Integrating mass spectrometry imaging and data-driven segmentation for spatial metabolic mapping of diabetic eye disease

Abstract

Diabetic eye disease (DED) is a leading cause of vision impairment worldwide, yet the molecular mechanisms underlying its progression remain incompletely understood. In this study, we applied a dual-platform spatial metabolomics strategy integrating air flow-assisted desorption electrospray ionization mass spectrometry imaging (AFADESI-MSI) and matrix-assisted laser desorption ionization mass spectrometry imaging (MALDI-MSI) to characterize spatial metabolic alterations in the eyes of diabetic rats. Data-driven segmentation of retinal micro-regions using SCiLS Lab software enabled fine-scale mapping of metabolic heterogeneity. Physiological, biochemical, and histopathological analyses were combined with spatial metabolite mapping to construct a metabolic atlas and evaluate the regulatory effects of ferulic acid. We established a comprehensive spatial metabolome atlas of the rat eye, identifying 135 annotated metabolites and revealing significant region-specific metabolic heterogeneity. Unsupervised k-means clustering was further applied to the high-resolution MALDI-MSI data, successfully delineating distinct functional micro-regions of the retina solely based on endogenous metabolic profiles, demonstrating the power of data-driven tissue segmentation. In diabetic eyes, 39 metabolites were significantly dysregulated, involving amino acid, glucose, lipid, and redox metabolism. Notably, lysine, arginine, carnitine, and glutathione (GSH) were depleted, while glucose-6-phosphate (G6P), glycerol-3-phosphate (G3P), and pro-inflammatory lipids were elevated, highlighting profound metabolic reprogramming across ocular compartments. Ferulic acid treatment restored nine key metabolites, alleviated oxidative stress, normalized lipid and glucose metabolism, and improved retinal structural integrity in a dose-dependent manner. This study shows that integrating mass spectrometry imaging with data-driven tissue segmentation reveals spatial metabolic reprogramming in DED and highlights ferulic acid as a promising therapeutic candidate.

Graphical abstract

Highlights

• •AFADESI-MSI and MALDI-MSI were used for mapping rat eye metabolic atlas.
• •Regional metabolic heterogeneity in the rat eye was revealed.
• •Identified and spatially resolved 39 key metabolites in diabetic eye disease eyes.
• •Provided novel insights into diabetic eye disease pathology and therapeutic potential of ferulic acid.

1. Introduction

Diabetic eye disease (DED), encompassing a spectrum of ocular complications such as diabetic retinopathy (DR) and macular edema, represents one of the most prevalent and debilitating microvascular complications of diabetes mellitus [1,2]. As the leading cause of vision impairment and blindness in working-age populations worldwide, DED poses a growing public health challenge [3,4]. Chronic hyperglycemia and associated systemic disturbances drive complex biochemical and structural changes in ocular tissues, ultimately impairing visual function [5,6]. Despite progress in clinical management, including anti-vascular endothelial growth factor (VEGF) therapy and laser photocoagulation, effective strategies for early detection and metabolic intervention remain limited [7]. Thus, unraveling the molecular underpinnings of DED is essential for advancing both diagnostic and therapeutic approaches.

Over the past decade, metabolomics has emerged as a powerful tool for elucidating disease-associated biochemical perturbations [[8], [9], [10]]. Traditional mass spectrometry-based metabolomics has provided valuable insights into systemic metabolic dysregulation in diabetes, identifying alterations in amino acids, lipids, nucleotides, and redox metabolites [[11], [12], [13]]. However, conventional metabolomics approaches largely overlook tissue-specific heterogeneity, which is particularly critical in highly compartmentalized organs like the eye. Mass spectrometry imaging (MSI) offers a unique solution by enabling spatially resolved metabolic profiling, linking biochemical changes to distinct anatomical regions [14,15]. Recent applications of MSI have uncovered metabolic signatures associated with retinal degeneration, corneal diseases, and diabetic complications [[16], [17], [18], [19]]. Nonetheless, comprehensive spatial metabolomic mapping of the eye under diabetic conditions remains scarce, and the interplay between spatial metabolic alterations and disease progression is poorly understood. Moreover, while previous studies have reported disruptions in glucose, amino acid, and lipid metabolism in diabetic complications, the precise spatial distribution and layer-specific reprogramming of these pathways within the eye are not well defined. The retina, in particular, exhibits marked metabolic compartmentalization across its cellular layers, yet it remains unclear how diabetes perturbs these finely tuned metabolic networks in situ. Furthermore, although natural compounds with antioxidant and metabolic regulatory properties, such as ferulic acid, have demonstrated protective effects in other diabetic complications [18,20], their roles in modulating spatial metabolic dysregulation in DED are underexplored.

In this study, we employed a dual-platform MSI strategy that integrates air flow-assisted desorption electrospray ionization mass spectrometry imaging (AFADESI-MSI) and matrix-assisted laser desorption ionization mass spectrometry imaging (MALDI-MSI) to overcome the limitations of single-modality imaging. AFADESI-MSI provides broad molecular coverage and high sensitivity for small polar metabolites under ambient conditions, enabling comprehensive detection of diabetes-related metabolic disruptions. In contrast, MALDI-MSI offers superior spatial resolution and complementary detection of lipids and other higher-mass species. Combining these two ionization techniques allows us to capture a wider range of metabolite classes while simultaneously achieving more precise layer-specific localization within the retina, thereby providing a more complete spatial metabolomic landscape of DED. In addition, to overcome the limitations of manual region selection in MSI studies, we incorporated a data-driven tissue segmentation approach using unsupervised spectral clustering in SCiLS Lab. This strategy enables objective and reproducible differentiation of retinal layers based on MSI spectral features, thereby improving the accuracy of spatial metabolic mapping in DED. By integrating MSI with data-driven segmentation, our study provides a more precise characterization of retinal metabolic heterogeneity and offers a robust framework for evaluating therapeutic interventions such as ferulic acid. The overall experimental design and technical workflow are summarized in Fig. 1.

