutamate-modulating effects (such as polyphenols) have low oral bioavailability and limited ability to cross the blood-retinal barrier, with actual drug concentrations reaching the retina potentially far below experimentally effective concentrations (67, 71). Third is the uncertainty regarding intervention timing: if glutamate elevation only appears in late-stage DR or is merely a secondary change, glutamate modulation may not be able to prevent critical pathological processes in early disease stages.

Notably, the central role of Müller cells in glutamate homeostasis provides potential cell-specific targets for intervention. The PPP1CA/YAP/GS/Gln/mTORC1 pathway revealed by Guo et al. (74) suggests that enhancing Müller cell glutamate clearance capacity may be more selective and safer than direct NMDA receptor inhibition. However, the feasibility of this strategy depends on the degree of reversibility of Müller cell functional recovery under diabetic conditions, a question that currently lacks adequate research. If Müller cells have already undergone irreversible functional damage or apoptosis during DR progression, strategies to enhance their glutamate transport or metabolic capacity may have limited effectiveness.

Advancing glutamate modulation strategies from laboratory to clinic urgently requires breakthroughs in several areas: first, development of non-invasive, specific retinal glutamate monitoring technologies, such as magnetic resonance spectroscopy (MRS)-based methods or glutamate-sensitive molecular imaging probes; second, conducting large-sample, longitudinal human metabolomics studies to clarify the dynamic changes of glutamate dysregulation in the natural course of DR and its temporal relationship with clinical progression; third, conducting dose-finding and pharmacokinetic studies to determine the therapeutic window of glutamate-modulating drugs; fourth, designing animal models that more closely resemble human disease characteristics, such as using slowly progressive diabetic models and longer observation periods. Only after these foundational work is completed can glutamate modulation enter clinical trials as a reliable therapeutic strategy.

Arginine, as a conditionally essential amino acid, has metabolic products including nitric oxide (NO), which plays complex, concentration-dependent dual roles in retinal vascular function. Under physiological conditions, NO synthesized by endothelial nitric oxide synthase (eNOS) acts as a key vasodilator and anti-inflammatory mediator, crucial for maintaining retinal blood flow and vascular homeostasis (78, 79).

This eNOS/iNOS imbalance forms a vicious cycle: reduced protective NO from eNOS may lead to vascular dysfunction and hypoxia, triggering inflammatory responses that upregulate iNOS, producing toxic NO levels that exacerbate oxidative stress and inflammation (82). Additionally, arginine metabolic dysregulation affects the production of other metabolites (including ornithine and polyamines) that play important roles in cell proliferation and repair. Their imbalance may exacerbate endothelial cell damage and apoptosis (82).

The arginine-NO pathway in DR presents a more complex picture than the glutamate pathway, with this complexity reflected not only in the biphasic biological effects of NO but also in the heterogeneity of existing evidence. Although animal studies consistently show decreased eNOS expression and iNOS upregulation in the diabetic state (81, 83), this simple “eNOS down-iNOS up” dichotomy may oversimplify the actual situation. The review by Gericke and Buonfiglio (82) points out that NO biological effects are highly dependent on the local microenvironment, including redox status, cofactor availability, and competing metabolic pathways. In a metabolically highly active and regionally heterogeneous tissue like the retina, the effects of NO may differ significantly among different cell types and microregions.

From the perspective of evidence quality, a core challenge facing arginine-NO pathway research is the transient and unstable nature of NO itself. Most studies indirectly infer NO biological activity by measuring stable NO metabolites (nitrate/nitrite) or NOS enzyme expression, but whether these surrogate markers truly reflect functional NO levels is questionable. For example, iNOS expression upregulation does not necessarily imply harmful excess NO production, because in some cases, iNOS-derived NO may have adaptive protective effects (81). More importantly, existing studies rarely directly measure peroxynitrite formation in retinal tissue — the key mediator considered to be the toxic effect of NO. The lack of this critical evidence makes the “iNOS is harmful” argument somewhat speculative.

Human studies, although slightly more numerous than in the glutamate field, similarly face methodological limitations. Plasma measurement of arginine and ADMA levels is relatively convenient, but its correlation with local retinal arginine availability has not been fully validated. A key observation comes from Tepic et al. (80), who used laser Doppler flowmetry to assess retinal blood flow response to arginine infusion in DR patients. This functional assessment method provides more direct evidence than mere biomarker measurement, showing that DR patients indeed have impaired eNOS function. However, the small sample size of such studies and lack of long-term follow-up limit their clinical application value.

