Molecular Responses of Anti-VEGF Therapy in Neovascular Age-Related Macular Degeneration: Integrative Insights From Multi-Omics and Clinical Imaging

Abstract

Purpose: The purpose of this study was to investigate the molecular mechanisms underlying anti-vascular endothelial growth factor (anti-VEGF) efficacy and response variability in neovascular age-related macular degeneration (nAMD) using longitudinal proteomic and metabolomic analysis alongside three-dimensional lesion measurements.

Methods: In this prospective study, 54 treatment-naive patients with nAMD underwent “3+ pro re nata” (3+PRN) anti-VEGF regimens followed for at least 12 weeks. Aqueous humors were collected pre- and post-treatment for proteomic and metabolomic analysis. Three-dimensional optical coherence tomography (OCT) and OCT angiography assessed different types of nAMD lesion volumes and areas.

Results: There were 1350 proteins and 1268 metabolites that were identified in aqueous humors, with 301 proteins and 353 metabolites significantly altered during anti-VEGF treatment, enriched in pathways of angiogenesis, energy metabolism, signal transduction, and neurofunctional regulation. Sixty-seven changes of (Δ) molecules significantly correlated with at least one type of ΔnAMD lesion. Notably, proteins FGA, TALDO1, and ASPH significantly decreased during treatment, with their reductions correlating with greater lesion regression in at least two lesion types. Conversely, despite that YIPF3 also showed significant downregulation, its decrease was associated with poorer regression in total nAMD lesion and subretinal hyper-reflective material.

Conclusions: This study identifies FGA, TALDO1, and ASPH as potential key molecules in the efficacy of anti-VEGF therapy, whereas YIPF3 may be a key factor in poor response. The integration of longitudinal three-dimensional lesion analysis with multi-omics provides valuable insights into the mechanisms and response variability of anti-VEGF treatment in nAMD.

Neovascular age-related macular degeneration (nAMD) is a principal cause of blindness globally, with its impact intensifying alongside an aging population.¹ This condition imposes significant health and economic burdens, underscoring the critical need for effective interventions. Anti-vascular endothelial growth factor (VEGF) therapy plays a pivotal role in managing nAMD by reducing vascular permeability and hindering the progression of macular neovascularization (MNV), thereby improving vision.² However, it is questioned whether the mechanism of anti-VEGF is limited solely to these effects. Moreover, despite that the efficacy of anti-VEGF treatments were identified in many patients, a significant subset exhibits non-responsiveness or develops resistance, with the underlying mechanisms of this variability still not fully understood.³ This underscores the necessity of a comprehensive and in-depth exploration into the mechanisms of anti-VEGF therapy, particularly in vivo, to understand not only the roots of resistance, but also the unexplored therapeutic pathways. 

Advancing into the exploration of ocular molecular dynamics, modern methodologies encompass a diverse array including genomics, proteomics, metabolomics, and other omics technologies.^4–6 Each of these approaches offers unique strengths and inherent limitations, yet proteomics and metabolomics are particularly pertinent in the context of anti-VEGF therapy, considering that the therapy is primarily mediated by proteins and potentially alters the dynamics of intraocular metabolites.^7,8 Consequently, an integrated omics approach, combining both proteomics and metabolomics, may facilitate a comprehensive exploration of molecular alterations in nAMD under anti-VEGF treatment. This approach is capable of revealing more potential mechanisms of anti-VEGF therapy and enhancing our understanding of the factors contributing to treatment nonresponsiveness. 

Three-dimensional optical coherence tomography (OCT) and OCT angiography (OCTA) are crucial for evaluating the prognosis of nAMD after anti-VEGF therapy.⁹ These approaches provide direct and precise quantification of morphological lesions, such as intraretinal fluid (IRF), subretinal hyper-reflective material (SHRM), and MNV, as well as structural damages in the outer nuclear layer (ONL), and the ellipsoid zone (EZ), etc.¹⁰ The integration of such three-dimensional imaging with molecular omics enables a holistic understanding of the relationship between ocular molecular changes and the therapeutic efficacy of nAMD lesions. 

Hence, our study combining three-dimensional imaging with multi-omics analysis, investigates the alterations of intraocular protein and metabolite, and their relationships with nAMD lesion changes during anti-VEGF therapy, aiming to uncover novel mechanisms of anti-VEGF and elucidate the factors contributing to poor therapeutic responses. 

Methods

Prospective Study Design

In this prospective study, conducted between January 1, 2022, and March 31, 2023, treatment-naïve patients with nAMD were recruited from the Zhongshan Ophthalmic Center and Guangdong Provincial People’s Hospital. The inclusion criteria were set for patients aged 50 years and above, diagnosed with either type 1 or type 2 MNV based on the 2020 Nomenclature Consensus.⁹ Exclusion criteria included eyes with complete scarring, severe refractive media opacities, extensive vitreoretinal or subretinal hemorrhage, MNV secondary to pathologies other than nAMD, and concurrent ocular disorders, such as glaucoma. Additionally, eyes with any previous vitreoretinal surgery or ocular surgery other than cataract surgery were excluded. Patients who had undergone cataract surgery within 6 months prior to the study were also excluded. Furthermore, patients with severe systemic conditions, such as uncontrolled hypertension, history of myocardial infarction or stroke, severe liver or kidney diseases, respiratory system diseases, or other severe systemic diseases were excluded from the study. All participants underwent a standardized three + pro re nata (3+PRN) regimen of anti-VEGF therapy, which consists of 3 initial 4 weeks apart injections (loading phase), followed by additional injections as needed based on disease activity and treatment response, and were continuously followed up for a minimum of 12 weeks. The overall experimental workflow is depicted in Figure 1. Ethical approval for this study was granted by the Zhongshan Ophthalmic Center Ethics Committee (ID: 2022KYPJ117) and Guangdong Provincial People's Hospital Ethics Committee (ID: 2018-081H-1). The study was registered under the clinical trial number ChiCTR2200063428. Informed consent was obtained from all patients. 