Fig. 1.

Schematic overview of the research strategy for investigating spatial metabolic alterations in diabetic eye disease (DED). A streptozotocin (STZ)-induced diabetic rat model was established to mimic DED, followed by ferulic acid intervention. Air flow-assisted desorption electrospray ionization mass spectrometry imaging (AFADESI-MSI) was employed for high-throughput metabolite discovery across whole-eye sections, while matrix-assisted laser desorption ionization mass spectrometry imaging (MALDI-MSI) provided high-spatial-resolution (20 μm) validation of key metabolites. Data-driven spatial segmentation of the retina was further applied to characterize region-specific metabolic reprogramming. MS: mass spectrometry; UMP: uridine 5′-monophosphate; FA: fatty acid; PC: phosphatidylcholine; L-FA: low-dose ferulic acid; H-FA: high-dose ferulic acid; 12-HETE: 12-hydroxyeicosapentaenoic acid.

2. Materials and methods

2.1. Chemicals and reagents

High performance liquid chromatography (HPLC)-grade methanol, acetonitrile, and formic acid were sourced from Fisher Scientific Co., Ltd. (Loughborough, Leicestershire, UK). Guaranteed reagent-grade hydrochloric acid was procured from Sinopharm Chemical Reagent Co., Ltd. (Shanghai, China) Ferulic acid, 1,5-Diaminothalene (1,5-DAN), citrate, and sodium carboxymethyl cellulose (Na-CMC) were obtained from Sigma-Aldrich Co., Ltd. (St. Louis, MO, USA). Ultrapure water was provided by Wahaha Co., Ltd. (Hangzhou, China). 2,5-Dihydroxybenzoic acid (DHB) was purchased from Aladdin Biochemical Technology Co., Ltd. (Shanghai, China).

2.2. Animal model

Six-week-old male Wistar rats (180−200 g) were supplied by Vital River Laboratory Animal Technology Co., Ltd. (Beijing, China). The animals were acclimatized for one week under a controlled environment with a 12-h light-dark cycle, a temperature of 20−26 °C, and a relative humidity of 40%−70%, during which they had ad libitum access to standard chow and water. Male rats were selected to minimize hormonal variability associated with the estrous cycle, which can influence glucose metabolism and retinal biochemical profiles.

Following acclimatization, rats were randomly divided into two cohorts: a control group (n = 6) maintained on a normal diet, and a model group (n = 18) fed a high-fat diet. After four weeks of dietary intervention, the model group received a single intraperitoneal injection of streptozotocin (STZ; 35 mg/kg body weight) dissolved in citrate buffer (0.1 mol/L, pH 4.4) at a volume of 3.5 mL/kg, while the control group received an equivalent volume of the vehicle buffer. The induction of DED was validated by the presence of sustained hyperglycemia, defined as fasting blood glucose (FBG) levels exceeding 16.7 mmol/L on three consecutive measurements, alongside the observation of bilateral corneal clouding and reddish ocular discharge.

Subsequently, the DED model rats were randomly subdivided into three treatment arms (n = 6 per group): the DED group, a low-dose ferulic acid (L-FA; 50 mg/kg, 10 mL/kg) group, and a high-dose ferulic acid (H-FA; 200 mg/kg, 10 mL/kg) group. Both the control and DED groups received daily intragastric administration of a 0.5% Na-CMC vehicle solution (10 mL/kg). Throughout the 20-week treatment period, the general health of the animals was monitored daily. Body weight, food intake, and water consumption were recorded weekly, and FBG levels were assessed biweekly. At the study's conclusion, all animals were euthanized. Blood samples and ocular tissues were promptly collected, snap-frozen in liquid nitrogen, and stored at −80 °C for subsequent analyses.

All animal procedures were conducted in accordance with the Guide for the Care and Use of Laboratory Animals and complied with relevant institutional and national guidelines for the care and use of laboratory animals. The experimental protocol was reviewed and approved by the Biological and Medical Ethics Committee of Minzu University of China (Approval No. 2017-01). All in vivo experiments were carried out at Beijing United Genius Pharmaceutical Technology Development Co., Ltd. (Beijing, China) and were additionally approved by its Animal Welfare Ethics Committee (Approval No. TS19100-YX; December 2, 2019).

2.3. Biochemical analysis and histopathological staining

Serum concentrations of glucose, total protein (TP), albumin (ALB), globulin (GLB), total bilirubin (TBIL), sodium (Na⁺), and chloride (Cl⁻) were quantified using an AU480 automatic chemistry analyzer (Beckman Coulter Inc., Brea, CA, USA). The level of glycated hemoglobin (HbA1c) was determined using a Quo-Test HbA1c analyzer (QUOTIENT Diagnostics Ltd., Walton-on-Thames, Surrey, UK).

For histopathological examination, ocular tissues were embedded in 3% Na-CMC and sectioned transversely at a thickness of 12 μm using a cryostat (Leica CM1860, Microsystems Ltd., Wetzlar, Hesse, Germany) maintained at −20 °C. The sections were thaw-mounted onto adhesive glass slides, air-dried at ambient temperature, and subjected to hematoxylin and eosin (H&E) staining. Following staining, the slides were digitized using a high-resolution slide scanner (Ningbo Jiangfeng Biomedical Information Co., Ltd., Yuyao, China) to facilitate detailed histopathological evaluation.

