Abstract

   Blood-retinal barrier (BRB) breakdown, a pivotal contributor to
   multiple retinal vascular diseases, manifests as a progressive increase
   in vascular permeability induced by various pathological stimuli. The
   functional plasticity of retinal endothelial cells can be intricately
   shaped by metabolic alteration. However, little is known about the
   mechanisms through which endothelial metabolic disorders trigger the
   dissolution of inter-vascular junctions and the selective approaches to
   targeting metabolic homeostasis. Herein, we identify AMPK-associated
   fatty acid oxidation (FAO) inhibition as a critical driver of vascular
   barrier dysfunction via exacerbating redox imbalance. Pharmacological
   facilitation of FAO by ginsenoside Rd (Rd) suppresses BRB collapse and
   other secondary retinal damage in diabetic retinopathy (DR).
   Mechanistically, Rd targets GPR30 to phosphorylate AMPK via the
   PKA-LKB1-AMPK kinase cascade. The AMPK activation induced by Rd
   revitalizes hyperglycemia-compromised FAO, and then sustains
   mitochondrial NADPH regeneration by emphasis on IDH2 at various levels,
   including substrate supply, transcription, and post-translational
   modifications. Therefore, Rd alleviates the disruption of BRB integrity
   driven by mitochondrial oxidative stress, with the vasculoprotection of
   Rd diminished by GPR30 knockdown and pharmacological attenuation of
   AMPK. These findings collectively reveal the previously-unanticipated
   role of endothelial FAO in heightened retinal vascular leakage, and
   highlight the potential translational application of GPR30 agonism with
   Rd to mitigate barrier dysfunction, providing a metabolic regulatory
   therapeutic strategy for DR.

Graphical Abstract

   [32]graphic file with name 12933_2025_2638_Figa_HTML.jpg

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s12933-025-02638-3.

   Keywords: Diabetic retinopathy, Ginsenoside Rd, Blood-retinal barrier,
   Fatty acid oxidation, GPR30, Mitochondrial oxidative stress

Introduction

   Diabetic retinopathy (DR) emerges as a prevalent and serious diabetic
   microvascular complication, affecting approximately one-third of
   individuals with diabetes mellitus [[33]1–[34]2]. As the prevalence of
   diabetes mellitus continues to rise, DR has gradually become the
   leading cause of blindness in adults [[35]3]. Although the clinical
   manifestations of DR are diverse and intricate, the progressive
   vascular barrier dysfunction and the subsequent compromise of
   blood-retinal barrier (BRB) integrity exhibit a close connection with
   all known clinical symptoms of DR [[36]4–[37]5]. The current approach
   to treating early DR primarily relies on strict metabolic control of
   diabetes until the disease progresses to the advanced stage,
   necessitating vitreous injections of vascular endothelial growth factor
   (VEGF) antibodies and other invasive and destructive therapies [[38]4,
   [39]6]. Characterized by its insidious onset, poor prognosis and
   limited treatment options, initial prevention of lesions in DR carries
   profound clinical and societal significance, before visual symptoms
   become irreversible.

   Endothelial cell (EC) metabolism plays a critical role in determining
   the vascular function [[40]7–[41]8]. Nevertheless, the extent to which
   abnormal retinal endothelial metabolism contributes to these phenotypic
   changes in DR remains largely unexplored. Clinical studies have
   frequently revealed dysregulated lipid metabolism resulting from
   hyperglycemia in cases of DR [[42]9]. Consequently, hyperlipemia has
   been proposed as a potential risk factor for DR, and the use of
   fibrates has shown protective effects against retinal vascular damage
   [[43]10]. Most findings around retinal vascular metabolism have
   centered on enhanced vessel sprouting through endothelial glycolysis
   [[44]11]. Further investigation is warranted to understand the specific
   mechanisms by which endothelial lipid metabolism maintains phenotypic
   homeostasis of DR, in consideration of the powerful vaso-regulatory
   activity of lipid metabolism in other diabetic complications [[45]12].

   AMP-activated protein kinase (AMPK) serves as a sensitive sensor of
   cellular energy levels, capable of being activated by AMP accumulation
   as well as various upstream kinases [[46]13]. In instances of depleted
   energy, the phosphorylated AMPK re-establishes ATP balance and adapts
   to nutrient availability by inhibiting anabolism and facilitating
   catabolism. The hyperglycemia-induced inhibition of AMPK fosters
   intraocular inflammation, pericyte apoptosis, and reduced retinal blood
   flow in diverse models of DR [[47]14]. Moreover, numerous clinical
   drugs and natural products improve neurovascular injury of DR in an
   AMPK-dependent manner [[48]14–[49]15]. GPR30, a member of the estrogen
   receptor family, affects vascular function and cell metabolism, but has
   not been investigated in retinal diseases [[50]16–[51]17]. Based on the
   preceding discoveries, it is plausible to hypothesize that AMPK and
   GPR30 may be involved in regulating the retinal endothelial dysfunction
   caused by metabolic disturbances in diabetic conditions.

   Ginsenoside Rd (Rd) is a natural tetracyclic triterpenoid, served as
   the principal medicinal constituent in Panax notoginseng (Burk.) F. H.
   Chen [[52]18]. Notoginseng Radix et Rhizoma (Sanqi) and its compound
   preparation are commonly prescribed in clinical settings for the
   treatment of DR [[53]19]. Previous pharmacological investigations have
   primarily concentrated on the neuroprotective effects of Rd in
   neurological and cerebrovascular disorders [[54]20]. Our prior research
   has demonstrated that Rd facilitates the interaction between AMPK and
   SIRT1 in ECs by increasing the levels of NAD^+/NADH and deacetylating
   LKB1. The emphasis on AMPK/SIRT1 interaction by Rd beneficially reduces
   ROS production, reverses endothelial mitochondrial intrinsic apoptosis,
   and ultimately ameliorates diabetes-driven retinal vascular damage
   [[55]21]. The specific upstream mechanisms by which Rd activates AMPK
   and the target molecules for treating DR are still not fully
   understood. Thereby, this study aims to explore the significance of
   AMPK-related lipid metabolism derangements in endothelial dysfunction
   induced by hyperglycemia. Additionally, a metabolic perspective is
   employed to further elucidate the vasculoprotective mechanism of Rd in
   DR and identify the potential therapeutic targets.

Materials and methods

Animals

   C57BL/6 male mice (No: SCXK [Su] 2018-0006), aged 6–8 weeks, were
   procured from Shanghai Sippr-BK Laboratory Animal Co. Ltd. (Shanghai,
   China). Animals were housed and provided with standard laboratory
   conditions, including a 12h light/dark cycle, ad libitum access to
   drinking water, and standard laboratory chow. All animal studies were
   conducted in accordance with the guidelines set by the National
   Institutes of Health, and reviewed and approved by the Institutional
   Animal Care and Use Committee of China Pharmaceutical University and
   the Institutional Ethics Committee of China Pharmaceutical University
   (approval number: 2023-08-017). A deliberate endeavor was undertaken to
   mitigate the harm caused to animals and reduce the quantity of animals
   utilized. The animal studies were methodically designed to ensure the
   formation of groups of equal size through the implementation of
   randomized and blinded analysis techniques.

   For knockdown, recombinant AAV8 vectors carrying shRNA targeting mouse
   GPR30 (GenePharma, Shanghai, China) were used. In brief, C57BL/6 male
   mice were subjected to intravitreal injection of 5 µl AAV8-shGPR30
   (virus titer > 10^12 viral genomes/mL) one week prior to STZ
   administration. Control mice were injected with an equal volume of
   AAV8-control suspension. The efficiency of AAV transduction in vivo was
   assessed through Western blot analysis.

STZ-induced diabetic model

   The DR model was established in C57BL/6 mice via intraperitoneal
   injection of STZ. The fasted mice received 55 mg/kg of STZ via
   intraperitoneal injection for five consecutive days. Control animals
   were administered with an equivalent volume of sterile citrate buffer.
   Fasting blood glucose levels were measured using a glucometer (Roche,
   Mannheim, Germany) one week after the final injection. Mice exhibiting
   blood glucose levels above 16.7 mM, along with other characteristic
   symptoms of diabetes, were considered to have successfully undergone
   modeling.

