Comprehensive Proteomic Profiling of Exfoliation Glaucoma Via Mass Spectrometry Reveals SVEP1 as a Potential Biomarker

Abstract

Purpose: This study investigated the proteomic landscape of exfoliation glaucoma to find potential biomarkers.

Methods: The study enrolled 34 patients diagnosed with either exfoliation syndrome with/without glaucoma or age-related cataract. Plasma proteins were analyzed through mass spectrometry and Mendelian randomization (MR) based on data from deCODE, FinnGen, Atherosclerosis Risk in Communities (ARIC), eQTLGen, and UK Biobank (UKB) cohorts to infer relationships.

Results: Among 2025 plasma proteins analyzed, 130 were differentially expressed in the exfoliation glaucoma group, which exhibited elevated intraocular pressure. Our proteomics data suggested that infection, immune responses including intestinal immune network, endocrine hormones, and complement and coagulation cascades are involved in the development of exfoliation glaucoma. Notably, there was a significant correlation between SVEP1 and exfoliation glaucoma (odds ratio [OR] = 1.20, 95% confidence interval [CI] = 1.10 to 1.31, P = 0.0000428), with findings corroborated in an independent cohort. Further analysis predicted a protective role of LOXL1-AS1 in exfoliation glaucoma through its regulation of SVEP1 expression. In MR phenome-wide association studies, SVEP1 was associated with complications of exfoliation glaucoma. After multiple testing corrections, there was a tendency for SVEP1 to be associated with glaucoma (OR = 1.14, 95% CI = 1.11 to 1.16, P = 0.0000003) and type 2 diabetes (OR = 1.07, 95% CI = 1.05 to 1.08, P = 0.0000067).

Conclusions: Plasma proteomic analysis reveals that high expression of SVEP1 is a risk factor for exfoliation glaucoma, which potentially affects diabetes and is affected by estradiol or LOXL1-AS1. However, further research is needed to establish causality.

Glaucoma, as a neurodegeneration disease, is the leading cause of irreversible blindness globally affecting approximately 95 million people, thus early detection and effective treatment are essential.¹ Exfoliation glaucoma (XFG), the predominant cause of secondary glaucoma, associated with principal susceptibility gene lysyl oxidase like 1 (LOXL1), appears to have worse prognosis.^1,2 XFG is the ocular manifestation of exfoliation syndrome (XFS), which is a systemic disease associated with increased risk of vascular disease.^3,4 In addition, the polymorphisms in the long non-coding RNA LOXL1-antisense 1 (LOXL1-AS1) is a risk factor for XFG.⁵ However, it is still unclear what the underlying mechanisms of XFG are. Therefore, we need the results of further experimental and clinical studies to elucidate the potential pathophysiological processes of XFG. 

Recently, human plasma proteomics, the key biomarkers for diagnosis and prognosis in precision medicine, have made great achievements, such as the maturation of new high-throughput technologies and updates of the Human Plasma Proteome Project (HPPP).⁶ In recent years, there are many researches about plasma protein and eye disease. By investigating the plasma proteomics, it is beneficial for us to understand the mechanism of XFG pathogenesis. Moreover, Mendelian randomization (MR) is a robust computational methodology that uses single nucleotide polymorphisms (SNPs) from quantitative trait loci (QTL) and genome-wide association studies (GWAS) summary statistics to analyze relationships between exposures and outcomes.⁷ 

Here, we aimed to identify comprehensive plasma proteome profiling of XFG by the data-independent acquisition mass spectrometer (DIA-MS). With the proteomics data, we analyzed plasma proteins significantly associated with incident XFG by our observational studies and the biologic processes and pathways of these proteins regarding XFG pathogenesis. Furthermore, we explored the relationships between specific proteins and diseases by the MR approach, and the potential mechanisms led to XFG. In addition, we conducted an MR phenome-wide association study (MR-PheWAS) and a review of current studies of biomarkers for XFG in plasma or serum. Consequently, we undertook an integrated analysis of observational and genetic data of proteins that could serve as potential drug targets, providing new insights for the clinical prevention and treatment of XFG. 

Methods

Study Approval

The First People’s Hospital of Kashi Prefecture Ethics Committee approved this study that was conducted in accordance with the guidelines of the Declaration of Helsinki. This article does not contain identifiable data for any participant. All publicly available summarized GWAS data sources in this study were approved by the respective institutional ethics committees. We summarized the methodological framework in Figure 1. 

Figure 1.