2.4. Sample preparation for AFADESI- and MALDI-MSI analysis

Frozen ocular tissues were cryo-sectioned to a thickness of 12 μm. These sections were then thaw-mounted onto either standard adhesive microscope slides for AFADESI-MSI or onto indium tin oxide (ITO)-coated glass slides for MALDI-MSI. All slides were stored at −80 °C until analysis. Immediately prior to mass spectrometry imaging, the tissue sections were placed in a vacuum desiccator for 30 min at room temperature to ensure complete dryness.

For MALDI-MSI, a matrix solution was applied uniformly onto the tissue surface using an automated spraying device (TM-Sprayer, HTX Technologies LLC, Carrboro, NC, USA). A mixture of 1,5-DAN hydrochloride and DHB was employed as the matrix for analyses in both positive and negative ionization modes. The detailed matrix preparation protocol and specific spraying parameters are documented in the Supplementary data.

2.5. AFADESI- and MALDI-MSI analysis of eye tissue sections

Spatial metabolomic profiling of eye tissue sections was conducted using an in-house developed AFADESI-MSI platform. This platform coupled an AFADESI ambient ion source (Viktor (Beijing) Technology Co., Ltd., Beijing, China) with a quadrupole-Orbitrap-quadrupole ion trap (Q-OT-qIT) hybrid mass spectrometer (Orbitrap Fusion Lumos, Thermo Fisher Scientific Co., Ltd., San Jose, CA, USA). Analyses were performed in both positive and negative ion modes. The instrumental parameters were optimized as follows: a solvent system of MeOH:H₂O (8:2, v/v) was delivered at a flow rate of 5 μL/min. The nebulizing gas pressure was set to 0.6 MPa, with an extraction gas flow rate of 45 L/min. The tissue surface was raster-scanned at a speed of 0.1 mm/s in the x-direction with a step size of 0.2 mm in the y-direction. Mass spectra were acquired over a mass range of m/z 100−1000 at a resolution of 120000 full width at half maximum (FWHM) at m/z 200. The capillary temperature was maintained at 350 °C, and the spray voltage was set to ±3 kV. Data acquisition was controlled by Xcalibur software (Version 4.0, Thermo Scientific Co., Ltd., San Jose, CA, USA).

Complementary high-resolution imaging was performed using an Autoflex Speed MALDI TOF/TOF mass spectrometer (Bruker Daltonics Co., Ltd., Billerica, MA, USA), equipped with a 2000-Hz Smartbeam Nd:YAG laser (355 nm). Data were collected in reflectron mode over an m/z range of 100−1000 with a mass resolution of approximately 10000 FWHM at m/z 400–600, achieving spatial resolutions of 50 μm and 20 μm for positive and negative ion modes, respectively. The laser spot size was set to “small” and the laser energy was optimized at the start of each run and held constant thereafter. External calibration was performed prior to each data acquisition session to ensure mass accuracy. Data acquisition and image generation were managed using FlexAnalysis 3.4 and FlexImaging 4.1 software (Bruker Daltonics Co., Ltd., Billerica, MA, USA).

2.6. Liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis

To validate the imaging findings, eye tissue homogenates were subjected to LC-MS/MS analysis. The system consisted of an UltiMate 3000 series HPLC system (Thermo Fisher Scientific Co., Ltd., San Jose, CA, USA) coupled to a Q-OT-qIT hybrid mass spectrometer (Orbitrap Fusion Lumos, Thermo Fisher Scientific Co., Ltd., San Jose, CA, USA). Analyses were conducted in both positive and negative ionization modes. The comprehensive protocols for sample preparation and the detailed instrumental parameters for LC-MS/MS are provided in the Supplementary data.

2.7. Data processing and analysis

The raw data from AFADESI-MSI experiments were processed according to a workflow established in our previous study [21]. In brief, the raw files were converted to.cdf format using Xcalibur software and then imported into MassImager 2.0 for image reconstruction and preprocessing. This workflow included background subtraction and the definition of regions of interest (ROIs) by co-registration with adjacent H&E-stained sections. The spectral data from each ROI were exported as two-dimensional matrices (m/z vs. intensity). Peak detection and alignment were performed using MarkerView 1.2.1 (AB SCIEX LLC, Framingham, MA, USA). Ion intensities within each ROI were normalized to the total ion current (TIC).

For MALDI-MSI data, relative ion intensities within predefined ROIs were extracted and normalized to the TIC using FlexImaging 4.1 software (Bruker Daltonics Co., Ltd., Bremen, Germany).

2.8. Metabolite identification

A multi-strategy approach was employed for the confident identification of metabolites. Initially, the accurate m/z values detected by AFADESI-MSI were matched against an in-house custom-built database (integrated from HMDB, METLIN, and other curated sources) using a mass error tolerance of ±5 ppm for the rapid annotation of known compounds. Where applicable, isotopic pattern matching was also required to support the putative assignments. To further verify these identifications, eye tissue homogenates were analyzed by LC-MS/MS. The resulting MS/MS spectra were interpreted according to their characteristic fragmentation patterns, and identifications were confirmed when at least two diagnostic fragment ions matched with reference spectra from public databases. In addition, theoretical m/z values for possible adducts and in-source fragments were calculated using the R programming language to avoid misannotation and were cross-checked against both MSI datasets. A high degree of concordance in the spatial localization of the same metabolite detected by both AFADESI-MSI and MALDI-MSI served as a supportive criterion to validate the accuracy of the identification.

2.9. Statistical analysis

Multivariate segmentation of MALDI-MSI datasets was performed using SCiLS Lab (Bruker Daltonics, Co., Ltd., Bremen, Germany) to distinguish retinal subregions based on spectral similarity. Segmentation was carried out using the k-means clustering algorithm with correlation distance as the spectral similarity metric and spatial smoothing (3 × 3 pixels) to reduce local noise. The number of clusters was optimized empirically to correspond to major anatomical retinal layers (retinal pigment epithelium, outer nuclear layer, inner nuclear layer, and ganglion cell layer). The resulting segmentation maps were validated by overlaying with corresponding H&E-stained tissue images to ensure anatomical accuracy.