   In the pharmacodynamic studies, the mice were separated into five
   groups: normal control (A1), DR model (B1), 0.5% Rd eye drops (C1), 1%
   Rd eye drops (D1), and 100 mg/kg CaD (F1). Mice in groups A1 and B1
   were given an equal volume of vehicle eye drops. Groups C1 and D1 were
   subjected to daily administration of 5 µl of 0.5 or 1% (w/v) Rd eye
   drops. 100 mg/kg CaD was administered daily to mice in group F1 by oral
   administration.

   In mechanistic studies for lipid metabolism, the mice were separated
   into five groups: normal control (A2), DR model (B2), 1% Rd eye drops
   (C2), 1% Rd eye drops + Eto (D2), and Mito-TEMPO (E2). Groups C2 and D2
   were administered with 5 µl 1% (w/v) Rd eye drops every day. 1 µl of
   Eto (200 µM) or Mito-TEMPO (200 µM) was injected into vitreous of mice
   in group D2 or E2 once a week.

   In mechanistic studies for GPR30, the mice were separated into six
   groups: normal control (A3), DR model (B3), 1% Rd eye drops (C3), 1% Rd
   eye drops + AAV8-shGPR30 (D3), 1% Rd eye drops + AAV8-NC (E3), and 100
   µM G1 (F3). Groups C3, D3 and E3 were administered with 5 µl 1% (w/v)
   Rd eye drops every day. 1 µl of G1 (100 µM) was injected into vitreous
   of mice in group F3 once a week.

   Following three months of uninterrupted administration, the retinal
   tissue samples were collected for subsequent analysis. The preparation
   of eye drops and other drug dosages were conducted according to the
   previous report [[56]21].

Evans blue dye leakage

   Evans blue dye leakage assay was carried out as detailed previously
   [[57]21]. Briefly, the left heart ventricle of mice was cannulated and
   perfused with PBS, after injection of 50 mg/kg Evans blue (Sigma) and
   full circulation. After fixation, the retinas were removed from eye
   cups. The dye concentration in the retina was measured with a
   multi-mode microplate reader (Biotek Epoch, Winooski, VT, USA), and the
   representative pictures were taken with a fluorescence microscope
   (Leica DMi8, Wetzlar, Heessen, Germany).

Histological analysis

   Retinal trypsin digestion assay and Hematoxylin & Eosin (HE) staining
   assay were performed as previously reported [[58]21]. The retinas were
   fixed in 10% neutral formalin over 24 h. For retinal trypsin digestion,
   the retinas were incubated in trypsin until the medium became cloudy
   and the peeled vessel networks were stained with PAS-hematoxylin. For
   HE staining, the paraffin sections of eye tissues were dewaxed and
   dehydrated, and then stained with Hematoxylin and Eosin. Representative
   pictures were obtained by an optical microscope (Leica DMi8, Wetzlar,
   Heessen, Germany).

Western blot

   The mice retina and cells were lysed in RIPA buffer with PMSF to
   extract total protein. The cytosolic and nuclear proteins were isolated
   as described in the nucleoprotein extraction kit (Beyotime
   Biotechnology, Nanjing, Jiangsu, China). The protein estimation was
   conducted using BCA protein assay. Equal amounts of the collective
   proteins were separated by SDS-PAGE gels and transferred onto PVDF
   membranes. After blocked with 5% BSA, the membranes were incubated with
   specific primary antibodies overnight at 4 °C. Following removal of
   unbound antibody, the membranes were probed with HRP-conjugated
   secondary antibodies. The protein imaging was performed on Tanon 5200
   multi chemiluminescent substrate system (Tanon, Shanghai, China) and
   analyzed by Image-Pro Plus 6.0 (IPP 6.0) software.

RNA isolation and quantitative real-time PCR

   The total RNA from cells was extracted using Trizol reagent (Takara,
   Otsu, Japan) according to the manufacturer’s instructions. cDNA
   transcribed from the templet RNA was synthesized using HiScript II Q RT
   SuperMix (Vazyme, Nanjing, Jiangsu, China). Real-time qPCR was
   performed using Ace qPCR SYBR Green Master Mix (Vazyme, Nanjing,
   Jiangsu, China) with a StepOne Real-Time PCR System (Applied
   Biosystems). The primer sequences used in the study were shown in
   Supplementary file [59]1: Table S1. The mRNA relative expression of
   target genes was normalized to GAPDH and analyzed using the 2^−ΔΔCt
   method.

FITC-dextran leakage assay

   Endothelial cell permeability was monitored via the leakage of
   FITC-dextran. Primary rat retinal vascular endothelial cells (RVECs)
   were grown to confluence on transwell inserts (0.4 μm; Millipore).
   After treatment, 1 mg/ml FITC-dextran (70 kDa) was added into the upper
   wells of the transwell inserts. Following 4 h incubation, the
   fluorescence intensity of FITC-dextran in the medium of the lower wells
   was measured by multi-mode microplate reader (Biotek Epoch). The
   fluorescence intensity was converted into concentration through the
   standard curve.

Transendothelial electrical resistance assay

   RVECs were cultured on transwell inserts (0.4 μm; Millipore), and
   received different treatments. Transendothelial electrical resistance
   (TER) was determined by Millicell-ERS voltohmmeter together with an
   electrode (Millipore, USA). The TER value was expressed as the common
   unit (Ω·cm^2) after background-correction by subtraction of the value
   of a blank cell-free insert.

Detection of oxygen consumption rate

   FAO was measured using XF palmitate oxidation stress test kit (Seahorse
   Bioscience, MA, USA) according to the manufacturer recommended
   protocols. Briefly, RVECs (8000 per well) were seeded in XF96 cell
   culture microplates precoated with fibronectin. After starvation in
   substrate limited medium (500 µM glucose, 1 mM L-glutamine, 0.5 mM
   L-carnitine and 1% FBS), the cells were replaced with FAO assay medium
   (111 mM NaCl, 4.7 mM KCl, 1.25 mM CaCl[2], 2 mM MgSO[4], 1.2 mM
   NaH[2]PO[4], 2.5 mM glucose, 500 µM L-carnitine, and 5 mM HEPES).
   Immediately before the assay, either BSA or palmitate-BSA FAO substrate
   was added to the predetermined wells. The oxygen consumption rate (OCR)
   was continuously measured in real-time following injections of
   oligomycin (1 µg/ml), FCCP (1 µM), rotenone (0.5 µM), and antimycin A
   (0.5 µM). The OCR were normalized according to protein concentrations.

Microscale thermophoresis assay

   The binding affinity of Rd for GPR30 was determined using microscale
   thermophoresis assay (MST). HEK293T cells were transfected with the
   plasmids encoding EGFP-tagged GPR30 (GenePharma) and the lysate cell
   supernatant was collected for MST experiments. The samples were loaded
   in MST standard capillaries and analyzed by NT.115 NanoTemper Monolith
   Instrument (NanoTemper, Germany).

Statistical analysis

   Data were presented as the mean ± standard error of the mean (SEM). All
   statistical analyses and plots were generated using GraphPad Prism
   software (San Diego, CA, USA). Statistical variation between two groups
   by using the two-tailed Student’s t-test. Multi-group comparisons were
   evaluated utilizing one-way ANOVA with Tukey’s test. p-value < 0.05 was
   considered statistically significant.

Results

Ginsenoside Rd attenuated diabetes-related endothelial barrier dysfunction in
vitro and in vivo

   Notoginseng Radix et Rhizoma and its primary components i.e. Panax
   notoginseng saponins (PNS) had been found to strengthen the barrier
   function of the retinal vessels in DR [[60]19, [61]22]. In this study,
   an in vitro model of DR was established in high glucose-induced RVECs.
   The EC permeability and immunoblot assays were conducted to screen the
   main ginsenosides in PNS (Fig. [62]1a and Supplementary file [63]1:
   Fig. S1b-c). Subsequently, four saponins (ginsenoside Rg1, Rb1, Rd, and
   Rc) with superior in vitro activities were topically administered to
   mice intravitreally injected with VEGF. Among these saponins,
   ginsenoside Rd efficiently inhibited the acute augmentation of vascular
   permeability in response to VEGF (Fig. [64]1b). Accordingly, Rd was
   selected for the further investigation.

Fig. 1.