Figure 1.

 

Framework for longitudinal three-dimensional imaging and multi-omics analysis of nAMD during anti-VEGF treatment. 3D, three-dimensional; DIA, data-independent acquisition; LC-MS/MS, liquid chromatography-tandem mass spectrometry; nAMD, neovascular age-related macular degeneration; OCT, optical coherence tomography; VEGF, vascular endothelial growth factor.

Clinical Image Analyses

OCT (Spectralis OCT, Heidelberg Engineering, Germany) was used to image an 8 × 6 mm macular area (61 dense scans), and OCTA (RTVue-XR Avanti; Optovue, Fremont, CA, USA) was used for a 6 × 6 mm macular region. The 3D slicer software (version 5.4.0; http://www.slicer.org; provided in the public domain by Harvard University, Cambridge, MA, USA) was utilized to calculate the volumes of 6 distinct morphological lesions: MNV (heterogeneous reflectivity and confirmed blood flow on OCTA), subretinal fluid (SRF) hyporeflective space between the neurosensory retina and retinal pigment epithelium (RPE), IRF (hyporeflective spaces within the neurosensory retina), SHRM (hyper-reflective material without blood flow signal between the neurosensory retina and RPE), and pigment epithelial detachment (PED; hyporeflective RPE elevations), with their combined volumes constituting the total lesion volume. Areas of structural damages were also quantified, including EZ and external limit membrane (ELM) damages defined as discontinuous reflectivity, and ONL damage defined as ONL thinning exceeding 50% compared to adjacent areas. 

Quantitative measurements of each lesion within the scanning range were conducted using the following steps. (1) All 61 B-scans of OCT and OCTA were imported into 3D Slicer software, using the Slicer Morph Extension. (2) Volume measurement: The contours of MNV, SRF, IRF, SHRM, and PED were manually outlined in 61 cross-sectional images. The volume statistics of these five components were then automatically calculated by the software (see Fig. 1). The total lesion volume was obtained by summing the volumes of these five morphological lesions. (3) Area measurement: The outer structures of the retina were measured using a paint tool with a fixed diameter/width. The desired stripes were outlined without changing the image size, and this process was repeated for all 61 B-scans. The software automatically generated the volume based on these outlines. The area of the target layers was calculated by dividing the obtained volume by the pre-set fixed diameter/width of the paint tool (see Fig. 1). 

These assessments were accurately conducted by two experienced independent readers, Xuenan Zhuang and Jiaxin Pu, with adjudication by author Feng Wen if their measurements differed by more than 10%. Notably, the agreement between the readers was excellent (kappa > 0.80), led to the adoption of their averaged measurements for further analysis. Alterations in volume and area were determined using the formula: post-treatment value – pre-treatment value. 

Treatment Protocol and Samples Collection

All patients were administered a 3+PRN anti-VEGF treatment regimen. The initial choice of anti-VEGF agent (ranibizumab, aflibercept, or conbercept) was determined by the diagnosing physician and remained consistent for the subsequent 12-week period. For each injection, 0.05 mL of the anti-VEGF drug was delivered into the vitreous cavity using a 30 G needle, opting for the superotemporal quadrant approximately 3.5 to 4 mm from the limbus. Concomitant with this treatment, aqueous humor (AH) samples were collected at two critical timepoints: prior to the first injection (pretreatment) and the third injection (post-treatment). Then, 80 to 150 µL of AH were extracted with an insulin syringe and promptly transferred into an Eppendorf tube, then immediately stored at −80°C. 

Sample Preparation and Extraction for Proteomic Analysis

PI and AH samples were gently thawed on ice. A volume of 30 µL from one sample was orderly mixed with 270 µL of PBS and 1 mM of PMSF. This mixture was then agitated at room temperature for 5 minutes. Following protein quantification via the bicinchoninic acid (BCA) assay, the protein-rich mixture was processed for tryptic digestion as per the manufacturer’s instructions. An equal amount of protein from each sample was subjected to digestion. To the supernatants, 8M urea was added to a final volume of 200 µL, followed by reduction with 10 mM DTT for 45 minutes at 37°C and alkylation with 50 mM iodoacetamide (IAM) for 15 minutes in darkness at room temperature. A 4 × volume of chilled acetone was introduced, and the mixture was left to precipitate at −20°C for 2 hours. Post-centrifugation, the protein precipitate was air-dried and reconstituted in 200 µL of 25 mM ammonium bicarbonate solution with 3 µL of trypsin (Promega), digesting overnight at 37°C. The digested peptides were then desalted using a C18 Cartridge, dried via a vacuum concentrator, concentrated by vacuum centrifugation, and finally redissolved in 0.1% (v/v) formic acid. 