Multivariate statistical analysis of the processed AFADESI-MSI data was performed using partial least squares discriminant analysis (PLS-DA) in SIMCA-P+ 14.0 (Umetrics AB, Umeå, Västerbotten County, Sweden). Group differences were assessed using the Student's t-test, with P < 0.05 considered statistically significant. Metabolic network analysis was carried out using the MetaboAnalyst online platform [22].

3. Results and discussion

3.1. Physiological, biochemical, and histopathological analyses

Physiological assessments revealed that rats in the DED group exhibited a hallmark diabetic phenotype, characterized by a significant reduction in body weight (P < 0.01), a marked increase in water-intake (P < 0.05), and a decrease in food-intake compared to the control group (Figs. S1A–C). These alterations are consistent with the systemic manifestations of chronic hyperglycemia. Biochemical profiling of serum samples corroborated these findings, demonstrating significantly elevated levels of serum glucose (P < 0.01) and glycated hemoglobin (HbA1c, P < 0.01) in the DED group (Figs. S1D and E), indicative of a persistent hyperglycemic state. Furthermore, the DED group showed significant reductions in TP, ALB, GLB, TBIL, Na⁺, and Cl⁻ levels (P < 0.01, Figs. S1F–K), suggesting impaired protein synthesis and electrolyte imbalances.

Histopathological examination via H&E staining provided further validation of these systemic improvements (Fig. S2). The control group displayed a well-organized retinal architecture with clearly defined layers. In contrast, the DED group exhibited severe retinal damage, including disorganization of the nerve fiber layer, structural disruption, vacuolar degeneration, and a reduced density of ganglion cells, alongside abnormal thicknesses of the inner and outer limiting membranes. Collectively, these systemic metabolic disturbances are consistent with features commonly observed during early-stage diabetic ocular involvement, with particular relevance to early non-proliferative retinal pathology.

Administration of ferulic acid resulted in a dose-dependent amelioration of these physiological and biochemical abnormalities. Both the L-FA and H-FA treatment groups showed a significant increase in body weight (P < 0.01 vs. DED), with water intake and food consumption trending towards normalization (Figs. S1A–C). Corresponding improvements were observed in biochemical parameters. The H-FA group exhibited a significant reduction in blood glucose (P < 0.05 vs. DED) and a pronounced decrease in HbA1c levels (Figs. S1D and E). Levels of TP, ALB, GLB, TBIL, Na⁺, and Cl⁻ were restored to near-normal levels, with the effects in the H-FA group being more pronounced than those in the L-FA group, thus demonstrating a clear dose-response relationship (Figs. S1F–K).

These systemic improvements were paralleled by pronounced retinal protection at the histological level. Ferulic acid treatment substantially attenuated retinal disorganization, vacuolar degeneration, and ganglion cell loss observed in the DED group. The L-FA group showed partial recovery of retinal laminar integrity, whereas the H-FA group displayed a retinal architecture closely resembling that of control animals, with preserved retinal stratification and reduced pathological alterations (Fig. S2). Together, these findings demonstrate that ferulic acid effectively mitigates early diabetic metabolic and structural insults to the eye, supporting its protective role in early-stage diabetic ocular pathology.

3.2. Metabolic profiling of diabetic rat eyes via AFADESI-MSI and MALDI-MSI

The eye, as a complex and highly compartmentalized sensory organ, possesses distinct biochemical compositions and physiological functions across its various anatomical regions, including the cornea, lens, retina, and vitreous body. To systematically elucidate its intricate spatial metabolic landscape, this study employed an innovative, integrated approach by combining AFADESI-MSI and MALDI-MSI for the analysis of rat eye sagittal sections. Initially, AFADESI-MSI served as a high-throughput discovery platform, detecting 2474 and 2208 ion features in positive and negative ion modes, respectively. Through a rigorous identification workflow, 135 metabolites—including amino acids, nucleotides, and lipids, etc.—were successfully annotated, with 15 additionally confirmed by LC–MS/MS, establishing a comprehensive spatial metabolite database for the rat eye (Table S1).

To validate findings and achieve higher spatial resolution, MALDI-MSI was subsequently applied at 50 μm and 20 μm pixel sizes. Of the 135 annotated metabolites, 72 were also detected by MALDI-MSI, and their spatial distributions showed good agreement with AFADESI-MSI results, reinforcing the reliability of the observed patterns. Whereas AFADESI-MSI excels in metabolite coverage and throughput, MALDI-MSI provides superior spatial resolution and sharper delineation of microstructural boundaries, making the two platforms highly synergistic. Representative ion images acquired by AFADESI-MSI and MALDI-MSI are shown in Fig. 2.

Fig. 2.

Spatial heterogeneity of the rat eye. (A) Hematoxylin and eosin (H&E) stain of the rat eye. (B) structural diagram of the rat eye. (C) Ion images of representative metabolite ions in positive and negative modes by air flow-assisted desorption electrospray ionization mass spectrometry imaging (AFADESI-MSI). (D) Ion images of representative metabolite ions in positive and negative modes by matrix-assisted laser desorption ionization mass spectrometry imaging (MALDI-MSI). The eye region outlined with the white dashed line was acquired at 50 μm spatial resolution, whereas the region outlined with the purple dashed line was acquired at 20 μm spatial resolution. F1,6P: fructose 1,6-bisphosphate; UMP: uridine 5′-monophosphate; ADP: adenosine diphosphate; FA: fatty acid; FAHFA: hydroxy fatty acid; PA: phosphatidic acid; PC: phosphatidylcholine; PE: phosphatidylethanolamine; PG: phosphatidylglycerol; PI: phosphatidylinositol; LPS: lysophosphatidylserine.