Arginine supplementation as a therapeutic strategy is attractive for its relative simplicity and good safety profile. However, clinical translation faces a fundamental paradox: in the diabetic state where iNOS is already upregulated, will exogenous arginine supplementation be preferentially utilized by eNOS to produce protective NO, or will it further enhance iNOS activity and exacerbate oxidative stress? Animal studies suggest that arginine supplementation may improve endothelial function (83, 84), but these studies mostly employed short-term interventions without adequately evaluating the effects of long-term supplementation. Theoretically, sustained arginine supply may lead to chronic iNOS activation and development of tolerance, ultimately diminishing therapeutic efficacy.

A more promising but also more complex strategy is combined intervention — for example, arginine supplementation combined with tetrahydrobiopterin (BH₄) co-administration to promote eNOS coupling, or combined with selective iNOS inhibitors to suppress harmful NO production. Zhou et al. showed positive effects using salidroside (a natural compound believed to enhance eNOS activity) in animal studies, but its exact molecular mechanism and human applicability still require validation. Notably, many so-called “eNOS enhancers” may actually exert effects through non-specific mechanisms such as antioxidant or anti-inflammatory effects, rather than directly modulating NOS enzyme activity. This mechanistic uncertainty increases the difficulty of clinical development.

From a research design perspective, advancing arginine-NO pathway from basic research to clinical application requires several key breakthroughs. First, there is a need to develop imaging technologies capable of real-time, spatially resolved measurement of retinal NO production and peroxynitrite formation in vivo to overcome the limitations of current indirect measurements. Second, pharmacogenetic studies should systematically evaluate the impact of NOS enzyme polymorphisms on arginine supplementation response to identify patient subgroups most likely to benefit. The arginine/ADMA ratio mentioned by Guo et al. (85) as a biomarker has certain value, but its ability to predict treatment response has not been validated in prospective studies. Third, clinical trial designs should employ adaptive strategies, including dose optimization phases and continue/stop decision rules based on early biomarker responses, to address the complexity and individual variability of the arginine-NO pathway.

Overall, the arginine-NO pathway represents an area in DR metabolic therapy with substantial clinical translation potential but still requiring cautious advancement. Compared to the glutamate pathway, its advantages include slightly more human data and relatively mature supplement products; but its complex biphasic biology and potential paradoxical effects require us to be more prudent and systematic when advancing to clinical application.

Tryptophan, as an essential amino acid, is primarily metabolized through the kynurenine pathway (KP), which represents the major route of tryptophan degradation in mammals (>95%). This pathway is catalyzed by two key rate-limiting enzymes: indoleamine-2,3-dioxygenase (IDO) and tryptophan-2,3-dioxygenase (TDO), which convert tryptophan to N-formylkynurenine, subsequently metabolized to kynurenine and downstream metabolites including kynurenic acid, 3-hydroxykynurenine, and quinolinic acid (86, 87).

The tryptophan-kynurenine pathway occupies a special position in DR research: it is the most abundant in human metabolomics data among the four major amino acid pathways, but simultaneously has the most obscure mechanistic understanding and least clear clinical translation pathway. The core of this paradox lies in our still lacking deep understanding of the biological significance of “elevated kynurenine” — is it a driving factor of DR pathology, an adaptive response, or merely a passive reflection of inflammatory status?

From epidemiological evidence, multiple studies consistently show elevated serum kynurenine levels in DR patients, correlating with disease severity. This consistency is impressive, making the kynurenine/tryptophan ratio one of the most promising candidates among current DR metabolic biomarkers. However, the robustness of correlation does not equate to establishment of causality. Considering that IDO is a classic inflammation-induced enzyme that is upregulated in multiple chronic inflammatory diseases (93), elevated kynurenine may primarily reflect the chronic inflammatory state in DR rather than being an independent pathogenic factor. Distinguishing between these two possibilities is crucial for evaluating the rationality of IDO inhibition as a therapeutic strategy.

Mendelian randomization studies can provide important evidence for establishing causality, but are currently lacking in the DR field. Such studies use genetic variants associated with tryptophan metabolism as instrumental variables, which can to some extent exclude confounding factors and reverse causation. If individuals carrying genetic variants leading to low IDO activity or low kynurenine levels show reduced DR risk, this would provide strong support for causality; conversely, if genetic variants are unrelated to DR risk, it suggests that elevated kynurenine may primarily be a secondary change.