 

Flowchart of the identification and integration of the differentially expressed proteins by DIA-MS and MR. ARIC = Atherosclerosis Risk in Communities Study; DIA-MS = data-independent acquisition mode mass spectrometry; LOXL-AS1 = lysyl oxiase-like 1 antisense RNA 1; MR = Mendelian randomization; OR = odds ratio; PheWAS = Phenome-Wide Association Studies; pQTL = protein quantitative trait locus; SVEP1 = Sushi, von Willebrand factor type A, EGF and pentraxin domain containing 1; UKB-PPP = UKB Pharma Proteomics Project.

Patients and Specimens

In total, we collected 34 blood samples of patients with XFS and age-related cataract (ARC) controls between February and May in 2023 in the First People’s Hospital of Kashi Prefecture in Northwest China which were categorized as follows: XFG (n = 10), XFS (n = 12), and ARC (n = 12). All patients were admitted to the hospital for surgery on similar dates. The inclusion criteria are described in the Supplementary Methods. 

Blood samples were collected after the patients’ overnight fasting under routine clinical blood guidance using Vacutainer K2 EDTA tubes (Becton Dickinson, Franklin Lakes, NJ, USA). We centrifuged blood samples at 2000 relative centrifugal force (rcf) for 15 minutes at 4°C after clotting at room temperature and then immediately aliquoted and stored plasma into sterile centrifuge tubes at −80°C for future analysis. 

Data-Independent Acquisition Mode Mass Spectrometry Analysis

We lysed plasma samples with Dissolution Buffer lysis buffer (8 M Urea, 100 mM triethylammonium bicarbonate; Sigma/T7408-500ML, pH 8.5) and then centrifuged the samples at 12,000 g for 15 minutes at 4°C and the supernatant was reduced with 1M DL-Dithiothreitol (Sigma/D9163-25G) for 1 hour at 56°C, and subsequently alkylated with sufficient iodoacetamide (Sigma/I6125-25G) for 1 hour at room temperature in the dark followed by an ice-bath for 2 minutes. After protein quality test (described in the Supplementary Methods), we performed tryptic digestion at 37°C for 4 hours and then added trypsin (Promega/V5280) and CaCl₂ for digesting overnight. We mixed formic acid with the digested sample, adjusted pH < 3, and centrifuged at 12,000 g for 5 minutes at room temperature. We loaded the supernatant to the C18 desalting column slowly, washed with washing buffer (0.1% formic acid and 3% acetonitrile) 3 times, then we added elution buffer (0.1% formic acid and 70% acetonitrile). We collected and lyophilized the eluents of each sample. We dissolved the lyophilized powder using 0.1% formic acid, centrifuged at 14,000 g for 20 minutes at 4°C, and injected 200 ng of the supernatant sample into the sample for liquid-quality detection. 

We used a Vanquish Neo upgraded ultra-HPLC (UHPLC) system with a C18 pre-column of 174500 (5 mm × 300 µm, 5 µm, thermo) heated at 50°C in a column oven, and a C18 analytical column of ES906 (PepMap TM Neo UHPLC 150 µm × 15 cm, 2 µm, thermo), an Orbitrap Astral mass spectrometer (Thermo Fisher Scientific, USA) in Novogene Co. Ltd. using DIA and an Easy-spray ion (ESI) source with the ion spray voltage 1.9 kilovolt (kV), the ion transfer tube temperature was 290°C, and a full first-stage mass spectrometry scanning range of m/z 380 to 980. The elution conditions of the liquid chromatography were as shown in the Supplementary Methods. 

We set the primary MS resolution to 240,000 (200 m/z), AGC to 500%, the parent ion window size to 2-Th, the number of DIA windows to 300, the NCE to 25%, the secondary m/z acquisition range from 150 to 2000, the sub-ion resolution Astral to 80000, and the maximal injection time was 3 ms. 

The mass spectrometry detection raw files (Project ID: IPX0008349000, submitted at iProX^8,9) were searched and analyzed using the DIA-NN library search software. More details are described in the Supplementary Methods. 

Statistics

Protein quantification results were statistically analyzed using the t-test, and if the data were found not to follow a normal distribution, then the Mann-Whitney U test was used instead. Differentially expressed proteins (DEPs) were defined as those with significant quantitative differences between the case and control groups (P < 0.05, fold change > 1.2 or fold change < 0.83). The comparison among groups was performed using the Chi-square test, Kruskal–Wallis tests, or 1-way ANOVA. 

Pathway/Network Analysis

We performed comprehensive functional analysis using the Gene Ontology (GO) and InterPro (IPR) databases through the InterProScan program. Additionally, we utilized the Clusters of Orthologous Groups (COG) and Kyoto Encyclopedia of Genes and Genomes (KEGG) databases to investigate protein families and pathways. The DEPs were used for volcano plot analysis, cluster heatmap analysis, and enrichment analysis of GO, IPR, Gene Set Enrichment Analysis (GSEA), and KEGG terms. The protein-protein interaction (PPI) analysis network was established by the Search Tool for the Retrieval of Interacting Genes (STRING) database and was visualized via Cytoscape. In addition, we used GOsemsim to predict the key proteins.¹⁰ 

Data Source

We used protein QTL (pQTL) in the deCODE and UK Biobank Pharma Proteomics Project (UKB-PPP) cohort for discovery study and then the cohort of Atherosclerosis Risk in Communities (ARIC) study for replication validation.^11–13 In addition, we used cis-expression QTL(eQTL) summary statistics from the eQTLGen Consortium.¹⁴ We used microbiota quantitative trait loci (mbQTL) of genus Senegalimassilia from the MiBioGen consortium.^15,16 We obtained GWAS summary statistics about XFG, POAG, ARC, and other cataracts from the DF9 release of the FinnGen Study.¹⁷ More details are described in the Supplementary Methods. 

Mendelian Randomization

We used a bidirectional two-sample MR approach. We excluded the palindromic or ambiguous SNP. The SNPs with potential confounding factors (P < 5 × 10⁻⁸) were screened by the Gene ATLAS database (http://geneatlas.roslin.ed.ac.uk/phewas/).¹⁸ We used the inverse variance weighted (IVW) method as our primary source of MR estimates, and when the number of inverse variances (IVs) was only one, we used the Wald ratio instead of the IVW method. We selected the one with the lowest P value for the following analysis if different SOMAmers or Olink antibodies shared the same pQTL.¹⁹ More details are described in the Supplementary Methods. 

Sensitivity Analysis

We performed sensitivity analyses using complementary approaches that are robust to violation of MR assumptions, including MR Egger intercept test and Mendelian Randomization Pleiotropy RESidual Sum and Outlier test (MR-PRESSO) to evaluate the potential presence of horizontal pleiotropy. In addition, a leave-one-out analysis was conducted to assess the influence of potentially pleiotropic IVs on the estimates. We also used the Radial MR method to detect outliers and repeated MR analysis after excluding outliers. We analyzed Steiger filtering directionality test to explore the robustness of the effect direction. The Cochran's Q heterogeneity test was used to test the heterogeneity between casual estimates. The strength of the IVs was evaluated with F statistics, whereas if the value of the F statistic was > 10 it was considered that there was no significant weak instrumental bias of the estimates. 

Mediation Analysis

We used a mediation analysis to investigate the effect of an exposure variable on an outcome variable through a mediating variable. We initially extracted relevant coefficients and standard errors from two separate datasets. To ensure consistency in our analysis, we only included effect sizes calculated using the IVW method. We calculated Mediated Effect = Coefficient_{Exposure-Mediator} × Coefficient_{Mediator-Outcome}, and further calculated the Z-score = mediated effect/ standard error of the mediated effect. If P = 2 × (1-normal distribution cumulative probability function at |Z|) was less than 0.05, the mediated effect was considered significant. 

RNA-Sequencing of the LOXL1-AS1/SVEP1 Pathway

We collected and analyzed the data of RNA-sequencing from a previous study that contained a LOXL1-AS1 targeted siRNA knockdown model and an overexpression model built by vectors carrying LOXL1-AS1, Δ14-LOXL1-AS1, or the compliment sequence, pENTR1A plasmids containing the LOXL1 promoter, which were transfected into immortalized human lens epithelial (HLE-B3) cells.²⁰ 

Mendelian Randomization Phenome-Wide Association Study

To identify comorbidities caused by SVEP1, we performed an MR-PheWAS using phenome-wide association study summary statistics based on the UKB cohort and the FinnGen cohort (R9). The UK Biobank data supplied 1403 electronic health record (EHR)-derived International Classification of Diseases (ICD)-based phenotypes for 20 million imputed variants in 400,000 White British individuals (https://pheweb.org/UKB-SAIGE/).²¹ In addition, the FinnGen data (R9) supplied 2272 phenotypes. In the MR analysis, we used the IVW method as the main outcome indicator and P < 0.05 for the IVW method was suggested for the nominal significant. Furthermore, we used multiple testing corrections and considered P < 0.05/3675 as a threshold to determine the statistical significance. 

Results

Clinical Characteristics of Patients

Our study included 34 patients without significant differences in age, gender, body mass index, and axial length (Supplementary Table S1). The median IOP at XFG was 28 millimeters of mercury (mm Hg) (interquartile range [IQR] = 26 to 42) was highest (P < 0.001), whereas in XFS it was 17 mm Hg (IQR = 15 to 18 mm Hg) and in ARC it was 16 mm Hg (IQR = 15 to 20 mm Hg). The median cornea endothelial cell density at XFG was 1847 mm² (IQR = 1752 to 2043), and was the lowest (P = 0.013), whereas in XFS it was 2222 mm² (IQR = 2145 to 2508), and in ARC it was 2328 mm² (IQR = 2265 to 2577). 

Plasma Proteomic Landscape of XFG/XFS

We identified 1668 proteins in XFG, 1738 proteins in XFS, and 1734 proteins in ARC (Fig. 2A). In total, we identified 2025 proteins which were distributed at approximately 7 orders of magnitude in plasma (range from 0.011 to 440,000 ng/mL), which included 717 protein biomarkers (Human Disease Plasma Protein Biomarker Database, online at http://biokb.ncpsb.org.cn/hdpp/#/) and according to Human Protein Atlas version 23.0 (https://www.proteinatlas.org/) 594 low abundance proteins (< 10 ng/mL) of 1319 proteins matched with the HPPP (Fig. 2B, Supplementary Table S2).²² The heatmap showed the protein xpression characteristics of each group (Fig. 2C). We found 1339 proteins between XFG and XFS (43 upregulated and 38 downregulated DEPs; Supplementary Table S3, Supplementary Fig. S1A), 1317 proteins between XFG and ARC (36 upregulated and 62 downregulated DEPs; Supplementary Table S4, Supplementary Fig. S1B) and 1362 proteins between XFS and ARC (27 upregulated and 55 downregulated DEPs; Supplementary Table S5, Supplementary Fig. S1C). Comparing XFS and ARC, 130 proteins were differentially expressed in the XFG group, which exhibited elevated intraocular pressure. 

Figure 2.

 

Plasma proteome determined by DIA-MS. (A) Venn diagram analysis of the number of plasma proteins in groups XFG, XFS, and ARC detected by DIA-MS. (B) Venn diagram analysis of the proteins identified by DIA-MS compared with the HPPP database. (C) Heatmap of the identified proteins in groups XFG, XFS, and ARC. (D) Series test of cluster for all significantly differentially expressed proteins in groups XFG, XFS, and ARC. ARC = age-related cataract; DIA-MS = data-independent acquisition mode mass spectrometry; HPPP = Human Plasma Proteome Project; XFG = exfoliation glaucoma; XFS = exfoliation syndrome.

Comparing XFG and XFS, the top 5 upregulated DEPs were protein HEG homolog 1, transforming growth factor beta-1 proprotein, immunoglobulin kappa variable 1D-33, immunoglobulin kappa variable 1–33, and dynein heavy chain 11, axonemal, whereas the top 5 downregulated DEPs were cDNA FLJ51597, highly similar to C4b-binding protein alpha chain, serum amyloid P-component, transthyretin, properdin, and cDNA FLJ58124, highly similar to complement factor I (Fig. 3A). In addition, four low abundance DEPs (fibroblast growth factor receptor 4, mannosyl-oligosaccharide 1,2-alpha-mannosidase IC, eukaryotic translation initiation factor 5, and protocadherin Fat 2) were detected (Fig. 3B). 

Figure 3.

 

Analysis of the differentially expressed proteins identified by DIA-MS. (A) The significance of differentially expressed proteins between groups XFG and XFS. (B) Low-abundance proteins identified between groups XFG and XFS. (C) The significance of differentially expressed proteins between groups XFG and ARC. (D) Low-abundance proteins identified between groups XFG and ARC. (E) The significance of differentially expressed proteins between groups XFS and ARC. (F) Low-abundance proteins identified between groups XFS and ARC.ARC = age-related cataract; DIA-MS = data-independent acquisition mode mass spectrometry; XFG = exfoliation glaucoma; XFS = exfoliation syndrome.

Comparing XFG and ARC, the top 5 upregulated DEPs were dynein heavy chain 11, axonemal, immunoglobulin kappa variable 1D-33, B cell receptor heavy chain variable region (Fragment), cadherin-13, and uncharacterized protein DKFZp779H1622 (fragment), whereas the top 5 downregulated DEP were Ig heavy chain variable region (fragment 1), Ig heavy chain variable region (fragment 2), glutamate dehydrogenase 2, mitochondrial, alpha globin, and putative hydroxypyruvate isomerase (Fig. 3C). In addition, 8 low abundance DEPs (fibroblast growth factor receptor 4, eukaryotic translation initiation factor 5, protocadherin, EGF-containing fibulin-like extracellular matrix protein 2, ATP synthase subunit alpha, mitochondrial, eukaryotic initiation factor 4A-I, T-complex protein 1 subunit delta, and enoyl-CoA hydratase domain-containing protein 2, mitochondrial) were detected (Fig. 3D). 

Comparing XFS and ARC, the top 5 upregulated DEPs were apolipoprotein F, APOB protein, lysosome-associated membrane glycoprotein 1, 14-3-3 protein gamma, and apolipoprotein L1 (fragment), whereas the top 5 downregulated DEPs were protein HEG homolog 1, Ig heavy chain variable region (fragment 1), Ig heavy chain variable region (fragment 2), ribonuclease pancreatic, and glutaminyl-peptide cyclotransferase (see Fig. 3E). In addition, 6 low abundance DEPs (antileukoproteinase, integral membrane protein 2B, C-C motif chemokine 18, actin-related protein 2/3 complex subunit 5, glutaminyl-peptide cyclotransferase, and exportin-2) were detected (Fig. 3F). 

To evaluate the expression trend of proteins in different degrees of the disease, we then performed a series test of clusters for all significant DEPs in groups XFG, XFS, and ARC (Fig. 2D, Table 1). The proteins were classified into 6 clusters, wherein the Sushi, von Willebrand factor type A, EGF, and pentraxin domain-containing protein 1 (SVEP1; see Fig. 3C), transforming growth factor beta-1 proprotein, transferrin (isoform CRA_c), heat shock protein (HSP) 90-alpha, heat shock 70 kDa protein 6, complement C4-B, and protein C (fragment) were categorized in cluster 2 (the levels of proteins increased as the disease worsened). In addition, anti-staphylococcal enterotoxin E heavy chain variable region (fragment), cathelicidin antimicrobial peptide, complement C1r subcomponent, complement C1q subcomponent subunit B, variable immunoglobulin anti-estradiol heavy chain (fragment), cystatin-A, cholesteryl ester transfer protein, thrombospondin-4, and glutamate dehydrogenase 2 (mitochondrial) were in cluster 4 (the levels of proteins decreased as the disease worsened). 

Table 1.

 

Series Test of Cluster for all Significantly Differentially Expressed Proteins

 

To compare with the results in previous plasma/serum proteome study, we searched in PubMed on August 25, 2024, by searching the texts filtered by (pseudoexfoliation glaucoma) OR (pseudoexfoliation syndrome) OR (exfoliative glaucoma) OR (pseudoexfoliative glaucoma) OR (exfoliation syndrome) OR (exfoliation glaucoma) AND (protein* OR proteom*) and published before August 25, 2024. We manually reviewed all 516 results and there remained 27 articles related to plasma/serum protein assays for XFG/XFS (Table 2). 

Table 2.

 

Summary Based on Plasma/Serum Protein Assays

 

Pathway/Network Analysis

The three most significant biological process categories identified in the GO enrichment analysis were proteolysis, oxidation-reduction process, and homophilic cell adhesion via plasma membrane adhesion molecules; cellular components were extracellular region, membrane, and integral components of membrane; molecular functions were protein binding, calcium ion binding, and ATP binding (Supplementary Fig. S2). SVEP1 involved calcium ion binding and protein binding. The KEGG pathway annotations suggested that the transport and catabolism, signal transduction, neurodegenerative diseases, and immune system played a vital role in the disease. SVEP1 was annotated as fibrillins and related proteins containing Ca2+-binding EGF-like domains and classified in signal transduction mechanisms which was the most significant function classification in KOG analysis. In IPR annotations, the immunoglobulin V-set domain was most significant. In addition, SVEP1, the extracell protein, had the following characteristics: CUB and sushi domain-containing protein, IPR002035 (von Willebrand factor, type A), IPR000436 (Sushi/SCR/CCP domain), IPR000742 (EGF-like domain), IPR001881 (EGF-like calcium-binding domain), IPR013032 (EGF-like, conserved site), IPR001759 (Pentraxin-related), IPR003410 (HYR domain), and IPR011641 (Tyrosine-protein kinase ephrin type A/B receptor-like). Comparing XFG with XFS and ARC, KEGG enrichment analysis illustrated complement and coagulation cascades, D-glutamine and D-glutamate metabolism, inflammatory bowel disease, Th17 cell differentiation, proximal tubule bicarbonate reclamation, and staphylococcus aureus infection involved functional pathways, whereas malaria, staphylococcus aureus infection, intestinal immune network for IgA production, and systemic lupus erythematosus involved comparing XFG and XFS with ARC (Supplementary Figs. S3–S5). The thyroid or estrogen hormone and Alzheimer’s disease might also involve in XFG. In GSEA analysis (Supplementary Fig. S6), the result was similar in IPR and serine proteases, the trypsin domain may play a vital role comparing XFG and XFS with ARC (P < 0.05). As for subcellular localization, the three most significant types were the extracell protein, the plasma membrane protein, and the cytoplasm protein. We built a PPI network to investigate the interactions between these DEPs (Supplementary Fig. S7). To predict key proteins in clusters 2 and 4, we used GOsemsim and found that none met with the cutoff value (> 0.75; Supplementary Fig. S8). 

MR Analyses and Sensitivity Analysis

We performed an MR to verify the associations between proteins in clusters 2 and 4 and XFG. In phase I, we overlapped 16 pQTL in the deCODE cohort and our cohort and found 2 proteins (SVEP1 and HP) may be related to XFG in the primary analysis (Supplementary Table S6). There was a significant correlation between SVEP1 and XFG (OR = 1.20, 95% CI = 1.10 to 1.31, P = 4.28 × 10⁻⁵). We found that the SVEP1 in the DEPs had a robust result after quality control steps were conducted including MR-Egger, Simplemode, Weighted median, and Weighted mode, whose beta was in the same direction and P < 0.05 in all methods (Fig. 4, Supplementary Fig. S9). Then, we repeated the process in another 2 cohorts and overlapped 8 pQTL in UKB-PPP and 7 pQTL in ARIC. We found that the association between SVEP1 and XFG remained significant (OR = 1.49, 95% CI = 1.01 to 2.19, P < 0.05 by IVW) in the ARIC cohort, whereas the HP was nonsignificant. 

Figure 4.

 

Mendelian randomization estimates for SVEP1 on XFG. (A) Forest plot. (B) Leave-one-out plot. MR = Mendelian randomization; SVEP1 = Sushi, von Willebrand factor type A, EGF and pentraxin domain containing 1; XFG = exfoliation glaucoma.

In phase II, we overlapped 13 pQTL in the deCODE cohort and previously reported plasma/serum proteins and found only endothelin 1 may have an impact on XFG in the primary analysis, which could not be found in UKB-PPP and ARIC (Supplementary Table S7). Besides, we performed the MR-Egger intercept and MR-PRESSO tests to identify potential pleiotropy and Cochran's Q tests to evaluate the heterogeneity (all P > 0.05). What's more, SVEP1 passed the leave-one-out analysis and Steiger filtering directionality test (see Fig. 4, Supplementary Table S8). The results of the sensitivity analysis guaranteed the association between SVEP1 and XFG. Then, we found that XFG was associated with SVEP1 (OR = 1.02, 95% CI = 1.00 to 1.04, P = 0.02 by IVW; see Supplementary Fig. S8, Supplementary Table S8) in reverse MR, whereas the result failed to pass through leave-one-out analysis (Supplementary Fig. S10). To determine whether protein SVEP1 involved the development of senile cataract, complicated cataract and POAG, or a reverse influence existed, we performed the two-sample bidirectional MR and found no potential influence existed between SVEP1 and these diseases (see Supplementary Table S8, Supplementary Figs. S11–S16). 

Potential Pathway Between SVEP1 and XFG

To explore whether LOXL1-AS1 played a role in the effect of SVEP1 on XFG, we confirmed that LOXL1-AS1 had a protective role in XFG (OR = 0.66, 95% CI = 0.63 to 0.70, P < 0.001 by IVW, see Supplementary Table S8, Supplementary Fig. S17) and SVEP1 (OR = 0.99, 95% CI = 0.98 to 0.99, P < 0.001 by IVW; see Supplementary Table S8, Supplementary Fig. S18) by MR. After mediation analysis, we found LOXL1-AS1 had a protective role in XFG mediated by SVEP1 (mediated effect = −0.002, Z-score = −3.07, P = 0.002). In the cell level by RNA-sequencing, we validated that the expression of SVEP1 was elevated when LOXL1-AS1 was knocked down (fold change = 1.24, adjusted P value < 0.05; see Supplementary Table S9, Supplementary Fig. S19A), whereas there was no change when LOXL1-AS1 was overexpressed (fold change = 0.96, adjusted P value > 0.05; see Supplementary Table S9, Supplementary Fig. S19B). 

Our previous study found that genus Senegalimassilia may influence XFG.²³ To explore the interaction between SVEP1 and genus Senegalimassilia, we performed two-sample bidirectional MR and found no associations between genus Senegalimassilia and SVEP1 in reverse MR (see Supplementary Table S8, Supplementary Fig. S20) and SVEP1 was associated with genus Senegalimassilia in forward MR (OR = 1.25, 95% CI = 1.01 to 1.55, P = 0.04 by IVW; see Supplementary Table S8, Supplementary Fig. S21), whereas the result failed to pass through the leave-one-out analysis. 

Mendelian Randomization Phenome-Wide Association Study

We performed MR-PheWAS to identify the associations between SVEP1 with 3675 phenotypes. In addition, we found that 131 phenotypes in the UKB cohort and 262 phenotypes in the FinnGen cohort were suggested to be nominally significant, including alcohol-related diseases, congenital anomalies, infectious diseases, injuries and poisonings, mental disorders, neoplasms, pregnancy complications, sense organs, and dermatologic, digestive, endocrine/metabolic, genitourinary, musculoskeletal, neurological, hematopoietic, respiratory, or circulatory system problems (see Supplementary Table S10). 

In the ocular area, SVEP1 was associated with ocular pain, Graves’ disease, conjunctivitis, keratoconus, glaucoma, visual field defects, macular cyst, age-related macular degeneration (whether dry or wet), and diabetes retinopathy. In addition, in other areas, SVEP1 was associated with Alzheimer's disease or diseases including Alzheimer's disease, Parkinson’s disease, immune bowel disease, hypertension or gestational hypertension, diabetes and its complications, hyperlipidemia, hypercholesterolemia, hemorrhagic disorder due to intrinsic circulating anticoagulants, diseases of white blood cells, aortic aneurysm, hernia, prolapse of vaginal walls, genital prolapse, preeclampsia and eclampsia, peptic ulcer, and acanthosis nigricans. After multiple testing corrections, SVEP1 appeared to have an impact on glaucoma (OR = 1.14, 95% CI = 1.11 to 1.16, P = 0.0000003), diabetes insulin treatment (OR = 1.07, 95% CI = 1.06 to 1.09, P = 0.000002), diabetic ketoacidosis (OR = 1.15, 95% CI = 1.12 to 1.17, P = 0.0000026), and type 2 diabetes (OR = 1.07, 95% CI = 1.05 to 1.08, P = 0.0000067; see Fig. 1). 

Discussion

We adopted a DIA-MS approach using in-depth plasma proteomics and pictured the largest wide-angled proteomic landscape mapping and identified deeper to identify comprehensive plasma proteome profiling of XFG/XFS. Plasma is the main clinical sample and is easy to obtain and standardize and is responsible for circulating proteins throughout the body that can reflect a variety of disease problems. However, plasma is the most complex and diverse sample in humoral analysis. The difference between the highest and lowest abundance protein concentrations in plasma is greater than 10 orders of magnitude, whereas the high-abundance proteins, such as albumin and immunoglobulin, account for 80% of the total plasma proteins, and the rest are present in very low concentrations, but they are associated with the evolution of many diseases and are potential biomarkers. Thus, we conducted this study by in-depth mass spectrometry technology to quantitatively identify proteins with high sensitivity and high accuracy. 

With our proteomics data, our results suggested that infection, immune responses, including intestinal immune network, endocrine hormones, and complement and coagulation cascades, may be involved in the development of XFG. As the disease progressed, enrichment signals were biological pathways involved in the downregulation of anti-infection factors, upregulation of oxidative stress protein, and dysregulation of complement and coagulation cascades, and dysregulation of degradation of glutamine metabolism. 

The past study that identified the most proteins identified 149 proteins and 17 studies focused only on one protein/peptide. There were six studies that found nonsignificant differences in C-reactive protein between XFG/XFS and the healthy controls and one found an increase in XFS. Our study found nonsignificant differences in C-reactive protein between the groups by DIA-MS and MR. In addition, our study also supported that endothelin 1 was related to XFG/XFS by MR. 

We first found that high expression of SVEP1 was a risk factor for XFG and was validated in multiple large cohorts by the MR approach. This conclusion was robust across multiple ethnicities. In addition, SVEP1 was recently linked to coronary heart disease, blood pressure, dementia, and type 2 diabetes.^24–26 This may explain the onset of complications associated with XFG, such as cardiovascular disease and Alzheimer’s disease. What's more, XFS was reported to be associated with a hernia and prolapse of the pelvic organs, and, in our results of MR-PheWAS, we validated these phenomena by SVEP1 at the nominal significance level.^27,28 The SVEP1 missense variant could be deleterious in TEK-related primary congenital glaucoma.²⁹ According to an open-angle glaucoma GWAS, rs61751937, a missense variant in SVEP1, could be pathogenic.³⁰ In our study, SVEP1 appeared to be related with diabetes and complications, revealing that SVEP1 is also a potential biomarker of diabetes or diabetes retinopathy. We need to validate this in further studies. 

Plasma proteomics offers promising insights into pathogenesis and therapeutic opportunities. Our data triangulation and MR estimates integrating genetics, plasma protein levels, and diseases, indicate that SVEP1 may be a therapeutic target for XFG. We searched candidate drugs from a drug signatures’ database for gene set analysis containing 22,527 gene sets (DSigDB, http://dsigdb.tanlab.org/DSigDBv1.0/).³¹ We scanned DSigDB and found four potential candidate drugs based on computational drug signatures from the comparative toxicogenomics database. Cytarabine with the US Food and Drug Administration (FDA), the World Health Organization (WHO), Indian and China approval was predicted to decrease the expression of SVEP1; decitabine with the FDA’s approval was predicted to decrease expression and reaction of SVEP1; cobaltous chloride was predicted to decrease the expression of SVEP1; and estradiol with the FDA, the WHO, Indian, China and Traditional Herbal approval was predicted to decrease SVEP1 and increase LOXL1. Estradiol is a naturally occurring hormone circulating endogenously in female subjects as an estrogen receptor agonist, causing an increase in hepatic synthesis of various proteins, which include sex hormone binding globulin and thyroid-binding globulin. However, SVEP1 expression is regulated in an estrogen-dependent manner, whereas 17beta-estradiol can increase the level of the SVEP1 expression.^32–34 In our study, downregulation of immunoglobulin anti-estradiol indicated that the function of estradiol to SVEP1 may increase. This suggested there may exist potential gender differences on XFG and the effects of female hormone fluctuations, such as menopause, may also have an impact on XFG. Further studies are needed to determine the sex difference in the onset of the disease caused by SVEP1. What's more, our results suggested SVEP1 expression might be regulated by LOXL1-AS1. The downregulation of LOXL1-AS1 and subsequent upregulation of SVEP1 expression were implicated in XFG pathogenesis. However, the primary protective function of LOXL1-AS1 may lie in other mechanisms. This study utilized mass spectrometry data to identify significant differential proteins and reinforced the conclusions by MR, but validation for the findings should be carried out in a large cohort population and further experiments are still needed to understand the effect of SVEP1 on XFG. 

Acknowledgments

The authors thank the participants and investigators of the FinnGen, UK Biobank, deCODE, and the ARIC study for their important contributions. This work is supported by Extreme Smart Analysis platform (https://www.xsmartanalysis.com/). 

Author Contributions: Study concept and design: H.W., G.X., and L.J. Acquisition, analyses, or interpretation: L.J., M.Y., X.L., Z.K., H.Y., M.X., Z.S., G.S., T.Y., M.G., A.M., Z.D., W.A., Z.X., T.C., W.W., H.W., and G.X. Drafting of the manuscript: L.J. Critical revision of the manuscript for important intellectual content: L.J., M.Y., X.L., Z.K., H.Y., M.X., Z.S., G.S., T.Y., M.G., A.M., Z.D., W.A., Z.X., T.C., W.W., H.W., and G.X. Statistical analyses: L.J. and Z.X. Obtained funding: G.X. Administrative, technical, or material support: H.W., G.X., and W.A. Study supervision: H.W. and G.X. 

Supported by Guangdong Basic and Applied Basic Research Foundation (2022A1515012257), the Fundamental Research Funds of the State Key Laboratory of Ophthalmology, and an award from Guangdong Provincial Department of Science and Technology (KTPYJ2022031). 

Disclosure: J. Li, None; Y. Ma, None; L. Xie, None; K. Zhuo, None; Y. He, None; X. Ma, None; S. Zheng, None; S. Guo, None; Y. Tang, None; G. Muhetaer, None; M. Aizezi, None; D. Zhang, None; A. Wumaier, None; X. Zhang, None; C. Tang, None; W. Wang, None; W. Huang, None; X. Gao, None