This integrated approach revealed distinct compartment-specific distribution patterns across the ocular anatomy. For instance, taurine and choline were abundant in the retina, vitreous body, and cornea; glucose was widely distributed throughout ocular tissues; fructose 1,6-bisphosphate (F1,6P) was predominantly localized to the choroid; uridine 5′-monophosphate (UMP) was concentrated in the anterior chamber and lens; adenosine diphosphate (ADP) and glutathione (GSH) were mainly found in the lens, anterior chamber, and vitreous body; pseudouridine was distributed across the lens, retina, and optic nerve; deoxyguanosine and riboflavin were primarily localized to the cornea; and 6-hydroxynicotinic acid was enriched in the anterior chamber and vitreous body. In contrast, the majority of lipid species were concentrated in the retina and optic nerve regions.

Furthermore, data-driven segmentation of retinal micro-regions was achieved using SCiLS Lab software through unsupervised clustering of MSI spectral profiles applied to high-resolution MALDI-MSI datasets (20 μm spatial resolution). This approach successfully delineated distinct metabolic domains within the retina, uncovering subtle layer-specific biochemical differences. These segmented micro-regions revealed highly selective spatial distributions of phospholipid species across ocular tissues (Fig. 3). For example, phosphatidylcholine (PC)(32:0) was primarily distributed in the photoreceptor cell layer (PRL) and retinal pigment epithelial cells; PC(34:1) was enriched in the inner plexiform and inner nuclear layers; PC(38:4) was enriched in the PRL. PC(38:7) was specifically localized to the sclera, and PC(42:9) was mainly found in the choroid. These findings collectively demonstrate significant spatial metabolic heterogeneity within the rat eye. The unique distribution patterns of these metabolites not only suggest their critical roles in maintaining the specialized structure and function of different ocular regions but also highlight their potential to serve as molecular markers for specific anatomical structures or cellular layers. These findings demonstrate that data-driven segmentation of MSI data enables precise mapping of molecular heterogeneity at the cellular layer level, providing critical spatial context for interpreting metabolic alterations in DED. Collectively, our results highlight the complementary strengths of AFADESI-MSI for broad metabolite discovery and high-resolution MALDI-MSI for both targeted validation and precise spatial localization of key metabolites, offering a flexible and adaptable framework that is potentially applicable to diverse biological systems and research objectives in spatial metabolomics.

Fig. 3.

Schematic of data-driven retinal segmentation and the spatial distribution of metabolites (20 μm resolution). (A) Data-driven spatial segmentation map of retinal micro-regions and the corresponding hematoxylin and eosin (H&E)-stained optical image. (B) Distribution of selected metabolites across different retinal layers. GCL: ganglion cell layer; INL: inner nuclear layer; IPL: inner plexiform layer; NFL: nerve fiber layer; ONL: outer nuclear layer; OPL: outer plexiform layer; PRL: photoreceptor cell layer; RPE: retinal pigment epithelial cells; PC: phosphatidylcholine.

3.3. Spatial-metabolic alterations in DED

To elucidate the spatial metabolic characteristics during the progression of DED, the AFADESI-MSI profiles of rat eye sections from the control and DED groups were compared by supervised multivariate PLS-DA to achieve the maximum separation. The PLS-DA scatter plot (Fig. S3) showed a clear distinction between the DED and control groups, indicating substantial metabolic divergence associated with diabetic injury. Employing a statistical threshold of Student's t-test (P < 0.05), a total of 39 metabolites were identified as being significantly altered in the DED state (Table S2). These dysregulated metabolites are primarily involved in core metabolic pathways, including amino acid, lipid, and carbohydrate metabolism. To gain a systems-level understanding of the intrinsic connections and synergistic dysregulation among these pathways, we constructed a comprehensive metabolic perturbation network for DED, based on the direct and indirect interactions of these metabolites (Fig. 4). This network provides a macroscopic overview of the global metabolic dysregulation in DED, offering a critical perspective for a deeper understanding of its underlying pathological mechanisms.

Fig. 4.

Comprehensive metabolic perturbation network in the eyes of diabetic eye disease (DED) rats. The network was constructed based on curated metabolic pathway information from the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Human Metabolome Database (HMDB). Nodes represent significantly altered metabolites, and arrows indicate biochemical relationships between metabolites, with solid and dashed lines denoting single-step and multi-step reactions, respectively. Pathway connectivity and directionality were inferred from established metabolic annotations rather than de novo mechanistic modeling. 3-MLA: 3-methyladipic acid; GPC: glycerylphosphorylcholine; FA: fatty acid; FAHFA: hydroxy fatty acid; PE: phosphatidylethanolamine; PC: phosphatidylcholine; G3P: glycerol-3-phosphate; S7P: sedoheptulose-7-phosphate; SM: sphingomyelin; MAG: monoacylglycerol; G6P: glucose-6-phosphate; TUDCA: tauroursodeoxycholic acid.

3.3.1. Dysregulation of amino acid metabolism

Amino acids, as fundamental building blocks of life, are indispensable for a myriad of physiological processes, including protein synthesis, energy production, signal transduction, and the response to oxidative stress. In this study, spatial metabolomics analysis revealed a profound disruption in the metabolic network of several key amino acids and their derivatives within the ocular tissues of DED rats. Specifically, lysine, arginine, carnitine, and 3-methyladipic acid (3-MLA) were significantly downregulated across multiple anatomical regions of the eye, as illustrated in Figs. 5A–D and S4A–D.

Fig. 5.