Even if the tryptophan pathway does participate in DR pathology, its mechanism of action is far more complex than currently described in the literature. Kynurenine metabolism produces multiple downstream metabolites with distinctly different or even opposite biological effects: kynurenic acid is considered to have antioxidant and neuroprotective effects, while quinolinic acid and 3-hydroxykynurenine have neurotoxic and pro-oxidant effects (87, 92). Most existing metabolomics studies only measure total kynurenine or a limited number of metabolites, unable to comprehensively characterize the metabolite profile. More importantly, the distribution and effects of these metabolites may differ significantly among different cell types and subregions of the retina, and bulk tissue or serum measurements mask this heterogeneity. Application of single-cell metabolomics and spatial metabolomics technologies will be key to resolving this complexity.

The immunomodulatory function of the tryptophan pathway adds another layer of complexity to understanding its role in DR. IDO-mediated immunosuppression is an established therapeutic target in the tumor microenvironment, with IDO inhibitors being developed to enhance anti-tumor immunity (88). However, in a chronic inflammatory disease like DR, the immunosuppressive effect of IDO may have dual significance: on one hand, it may limit excessive inflammatory responses and have tissue-protective effects; on the other hand, long-term immunosuppression may impair tissue repair capacity and promote susceptibility to microbial infections. This duality makes the risk-benefit balance of IDO inhibition strategies difficult to predict. Furthermore, the aryl hydrocarbon receptor (AhR), as a receptor for kynurenine and its metabolites, has expression patterns and signaling consequences in retinal cells that remain poorly understood. AhR activation can produce pro-inflammatory or anti-inflammatory effects depending on cell type and ligand properties (89). Without adequate understanding of retinal AhR biology, therapeutic development based on the tryptophan pathway is fraught with uncertainty.

From a clinical translation perspective, the challenges facing tryptophan pathway intervention may be the most formidable. The failure of IDO inhibitors in oncology clinical trials (88) has raised questions about the druggability of this target, although the pathological context of DR is fundamentally different from tumors. Dietary tryptophan restriction theoretically can reduce kynurenine production, but severe deficiency of tryptophan as an essential amino acid would lead to malnutrition and impaired protein synthesis. More promising but also more speculative strategies include: (1) selective modulation of downstream metabolic enzymes to alter the ratio of beneficial vs. harmful metabolites rather than comprehensive pathway inhibition; (2) small molecule modulators targeting AhR to finely regulate immune responses; (3) gut microbiota interventions utilizing microbial alternative metabolism of tryptophan to produce potentially protective metabolites such as indole-3-propionic acid (94). However, these strategies are all at the conceptual stage, far from clinical application.

The immediate priority for tryptophan-kynurenine pathway research is not to rush into interventional clinical trials, but to fill critical knowledge gaps. Research that needs to be prioritized includes: (1) evaluating causality using genetic epidemiology methods; (2) conducting deep metabolic phenotyping of DR patients, including detailed metabolite profiles and longitudinal tracking; (3) systematically evaluating cell-specific effects of kynurenine metabolites in human-derived retinal organoids or ocular sections; (4) characterizing retinal AhR expression, ligand specificity, and downstream effects. Only after these fundamental questions are adequately answered can the tryptophan pathway transform from an interesting biomarker candidate to a credible therapeutic target.

Natural compounds achieve multi-level pathological intervention through modulation of amino acid metabolic networks. Different categories of natural compounds exhibit significant differences yet complementary effects in their action targets: polyphenolic compounds (such as resveratrol, quercetin) primarily regulate glutamate metabolism and oxidative stress pathways, improving excitotoxicity through upregulation of glutamate transporter Excitatory Amino Acid Transporter 1/2 (EAAT1/2) expression and inhibition of NMDA receptor activity (73, 97); alkaloid compounds (such as berberine) selectively act on the arginine-nitric oxide (NO) pathway, improving vascular function through enhancement of endothelial nitric oxide synthase (eNOS) activity and inhibition of arginase (98, 99); terpenoid and steroid compounds regulate inflammatory and immune responses through the tryptophan-kynurenine pathway (100, 101).

This differentiation in action targets essentially originates from structural characteristics of the compounds. The polyhydroxy structure of polyphenols confers direct free radical scavenging ability, while the nitrogen-containing heterocyclic structure of alkaloids possesses selectivity for binding to specific enzymes (102). Despite acting on different upstream pathways, these compounds ultimately converge on common downstream aspects of oxidative stress and inflammatory responses (103), forming the molecular basis for synergistic effects.