   [65]Fig. 1
   [66]Open in a new tab

   Ginsenoside Rd attenuated diabetes-related endothelial barrier
   dysfunction in vitro and in vivo. a Permeability of EC monolayer was
   assessed by FITC-dextran. All saponins were added at a concentration of
   30 µM (n = 3). b Evans blue dye leakage assay was used to quantify
   vascular leakage in VEGF-induced acute retinal vascular
   hyperpermeability model by a microplate reader (n = 6). c-d Mean
   DCFH-DA and DHE fluorescent intensity levels in RVECs treated with
   vehicle, high glucose (33 mM), Rd and CaD (100 µM) (n = 3). e-f
   Endothelial monolayer integrity and polarity were examined by
   FITC-dextran leakage and TER assay (n = 3). g Immunoblot analysis of
   VE-cadherin, ZO-1, Occludin and Cx43 expressions in RVECs. GAPDH was
   used as the loading control (n = 3). h Immunofluorescent staining of
   VE-cadherin and ZO-1 in RVECs. Scale bars: 40 μm (n = 3). i Retinal
   section stained with H&E. Scale bars: 25 μm (n = 5). j MDA
   concentration of retina from vehicle, hyperglycemia, Rd-treated and
   CaD-treated mice (n = 6). k Retinal trypsin digestion was stained with
   Periodic-acid Schiff’s reagent. Acellular capillaries were denoted by
   red arrows in representative graphs. Scale bars: 50 μm (n = 5). l
   Retinal vascular leakage in STZ-induced DR model was detected by Evans
   blue dye. Representative images were observed by fluorescence
   microscope. Scale bars: 50 μm (n = 5). m Protein expressions of
   VE-cadherin, ZO-1, Occludin and Cx43 in retina were determined by
   Western blot (n = 6). ##p < 0.01 vs. the control, *p < 0.05, **p < 0.01
   vs. DR group or high glucose group

   Our previous research demonstrated that Rd exerted anti-apoptotic
   effects on retinal ECs via preserving redox homeostasis [[67]21]. The
   involvement of oxidative stress in the BRB impairment and DR had been
   extensively established [[68]5, [69]23]. The collapse of BRB was
   characterized through the FITC-dextran flux across EC monolayer and the
   transendothelial electrical resistance (TER) analysis in vitro. High
   glucose exposure elevated the levels of total ROS and superoxide anion
   in RVECs, which resulted in a disruption of endothelial monolayer
   integrity and polarity (Fig. [70]1c-f). The permeability of vascular
   barrier was mainly controlled by inter-endothelial adherens junctions,
   tight junctions, and gap junctions [[71]23]. High glucose blunted the
   expression of proteins associated with adherens junctions
   (VE-cadherin), tight junctions (ZO-1 and Occludin) and gap junctions
   (Cx43), thereby impeding the function of BRB (Fig. [72]1g and
   Supplementary file [73]1: Fig. S1d-g). In normal RVECs, the
   immunostaining for VE-cadherin and ZO-1 appeared as a continuous band
   at the junctions of retinal EC, whereas high glucose dampened and broke
   the immunostaining pattern (Fig. [74]1h and Supplementary file [75]1:
   Fig. S1h-i). In high glucose-treated RVECs, Rd attenuated the
   fluorescence intensities of DCFH and DHE (Fig. [76]1c-d). The loss of
   endothelial monolayer integrity and polarity was reversed by Rd in a
   concentration-dependent manner (Fig. [77]1e-f). Rd boosted the level of
   vascular barrier function-related proteins in RVECs (Fig. [78]1g-h and
   Supplementary file [79]1: Fig. S1d-i). Additionally, calcium dobesilate
   (CaD), a common drug for diabetic microvascular complications,
   partially restored the aforementioned alterations in endothelial
   function (Fig. [80]1c-h and Supplementary file [81]1: Fig. S1d-i). To
   further validate the screening findings and elucidate the in vivo
   efficacy of Rd, a mouse model with STZ-induced hyperglycemia was
   employed. HE staining highlighted the decreased retinal thickness,
   disorganized retinal architecture and neuronal cell loss in the DR
   mice. The above pathological changes were significantly alleviated by
   Rd in a dose-dependent manner (Fig. [82]1i and Supplementary file
   [83]1: Fig. S1j). Meanwhile, Rd decreased the content of lipid
   oxidation product MDA in the retina affected by DR (Fig. [84]1j). The
   retinal vessel morphology and perfusion were the main measures of
   retinal constitutive endothelial barrier stabilization, and could be
   detected by retinal vasculature digestion and Evans blue dye leakage
   assay. The administration of Rd and CaD reduced diabetes-induced
   retinal acellular capillary formation, and improved the vascular
   derangements and tortuosity (Fig. [85]1k and Supplementary file [86]1:
   Fig. S1k). Furthermore, there was more Evans blue extravasation in the
   model group, which was significantly blocked after treatment with Rd
   (Fig. [87]1l and Supplementary file [88]1: Fig. S1l). Moreover,
   therapeutic effects of Rd on in vivo BRB repair were also recapitulated
   in the detection of junctional proteins in the retina (Fig. [89]1m and
   Supplementary file [90]1: Fig. S1m-p). Similar effects were observed
   with the positive drug CaD, in which Rd with superior pharmacological
   activity and lower doses might have the better clinical potential.
   Above results collectively supported that Rd effectively prevented
   hyperglycemia-mediated retinal endothelial barrier breakdown and
   subsequent vascular damage both in vitro and in vivo.

Acceleration of FAO by ginsenoside Rd alleviated retinal endothelial barrier
function in vivo and in vitro

   Fatty acid metabolism played a multifaceted role, not only in providing
   energy, but also in regulating signaling pathway and endothelial
   function [[91]8]. To probe the metabolic susceptibility of lipid in
   RVECs, a PCR screening of a panel of crucial enzymes was conducted.
   High glucose elicited minimal impact on fatty acid uptake (CD36, FATP3
   and FATP4), fatty acid transport (FABP4 and FABP5) and fatty acid
   synthesis (FASN and SCD1) (Fig. S2a), but reduced transcript levels of
   fatty acid oxidation (FAO) metabolic enzymes (CPT1A and CPT2) and their
   regulatory factors (PPARα and PGC-1α) (Fig. [92]2a). Mannitol was used
   as an osmolality control. To explore whether these transcriptional
   changes were accompanied by corresponding alterations in cellular
   metabolism, the measurements were taken by both the indirect indicator
   of mitochondrial FAO capacity (TMRE assay) and the direct indicator of
   FAO capacity (seahorse assay). There was a lower mitochondrial membrane
   potential in RVECs stimulated with high glucose (Fig. [93]2b and Fig.
   S2b). High glucose milieu exerted a negative effect on the palmitate
   oxidation rate of RVECs (Fig. [94]2c-d). Notably, the transcriptome
   data from the public database ([95]GSE53257) supported these results,
   showing a downregulation of fatty acid metabolism-related processes in
   the retinas of DR patients (Fig. [96]2e). These results provided the
   compelling evidence implicating a decrease in FAO of ECs in response to
   the high glucose environment.

Fig. 2.

   [97]Fig. 2
   [98]Open in a new tab

   High glucose diminished the endothelial capacity of FAO. a q-PCR
   analysis of CPT1A, CPT2, PPARα and PGC-1α mRNA level in RVECs treated
   with vehicle, high glucose and 33mM Mannitol. GAPDH was used as the
   internal reference (n = 3). b Mitochondrial membrane potential was
   measured by TMRE fluorescent staining. Scale bars: 50 μm (n = 3). c-d
   FAO was determined by OCR using palmitate as a substrate with Seahorse
   XFp Analyser (n = 3). e GO and KEGG pathway enrichment analysis.
   ##p < 0.01 vs. the control

   ECs relied on FAO to facilitate nucleotide synthesis for angiogenesis
   or maintain cell quiescence [[99]24–[100]25]. However, limited research
   had been performed on the relationship between FAO and endothelial
   permeability. To ascertain whether weakened FAO was an indispensable
   prerequisite of BRB impairments, RVECs were treated with the PPARα
   agonist fenofibrate (Feb) to promote FAO. Feb inhibited ROS and MDA
   production (Fig. [101]3a-c), while preserved endothelial barrier
   stabilization in high glucose-induced RVECs (Fig. [102]3d-e).
   Consistently, Feb limited the loss of BRB structural proteins
   (VE-cadherin, ZO-1, Occludin and Cx43) (Fig. [103]3f-g and Fig. S2c-h).
   Thus, the augmented barrier permeability might be mediated by FAO in
   ECs.

Fig. 3.