High-pH Reverse Phase Fractionation for Proteomic Analysis

For a comprehensive library suitable for diaPASEF experiments, peptides were fractionated using a reversed-phase column (XBridge BEH300 C18, 4.6 µm × 250 mm, 3.5 µm 100 Å; Waters, Milford, MA, USA) at pH 10, utilizing a Vanquish Core system (Thermo Fisher Scientific). Approximately 50 µg of purified peptides were separated over 60 minutes. Peptides eluting were collected at 60-second intervals into a total of 52 fractions, later cross-concatenated into 10 fractions. These were vacuum-centrifuged to dryness and reconstituted in double-distilled water with 0.1 vol% formic acid for liquid-chromatography mass spectrometry (LC-MS) analysis. 

Liquid-Chromatography Tandem Mass Spectrometry Conditions for Proteomic Analysis

Liquid chromatography (LC) utilized a nanoElute UHPLC system (Bruker Daltonics, Germany), separating approximately 200 ng of peptides over 60 minutes at a flow rate of 0.3 µL/min through a reverse-phase C18 column equipped with an integrated CaptiveSpray emitter (25 cm × 75 µm ID, 1.6 µm, Aurora Series with CSI, IonOpticks, Australia). The separation temperature was maintained at 50°C by an integrated column oven. Mobile phases A and B consisted of 0.1 vol.% formic acid in water and 0.1% formic acid in ACN, respectively. The proportion of mobile phase B was increased from 2% to 22% over the first 45 minutes, then to 35% over the next 5 minutes, further to 80% over another 5 minutes, and maintained at 80% for the final 5 minutes. 

The LC was directly connected to a hybrid timsTOF Pro2 system (Bruker Daltonics, Germany) via a CaptiveSpray nano-electrospray ion source. The timsTOF Pro2 operated in Data-Dependent Parallel Accumulation-Serial Fragmentation (PASEF) mode, performing 10 PASEF tandem mass spectrometry (MS/MS) frames within a single complete cycle. The capillary voltage was set to 1400 V, capturing MS and MS/MS spectra from 100 to 1700 m/z. The ion mobility range (1/K0) spanned from 0.7 to 1.4 Vs/cm2. TIMS accumulation and ramp times were both set at 100 ms, facilitating operation at nearly 100% duty cycles. A “target value” of 10,000 was applied for a repeated schedule, with an intensity threshold of 2500. The collision energy was adjusted linearly with mobility, from 59 eV at 1/K0 = 1.6 Vs/cm2 to 20 eV at 1/K0 = 0.6 Vs/cm2. Quadrupole isolation widths were set at 2Th for m/z < 700 and 3Th for m/z > 800. 

In diaPASEF mode, instrument control software enhancements allowed for the definition of quadrupole isolation windows based on TIMS scan time. This ensured seamless and synchronous ramping of all applied voltages through modifications in the instrument control electronics. We defined 25 Th isolation windows from m/z approximately 400 to 1200, with a total of 64 windows specified. All other parameters remained consistent with the DDA-PASEF mode. 

Database Search and Quantification for Proteomic Analysis

MS raw data were analyzed using DIA-NN version 1.8.1 using a library-free method. The Homo sapiens SwissProt database (20,425 entries) facilitated the creation of a spectral library using deep learning algorithms of neural networks. The MBR parameter option was leveraged to generate a spectral library from DIA data, which was then reanalyzed using this library. Protein groups were identified with a false discovery rate (FDR) of 1% at both peptide and protein levels. 

Sample Preparation and Extraction for Metabolomics Analysis

Quantification of PI and AH metabolites was conducted using a targeted metabolomics approach on the ABSciex QTRAP 6600+ LC-MS/MS system. Initially, samples were defrosted on ice and subjected to a 10-second vortex. A mixture of 50 µL sample and 300 µL extraction solvent (ACN: methanol in a 1:4 ratio, V/V) with added internal standards was prepared in a 2 mL Eppendorf tube. This mixture was then vortex-mixed for 3 minutes, followed by centrifugation at 12,000 g for 10 minutes at 4°C. Subsequently, 200 µL of the supernatant for PI samples and 150 µL for AH samples were collected and chilled at −20°C for 30 minutes before a final centrifugation at 12,000 g for 3 minutes at 4°C. The resulting supernatants (180 µL for PI samples and 120 µL for AH samples) were then readied for LC-MS analysis. 

HPLC for Metabolomic Analysis

Chromatographic separation was achieved using a Waters ACQUITY UPLC BEH C18 column (1.8 µm, 2.1 mm × 100 mm) with a maintained column temperature of 40°C. The flow rate was set at 0.4 mL/min, and 2 µL of each sample was injected. The mobile phase comprised water (0.1% formic acid) and acetonitrile (0.1% formic acid) in a gradient elution program: starting from 95:5 (A/B, V/V) at 0 minutes, changing to 80:20 at 2.0 minutes, 40:60 at 5 minutes, and 1:99 at 6 minutes, maintaining at 1:99 until 7.5 minutes, and finally returning to 95:5 at 7.6 minutes and holding until 10.0 minutes. 