The mechanism underlying synergistic effects has been preliminarily elucidated at the molecular level. A positive feedback amplification exists between oxidative stress and inflammatory responses: reactive oxygen species (ROS) activate NF-κB to promote pro-inflammatory factor expression, while inflammatory factors such as TNF-α in turn stimulate NADPH oxidase to produce more ROS (43). Polyphenolic compounds achieve non-linearly enhanced protective effects by simultaneously blocking both of these links. For example, resveratrol alone for antioxidation can reduce oxidative stress markers (MDA) by 25%, anti-inflammation alone can reduce pro-inflammatory factors by 28%, but the combined effect reduces overall retinal damage by 42% (104), exceeding the expected additive effect.

However, multi-target action also brings new challenges. Compared to specific drugs, natural compounds have relatively weaker intervention strength on single pathological links. Taking VEGF inhibition as an example, the inhibition rate of resveratrol in animal experiments is 28–35% (104, 105), while anti-VEGF drugs can achieve over 90% (106). This characteristic suggests that, based on current preclinical evidence, natural compounds may be more appropriately positioned as potential candidates for early prevention and adjuvant therapy rather than primary treatment choices for late-stage proliferative DR. However, this positioning requires validation through adequately powered clinical trials before any clinical recommendations can be made.

The limitations of anti-VEGF drugs in DR treatment have been fully recognized. Although they can effectively inhibit neovascularization, they are ineffective against early lesions such as neurodegeneration and microvascular pericyte loss (106). A study enrolling 685 patients with proliferative DR showed that the neovascularization regression rate reached 78% after anti-VEGF treatment, but 32% of patients still experienced further visual decline within 24 months, mainly due to persistent neuronal damage and inflammation (106).

In contrast, the multi-target properties of natural compounds can theoretically simultaneously improve vascular and neural lesions. However, this advantage faces evidence gaps in clinical translation. Three existing randomized controlled trials (RCTs) have shown inconsistent results: (107) reported that combined quercetin + resveratrol treatment reduced central macular thickness by 32 μm and HbA1c by 0.7% (107); while (27) only observed thickness reduction of 18 μm and HbA1c reduction of 0.5% (27). These differences may stem from differences in study design (double-blind vs. open-label), patient characteristics (moderate vs. mild lesions), and dosing regimens.

In terms of safety, natural compounds demonstrate clear advantages. The adverse event incidence in the three RCTs was 10–20%, mostly mild gastrointestinal reactions (27, 107, 108), while procedure-related complication rates for anti-VEGF intravitreal injection were 1–3% (106). However, the lack of long-term safety data remains a major shortcoming. Existing RCTs have follow-up periods of only 3–6 months, unable to assess cumulative risks of long-term use.

Based on the grading of recommendations assessment development and evaluation (GRADE) evidence quality assessment framework, the current evidence quality for natural compounds in DR treatment is “low to very low” level, mainly limited by small sample sizes, short follow-up, and use of surrogate outcome measures. Nevertheless, considering their good safety profile, convenience of oral administration, and low cost (approximately $30–50/month vs. anti-VEGF $1500–2000/injection), natural compounds can be may warrant further investigation as adjuvant treatment choices for patients with early non-proliferative DR.

The regulation of glutamate metabolism by polyphenolic compounds involves three key levels: transport, receptors, and metabolic enzymes, and this multi-level regulatory mechanism has been systematically verified in both in vitro and in vivo experiments. At the transport level, quercetin (50 μM) can upregulate Müller cell glutamate transporter recombinant excitatory amino acid transporter(EAAT1)expression by 2.1-fold, reducing extracellular glutamate concentration by 38%, enhancing glutamate clearance capacity. At the receptor level, epigallocatechin gallate (EGCG) inhibits N-Methyl aspartic acid(NMDA) receptors in a non-competitive manner (IC50 = 84 μM), blocking glutamate-induced calcium influx by up to 65% (109), directly reducing excitotoxicity. At the metabolic level, quercetin intervention can increase the glutamine/glutamate ratio from 1.2 to 1.8 (healthy control is 2.0) (110), suggesting partial recovery of glutamine synthetase (GS) activity.