   [104]Fig. 3
   [105]Open in a new tab

   High glucose drove mitochondrial oxidative stress-related endothelial
   dysfunction via blunting FAO in vitro. a-c Analysis of DCFH-DA, DHE and
   MDA levels in RVECs treated with vehicle, high glucose and Feb (50 µM)
   (n = 3). d-e FITC-dextran flux and TER of the confluent endothelial
   monolayer (n = 3). f Immunoblot analysis of VE-cadherin, ZO-1, Occludin
   and Cx43 expressions in RVECs. (n = 3). g Immunofluorescent staining of
   VE-cadherin and ZO-1 in RVECs. Scale bars: 40 μm (n = 3). h
   Mitochondrial ROS was assessed by MitoSOX staining in RVECs treated
   with vehicle, high glucose, Feb (50 µM) and Mito-TEMPO (50 µM). Scale
   bars: 50 μm (n = 3). i-k Oxidative stress assessment in RVECs (n = 3).
   l-m Integrity and polarity of ECs monolayer (n = 3). n Representative
   Western blot of VE-cadherin, ZO-1, Occludin and Cx43 in RVECs (n = 3).
   o Representative images of immunofluorescence for VE-cadherin and ZO-1
   in RVECs. Scale bars: 40 μm (n = 3). #p < 0.05, ##p < 0.01 vs. the
   control, *p < 0.05, **p < 0.01 vs. high glucose group

   Since the mitochondria served as the principal site for the cellular
   ROS generation and FAO, it was speculated that the barrier-protective
   activity of FAO was inextricably connected to endothelial mitochondrial
   ROS. High glucose contributed considerably to mitochondrial oxidative
   stress, indicated by MitoSOX, which was subsequently reversed by Feb
   and Mito-TEMPO (Fig. [106]3h and Fig. S2i). The mitochondrial-specific
   ROS scavenger, Mito-TEMPO, was found to reduce the contents of totol
   ROS, superoxide anion and MDA in RVECs (Fig. [107]3i-k), suggesting the
   predominant role of mitochondria-derived ROS in oxidative stress
   induced by high glucose. Furthermore, Mito-TEMPO enhanced the integrity
   and polarity of the endothelial monolayer, evidenced by the
   FITC-dextran leakage and TER analysis (Fig. [108]3l-m). Mito-TEMPO
   increased the expressions of junctional proteins in RVECs
   (Fig. [109]3n-o and Fig. S2j-o). The outcomes of in vivo experiments
   were also in agreement with the previously mentioned in vitro results,
   indicating the ameliorative effects of Mito-TEMPO on the retinal
   endothelial injury of DR (Fig. [110]4k-o and Fig. S3f-l). The oxidative
   stress, primarily caused by mitochondrial ROS, led to phenotypic and
   functional transitions. The above results obviously illustrated that
   high glucose drove mitochondrial oxidative stress-related endothelial
   dysfunction through the suppression of FAO.

Fig. 4.

   [111]Fig. 4
   [112]Open in a new tab

   Ginsenoside Rd alleviated impaired retinal endothelial barrier function
   via acceleration of FAO in vivo and in vitro. a-c Quantitative analysis
   of DCFH-DA, DHE and MDA levels in RVECs treated with vehicle, high
   glucose, Rd (30 µM) and Eto (40 µM) (n = 3). d Mitochondrial ROS was
   assessed by MitoSOX staining. Scale bars: 50 μm (n = 3). e-f Integrity
   and polarity of ECs monolayer (n = 3). g Immunoblot analysis of
   VE-cadherin, ZO-1, Occludin and Cx43 expressions in RVECs (n = 3). h-j
   Immunofluorescent staining of VE-cadherin and ZO-1 in RVECs. Scale
   bars: 40 μm (n = 3). k Retinal section stained with H&E. Scale bars:
   25 μm (n = 5). l MDA concentration of retinal sections (n = 5). m
   Retinal trypsin digestion and staining. Acellular capillaries were
   denoted by red arrows in representative graphs. Scale bars: 50 μm
   (n = 5). n Retinal vascular leakage of Evans blue dye. Scale bars:
   50 μm (n = 5). o Proteins levels of VE-cadherin, ZO-1, Occludin and
   Cx43 in retina were determined by Western blot (n = 6). ##p < 0.01 vs.
   the control, *p < 0.05, **p < 0.01 vs. DR group, high glucose group or
   the indicated group

   To further identify the significance of FAO in the effectiveness of Rd
   against DR, the carnitine palmitoyltransferase 1 (CPT1) inhibitor
   etomoxir (Eto) was utilized to block FAO. Rd reciprocally mitigated the
   generation of total ROS, superoxide anion, MDA, and mitochondrial ROS
   (Fig. [113]4a-d and Fig. S3a). The increased endothelial monolayer
   permeability and the loss of junctional proteins were ameliorated by Rd
   in RVECs after high glucose challenge (Fig. [114]4e-j and Fig. S3b-e).
   The efficacy of Rd on mitochondrial oxidative stress-related
   endothelial dysfunction was largely abrogated when Eto was concurrently
   administered. Concordantly, Rd counteracted the retinal
   histopathological alterations and reinforced the expressions of
   VE-cadherin, ZO-1, Occludin and Cx43 in the retina of DR group. The
   effects of Rd were partially abolished by Eto in vivo (Fig. [115]4k-o
   and Fig. S3f-l). In accordance with these results, Rd alleviated
   mitochondrial oxidative stress through a FAO-associated mechanism,
   thereby sustaining retinal endothelial barrier function.

AMPK was activated by ginsenoside Rd through GPR30-cAMP-PKA-LKB1 pathway

   Based on our previous study, Rd activated AMPK, a crucial regulatory
   protein for FAO, in the retinal ECs [[116]21]. As a necessary extension
   of the initial study, it was postulated that AMPK might serve as the
   pivotal upstream regulatory molecule responsible for the vascular
   barrier-protective efficacy of Rd. Therefore, we sought to investigate
   the molecular mechanism and potential target through which Rd could
   activate AMPK in RVECs. Considering the structural resemblance of Rd to
   steroid hormones (Fig. S4l), it was plausible that Rd directly bound to
   steroid receptors, which could in turn phosphorylate AMPK at Thr172.

   Initially, a range of steroid hormones were screened to determine the
   capacity for AMPK activation in RVECs. β-Estradiol, progesterone and
   cortisol exhibited an augmentation in AMPK phosphorylation, whereas
   dihydrotestosterone executed a negligible impact (Fig. [117]5a). RU486
   (progesterone and glucocorticoid receptor antagonist) was found to have
   no effect on the ability of Rd to activate AMPK (Fig. [118]5b).
   Unexpectedly, the estrogen receptor (ER) antagonist ICI182780
   intensified Rd’s facilitation of phosphorylated AMPK (Fig. [119]5b). ER
   comprises both the nuclear receptor family (ERα and ERβ) and membrane
   receptor family (GPR30). Previous studies indicated that ICI182780 was
   commonly used as an antagonist for ERα and ERβ, but it could also act
   as an agonist for GPR30 [[120]26]. Despite the fact that G1 (a
   GPR30-specific agonist) promoted AMPK activation, the GPR30 antagonist
   G15 (1 µM) was not available to attenuate the activity of Rd
   (Fig. [121]5c). G15 was reported as a partial agonist for ERα at high
   concentrations, and β-estradiol increased p-AMPK levels through its
   binding to ERα [[122]26]. As anticipated, there was a marked inhibition
   of Rd-induced AMPK activation when the concentration of G15 was
   decreased to 10 nM (Fig. [123]5d and Fig. S4a). In addition, the
   ERα-specific antagonist MPP or the ERβ-specific antagonist PHTPP failed
   to diminish the effects of Rd (Fig. [124]5d and Fig. S4a). The
   effective concentrations of a tested drug might result in nonclassic or
   biphasic dose responses in pharmacological reactions related to steroid
   receptors, involving receptor selectivity, cross-reactivity, and
   multiple signaling pathways. These results proved that AMPK
   phosphorylation induced by Rd was dependent on the activation of GPR30.

Fig. 5.