MS Conditions for Metabolomics Analysis

Metabolic extracts were subjected to analysis using reversed-phase liquid chromatography (RPLC) coupled with mass spectrometry (MS) in both positive and negative ionization modes. The AB Sciex TripleTOF 6600 mass spectrometers were utilized, operating in full MS-scan mode for data collection. Data acquisition was conducted in information-dependent acquisition (IDA) mode, using Analyst TF 1.7.1 Software (Sciex, Concord, ON, Canada). Source parameters were adjusted as follows: ion source gas 1 (GAS1) at 50 psi; ion source gas 2 (GAS2) at 50 psi; curtain gas (CUR) at 25 psi; temperature (TEM) set to 550°C; declustering potential (DP) at 60 V or −60 V for positive or negative modes, respectively; and ion spray voltage floating (ISVF) at 5000 V or −4000 V for positive or negative modes, respectively. Time of flight (TOF) MS scan settings included a mass range of 50 to 1000 Da, an accumulation time of 200 ms, and dynamic background subtraction activated. Product ion scan settings comprised a mass range of 25 to 1000 Da, accumulation time of 40 ms, collision energy of 30 or −30 V for positive or negative modes, respectively, collision energy spread of 15, resolution set to UNIT, charge state from 1 to 1, intensity at 100 cps, isotopes excluded within 4 Da, mass tolerance at 50 ppm, and a maximum of 18 candidate ions monitored per cycle. 

Identification of Metabolomic Signatures

Metabolomic data were first transformed into mzXML format using ProteoWizard¹¹ (version 3.0.9134, November 11, 2015, http://proteowizard.sourceforge.net). Following this conversion, the XCMS software suite was used for data handling, including peak detection and data alignment based on mass-to-charge ratio (m/z ± 25 ppm) and retention times (RTs ± 6 secondds). This process facilitated the extraction of metabolomic signatures characterized by distinct m/z values and retention periods. Adjustments to the XCMS outputs were made to ensure metabolites quantified in fewer than 50% of samples within a group were normalized, using the Support Vector Regression (SVR) method for peak area calibration. 

Metabolite identification utilized a comprehensive four-stage protocol. Initially, an in-house metabolite database, featuring both chemical standards and a manually verified list of compounds distinguished by their accurate mass (m/z, ±25 ppm), retention times, and spectral signatures, served as the foundation for level 1 identification as per Microsatellite Instability (MSI)¹² criteria. Subsequently, accurate mass and MS/MS spectra comparisons with public repositories, such as HMDB (http://www.hmdb.ca/), MassBank (http://www.massbank.jp/), and MoNA (http://mona.fiehnlab.ucdavis.edu/) facilitated additional metabolite recognitions. Using a forward dot-product algorithm¹³ with a similarity score threshold of 0.5 enabled this, where matches underwent manual verification for level 2 identification following MSI¹² guidelines. Unmatched metabolic peaks in public databases were further analyzed using MetDNA¹⁴ and the in silico tool CFM-ID,¹⁵ which were assigned as level 3 identifications. Peaks identified solely through MS1 data were tagged as level 4 identifications. In cases of disparate identification methods yielding matches for a single metabolic feature, prioritization was given according to the identification hierarchy: standards > MS/MS > MetDNA. 

Statistical Analyses

For statistical analyses, Python version 3.6 and R version 4.2.1 were used. The paired t-test analyzed differences between groups for normally distributed data, whereas the Wilcoxon signed-rank test was used for non-normal data. The P values underwent adjustment through the Benjamini-Hochberg FDR method in omics’ comparison. Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA) was applied, focusing on Variable Importance in Projection (VIP) values. Metabolites with VIP > 1 and P value < 0.05 were identified as differentially expressed metabolites (DEMs). Similarly, proteins with a fold change > 1.2 or < 0.8, and P value < 0.05 were classified as differentially expressed proteins (DEPs). Both DEMs and DEPs were analyzed for biological pathway enrichment using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. Spearman correlation analysis was applied to investigate the associations between pretreatment and post-treatment “changes of” (Δ) nAMD lesions measurements and omics molecules metrics, considering statistical significance at an adjusted P value (adjusted by Storey-q method) of less than 0.05. To account for potential confounding factors, we conducted additional multivariable and univariate regression analyses, adjusting for systemic conditions, lifestyle factors, and anti-VEGF types (Datasets 1 and 2). 

Results

Patient Characteristics

As shown in Table 1, this study included 54 patients with nAMD with an average age of 64.80 ± 7.05 years, comprising 11 female patients (20.37%). The average disease duration was 11.09 ± 17.58 months, with 30 patients (55.56%) having type 1 MNV and 24 (44.44%) type 2 MNV. All patients received anti-VEGF 3+PRN treatment, including aflibercept (11 patients, 20.37%), conbercept (22 patients, 40.74%), and ranibizumab (21 patients, 38.89%). Besides, baseline best-corrected visual acuity (BCVA) averaged at 0.70 ± 0.43 logMAR and intraocular pressure (IOP) at 15.02 ± 1.83 mm Hg. Regarding systemic conditions, 20 patients (37.04%) had hypertension, 6 (11.11%) had diabetes, and 7 (12.96%) had hyperlipidemia. Lifestyle factors were reported in smoking (31 patients, 57.41%) and alcohol consumption (26 patients, 48.15%). 

Table 1.

 

Baseline Clinical Characteristics of Participants

 

Imaging Characteristics and Alteration During Anti-VEGF

Imaging alterations of nAMD observed between pre- and post-treatment are detailed in Table 2. Notably, there was a significant decrease in the volume of various morphologic lesions, except ΔPED, which remained unchanged (median interquartile range [IQR] = 0.000, IQR = 0.148 mm³, P = 0.813). Specifically, the changes were Δtotal lesions = −0.609 (1.689) mm³, ΔMNV = −0.116 (0.393) mm³, ΔSRF = −0.213 (0.503) mm³, ΔIRF = −0.017 (0.083) mm³, and ΔSHRM = −0.522 (1.300) mm³, all with P < 0.001. Additionally, the structural damages involving the EZ, ELM, and ONL demonstrated significant improvements post-treatment. Changes in these areas were quantified as −1.522 (4.028) mm² for ΔEZ, −0.571 (1.956) mm² for ΔELM, and −1.427 (5.823) mm² for ΔONL, all with P < 0.001. 