Dysregulation of metabolites associated with amino acid and glucose metabolism in the diabetic eye disease (DED) rat eye as revealed by air flow-assisted desorption electrospray ionization mass spectrometry imaging (AFADESI-MSI). (A) Lysine. (B) Arginine. (C) Carnitine. (D) 3-Methyladipic acid (3-MLA). (E) Glucose. (F) Glucose-6-phosphate (G6P). (G) Sedoheptulose-7-phosphate (S7P). (H) Glycerol-3-phosphate (G3P).

Lysine, an essential amino acid, serves as a core component for protein synthesis and is critical for maintaining the structural integrity of ocular tissues and facilitating cellular repair. Its significant depletion in the DED group can be attributed to two interconnected pathological mechanisms. On one hand, the hyperglycemic milieu promotes the non-enzymatic glycation of proteins, leading to the irreversible consumption of the free lysine pool for the formation of advanced glycation end-products (AGEs). On the other hand, the accumulation of AGEs and the ensuing oxidative stress exacerbate tissue damage, thereby increasing the demand for lysine in reparative processes and further depleting its available reserves [23].

Arginine plays a pivotal role in ocular physiology, primarily as the precursor for nitric oxide (NO), which is essential for regulating retinal vasodilation and maintaining hemodynamic homeostasis in the eye. The significant reduction in arginine levels observed in the DED group is largely a consequence of an imbalance in its metabolic pathways. Firstly, the diabetic state induces the upregulation of both the expression and activity of inducible nitric oxide synthase (iNOS) and arginase in the retina, leading to enhanced catabolism of arginine. Secondly, insulin resistance may suppress its endogenous biosynthesis. The combined effect of these processes results in a marked decline in arginine availability [24].

Carnitine, a quaternary ammonium compound biosynthesized from the amino acids, is crucial for transporting long-chain fatty acids into the mitochondria for β-oxidation, thereby providing energy for highly metabolically active ocular tissues such as photoreceptors [25,26]. It also possesses notable antioxidant properties [27]. The observed decrease in carnitine levels in the DED group reflects a profound shift in ocular energy metabolism. The hyperglycemic environment suppresses mitochondrial β-oxidation, which may be related to the decrease in carnitine [28,29]. Concurrently, oxidative stress and general metabolic dysregulation may directly impair its biosynthesis, collectively contributing to its diminished concentration [30]. In patients with diabetic retinopathy, abnormal carnitine metabolism may exacerbate oxidative stress and inflammation in the retina and promote the development of the disease [31].

3-MLA, a downstream metabolite in the lysine degradation pathway, exhibited a concurrent downregulation with lysine. This strong correlation strongly suggests a systemic suppression of the entire lysine metabolic flux in the context of DED.

3.3.2. Dysregulation of glucose metabolism

Glucose, the primary energy substrate for cellular activities, plays a central role in the pathogenesis of DED, with aberrant activation of its metabolic pathways being a key driver of disease progression [32]. Our spatial metabolomics analysis demonstrated a significant upregulation of several key glucose metabolic intermediates in the ocular tissues of DED rats, including glucose, glucose-6-phosphate (G6P), sedoheptulose-7-phosphate (S7P), and glycerol-3-phosphate (G3P), as shown in Figs. 5E–H and S4E–H. This accumulation indicates that ocular tissues undergo substantial metabolic reprogramming under diabetic conditions, a process closely linked to both systemic hyperglycemia and local compensatory responses.

The accumulation of glucose, the initial substrate of glycolysis, directly reflects the persistent impact of hyperglycemia on ocular tissues. G6P stands at a critical junction, serving as a precursor for both glycolysis and the pentose phosphate pathway (PPP), while S7P is a characteristic intermediate of the non-oxidative phase of the PPP. The abnormal accumulation of these metabolites suggests a combined effect of impaired glycolytic flux and a compensatory activation of the PPP. Although the upregulation of the PPP can bolster the cellular supply of nicotinamide adenine dinucleotide phosphate hydrogen (NADPH) to counteract oxidative stress, its chronic activation is implicated in promoting inflammatory responses and pathological angiogenesis, thereby accelerating the progression of DED [33].

G3P acts as a crucial metabolic node, linking glycolysis with lipid metabolism. It is generated from the reduction of dihydroxyacetone phosphate and serves as a vital precursor for the synthesis of phospholipids and triglycerides. The marked elevation of G3P levels in the DED group points to a close interplay and dysregulation between glycolytic and lipid metabolic pathways [34]. The excess accumulation of G3P may drive increased lipid synthesis, which in turn can induce lipotoxicity, trigger inflammatory cascades, and cause cellular damage [35], further exacerbating the pathological landscape of DED.

3.3.3. Dysregulation of lipid metabolism

Lipid metabolism dysregulation plays a pivotal role in the pathogenesis and progression of DED. In this study, AFADESI-MSI-based spatial metabolomics analysis revealed in situ alterations in 26 lipid species, encompassing multiple synthetic and catabolic pathways of free fatty acids (FAs), fatty acid esters of hydroxy fatty acids (FAHFAs), phospholipids, sphingolipids, and monoacylglycerols (MAGs), as illustrated in Figs. 6, 7 and S5–S8.

Fig. 6.

Dysregulation of metabolites associated with fatty acid metabolism in the diabetic eye disease (DED) rat eye as revealed by air flow-assisted desorption electrospray ionization mass spectrometry imaging (AFADESI-MSI). (A) Fatty acid (FA)(18:1). (B) FA(20:1). (C) Hydroxy fatty acid (FAHFA)(34:0). (D) FAHFA(36:1). (E) FAHFA(36:2). (F) FA(14:0). (G) FA(14:1). (H) FA(16:1). (I) FA(20:5). (J) FAHFA(21:1). (K) FAHFA(32:0). (L) 12-Hydroxyeicosapentaenoic acid (12-HETE).