   [125]Fig. 5
   [126]Open in a new tab

   AMPK was activated by ginsenoside Rd through GPR30-cAMP-PKA-LKB1
   pathway. a Immunoblot analysis of p-AMPK and AMPK in RVECs treated with
   vehicle, 10 nM β-estradiol (E2), 10 nM dihydrotestosterone (DHT), 100
   nM progesterone (PRO) and 100 nM cortisol (hydrocortisone, HC) (n = 3).
   b Immunoblot analysis of p-AMPK and AMPK in RVECs treated with vehicle,
   30 µM Rd, 1 µM ICI182780 and 1 µM RU486 (n = 3). c Protein level of
   p-AMPK and AMPK were determined by Western blot in RVECs treated with
   vehicle, 30 µM Rd, 1 µM G15 and 1 µM G1 (n = 3). d Representative
   Western blot of p-AMPK, AMPK, p-LKB1, LKB1, p-PKA and PKA in RVECs
   treated with vehicle, 30 µM Rd, 1 µM MPP, 1 µM PHTPP, 10 nM G15 and 1
   µM G1 (n = 3). e-f cAMP levels were analyzed in RVECs (n = 3). g
   Western blot analyzed the protein level of p-AMPK, AMPK, p-LKB1 and
   LKB1 in RVECs treated with vehicle, 30 µM Rd, 100 µM SQ22536, 10 µM
   Rp-cAMPs, 10 µM H-89 and 10 µM forskolin (n = 3). h Immunoblot analysis
   of p-AMPK and AMPK in RVECs treated with vehicle, 30 µM Rd, NC siRNA
   and LKB1 siRNA (n = 3). i Representative Western blot of p-AMPK, AMPK,
   p-LKB1, LKB1, p-PKA and PKA in RVECs treated with vehicle, 30 µM Rd, NC
   siRNA and GPR30 siRNA (n = 3). j-l Protein level of p-AMPK and AMPK
   were determined by Western blot in RVECs, MCF-7, and HEK293 (n = 3).
   #p < 0.05, ##p < 0.01 vs. the control, *p < 0.05, **p < 0.01 vs. Rd
   group or the indicated group

   The activation of GPR30 exerted biological effects through various
   kinase-signaling cascades. Both Rd and G1 elevated intracellular cAMP
   levels in RVECs, whereas the cAMP accumulation resulting from Rd was
   regressed by G15 or GPR30 knockdown (Fig. [127]5e-f). Concomitantly, Rd
   and G1 induced notable phosphorylation of PKA, which was then believed
   to activate LKB1 and hence stimulate AMPK activity (Fig. [128]5d and
   Fig. S4b-c). Rd enhanced the activation of LKB1 and AMPK, but this
   effect was abrogated when RVECs were pretreated with adenylate cyclase
   inhibitor (SQ22536), cAMP competitive antagonist (Rp-cAMPs) or PKA
   inhibitor (H-89). As a positive control, the adenylate cyclase
   activator forskolin elicited a similar effect compared to Rd
   (Fig. [129]5g and Fig. S4d-e). Furthermore, knockdown of LKB1 decreased
   the action of Rd on phosphorylation of AMPK (Fig. [130]5h). Rd improved
   the levels of p-PKA, p-LKB1, and p-AMPK in RVECs, regardless of the
   presence of MPP or PHTPP. By contrast, the pharmacological inhibition
   or genetic suppression of GPR30 reversed these alternations caused by
   Rd (Fig. [131]5d, i, Fig. S4a-c and f-h). To further substantiate the
   essential of GPR30 in the AMPK pathway activated by Rd, we examined the
   time-dependent induction of AMPK activation in RVECs (low GPR30
   expression), MCF-7 (high GPR30 expression), and HEK293 (no GPR30
   expression) (Fig. S4i). The increase in p-AMPK generated by Rd was
   directly proportional to the abundance of GPR30 in the tested cell
   lines (Fig. [132]5j-l). Validations of siRNA knockdown for GPR30 and
   LKB1 were shown in Fig. S4j-k. These aforementioned pharmacological
   studies unequivocally suggested that Rd activated the signaling pathway
   composed of GPR30-cAMP-PKA-LKB1-AMPK.

GPR30 was identified as a target protein binding with ginsenoside Rd

   Because Rd had no effect on the endogenous expression of GPR30 in RVECs
   (Fig. S5a), the investigation aimed to determine the binding
   characteristics of Rd and GPR30 using the thermodynamic and kinetic
   experiments. Upon binding to small molecule, conformational changes of
   the target protein occurred, leading to the alterations in protein
   stability. Firstly, the incubation with Rd stabilized GPR30 in
   heat-denatured intact RVECs and cell lysates for the cellular thermal
   shift assay (CETSA) (Fig. [133]6a-b). On the drug affinity responsive
   target stability assay (DARTS), the stability of GPR30 against
   enzymatic hydrolysis in RVECs lysates was improved by Rd in a
   concentration-dependent manner (Fig. [134]6c). Another indirect method
   for assessing binding was affinity ultrafiltration mass spectrometry.
   This technique involved separating ligand-protein complexes from free
   components using ultrafiltration, and then analyzing the released
   ligands (Rd) using LC-MS. The resultant increase in the MS response
   indirectly confirmed the binding between Rd and GPR30 (Fig. [135]6d and
   Fig. S5b). On the microscale thermophoresis assay (MST), the recorded
   equilibrium dissociation constant was rated at approximately 5.46 µM,
   indicative of Rd’s feasible binding affinity with GRP30, and Rd failed
   to combine with negative control with EGFP (Fig. [136]6e). The results
   of molecular docking predicted that Rd and the amino acid residues of
   Glu 115, Asp 120, Asp 210, Glu 275, and Asn 310 of GPR30 could form
   hydrogen bonds, in which these predicted amino acid residues were
   located in the native ligand binding pocket of GPR30 (Fig. [137]6f). To
   further validate these sites, the site-directed mutations were
   conducted and the overexpression of GPR30 was verified (Fig. [138]6g).
   The results of DARTS assay demonstrated that the mutating sites of Asp
   120, Glu 275, and Asn 310 by alanine did not affect the interaction,
   but the mutating sites of Glu 115 and Asp 210 by alanine diminished the
   binding between Rd and GPR30 (Fig. [139]6h). Meanwhile, the mutations
   of Glu 115 and Asp 210 compromised the AMPK activation induced by Rd
   (Fig. [140]6i), and Glu 115 and Asp 210 of GPR30 are conserved in
   species ranging from Homo sapiens to various mammals (Fig. S5c).
   Therefore, the key residues of Glu 115 and Asp 210 were responsible for
   the interaction of Rd and GPR30. Collectively, the accumulated data
   definitively situated GPR30 as a direct target of Rd.

Fig. 6.

   [141]Fig. 6
   [142]Open in a new tab

   GPR30 was identified as a target protein binding with ginsenoside Rd.
   a-b Thermal stability of GPR30 in intact RVECs and cell lysates treated
   with vehicle and 30 µM Rd was measured by the CETSA (n = 3). c
   Stability of GPR30 against enzymatic hydrolysis in RVECs lysates
   treated with vehicle and Rd was measured by the DARTS (n = 3). d
   Affinity between GRP30 and Rd was tested by HPLC-QQQ-MS/MS (n = 3). e
   Binding affinity of Rd with GPR30 was determined by the MST. An empty
   vector that only expresses EGFP was included as negative controls
   (n = 3). f Docking of protein GPR30 and Rd. g Overexpression of GPR30
   was verified by Western blot (n = 3). h Immunoblotting analysis of the
   binding between 30 µM Rd and GPR30 with different mutations by DARTS in
   HEK293T (n = 3). i Immunoblot analysis of p-AMPK and AMPK (n = 3).
   #p < 0.05, ##p < 0.01 vs. the control, **p < 0.01 vs. Pronase group or
   Rd-treated group

Ginsenoside Rd restored high glucose-induced FAO defect and endothelial
barrier dysfunction via GPR30-AMPK pathway