Animal experiments further confirmed the in vivo relevance of these mechanisms. Analysis of three high-quality quercetin studies showed that glutamate levels decreased by an average of 32–38% in DR models, consistent with in vitro observations (111, 112). Liu et al. (113) used quercetin (50 mg/kg/day, 12 weeks) in STZ-induced DR rats and observed not only 35% reduction in retinal glutamate but also confirmed through immunohistochemistry a 2.3-fold increase in EAAT1 protein expression and 52% reduction in NMDA receptor NR2B subunit phosphorylation (110), establishing a direct link between mechanism and effect. More importantly, these molecular changes translated into significant pathological improvements: 52% reduction in retinal ganglion cell apoptosis and 45% improvement in nerve fiber layer thickness preservation (110).

The relative importance of these mechanisms may vary with disease stage. In early non-proliferative DR, when Müller cell function is still partially preserved, enhancing glutamate transport may be more effective; while in late proliferative DR, where Müller cells have widespread dysfunction (55), receptor-level protection may be more critical. Notably, polyphenolic compounds also protect GS activity through ROS scavenging, forming an indirect metabolic regulatory mechanism, as ROS can directly oxidize the active center cysteine residues of GS leading to enzyme inactivation (114).

Catechins regulate cellular redox status through both direct and indirect mechanisms, and the temporal characteristics of this dual mechanism are crucial for understanding their pharmacological action. The direct mechanism manifests as rapid scavenging of superoxide anions and hydroxyl radicals, with scavenging rates comparable to vitamin C (115). The indirect mechanism involves activation of the nuclear factor erythroid 2-related factor 2/antioxidant response element(Nrf2-ARE) pathway, upregulating endogenous antioxidant enzyme expression: EGCG (10–50 μM) can increase heme oxygenase-1 (HO-1) expression by 3.2-fold and quinone oxidoreductase 1 (NQO1) by 2.8-fold (116).

Time-course studies have revealed the relative contributions of these two mechanisms: in high glucose-treated endothelial cells, green tea polyphenols reduced ROS by 22% within 30 minutes, 45% after 6 hours, and 52% after 24 hours. This biphasic kinetics suggests that early effects mainly originate from direct scavenging (contributing 22%), while sustained effects depend on activation of endogenous antioxidant systems (additional contribution of 30%). Time-course analysis of animal experiments further supports this pattern: resveratrol reduced ROS by 19% at 1 week, 33% at 4 weeks, and reached a plateau (38%) at 8 weeks in DR rats (104). This mechanistic difference has important implications for clinical application: acute hyperglycemic events require rapid protection from direct antioxidant action, while long-term prevention relies on sustained activation of endogenous systems.

In promoting glutathione (GSH) synthesis, polyphenolic compounds simultaneously act on both rate-limiting enzymes and substrate supply. Catechins increase glutamate-cysteine ligase catalytic subunit (GCLC) expression by 2.5-fold through Nrf2 activation (117), enhancing GSH synthesis capacity. Meanwhile, quercetin metabolites promote cysteine uptake by upregulating the cystine/glutamate antiporter, increasing intracellular cysteine levels by 35% (118), alleviating substrate limitation. Meta-analysis of six high-quality animal studies showed that GSH levels increased by an average of 42% (95% CI: 38–46%), highly correlated with the magnitude of GCLC upregulation (r = 0.89) (55, 104, 105, 111, 112, 119).

Polyphenolic compounds have demonstrated multi-dimensional protective effects in animal experiments, but translation to clinical efficacy faces critical barriers. The comprehensive study by Do et al. (104) showed that resveratrol (50 mg/kg/day, 8 weeks) in STZ-induced DR rats reduced glutamate levels by 32%, oxidative stress markers by 35–40%, pro-inflammatory factors by 30–35%, ultimately leading to 40% reduction in neuronal apoptosis and 35% reduction in blood-retinal barrier leakage (104). Meta-analysis of six independent studies further confirmed the robustness of these effects: weighted mean ROS reduction rate was 37.8% (95% CI: 34.2–41.4%), with moderate heterogeneity (I² = 42%) mainly derived from different model types and intervention durations (105, 111, 112, 119).

However, the reproducibility of these encouraging animal experimental results in clinical trials has been limited. Three RCTs provide preliminary clinical evidence, but effect sizes are notably smaller than expected from animal experiments. The double-blind RCT by Callan et al. (107) (n = 85) showed that combined quercetin (500 mg/day) + resveratrol (250 mg/day) treatment for 6 months reduced central macular thickness by 32 μm (p < 0.001), HbA1c by 0.7% (p < 0.01), and improved visual acuity by 0.15 logarithm of the minimum angle of resolution(logMAR) (107). The open-label trial by Cappellani et al. (27) (n = 62) showed weaker effects: thickness reduction of 18 μm, HbA1c reduction of 0.5% (27). The small-sample trial by Li et al. (38) (n = 30) was intermediate: thickness reduction of 25 μm (108).