Table 2.

 

Alteration of Imaging Characteristics Between Pre- and Post-Treatment

 

AH Proteomics and Metabolomics Profiling and Alteration During Anti-VEGF

In our analysis, the AH samples revealed a total of 1350 proteins and 1268 metabolites. Using PCA and OPLS-DA, as depicted in Figures 2A to 2D, significant changes were noted in both protein and metabolite profiles when comparing pre- and post-treatment samples. The study identified 301 DEPs, with 129 showing increased expression and 172 decreased post-treatment (Fig. 2E). In the case of DEMs, 353 were found, including 272 with increased levels and 81 with decreased levels post-treatment (Fig. 2F). KEGG pathway analysis of DEPs indicated that anti-VEGF treating nAMD mainly involves pathways including cGMP-PKG signaling pathway, citrate cycle, cellular senescence, and phototransduction (Fig. 2G). For DEMs, the pathways impacted included glycolysis/gluconeogenesis, citrate cycle, HIF-1 signaling pathway, and AMPK signaling pathway (Fig. 2H). The combined analysis of DEPs and DEMs primarily highlighted the roles of anti-VEGF therapy of nAMD in angiogenesis regulation, energy metabolism, signal transduction, and neurofunctional regulation pathways (Fig. 2I). Figure 2J features a network graph that maps significant correlations between selected DEPs and DEMs. 

Figure 2.

 

Differential analysis and functional enrichment of aqueous humor molecules in nAMD during anti-VEGF treatment. (A, C) PCA 3D plots for AH proteins and metabolites of “pretreatment versus post-treatment,” respectively. (B, D) OPLS-DA plots for AH proteins and metabolites of “pretreatment versus post-treatment,” respectively. (E, F): Volcano plots highlighting significant differences in AH proteins and metabolites of “pretreatment versus post-treatment,” respectively. (G, H) KEGG analyses of differential AH proteins and metabolites, respectively, between pretreatment and post-treatment. (I) Integrated KEGG pathway analysis combines both differential AH proteins and metabolites between pretreatment and post-treatment.” (J) Network plot illustrates significant correlations between selected AH DEPs and DEMs; ellipse denotes AH DEPs and rhombus indicates AH DEMs; pink denotes upregulation and green indicates downregulation. Abbreviations: AH, aqueous humor; nAMD, neovascular age-related macular degeneration. Full names of selected molecules: A1BG, alpha-1-B glycoprotein; ARHGDIA, rho GDP dissociation inhibitor alpha; ASPH, aspartate beta-hydroxylase; CST3, cystatin C; EPDR1, ependymin related 1; FGA, fibrinogen alpha chain; IGHD, immunoglobulin heavy constant delta; IGHV1-8, immunoglobulin heavy variable 1-8; ITM2B, integral membrane protein 2B; MW0008696, N-(2,2,2-Trifluoroethyl)-N-{4-[2,2,2-trifluoro-1-hydroxy-1-(trifluoromethyl)ethyl]phenyl}benzenesulfonamide; MW0149286, (2S,3S)-2-(dimethylamino)-N-[(2Z,6S,9S,10S)-6-isobutyl-10-isopropyl-5,8-dioxo-11-oxa-4,7-diazabicyclo[10.2.2]hexadeca-1(14),2,12,15-tetraen-9-yl]-3-methyl-pentanamide; MW0169901, epipodophyllotoxin, 4'-demethyl-, 9-(4,6-O-2-thenylidene-beta-D-glucopyranoside); RAB1A, RAB1A, member RAS oncogene family; SKP1, S-phase kinase-associated protein 1; SYNCRIP, synaptotagmin binding cytoplasmic RNA interacting protein; TALDO1, transaldolase 1; YIPF3, Yip1 domain family member 3.

Spearman Correlation of ΔnAMD Lesion and AH ΔProteins/Metabolites

Figure 3A presents the Spearman correlation analysis between nAMD lesion change and AH proteins/metabolites change (each change calculated as post-treatment values subtracted from pretreatment values). The correlation heatmap featuring 65 proteins and 2 metabolites changes, displays adjusted statistically significant correlation with at least one type of nAMD lesion changes. In the heatmap, deeper colors denote stronger correlations, indicating that the changes in proteins/metabolites are primarily associated with changes in total lesion and SHRM. 

Figure 3.

 

Integrative analysis of aqueous humor δmolecules in relation to ΔnAMD lesion. (A) Spearman correlation heatmap between ΔnAMD lesion and AH Δproteins/metabolites, featuring 65 proteins and 2 metabolites changes adjusted statistically significant correlation with at least one type of ΔnAMD lesion; deeper red denotes stronger positive correlations and deeper blue indicates stronger negative correlations. (B) Concordant Venn diagram of AH proteins/metabolites significant differences and correlation with ΔnAMD lesions. (C) Contradictory Venn diagram of AH proteins/metabolites significant differences and correlation with ΔnAMD lesions. Abbreviations: AH, aqueous humor; ELM, external limit membrane; EZ, ellipsoid zone; IRF, intraretinal fluid; MNV, macular neovascularization; nAMD, neovascular age-related macular degeneration; ONL, outer nuclear layer; PED, pigment epithelial detachment; SHRM, subretinal hyper-reflective material; SRF, subretinal fluid; VEGF, vascular endothelial growth factor; Δ, change. Full names of selected molecules: A0A0G2JRQ6, Ig-like domain-containing protein; ACTB, actin, cytoplasmic 1; AEBP1, adipocyte enhancer-binding protein 1; A1BG, alpha-1-B glycoprotein; APLP1, amyloid beta precursor-like protein 1; APOA4, apolipoprotein A-IV; ARHGDIA, rho GDP dissociation inhibitor alpha; ASPH, aspartate beta-hydroxylase; ATIC, 5-aminoimidazole-4-carboxamide ribonucleotide formyltransferase/IMP cyclohydrolase; BLVRB, biliverdin reductase B; CAPG, macrophage capping protein; CLU, clusterin; CLSTN3, calsyntenin 3; COLEC12, collectin sub-family member 12; CPQ, carboxypeptidase Q; CST3, cystatin C; Cys-Gly, cysteinylglycine; DAG1, dystroglycan 1; EPB41, erythrocyte membrane protein band 4.1; EPDR1, ependymin related 1; EFEMP2, EGF containing fibulin extracellular matrix protein 2; F12, coagulation factor XII; F13B, coagulation factor XIII B chain; F5, coagulation factor V; FGA, fibrinogen alpha chain; FGG, fibrinogen gamma chain; FN1, fibronectin 1; FSTL5, follistatin-like 5; GM2A, GM2 ganglioside activator; GPNMB, glycoprotein nmb; GPX3, glutathione peroxidase 3; GRXCR2, glutaredoxin, cysteine rich 2; GSN, gelsolin; HSPA13, heat shock protein 70 kDa family, member 13; IGHD, immunoglobulin heavy constant delta; IGHV1-8, immunoglobulin heavy variable 1-8; IGKV4-1, immunoglobulin kappa variable 4-1; IGLV2-8, immunoglobulin lambda variable 2-8; ITM2B, integral membrane protein 2B; KNG1, kininogen 1; LAMC1, laminin subunit gamma 1; LYPD2, LY6/PLAUR domain containing 2; MFAP2, microfibril associated protein 2; NCAM2, neural cell adhesion molecule 2; OLFM2, olfactomedin 2; PAPLN, papilin, proteoglycan-like sulfated glycoprotein; PGAM1, phosphoglycerate mutase 1; PFN1, profilin 1; PGLYRP2, peptidoglycan recognition protein 2; PKM, pyruvate kinase, muscle; PTGDS, prostaglandin D2 synthase; PZP, pregnancy zone protein; RAB1A, RAB1A, member RAS oncogene family; RELN, reelin; SKP1, S-phase kinase-associated protein 1; SYNCRIP, synaptotagmin binding cytoplasmic RNA interacting protein; TALDO1, transaldolase 1; YIPF3, Yip1 domain family member 3.

Concordant Molecules: Anti-VEGF Treatment-Induced Changes Align With nAMD Lesion Regression

To identify molecules potentially enhancing lesion regression in nAMD during anti-VEGF treatment, we performed intersection analyses of significantly altered proteins/metabolites with changes concordant to lesion regression. Specifically, two intersections were analyzed: the first involved upregulated proteins/metabolites that negatively correlated with lesion changes, yielding zero significant findings. The second intersection focused on downregulated proteins/metabolites that positively correlated with lesion changes, resulting in eight proteins: FGA, TALDO1, ASPH, IGHD, A1BG, CST3, IGHV1-8, and SKP1 (Fig. 3B, Table 3). Among these, three AH proteins showed significant correlations with at least two types of nAMD lesions: ΔFGA with ΔTotal Lesion, ΔSHRM and ΔONL; ΔTALDO1 with ΔTotal Lesion and ΔONL; and ΔASPH with ΔTotal Lesion and ΔSHRM. Figures 4A to 4C presents detailed information about these proteins. 

Table 3.

 

Concordant Molecules: Anti-VEGF Treatment-Induced Changes Align With nAMD Lesion Regression

 

Figure 4.

 

Changes in FGA, TALDO1, ASPH , and YIPF3 pre- and post-treatment and their correlation with ΔnAMD lesion. (A1-D1) Violin-box plots of AH FGA, TALDO1, ASPH, and YIPF3 levels pretreatment and post-treatment, showing a significant downregulation of these four proteins after anti-VEGF treatment. (A2-A4) Correlation scatter plots of ΔFGA with ΔTotal lesion, ΔSHRM, and ΔONL, illustrating that greater reductions in FGA are associated with more extensive regression of nAMD lesions. (B2-B4) Correlation scatter plots of ΔTALDO1 with ΔTotal lesion and ΔONL, indicating that greater downregulation of TALDO1 correlates with more pronounced nAMD lesion regression. (C2-C3) Correlation scatter plots of ΔASPH with ΔTotal lesion and ΔSHRM, suggesting that more substantial decreases in ASPH are linked to greater nAMD lesion reduction. (D2-D3) Correlation scatter plots of ΔYIPF3 with ΔTotal lesion and ΔSHRM, revealing that more significant downregulation of YIPF3 is associated with less nAMD lesion regression. Abbreviations: ONL, outer nuclear layer; SHRM, subretinal hyper-reflective material; SRF, subretinal fluid; VEGF, vascular endothelial growth factor; Δ, change. Full names of selected molecules: ASPH, aspartate beta-hydroxylase; FGA, fibrinogen alpha chain; TALDO1, transaldolase 1; YIPF3, Yip1 domain family member 3.

Contradictory Molecules: Anti-VEGF Treatment-Induced Changes Oppose nAMD Lesion Regression

Conversely, to identify molecules potentially inhibiting lesion regression during anti-VEGF therapy, we conducted intersection analyses for proteins/metabolites with changes contradictory to lesion regression. This analysis was also divided into two intersections: the first involved upregulated molecules positively correlating with lesion changes, identifying three proteins: RAB1A, ARHGDIA, and SYNCRIP. The second intersection included downregulated molecules negatively correlating with lesion changes, identifying three proteins: YIPF3, ITM2B, and EPDR1. These findings are detailed in Figure 3C and Table 4. Among these, one AH protein showed significant correlations with at least two types of nAMD lesions: ΔYIPF3 with ΔTotal Lesion and ΔSHRM and ΔONL (Fig. 4D). 

Table 4.

 

Contradictory Molecules: Anti-VEGF Treatment-Induced Changes Oppose nAMD Lesion Regression

 

Stable Molecules: Unchanged Post-Anti-VEGF Treatment Yet Correlated With nAMD Lesion Regression

Among the changes of 65 proteins and 2 metabolites significantly correlated with the changes of various nAMD lesions, 53 molecules remained stable, showing no significant alterations post-anti-VEGF treatment (Table 5). This group comprises 51 proteins and 2 metabolites. Notably, within this subset, changes of 10 molecules were found to be significantly correlated with both changes of total lesion and SHRM. These include nine proteins (OLFM2, EPB41, F12, VTN, GM2A, SDC1, GPX3, YWHAE, and CLSTN3) and one metabolite (m-Cresol). 

Table 5.

 

Stable Molecules: Unchanged Post-Anti-VEGF Treatment Yet Correlated With nAMD Lesion Regression

 

Discussions

Our study analyzing patients with nAMD pre- and post-anti-VEGF therapy, detected significant changes in 301 AH proteins and 353 metabolites, linked to pathways including angiogenesis regulation, energy metabolism, signal transduction, and neurofunctional regulation. Changes in 65 AH proteins and 2 metabolites correlated significantly with at least one type of nAMD lesion alteration. Among them, the decrease in proteins FGA, TALDO1, and ASPH may contribute to the mechanism of at least two types of nAMD lesion regression induced by anti-VEGF treatment, whereas the decrease in protein YIPF3 could be related to poor treatment response of at least two types of lesion regression. Additionally, several proteins and metabolites potentially involved in lesion regression, yet they were unaffected by anti-VEGF. This exploration enhances understanding of the intraocular molecular changes in response to nAMD lesions during anti-VEGF therapy, offering new insights into treatment action and resistance mechanisms, and guiding the identification of potential novel therapeutic targets. 

High-throughput proteomic and metabolomic analyses of AH samples revealed hundreds of proteins and metabolites that underwent significant changes during anti-VEGF therapy of patients with nAMD. KEGG enrichment analysis of these significantly altered molecules suggested that the therapeutic pathway of anti-VEGF in nAMD extends beyond angiogenesis regulation. It appears to be involved in energy metabolism, including pathways like the TCA cycle and AMPK signaling, indicating that anti-VEGF may promote the repair of damaged retinal cells by enhancing energy metabolism efficiency and regulating energy balance.^16,17 It also seems to participate in signal transduction, involving pathways such as the phosphatidylinositol signaling system, insulin signaling pathway, and cAMP signaling pathway, which are likely key routes through which anti-VEGF influences cell survival, proliferation, and etc.^18,19 Furthermore, it appears to regulate neurofunctional pathways, including dopaminergic synapse and GABAergic synapse, suggesting that anti-VEGF treatment impacts retinal neurofunction, potentially contributing to the protection and functional recovery of retinal neurons.²⁰ This comprehensive analysis provides a deeper understanding of the molecular mechanisms underlying anti-VEGF therapy in nAMD. 

In the nAMD lesions analyses, we used three-dimensional reconstruction of OCT and OCTA images to delineate different types of nAMD lesions. We observed a significant regression in the volume or area of seven out of eight types of lesions (except PED), and the total lesion during the anti-VEGF treatment, corroborating previous research findings.²¹ To investigate the molecular participants involved in lesion regression during treatment, we conducted a correlation analysis between changes in AH molecules and alterations in nAMD lesions. After adjusting for multiple comparisons, 65 proteins and 2 metabolites showed significant correlations with the regression of at least one type of nAMD lesion. Proteins play a pivotal role in biological functions, such as angiogenesis, inflammatory response, and apoptosis, and are direct targets of anti-VEGF therapy in nAMD, which may explain their strong correlation with lesion regression.²² The observed protein changes, which likely reflect the activity and modification of biological pathways during treatment, suggest a closer association with lesion alterations than metabolites.²³ In contrast, metabolites, which are often the end products of biological processes, may reflect the final changes in cellular states rather than the pathophysiological processes themselves. This suggests that alterations in metabolites could be an indirect result of the pathological process, rather than a direct driving force.⁶ Therefore, although metabolites are crucial in the overall disease landscape, their role in specific treatment responses may not be as direct or significant as that of proteins. 

In the concordant intersection analysis of AH molecules exhibiting both significant changes post-anti-VEGF therapy and significant correlations with nAMD lesion regression, eight proteins were identified as potential contributors to the treatment’s efficacy in lesion reduction. Notably, three of these significant proteins – FGA, TALDO1, and ASPH – were found to be downregulated and associated with the regression of at least two types of nAMD lesions. FGA plays a role in wound healing and inflammation, and its overexpression can stimulate angiogenesis via VEGF-A^24–26; thus, its marked reduction during anti-VEGF therapy, correlating with significant regression in total lesion, SHRM, and ONL damage, suggests FGA as a potential key protein in the mechanism of anti-VEGF action. TALDO1, a key enzyme in the pentose phosphate pathway, has been linked to adverse outcomes in certain cancers, implying its role in abnormal cell proliferation.^27–29 Its significant decrease during therapy correlating with the regression of total lesion and ONL damage, indicates its potential role in anti-VEGF therapy. ASPH, known for enhancing cell proliferation, migration, and invasiveness, also showed a significant decrease correlated with the regression of total lesion and SHRM, further suggesting its importance in the anti-VEGF mechanism.^30–32 However, whether these proteins indeed play a role in the therapeutic effect of anti-VEGF, or are merely concomitant protein changes during nAMD lesion regression, requires further investigation to elucidate these mechanisms. 

In our search for molecules potentially linked to poor response to anti-VEGF therapy, our analysis identified six proteins in the contradictory intersection of AH molecules. Of these, the protein YIPF3 exhibited a paradoxical relationship with the regression of at least two types of nAMD lesions. YIPF3, a small membrane protein, plays a role in membrane transport between the endoplasmic reticulum and Golgi apparatus.^33,34 Previous studies have shown that the knockout of YIPF3 leads to Golgi fragmentation, indicating its role in maintaining Golgi structure.³⁵ During anti-VEGF therapy for nAMD, a significant downregulation of intraocular YIPF3 was observed. The greater the reduction of YIPF3, the less pronounced the regression of total lesion and SHRM, suggesting that YIPF3 plays a reverse role in lesion regression. Given YIPF3’s critical role in maintaining Golgi structure and membrane transport, its downregulation may disrupt cellular processes essential for effective response to anti-VEGF therapy. Therefore, a decrease of YIPF could be involved in the mechanisms of poor response or resistance to anti-VEGF treatment, warranting further investigation to clarify the specific roles. Additionally, our study identified 51 proteins and 2 metabolites significantly correlated with nAMD lesion regression but unaffected by anti-VEGF treatment. Among these, 10 molecules were significantly related to the regression of at least two types of nAMD lesions. Although these molecules did not show significant changes during the treatment, they may still play a role in the process of lesion regression or be associated with treatment inefficacy, which also merits further exploration. 

Our study has several limitations. First, the small sample size may affect the generalizability of our findings. Second, our study included three different anti-VEGF medications, and participants had varying systemic conditions and lifestyles which could potentially influence the results. However, our additional analyses (Datasets 1 and 2) showed that these factors, including the choice of anti-VEGF type, had minimal impact on the main findings. Future studies specifically designed to compare the molecular mechanisms of different anti-VEGF drugs would be valuable to deepen our understanding of the heterogeneity in drug mechanism. Third, although we observed significant correlations between identified proteins and metabolites and nAMD lesion regression, these correlations do not necessarily imply causation. 

Despite these limitations, by integrating longitudinal three-dimensional lesion analysis and multi-omics analysis, our study provides valuable insights into the mechanisms underlying the efficacy and poor response to anti-VEGF treatment for nAMD. This research lays a solid foundation for future investigations, yet additional studies are essential to elucidate the causal relationships between the identified molecular changes and nAMD lesion regression. 

Acknowledgments

Supported by the National Natural Science Foundation of China (Grant No. 81970808). The authors are also grateful for the assistance provided by Wei Zhao and Xiqiang Hong from the Wuhan Metware Biotechnology Co., Ltd. for their kind help with proteomics and metabolomics analysis. 

Author Contributions: Xuenan Zhuang and Xiongze Zhang conceptualized the study. Data curation was performed by Xuenan Zhuang, Liang Zhang, Lan Mi, Hui Chen, Miaoling Li, Liwei Yao, Yuying Ji, and Honghua Yu. Funding acquisition was managed by Feng Wen. Investigation was conducted by Xuenan Zhuang, Yining Zhang, Guiqin He, Xuelin Chen, Yongyue Su, Yuhong Gan, Jiaxin Pu, and Xinlei Hao. Methodology was designed by Xuenan Zhuang and Feng Wen. Project administration was handled by Xuenan Zhuang and Feng Wen. The study was supervised by Xiongze Zhang and Feng Wen. Formal analysis and validation were done by Xiongze Zhang. Visualization was performed by Xuenan Zhuang. The original draft was written by Xuenan Zhuang and reviewed and edited by Xiongze Zhang, Miaoling Li, Yongyue Su, Yuhong Gan, Chengguo Zuo, and Feng Wen. 

Disclosure: X. Zhuang, None; M. Li, None; L. Mi, None; X. Zhang, None; J. Pu, None; G. He, None; L. Zhang, None; H. Yu, None; L. Yao, None; H. Chen, None; Y. Ji, None; C. Zuo, None; Y. Su, None; Y. Gan, None; X. Hao, None; Y. Zhang, None; X. Chen, None; F. Wen, None