   After unearthing the target and upstream regulatory mechanism of Rd,
   the subsequent objective was to elucidate the potential impact of the
   GPR30-AMPK cascade on downstream phenotypic and functional outcomes. Rd
   and G1 potentiated the phosphorylation of AMPK and ACC, the expressions
   of PGC-1α and PPARα, and the mRNA levels of CPT1A, CPT2, PGC-1α and
   PPARα in RVECs treated with high glucose (Fig. [143]7a-b and Fig.
   S6a-d). ACC converted acetyl-CoA to malonyl-CoA, which served as a
   physiological inhibitor of CPT1A, the decisive rate-limiting enzyme in
   FAO. AMPK-triggered ACC phosphorylation could inhibit the catalytic
   activity [[144]27]. Additionally, AMPK modulated the transcription of
   metabolic enzyme in FAO through the transcription factors, PGC-1α and
   PPARα. Rd and G1 magnified maximal oxygen consumption rate (OCR) change
   resulting from exogenous FAO and increased the mitochondrial membrane
   potential (Fig. [145]7c-d and Fig. S6e-f). However, the stimulation of
   Rd on fatty acid metabolism pathway and flux of FAO were diminished by
   the GPR30 antagonist G15 and AMPK inhibitor Compound C (C-C)
   (Fig. [146]7a-d and Fig. S6a-f). These results strongly provided
   evidence for the accelerative ability of Rd on high glucose-impaired
   FAO via the GPR30-AMPK-ACC-CPT1A/PGC-1α-PPARα axis. Consistently, Rd
   and G1 suppressed high glucose-driven mitochondrial oxidative stress
   and dissolution of inter-endothelial junctions. Conversely, the
   pharmacological blockage of GPR30 and AMPK using G15 and C-C elicited
   weakening effects on pharmacodynamics of Rd, including
   anti-mitochondrial oxidative stress (Fig. [147]7e-h and Fig. S6g),
   increased endothelial integrity (Fig. [148]7i-j) and emphasis on
   barrier proteins (Fig. [149]7k-l and Fig. S6h-m). Together, these
   results corroborated the function for GPR30-AMPK pathway in the
   efficacy of Rd in mitigating FAO deficiency and endothelial barrier
   dysfunction.

Fig. 7.

   [150]Fig. 7
   [151]Open in a new tab

   Ginsenoside Rd restored high glucose-induced FAO defect and endothelial
   barrier dysfunction via GPR30-AMPK pathway. a Immunoblot analysis of
   p-AMPK, AMPK, p-ACC, ACC, PGC-1α and PPARα expressions in RVECs treated
   with vehicle, high glucose, 30 µM Rd, 10 nM G15, 1 µM C-C and 1 µM G1
   (n = 3). b mRNA relative expressions of CPT1A, CPT2, PGC-1α and PPARα
   were measured by qRT-PCR (n = 3). c Bar chart showing maximal OCR
   alterations due to exogenous FAO (n = 3). d TMRE fluorescence images of
   RVECs treated with vehicle, high glucose, Rd, G15, C-C and G1. Scale
   bars: 50 μm (n = 3). e-g Quantitative analysis of DCFH-DA, DHE, and MDA
   levels in RVECs (n = 3). h Mitochondrial ROS levels were evaluated in
   RVECs using MitoSOX. Scale bars: 50 μm (n = 3). i-j Integrity and
   polarity of ECs monolayer (n = 3). k Immunoblot analysis of
   VE-cadherin, ZO-1, Occludin and Cx43 expressions in RVECs (n = 3). l
   Immunofluorescent staining of VE-cadherin and ZO-1 in RVECs. Scale
   bars: 40 μm (n = 3). ##p < 0.01 vs. the control, *p < 0.05, **p < 0.01
   vs. high glucose group or the indicated group

AMPK supported IDH2-catalyzed mitochondrial NADPH regeneration to maintain
redox homeostasis in a FAO-dependent manner

   The BRB injury was primarily contributed by the endothelial
   mitochondrial ROS, with FAO being the main source of endothelial
   mitochondrial respiration rather than glycolysis [[152]28]. It was
   speculated that compromised FAO might be the reason for endothelial
   mitochondrial oxidative stress after exposure to high glucose.
   Mitochondrial ROS levels were meticulously monitored by an intricate
   and orchestrated system of antioxidant defense mechanisms [[153]29].
   Therein, manganese superoxide dismutase (Mn-SOD), which was exclusively
   localized in mitochondrial matrix, catalyzed the conversion of
   superoxide into hydrogen peroxide (H[2]O[2]). Nicotinamide adenine
   dinucleotide phosphate (NADPH) was physiologically coupled with
   glutathione (GSH) pool and strictly scavenged peroxide. Rd preserved
   the contents of NADPH and GSH, as well as the activity of Mn-SOD in
   RVECs after high glucose challenge, while those alternations were
   abolished by AMPK inhibitor C-C (Fig. [154]8a-c). Due to the inability
   of NADPH to penetrate the mitochondrial membrane, the mRNA levels of
   the key enzymes responsible for mitochondrial NADPH regeneration were
   measured. There were no significant responses observed in MTHFD1, ME2,
   ME3, GLUD and NNT, but high glucose reduced the transcripts of MTHFD2
   and IDH2 in RVECs (Fig. [155]8d). Rd was ineffective to MTHFD2 mRNA
   levels, but revitalized the expression of IDH2 in an AMPK-dependent
   manner (Fig. S7a-b). IDH2 was crucial for Rd to sustain the
   mitochondrial NADPH pool.

Fig. 8.

   [156]Fig. 8
   [157]Open in a new tab

   AMPK supported IDH2-catalyzed mitochondrial NADPH regeneration to
   maintain redox homeostasis in a FAO-dependent manner. a-b Mitochondrial
   NADPH and GSH abundance in vehicle, high glucose, 30 µM Rd and 1 µM
   C-C-treated RVECs (n = 3). c Enzymatic activity of Mn-SOD in RVECs
   stimulated with vehicle, high glucose, Rd and C-C (n = 3). d Screening
   for critical enzymes of mitochondrial NADPH regeneration by qPCR
   (n = 3). e Immunoblotting analysis of nuclear and cytosolic FOXO1 in
   RVECs. Lamin B was used as the nuclear loading control and GAPDH as the
   cytosolic loading control (n = 3). f Analysis of FOXO1 on IDH2 promoter
   by ChIP-qPCR (n = 3). g Immunoprecipitation of IDH2 was performed to
   detect the combination of IDH2 and SIRT3 (n = 3). h Analysis of PGC-1α
   on SIRT3 promoter by ChIP-qPCR (n = 3). i Enzymatic activity of mIDH in
   RVECs stimulated with vehicle, high glucose, 30 µM Rd, 5 µM 3-TYP, 5 µM
   AS1842856 and 1 µM C-C (n = 3). j-l Quantitative analysis of DCFH-DA,
   DHE and MDA levels in RVECs (n = 3). m-n Mitochondrial NADPH and GSH
   contents in RVECs (n = 3). o Mitochondrial ROS levels were evaluated in
   RVECs. Scale bars: 50 μm (n = 3). p Mn-SOD activity was determined in
   RVECs stimulated with vehicle, high glucose, 30 µM Rd, 5 µM 3-TYP and 1
   µM C-A (n = 3). q Content of mitochondrial NADPH in RVECs with control
   transfection or IDH2 knockdown and treated with vehicle, high glucose
   and 30 µM Rd (n = 3). r-s Integrity and polarity of ECs monolayer with
   IDH2 knockdown (n = 3). t Immunoblot analysis of VE-cadherin, ZO-1,
   Occludin and Cx43 expressions in RVECs with IDH2 knockdown (n = 3). u
   Mitochondrial NADPH regeneration increased in RVECs transfected with
   IDH2 plasmid (n = 3). v-w Detection of FITC-dextran leakage and TER of
   RVECs monolayer with IDH2 plasmid (n = 3). x VE-cadherin, ZO-1,
   Occludin and Cx43 expressions were detected by Western blot in RVECs
   with IDH2 plasmid (n = 3). y Mechanism schematic of the AMPK-driven
   redox homeostasis in RVECs. ##p < 0.01 vs. the control, *p < 0.05,
   **p < 0.01 vs. high glucose group or the indicated group