Differences in effect size may stem from multiple translational barriers. First is the bioavailability issue: oral bioavailability of resveratrol is only 0.5–1% (120), and the 50 mg/kg dose in animal experiments converts to approximately 300 mg/day in humans. Considering bioavailability, the amount of active ingredient actually entering circulation may be insufficient to achieve tissue concentrations seen in animal experiments. The 250 mg/day dose used in the Zhang study, while higher than the 150 mg/day in the Wang study, may still be below the optimal therapeutic window. Second is patient heterogeneity: the Zhang study enrolled patients with moderate non-proliferative DR (mean disease duration 5.2 years, baseline thickness 387 μm), while the Wang study included more patients with mild disease (duration 3.8 years, baseline thickness 318 μm); the more severe the baseline lesion, the greater the room for improvement. Third is intervention duration: animal studies typically last 8–12 weeks (equivalent to several human years), while clinical trials were only 6 months, which may be insufficient to observe maximal effects after full mechanism activation.

Berberine regulates the arginine-nitric oxide pathway through multiple nodes, demonstrating action characteristics different from polyphenols. At the enzyme activity level, berberine (10 μM) increases eNOS Ser1177 phosphorylation by 2.3-fold and NO production by 38% through activation of the Adenosine 5’-monophosphate-activated protein kinase(AMPK-eNOS) signaling pathway (121). Simultaneously, berberine (5 μM) inhibits inducible nitric oxide synthase (iNOS) expression by 68% under inflammatory conditions (122), reducing iNOS gene transcription by inhibiting NF-κB nuclear translocation. This bidirectional regulatory pattern of “selectively activating eNOS while inhibiting iNOS” has important physiological significance: low concentrations of NO produced by eNOS have vascular protective effects, while high concentrations of NO produced by iNOS participate in inflammatory damage.

At the substrate supply level, berberine increases arginine bioavailability through three complementary mechanisms. First, upregulation of cationic amino acid transporter (CAT-1) expression increases arginine uptake by 20–30% (123). Second, inhibition of arginase activity reduces arginine metabolic consumption toward ornithine, increasing arginine retention. Third, reduction of asymmetric dimethylarginine (ADMA) levels by 28%; ADMA is an endogenous eNOS inhibitor, and its reduction can relieve competitive inhibition of eNOS (124).

However, the integrated effect of these mechanisms is controversial. Theoretically, if the effects of CAT-1 upregulation (+25%), arginase inhibition (+35%), and ADMA reduction (+15–20% eNOS activity) were independent of each other, total NO increase should approach 100% (multiplicative effect). But actual observed NO increase was only 42% (125), far below theoretical expectations. This suggests overlap among mechanisms, or the existence of other rate-limiting steps (such as insufficient eNOS cofactor tetrahydrobiopterin(BH4), or compensatory mechanisms in vivo offsetting part of the effects.

A more critical issue is in vivo drug concentration. Plasma concentrations of berberine are typically < 0.5 μM (126), far below the effective in vitro concentration (5–10 μM). This large gap suggests that the direct effects observed in vitro may not be the main mechanism in vivo; intestinal microbial metabolites or indirect signaling pathways may play more important roles.

Three independent studies validated the protective effects of berberine in different DR models, but also revealed important methodological heterogeneity. Zhang et al. (95) used 200 mg/kg dose in STZ rats, and after 8 weeks of intervention, serum NO increased by 38% and vascular leakage decreased by 42% (127). Li et al. (128) used 150 mg/kg in db/db mice, and after 12 weeks, retinal tissue NO increased by 42%, DR lesion score decreased by 35%, and electroretinogram b-wave amplitude improved by 28% (128). Yin et al. (129) compared two doses of 100 and 200 mg/kg in a STZ + high-fat diet mixed model; the high-dose group showed 40% NO increase and 40% reduction in acellular capillaries (129).

The consistency of these studies is mainly reflected in the magnitude of NO increase (38–42%) and degree of lesion improvement (35–42%), with weighted mean NO increase of 39% (95% CI: 36–42%). However, in-depth analysis reveals critical evidence gaps. First, measurement methods were inconsistent: Zhang et al. (95) used the Griess method to measure serum nitrite (a stable metabolite of NO), Li et al. used fluorescent probes to directly measure tissue NO; the comparability of the two is questionable. Serum levels reflect systemic effects while tissue levels reflect local effects. Yin et al.’s results using both methods simultaneously were most valuable, finding that retinal NO increase (40%) was slightly higher than serum (35%) (129), suggesting berberine may have tissue-specific effects in the retina.

Second, time dependence was not adequately explored. Time-course data from Li et al. showed that NO increase was 22% at 4 weeks, 35% at 8 weeks, and 42% at 12 weeks, showing a continuously increasing trend (125). This suggests that berberine’s effects require longer time to accumulate, possibly related to its action through regulation of gene expression and metabolic network remodeling. This characteristic has important implications for clinical trial design: at least 3 months of intervention is needed to observe full effects.

Third, dose-response relationship validation was insufficient. Only Yin et al. included dose comparison: 100 mg/kg increased NO by 25% and reduced lesions by 28%, while 200 mg/kg increased NO by 40% and reduced lesions by 40% (129). Doubling the dose increased effects by approximately 60%, showing a near-linear relationship but possibly approaching plateau. This non-linear characteristic needs to be considered when extrapolating human doses, avoiding simple pursuit of high doses.

The biggest challenge facing berberine for DR treatment is extremely low bioavailability (< 1%) (126). The 150–200 mg/kg doses used in animal experiments convert to approximately 900–1200 mg/day for adults based on body surface area. Considering bioavailability, the actual berberine entering circulation is only 9–12 mg, making it difficult to achieve effective concentrations shown in in vitro studies.

Strategies to improve bioavailability include nanoformulations, phospholipid complexes, and gut microbiota modulation, but clinical studies of these new formulations are extremely limited. A more realistic strategy may be to utilize berberine’s local intestinal effects and metabolite activity. High concentrations of berberine in the intestine may exert indirect effects through modulating gut microbiota and the gut-retina axis (130), while metabolites such as berberrubine, although less active, may have higher bioavailability.

Another critical issue is drug interactions. Berberine is an inhibitor of CYP2D6 and CYP3A4 (131), potentially increasing plasma concentrations of drugs metabolized by these enzymes (such as statins, some hypoglycemic agents). Although in vitro studies show inhibitory effects, clinical significance requires dedicated drug interaction studies for confirmation. This safety issue should not be ignored given that diabetic patients routinely use multiple medications.

Terpenoid compounds represent another important category of natural active ingredients, with their mechanism of action primarily focused on immunometabolic regulation of the tryptophan-kynurenine (Trp-Kyn) pathway. Unlike polyphenols and alkaloids, terpenoid compounds such as ginsenosides, asiaticoside, and glycyrrhizic acid primarily regulate inflammatory and immune responses by modulating indoleamine 2,3-dioxygenase (IDO) and tryptophan 2,3-dioxygenase (TDO) activity, affecting the metabolic conversion of tryptophan to kynurenine.

Ginsenoside Rg1 is one of the most extensively studied terpenoid compounds. Under DR pathological conditions, pro-inflammatory factors such as IFN-γ and TNF-α significantly upregulate IDO expression, causing excessive metabolism of tryptophan to kynurenine and its downstream products, leading to reduced synthesis of protective tryptophan metabolites serotonin and melatonin, while accumulating neurotoxic metabolites of the kynurenine pathway (such as quinolinic acid) (93). Ginsenoside Rg1 (20–40 μM) reduces IDO expression by 45% and the Kyn/Trp ratio from the pathological state of 0.082 to 0.051 (healthy control is 0.035) through inhibition of the JAK-STAT signaling pathway (132). This regulation not only reduces neurotoxic metabolite production but also partially restores tryptophan flow toward protective pathways.

Asiaticoside demonstrates an action pattern complementary to ginsenoside. In addition to inhibiting IDO activity, asiaticoside (10–30 μM) also promotes the conversion of kynurenine to kynurenic acid through activation of the aryl hydrocarbon receptor (AhR); kynurenic acid is a metabolite with antioxidant and neuroprotective effects (133). This dual regulatory strategy of “reducing upstream production while promoting harmless metabolism” may be more advantageous than simply inhibiting IDO.

The regulation of the Trp-Kyn pathway by terpenoid compounds is closely related to the inflammation-metabolism axis, forming a multi-level regulatory network. Glycyrrhizic acid, as the main active component of licorice, blocks the AGE-RAGE-inflammation axis by inhibiting high mobility group box 1 (HMGB1) release and RAGE (receptor for advanced glycation end products) activation (134). In STZ-induced DR rats, glycyrrhizic acid (30 mg/kg/day, 10 weeks) reduced serum HMGB1 by 42%, retinal RAGE expression by 48%, subsequently IDO expression by 35%, and inflammatory factors IL-1β and IL-6 by 40% and 38%, respectively (135).

This regulation of the inflammation-metabolism axis has cascade amplification effects. Inflammatory factors not only directly upregulate IDO but also enhance oxidative stress through activation of NF-κB and STAT pathways, while oxidative stress in turn promotes inflammatory factor release, forming a vicious cycle (136). Terpenoid compounds achieve multi-point blockade of this vicious cycle by simultaneously acting on multiple nodes (inflammatory factors, oxidative stress, metabolic enzymes). The study of ginsenoside Rb1 in db/db mice showed that, in addition to reducing IDO activity, it also reduced ROS by 32% and NF-κB nuclear translocation by 55%, ultimately leading to a 45% reduction in retinal microvascular lesion score (137).

Despite in vitro and animal experiments showing the potential of terpenoid compounds, their clinical translation faces unique challenges. Ahmed et al. (138) systematically evaluated the effects of ginsenoside Rg1 in STZ rats: 40 mg/kg/day for 12 consecutive weeks reduced retinal Kyn/Trp ratio by 38%, increased kynurenic acid/kynurenine ratio by 65%, accompanied by 42% improvement in ganglion cell layer thickness preservation and 47% reduction in blood-retinal barrier leakage (138). Similarly, the asiaticoside study by Li et al. (38) (50 mg/kg/day, 10 weeks) showed 40% reduction in DR lesion score and 35% improvement in electroretinogram oscillatory potentials (113).

However, these studies also exposed common limitations of terpenoid compounds. First is the metabolic stability issue: ginsenosides are rapidly deglycosylated in vivo, converting to metabolite ginsenoside aglycones, whose activity and safety differ from the parent compound. Second is tissue distribution specificity: terpenoid compounds generally have high plasma protein binding rates (> 95%), limiting distribution to retinal tissue (139). Pharmacokinetic studies show that after oral administration of ginsenoside Rg1, retinal tissue concentrations are only 8–12% of plasma concentrations, far below effective concentrations in in vitro experiments (140).

More importantly, clinical studies of terpenoid compounds are almost completely absent. Currently, no clinical trials of terpenoid compounds targeting DR have been published. This evidence gap severely limits clinical application recommendations for terpenoid compounds, although their unique mechanism of action suggests potential synergistic value when combined with polyphenols and alkaloids.

Despite limited evidence for use alone, terpenoid compounds may play a unique role in combination therapy strategies. Polyphenols primarily regulate glutamate metabolism and direct antioxidation, alkaloids primarily act on the arginine-NO pathway and vascular function, while terpenoids regulate immune-inflammatory responses through the Trp-Kyn pathway; the mechanisms of action of the three are highly complementary. Theoretically, this “antioxidant (polyphenols) + vascular protection (alkaloids) + anti-inflammatory (terpenoids)” combination may achieve more comprehensive pathological intervention.

Preliminary animal experiments support this hypothesis. Compared the effects of resveratrol alone, ginsenoside Rg1 alone, and their combination in db/db mice: resveratrol reduced ROS by 35% and lesion score by 32%; ginsenoside reduced inflammatory factors by 42% and lesion score by 30%; the combination group reduced ROS by 52%, inflammatory factors by 58%, and lesion score by 55%. The combined effect was notably greater than the sum of single-use effects, suggesting synergistic action. Mechanistic studies showed that oxidative stress reduced by resveratrol decreased inflammatory factor-induced IDO activation, while inflammatory factors reduced by ginsenoside decreased stimulation of ROS generation, forming positive synergy.

However, combination therapy also brings increased complexity. Issues such as pharmacokinetic interactions, dose optimization, and adverse reaction superposition require systematic study. More importantly, clinical trial design complexity increases exponentially: evaluating all possible combinations of three compound classes requires substantial samples and resources. A realistic strategy may be to first establish efficacy of single compounds, then gradually explore the most promising combination regimens.