Fig. 7.

Dysregulation of metabolites associated with phosphatidylcholine (PC) and phosphatidylethanolamine (PE) metabolism in the diabetic eye disease (DED) rat eye as revealed by air flow-assisted desorption electrospray ionization mass spectrometry imaging (AFADESI-MSI). (A) PC(32:1). (B) PC(33:4). (C) PC(34:4). (D) PC(38:4). (E) PC(39:6). (F) PC(42:10). (G) PE(P-18:1/18:4). (H) PE(P-18:1/22:5). (I) PE(22:6/P-16:0). (J) PE(44:12).

In the context of fatty acid metabolism, FA(18:1) and FA(20:1) were significantly upregulated in the DED group. This observation aligns with our previous findings in diabetic liver disease [20], thereby providing, from a metabolic perspective, scientific evidence supporting the traditional Chinese medicine theory that “the liver opens into the eyes”.

Notably, FAHFA lipids with established anti-inflammatory and anti-diabetic properties, such as FAHFA(34:0), FAHFA(36:1), and FAHFA(36:2), also exhibited an upward trend. This suggests a potential compensatory protective response of the organism under pathological conditions [36]. In contrast, several beneficial fatty acids, including FA(14:0), FA(14:1), FA(16:1), FA(20:5), as well as FAHFA(21:1) and FAHFA(32:0), were significantly downregulated. These lipids are crucial for energy supply, anti-inflammatory responses, and the improvement of insulin sensitivity [37,38]. The reduction in their levels may compromise the body's self-protective capacity, thereby exacerbating the pathological progression of DED.

Furthermore, this study identified a significant elevation in the pro-inflammatory mediator 12-hydroxyeicosatetraenoic acid (12-HETE) within the DED group. 12-HETE can induce reactive oxygen species (ROS) generation by activating NADPH oxidase, subsequently activating retinal cells—particularly vascular endothelial cells—and mediating multiple pathological changes in DR [39]. This finding further substantiates the clinical value of 12-HETE as a potential biomarker for DR [40].

Phospholipids and sphingolipids, as primary structural components of cellular membranes, are essential for maintaining membrane integrity and fluidity and broadly participating in cell signal transduction [41,42]. Under pathological conditions such as DED, the reduction in these lipid levels may lead to the disruption of membrane architecture, decreased membrane fluidity, and aberrant signal transduction [41,42]. Consequently, the normal functions of retinal and vascular endothelial cells may be impaired, aggravating visual signal transduction deficits and tissue damage [43,44].

Additionally, a marked upregulation of MAGs was observed in the DED group. MAGs are key intermediates in the catabolism of triglycerides and the metabolism of glycerophospholipids. Their elevated levels typically reflect enhanced lipid catabolism or impaired re-esterification within tissues. In the hyperglycemic milieu of diabetes, impaired insulin signaling leads to a relative increase in hormone-sensitive lipase activity. This may accelerate the hydrolysis of stored triglycerides in ocular tissues (e.g., the retina), resulting in the substantial generation of MAGs [45].

This study also revealed that choline and glycerophosphocholine were significantly downregulated in the DED group. This further indicates a widespread perturbation of phospholipid metabolic pathways during DED progression, potentially affecting membrane homeostasis and neurotransmitter synthesis, thereby exacerbating retinal dysfunction.

3.3.4. Dysregulation of other metabolic pathways

Riboflavin (Vitamin B₂), the precursor of flavin adenine dinucleotide and flavin mononucleotide, serves as a critical cofactor for numerous antioxidant enzymes, such as GSH reductase, playing a vital role in maintaining cellular redox balance. This study found that riboflavin levels were significantly upregulated in the ocular tissues of DED rats (Figs. S9A and S10A). This alteration may be associated with the compensatory activation of the body's antioxidant defense systems in response to a high-glucose environment.

Guanosine, an essential precursor for RNA and DNA synthesis, is also involved in energy metabolism and signal transduction. Our study demonstrated a significant decrease in guanosine levels within ocular tissues (Figs. S9B and S10B). The depletion of guanosine may compromise the energy supply to the eye, thereby aggravating the progression of DED. Interestingly, research has indicated that guanine is upregulated in the tears of patients with DR and could serve as a potential biomarker for diagnosing the disease [46].

GSH is the predominant endogenous antioxidant in ocular tissues, protecting retinal cells from oxidative damage by scavenging free radicals and maintaining redox homeostasis. This study revealed that GSH levels were significantly downregulated in the ocular tissues of DED rats (Figs. S9C and S10C), indicating a severe impairment of their antioxidant defense system. Chronic GSH deficiency not only leads to the death of retinal pigment epithelial cells but also exacerbates tissue damage by inducing mechanisms such as ferroptosis, autophagy, and premature cell senescence [47].

In addition, taurocholic acid (TCA) and tauroursodeoxycholic acid (TUDCA) were significantly upregulated in the DED group (Figs. S9D, S9E, S10D and S10E) that aligns with previous reports of bile acids functioning as stress-responsive molecules in diabetic models [20].

Recent metabolic studies in DED, including both DED and DR, have reported characteristic alterations in energy metabolism, amino acid homeostasis, lipid remodeling, and oxidative stress pathways. For example, elevated glucose and related glycolytic intermediates, disruptions in choline metabolism, and shifts in nucleotide turnover have been observed in human or rat aqueous humor, vitreous, and lens samples from patients or model animals with diabetic ocular complications [48,49]. Altered lipid profiles—particularly changes in phosphatidylcholines, lysophospholipids, and fatty acid oxidation products—have also been consistently associated with retinal dysfunction in clinical DR metabolomics studies [40,50]. In our work, several MSI-derived metabolites (e.g., glucose, choline, GSH, and 12-HETE) exhibit regulation patterns that align with these previously reported metabolic disturbances, indicating that the spatially resolved signatures identified here reflect physiologically meaningful alterations associated with diabetic pathology. At the same time, the high-resolution spatial perspective provided by MSI allowed us to uncover region-specific metabolic changes—such as choroid-localized glycolytic intermediates and retina-enriched lipid species—that have been less accessible in bulk tissue or biofluid studies.