   In cancer cells, IDH2 levels increased adaptively upon metabolic stress
   via the AMPK-FOXO1 pathway [[158]30]. Rd led to drastic AMPK
   activation, which in turn triggered the nuclear translocation of FOXO1
   (Fig. [159]8e and Fig. S7c). ChIP assay suggested that FOXO1 antibody
   could expectedly enrich the DNA of IDH2 promoter region, and Rd
   strengthened the binding of FOXO1 to IDH2 promoter, which was disrupted
   by C-C (Fig. [160]8f). Beyond the transcriptional control of gene
   expression, the enzymatic activity of IDH2 could also be regulated by
   post-translational modifications. High glucose decreased the expression
   of mitochondrial deacetylase SIRT3 and increased the levels of
   acetylated inactive form of IDH2. And Rd reduced the levels of
   acetylated IDH2 by enhancing the interaction with SIRT3 and IDH2
   (Fig. [161]8g and Fig. S7d-f). As reported in the literature, PGC-1α
   polymerized with ERRα enriched the SIRT3 promoter region [[162]31].
   After demonstrating the transcriptional augmentation of SIRT3 by Rd,
   C-C and Compound A (C-A) were employed to inhibit AMPK and ERRα
   respectively, and the treatment blocked the enhancing effects of Rd
   (Fig. [163]8h). Similarly, Rd reversed the activity of mitochondrial
   IDH in high glucose-stimulated RVECs, while this change was suppressed
   by SIRT3 inhibitor 3-TYP, FOXO1 inhibitor AS1842856 and C-C
   (Fig. [164]8i). Concordantly, the regulation of Rd on total ROS
   (Fig. [165]8j), superoxide anion (Fig. [166]8k), MDA (Fig. [167]8l),
   mitochondrial NADPH (Fig. [168]8m), mitochondrial GSH (Fig. [169]8n)
   and mitochondrial ROS (Fig. [170]8o and Fig. S7g) were lost when
   co-incubation with 3-TYP or AS1842856. Similar to IDH2, Mn-SOD could be
   deacetylated and potentiated by SIRT3. The activation of Mn-SOD was
   observed following the treatment with Rd, and further hampered by 3-TYP
   or C-A (Fig. [171]8p). The TCA cycle metabolite isocitrate served as
   the substrate for the oxidative decarboxylation reaction catalyzed by
   IDH2 to regenerate NADPH from NADP^+. Therefore, TCA cycle influenced
   the rate of NADP^+ reduction mediated by IDH2, and acetyl-CoA (Ac-CoA)
   from FAO could satisfy the demands of TCA pathway. As shown in the
   ELISA assay and metabolite determination by LC-MS, Rd accelerated
   AMPK-related FAO, which was accompanied by the elevated concentrations
   of TCA cycle intermediates (Ac-CoA, citric acid, cis-aconitic acid,
   α-ketoglutaric acid, succinic acid, fumaric acid, L-malic acid, and
   oxaloacetic acid) (Fig. S7h). Moreover, IDH2 knockdown partly weakened
   the ability of Rd to ameliorate mitochondrial oxidative stress
   (Fig. [172]8q and Fig. S8a-f), endothelial barrier compromise
   (Fig. [173]8r-s), and junctional proteins loss (Fig. [174]8t and Fig.
   S8g-m). The results of siRNA-mediated knockdown of IDH2 was seen in
   Fig. S8n. The IDH2 overexpression recapitulated the observed phenotype
   in Rd, encompassing mitochondrial redox homeostasis (Fig. [175]8u and
   Fig. S9a-f) and vascular barrier enhancement (Fig. [176]8v-x and Fig.
   S9g-m). The overexpression of IDH2 was validated by Western blot (Fig.
   S9n).

   As articulated in Fig. [177]8y, AMPK activated by Rd supported
   mitochondrial NADPH regeneration to maintain redox homeostasis and
   repair BRB breakdown by the FOXO1-dependent transcription of IDH2,
   SIRT3-dependent deacetylation of IDH2, and FAO-dependent growth of TCA
   cycle flux.

Ginsenoside Rd ameliorated lipid metabolism disorders and vascular barrier
damage by targeting GPR30 in the retina of diabetic mice

   To interrogate whether the activation of GPR30 underlay the therapeutic
   efficacy of Rd on BRB stabilization in vivo, a model of retinal
   GPR30-knockdown mice was created by intravitreal injection of
   adeno-associated virus (AAV8-shGPR30). The successful inhibition of
   GPR30 in the retina was confirmed through the Western blot
   (Supplementary file [178]1: Fig. S10a). Pathological staining revealed
   that Rd was effective in mitigating disorganization of retinal anatomy
   in response to hyperglycemia, and elicited minimal effects on
   AAV8-shGPR30 mice. In contrast, retinal structural features remained
   stable between Rd and AAV8-NC shRNA co-treatment groups (Fig. [179]9a
   and Supplementary file [180]1: Fig. S10b). Under prolonged
   hyperglycemic conditions, Rd triggered a decrease in capillary
   degeneration coupled with reduced extravasation of Evans blue, and a
   decline in the expression of junctional proteins (VE-cadherin, ZO-1,
   Occludin and Cx43) in the retina. However, these effects were notably
   attenuated in mice treated with AAV8-shGPR30, compared with those
   treated with AAV8-NC shRNA (Fig. [181]9b-c, h, Supplementary file
   [182]1: Fig. S10c-d and l-o). While observing retinal ultrastructure
   using transmission electron microscopy (TEM), control mice exhibited
   retinal capillaries with a normal and uninterrupted tight junctions
   (yellow arrow). Hyperglycemia disrupted the electron-dense region
   indicative of the tight junctions, and punctuated endothelial barrier
   (red arrow) (Fig. [183]9d). After diabetic induction, retinal ganglion
   cells displayed unclear and incomplete nuclear membranes (red star),
   chromatin dissolution (green star), and neuropil edema (blue star)
   (Fig. [184]9e). Additional anomalies including nuclear pyknosis of the
   bipolar cell (green arrow) and boosted cell cavity voids (red arrow)
   were identified in the retinal inner nuclear layer of the model mice
   (Fig. [185]9f). The outer segments of diabetic individuals revealed a
   reduction in photoreceptor extracellular membrane disks length,
   alongside a disorganized structure characterized by swelling,
   degeneration, and vacuoles (Fig. [186]9g). The above detrimental
   changes in the capillary, ganglion, inner nuclear layer and outer
   segments were uniformly prevented by the application of Rd in the
   retina of DR mice. Conversely, the advancement of such therapeutic
   alterations of Rd was absent during concurrent administration of Rd and
   AAV8-shGPR30 in the tested retinas (Fig. [187]9d-g).

Fig. 9.

   [188]Fig. 9
   [189]Open in a new tab

   Ginsenoside Rd ameliorated lipid metabolism disorders and vascular
   barrier damage by targeting GPR30 in the retina of diabetic mice. a
   Representative images of H&E staining. Scale bars: 25 μm (n = 5). b
   Representative images of PAS staining in retinal vascular network.
   Acellular capillaries were denoted by red arrows. Scale bars: 50 μm
   (n = 6). c Representative images of retinal Evans blue leakage in all
   experiment groups. Scale bars: 50 μm (n = 6). d Electron micrograph of
   retinal capillary. Scale bars: 1 μm (n = 6). e Electron micrograph of
   retinal ganglion cells. Scale bars: 5 μm (n = 6). f Electron micrograph
   of bipolar cells in inner nuclear layer. Scale bars: 5 μm (n = 6). g
   Electron micrograph of photoreceptor extracellular membrane disk. Scale
   bars: 5 μm (n = 6). h Immunoblot analysis of p-AMPK, AMPK, p-ACC, ACC,
   CPT1A, PPARα, PGC-1α, IDH2, SIRT3, VE-cadherin, ZO-1, Occludin and Cx43
   expressions in retina (n = 6). i-k MDA, NADPH and GSH concentrations in
   retinal sections (n = 6). ##p < 0.01 vs. the control, *p < 0.05,
   **p < 0.01 vs. DR group or the indicated group

   For in vivo mechanism validation, Rd expectedly phosphorylated AMPK and
   ACC, and preserved the levels of CPT1A, PPARα and PGC-1α (Fig. [190]9h
   and Supplementary file [191]1: Fig. S10e-i). Meanwhile, Rd reversed the
   elevated contents of MDA (Fig. [192]9i), and reduced levels of NADPH
   (Fig. [193]9j) and GSH (Fig. [194]9k) in retina. The downregulation of
   IDH2 and SIRT3 in response to hyperglycemia were restored by Rd
   (Fig. [195]9h and Supplementary file [196]1: Fig. S10j-k).
   Concordantly, the beneficial effects of Rd on FAO defect and oxidative
   stress were approximately abolished by GPR30 knockdown (Fig. [197]9h-k
   and Supplementary file [198]1: Fig. S10e-k). Similar to Rd, the GPR30
   agonist G1 exerted an increased retinal resistance to hyperglycemia
   stimulation and repaired the pathological changes in DR, providing
   additional evidence to support our hypothesis (Fig. [199]9a-k and
   Supplementary file [200]1: Fig. S10b-o). Taken together, these results
   underscored the critical role of Rd in ameliorating abnormal redox
   status and subsequent BRB collapse by intervening GPR30-driven FAO.

Discussion

   DR can be managed through various treatment modalities, including
   intraocular injection of glucocorticoids, vitrectomy, retinal laser
   photocoagulation and intravitreal injections of anti-VEGF agents. The
   current first-line therapies are primarily applicable for advanced DR,
   but which are associated with numerous notable limitations [[201]32].
   Among them, targeting VEGF treatment has demonstrated encouraging
   clinical advantages by mitigating the risk of neovascularization and
   retinal detachment. The blockade of single target may appear appealing,
   but could trigger compensatory regulation in a complex disease of
   multifactorial basis, which is the main reason for resistance or
   nonresponse [[202]33–[203]34]. Given that a significant portion of
   detrimental alterations in advanced DR are irreversible, early
   systematic detection and timely prevention are crucial for preventing
   vision impairment in DR patients. The sole recognized clinical
   intervention for early DR is the control of hyperglycemia,
   hyperlipidemia, and hypertension, whereas irregular dietary management
   and metabolic memory phenomenon attenuate the clinical benefit for
   patients [[204]1]. The effectiveness of simple metabolic control is
   limited when the mechanism underlying the metabolic disorders-related
   vascular pathology of DR remains unclear. This study investigates the
   potential therapeutic effects of Rd on vascular early injury in a mouse
   model of DR and high glucose-induced RVECs. Mechanistically, Rd
   specifically enhances AMPK-induced FAO by agonizing GPR30, and the
   metabolic reprogramming regulated by Rd effectively reverses the
   increased endothelial permeability and BRB breakdown. The findings not
   only bolster the robust theoretical foundation for advancing innovative
   therapeutic avenues targeting DR, but also pave the way for future
   clinical investigations.

   The common pathogenesis in blinding retinal diseases is retinal
   vascular leakage resulting from the loss of BRB integrity [[205]34]. To
   maintain physiological homeostasis in the retinal microenvironment, BRB
   selectively regulates material exchange between the retina and systemic
   circulation by intercellular physical barrier formed by ECs and other
   cells involved in BRB [[206]35]. Prior to the manifestation of clinical
   indications of DR, long-term hyperglycemia could alter the junctional
   stability resulting in vascular leakage. At this time, pre-clinical
   diabetic retinopathy (PCDR) is characterized by mild and diffuse BRB
   injury. As PCDR progresses to NPDR, leaky plasma constituents
   accumulate in the retina and subsequently form hard exudates. The
   localized profuse leakage is accompanied by microvascular abnormalities
   and local tissue ischaemia, which induces an increase in VEGF and
   further exacerbates barrier instability, promoting neovascularization
   in PDR. The progressive elevation of endothelial permeability
   ultimately leads to diabetic macular edema and endangers the vision in
   the absence of prompt intervention [[207]23, [208]36]. While the
   barrier structural and functional disruption is paramount in the
   initial stages of DR, the precise molecular events in BRB collapse are
   poorly understood, and various, often conflicting, viewpoints exist.
   Consequently, there has been little advancement in the approaches to
   vascular barrier repair, despite the clinical trials of small molecules
   like CaD, aspirin, and antihistamines [[209]37]. Recent research has
   shown that the functional roles of individual ginsenosides in vascular
   remodeling may even be contradictory, depending on their chemical
   structures and distinct pathophysiological environments [[210]38]. In
   the present study, Rd is firstly identified as a novel endothelial
   barrier remediation, effectively alleviating the acute BRB breakdown in
   a VEGF-induced mouse model and the chronic vascular leakage in a
   STZ-induced DR model. In vitro experiments further reveal that Rd
   rectifies vasopermeability of RVECs in response to high glucose through
   the reconstruction of adhering, tight and gap junctions.

   For the complex metabolic diseases, deciphering the deep causes of
   vascular damage should start with multiple intertwined metabolic
   disorder. There are systemic and local dysregulation of lipid
   metabolism in almost all diabetic microvascular complications
   [[211]39]. In the 1950s, the identification of lipid compositions in
   hard exudates opened the prelude to clinical and basic research on
   lipid metabolism in DR [[212]10]. A range of cross-sectional and
   longitudinal studies have consistently shown an association between the
   development and progression of DR and traditional serum lipid profiles
   [[213]10]. Cardiovascular-system-focused clinical trials also
   investigate the potential benefit of numerous lipid-lowering
   medications in safeguarding against DR [[214]9–[215]10]. Fenofibrate
   improves retinal pathology by exerting anti-inflammatory and
   anti-apoptotic effects on ECs, pericytes and retinal pigment epithelial
   cells [[216]9, [217]40]. Lipid metabolic disorder that favors retinal
   neurovascular injury is a prominent hallmark of DR. But, little efforts
   have been made to explore potential targetable nodes responsible for
   the lipid metabolic adaptation of retina under pathological stressors.
   Our observations reveal for the first time that FAO is a essential
   metabolic process required for maintenance of the endothelial barrier
   function in retina. Concurrently, the retinal vasoprotective activity
   of Rd is at least partially contingent upon revitalization for FAO. The
   findings align with several notable observations that yield comparable
   outcomes. Besides that diabetic heart exhibits a decreased
   mitochondrial β-oxidation, deficient FAO induces renal dysfunction in
   diabetic nephropathy [[218]41–[219]42]. Acute CPT1A inhibition disrupts
   vascular integrity by promoting calcium influx [[220]43]. FAO sustains
   the hexosamine biosynthesis pathway to support protein glycosylation,
   and then relieves the defects in airway epithelial barrier integrity
   [[221]44]. Therefore, FAO emerges as a novel druggable checkpoint in
   governing retinal endothelial permeability. Further understanding of
   the metabolic plasticity in DR contributes to the advancement of
   pharmacological strategies capable of covering complex pathological
   mechanisms and avoiding compensatory regulation.

   Although much is known about the role of metabolism in endothelial
   function, there remains a substantial gap in our knowledge on how
   metabolic adaptation compromises BRB under hyperglycemic circumstances.
   Multiple pathways are implicated in the metabolism-driven alterations
   of cellular function. Firstly, the coupling metabolic network controls
   biomass flux and metabolic flow. The Warburg effect supports
   uncontrolled proliferation of cancer cells, and FAO is crucial for de
   novo nucleotide synthesis in ECs [[222]25]. Secondly, certain metabolic
   intermediates have the ability to function as signaling molecules.
   Lactate, an inevitable byproduct of glycolysis, activates GPR81.
   Additionally, succinate, a TCA cycle organic acid, serves as the
   natural ligand for GPR91 [[223]45–[224]46]. Third, metabolic
   intermediates participate in post-translational modification. The
   ketolysis strengthens the TCA cycle-dependent Ac-CoA production and
   subsequently histone acetylation, thereby reinforcing CD8^+ T cell
   effector responses [[225]47]. Lastly, metabolic enzymes execute
   nonmetabolic functions. In cancer, PKM2, glycolytic enzymes, can
   phosphorylate STAT3 and interact with HIF-1α to promote target genes
   transcription [[226]48]. Our data indicates that the activated AMPK
   accelerates FAO and fuel the TCA cycle with Ac-CoA. On the basis of the
   increased flux in TCA cycle, AMPK facilitates FOXO1-dependent IDH2
   expression, and revitalizes the enzymatic activity of IDH2 by the
   PGC-1α/SIRT3 pathway. This collaborative action sustains NADPH
   regeneration, effectively reversing the damage caused by mitochondrial
   ROS on retinal endothelial barriers.

   Although NADPH does not exhibit permeability across various cellular
   compartments, recent evidence has indicated intercommunication between
   the mitochondrial and cytoplasmic NADPH pools. In ECs treated with
   TGF-β, excessive glycolysis diminishes cytoplasmic NADPH production,
   and triggers a compensatory efflux of mitochondrial NADPH through
   isocitrate/α-ketoglutarate shuttle [[227]49]. Our results cannot
   exclude the involvement of NADPH crosstalk between the cytosol and
   mitochondria in DR, and this aspect deserves clarification in following
   investigation by isotopic metabolic labeling technique.

   Excess FAO aggravates myocardial oxidative stress, and the inhibition
   of FAO has evolved into a recognized approach to treat heart failure
   associated with diabetes and obesity [[228]50]. This paradoxical
   phenomenon of FAO in redox regulatory systems may be attributed to
   distinct cellular energy metabolic preferences. ECs predominantly