3.4. Regulatory effects of ferulic acid on metabolic dysregulation in DED

Ferulic acid is a natural phenolic acid widely found in plants, possessing multiple biological activities, including antioxidant, anti-inflammatory, and regulatory effects on energy and lipid metabolism [[51], [52], [53], [54]]. The results of this study demonstrate that ferulic acid intervention significantly ameliorated the abnormal levels of nine key metabolites in DED (Figs. 8 and S11). Notably, ferulic acid restored the balance of amino acid metabolism by elevating depleted lysine and carnitine, thereby supporting protein synthesis, nitric oxide production, and mitochondrial fatty acid oxidation in ocular tissues. Correction of glucose level indicates that ferulic acid modulates hyperglycemia-driven metabolic overload and helps reestablish glucose–lipid metabolic crosstalk. Moreover, ferulic acid attenuated lipid metabolism disorders, reducing detrimental lipid accumulation and restoring beneficial glycerophospholipid and fatty acid profiles essential for membrane integrity and signal transduction.

Fig. 8.

Air flow-assisted desorption electrospray ionization mass spectrometry imaging (AFADESI-MSI) reveals the regulation of in situ disruptions by ferulic acid. DED: diabetic eye disease; L-FA: low-dose ferulic acid; H-FA: high-dose ferulic acid; FA: fatty acid; 12-HETE: 12-hydroxyeicosapentaenoic acid; GPC: glycerylphosphorylcholine; TCA: taurocholic acid; TUDCA: tauroursodeoxycholic acid.

The observed normalization of riboflavin, TCA, and TUDCA levels suggests that the therapeutic mechanism of ferulic acid is intimately associated with the regulation of redox homeostasis and the modulation of stress-responsive molecules in diabetes. Importantly, the dose-dependent effects seen in this study, particularly the superior efficacy of the H-FA group, suggest a clear pharmacological potential and support previous findings in other diabetic complication models, such as diabetic cardiomyopathy [18]. Together, these results highlight ferulic acid as a promising multi-target therapeutic candidate capable of intervening at several critical metabolic nodes implicated in diabetic ocular injury.

Note that ferulic acid itself was not detectable in ocular tissues or plasma by AFADESI-MSI or LC-MS/MS, indicating extensive metabolism and/or low ocular penetration of the parent compound.

From a translational perspective, the protective effects of ferulic acid extend beyond its role as a general antioxidant. By directly influencing metabolic reprogramming within the eye, ferulic acid demonstrates the potential to preserve tissue homeostasis and retinal architecture under diabetic conditions. This dual action—combining systemic metabolic regulation with localized tissue protection—suggests that ferulic acid could complement existing therapies, such as anti-VEGF or glucose-lowering treatments, by addressing underlying metabolic dysregulation rather than focusing solely on symptomatic vascular changes.

4. Conclusion

In summary, this study established a comprehensive spatial metabolome atlas of the rat eye using an integrated AFADESI-MSI and MALDI-MSI strategy. Notably, unsupervised k-means clustering of the MALDI-MSI datasets allowed automatic identification of retinal micro-regions based exclusively on their metabolic signatures, highlighting the potential of data-driven approaches to define functional tissue compartments in an unbiased manner. In addition, we uncovered region-specific metabolic reprogramming during the progression of DED. Significant alterations in amino acid, glucose, and lipid metabolism were identified, alongside with disruptions in redox balance and nucleotide metabolism, all of which were spatially localized to distinct ocular compartments and retinal layers. Importantly, treatment with ferulic acid effectively restored multiple dysregulated metabolites, alleviated biochemical imbalances, and improved histopathological features in a dose-dependent manner, highlighting its protective role against diabetic ocular injury.

These findings provide new insight into the molecular underpinnings of DED by directly linking metabolic alterations to structural pathology. The spatially resolved metabolomic data not only enhance our understanding of how diabetes perturbs ocular homeostasis at a systems level but also identify potential molecular markers associated with disease progression. Furthermore, the demonstration of ferulic acid's therapeutic efficacy underscores the translational potential of natural compounds as metabolic interventions for DED.

This study also has inherent limitations. The modest sample size and absence of external validation place the work in an exploratory stage; however, the findings provide a solid framework for future investigation. Additionally, all MSI-derived data are semi-quantitative, and absolute ion intensities across heterogeneous ocular compartments may be influenced by tissue-specific matrix effects; however, relative changes within the same anatomical region across experimental groups remain biologically reliable. To strengthen the translational relevance of these results, future work will focus on validating the spatial metabolic signatures in human ocular tissues or clinical samples. Moreover, incorporating larger cohorts and longitudinal studies—along with independent validation sets—will be essential to further substantiate the identified metabolites and to assess their potential as early biomarkers or therapeutic targets in DED.

CRediT authorship contribution statement

Shuohan Cheng: Writing – original draft, Visualization, Investigation, Formal analysis. Shuo Wang: Visualization, Methodology, Investigation, Formal analysis. Tianfang Lan: Resources, Methodology. Hongtao Jin: Resources. Zhi Zhou: Resources. Zhonghua Wang: Writing – review & editing, Writing – original draft, Supervision, Investigation, Funding acquisition, Conceptualization. Zeper Abliz: Supervision, Funding acquisition.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the supplementary data to this article: