Central subfield thickness of diabetic macular edema: Correlation with the aqueous humor proteome

Samples

This study was conducted in compliance with the Institutional Review Board of Kyoto Prefectural University of Medicine, from which approval for the study was obtained (permission RBMR-C-864–6). The study adhered to the tenets of the Helsinki Declaration. Informed consent to use samples from the biobank was obtained from all patients after an explanation of the nature and possible consequences of the study.

Aqueous humor samples from eyes with treatment-naïve clinically significant DME (n = 13) and age-matched controls (n = 11) were donated from the biobank of Kyoto Prefectural University Hospital, Kyoto, Japan (Table 1). Normal distribution was assessed in STATA 16.0 (StataCorp, College Station, TX) with histograms and the skewness and kurtosis test for normality referred to in STATA 16.0 as the sktest. The Student’s t test was used to confirm that there was no statistically significant difference in age between the two groups (Table 1). Fisher’s exact test was used for categorical data. All patients in the DME group had clinically significant treatment-naïve DME. In the DME group, the exclusion criteria were iris rubeosis, hyphema, glaucoma (including neovascular glaucoma), vitreous hemorrhage, previous anti-VEGF treatment, central laser, and periocular or intraocular corticosteroid injections. Control samples were obtained before cataract surgery from age-matched patients who had no ocular disease other than cataract. Best-corrected visual acuity (BCVA) was measured as the logarithm of the minimum angle of resolution (logMAR). Optical coherence tomography (OCT) was obtained with Swept Source OCT (DRI-OCT Triton; Topcon, Tokyo, Japan). The severity of DME was measured as central subfield thickness (CST) with the caliber tool of Topcon OCT software (DRI-OCT Triton; Topcon, Tokyo, Japan).

Additional aqueous humor samples from eyes with treatment-naïve DME (n = 15) and control samples (n = 8) were obtained from the biobank for validation by ELISA (Table 2). Normal distribution was assessed in STATA 16.0 (StataCorp, College Station, TX) with histograms and the skewness and kurtosis test for normality, referred to in STATA 16.0 as the sktest. The Mann–Whitney U test was used to confirm that there was no statistical difference in age between the groups (Table 2). Fisher’s exact test was used for categorical data. Samples obtained for validation by ELISA were selected according to the inclusion and exclusion criteria that were applied to samples attained for proteomic analysis by mass spectrometry.

Sample preparation for mass spectrometry

The samples were stored at −80 °C until sample preparation was initiated. The protein concentration was measured using an infrared spectrometer (Direct Detect, Darmstadt, Germany). Sample preparation was performed according to the S-Trap microspin column digestion protocol (ProtiFi, Huntington, NY, USA). The volume of each sample was measured, and an equal volume of 2 x SDS lysis buffer (10% SDS, 100 mM triethylammonium bicarbonate [TEAB], pH 8.5) was added. Reduction was performed by adding tris(2-carboxyethyl)phosphine hydrochloride (TCEP) to the protein solution in SDS for a final concentration of 10 mM, followed by heating for 10 min at 95 °C. The protein solution was cooled to room temperature, followed by alkylation by adding iodoacetamide. A 12% aqueous phosphoric acid solution at 1:10 was added to reach a final concentration of 1.2% phosphoric acid. S-Trap binding buffer (90% methanol, 100 mM TEAB) was added. The samples were transferred to microcolumns and centrifuged at 4,000 ×g until the buffer passed through the S-Trap column. S-Trap binding buffer was added, and centrifugation at 4,000g was performed three times. The S-Trap microcolumn was moved to a new 1.7 ml sample tube, and 20 µl of digestion buffer was added, followed by incubation overnight at 37 °C. Elution was first performed with 40 µl of 50 mM TEAB and then with 0.2% formic acid, followed by centrifugation at 4,000 ×g. The recovery of hydrophobic peptides was performed with an elution of 35 µl of 50% acetonitrile containing 0.2% formic acid. Elutions were pooled, and the peptide concentration was measured using a fluorescence-based technique with tryptophan as a standard, with an excitation at 295 nm and emission at 350 nm, as previously described [15]. The samples were dried in a vacuum centrifuge and stored at −80 °C until further use.

Quantitative mass spectrometry using label-free quantification nano liquid chromatography tandem mass spectrometry

The samples were re-suspended in 0.1% formic acid and analyzed using label-free quantification nano liquid chromatography tandem mass spectrometry (LFQ nLC-MS/MS). The samples were analyzed in duplicate. LFQ nLC-MS/MS was performed on an Orbitrap Fusion Tribrid mass spectrometer equipped with an EasySpray ion source coupled to a Dionex Ultimate 3000 RSLC nano system (Thermo Fisher Scientific Instruments, Waltham, MA). The sequence in which the samples were run was mixed, thereby distributing the samples from both groups throughout the entire sequence. Technical replicas were run with several days of intermission. Peptides were trapped on a µ-Precolumn (300 µm x 5 mm, C18 PepMap100, 5 µm, 100Å, Thermo Fisher Scientific) and separated on an EASY-spray analytical column (500 mm x 75 µm, PepMap RSCL, C18, 2 mm, 100 Å, Thermo Fisher Scientific).

An elution gradient of 122 min was applied by mixing Buffer A (99.9% water and 0.1% formic acid) and Buffer B (99.9% acetonitrile and 0.1% formic acid). The universal method setting was used to obtain full Orbitrap scans with a scan range (m/z) of 375 to 1,500. The automatic gain control (AGC) target was set to 4×10⁵. The maximum injection time was 50 ms. The cycle time was set to 3 s. The most intense precursors with a charge state of 2–7, an intensity threshold of 5 × 10³ and a maximum intensity of 1 × 10²⁰ were selected. MS² scans were obtained in the linear ion trap in auto-scan range mode with collision-induced dissociation energy at 35%, an AGC target of 2 × 10³, and a maximum injection time of 300 ms. Precursor ions were isolated with the quadrupole set using an isolation window of 1.6 m/z. The dynamic exclusion time was 60 s.

LFQ analysis was performed with MaxQuant software [16], Version 1.6.6.0. Files were searched against the UniProt Homo Sapiens database downloaded on 19 March, 2021. Carbamidomethyl (C) was used as a fixed modification. Oxidation (M) and acetyl (protein N-terminal) were used as variable modifications. The false discovery rates for PSM, protein, and the site were set to 1%. The LFQ minimum ratio was set to 1. Successful MS/MS was required for LFQ comparisons. Unique peptides and razor peptides unmodified and modified with oxidation (M) or acetyl (protein N-terminal) were applied for protein quantification. Matching between runs was performed, and a decoy search was performed using revert sequences. Contaminant sequences were included in all searches. The unfiltered results of the database search are provided in Appendix 1.

The MaxQuant output file was entered into Perseus (Version 1.6.2.3) [17]. Only proteins identified by post-translational modification were removed, followed by the removal of proteins identified from a peptide found to be part of a protein derived from the reversed part of the decoy database. Proteins identified as contaminants were also removed from the dataset. Label-free quantification (LFQ) values were log₂ transformed, and mean LFQ values were calculated. For successful protein identification, at least two unique peptides were required. Proteins identified by only one unique peptide were removed from the dataset. All proteins that were successfully identified and quantified are provided in Appendix 2. Before statistical analysis, the proteins were filtered by requiring successful identification and quantification in at least 70% of the samples in the DME group and the control group (Appendix 3). Statistical analysis by Student’s t test was performed in Perseus to compare DME with controls. The imputation of missing values was not performed. Corrections for multiple hypothesis testing were executed using the permutation-based method in Perseus [18], with the number of randomizations set to 250 and an S₀ parameter of 0.1. A false discovery rate (FDR) of 0.05 was applied. CST values were log₁₀ transformed before calculating correlations to achieve a normal distribution. Correlations were calculated in STATA 16.0 (StataCorp, College Station, TX) using Pearson’s correlation coefficient (r). Correlations were considered statistically significant if p < 0.05. STATA 16.0 was used to create scatterplots with predictions from a linear regression.

A gene ontology analysis of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways regarding regulated proteins was conducted with GeneCodis 4.0 software [19], as previously described [20]. The 10 pathways with the lowest adjusted p values were selected. Cluster analyses were performed with STRING 11.5 (string-db.org) [21–23], as mentioned [13], with the minimum required interaction score set to 0.40 and the MCL inflation parameter set to 3. KEGG pathway analysis was performed in STRING 11.5, with a false discovery rate set to 0.01. Principal component analysis (PCA) was performed in Perseus with default settings after the imputation of missing values from the normal distribution.

Enzyme-linked immunosorbent assay (ELISA)

For the quantification of VEGF, the samples were diluted 1:2, and VEGF was quantified as detailed in a previous study using the ab222510 Human VEGF SimpleStep enzyme-linked immunosorbent assay (ELISA) Kit (Abcam, UK) [13]. For the quantification of lipopolysaccharide-binding protein (LBP), the ab279407 Human LBP SimpleStep ELISA^® kit (Abcam, UK) was applied, and LBP was quantified according to the manufacturer’s instructions. The wash buffer of the kit was diluted at 1:10. The capture and detector antibody solution was diluted at 1:10 in the antibody diluent 4BI of the kit. Standards 1–8 were created with the Sample Diluent NS of the kit. The samples were centrifuged at 2,000 × g for 10 min and diluted 1:500 in the Sample Diluent NS of the kit. A volume of 50 µl sample or standard was added to the appropriate wells, and 50 µl of the antibody solution was added to each well. The plate was sealed and incubated for 60 min at room temperature on a plate shaker set to 400 rpm. The wells were washed three times with 350 µl wash buffer. A volume of 100 µl TMB Development Solution was added to each well, followed by incubation in the dark for 10 min on a plate shaker set to 400 rpm. A volume of 100 µl Stop Solution of the kit was added. The plate was shaken on a plate shaker for 1 min. The optical density was recorded at 450 nm. Quantitative data were log₁₀ transformed to obtain a normal distribution. ELISA assays were performed in duplicate for all samples and standards. The average of duplicate readings for each standard and sample was calculated, and the average zero standard optical density was subtracted. Subsequently, sample values were normalized by determining the ratio of the sample absorbance to the absorbance of the standards, ensuring a standardized comparison across all samples. Statistical analyses of the ELISA data were conducted using the Student’s t test and Pearson’s correlation coefficient (r). Differences and correlations were considered statistically significant if p < 0.05 [13].

Immunofluorescence

Human postmortem eyes were provided by Corneabank, Beverwijk, the Netherlands. All procedures were in accordance with the provisions of the Declaration of Helsinki for the use of human tissue in research. A 61-year-old female with diabetic retinopathy treated with laser with more than 12 years of Type-2 diabetes and insulin use and a 68-year-old female non-diabetic patient were selected. The postmortem delays were 13 and 17 h, respectively. The semiquantitative grading of PAL-E on the vascular endothelium depicted patchy to uniformly marked staining [24] in two independent sections of the donor with diabetic retinopathy. Sections of 10 μm thickness were cut and fixed in 4% (w/v) formaldehyde for 20 min and were washed once in 3x phosphate-buffered saline (PBS). Individual sections were encircled using a PAP pen to form a hydrophobic barrier. The sections were incubated for 1 h in 1x PBS supplemented with 10% normal goat serum and 0.1% Triton X-100. Next, the slides were washed three times with PBS and incubated overnight at 4 °C with anti-human monoclonal mouse antibody PAL-E (Pathologische Anatomie Leiden–Endothelium; 1:200) and fluorescein isothiocyanate (FITC)-conjugated polyclonal rabbit anti-human fibrinogen (F0111, DAKO; 1:200). Antibodies were diluted in normal antibody diluent (AB999, Scytek). Sections were subsequently washed three times with PBS and incubated with goat anti-mouse Alexa Fluor™ 633 antibody (A21052, Life Technologies; 1:100). The sections were again washed three times with PBS and mounted with Vectashield antifade mounting medium with DAPI (H-1200–10, Vector Laboratories). Sections were imaged using a confocal laser scanning microscope (Leica, SP8) with settings kept constant between conditions.

Proteome analysis of diabetic macular edema (DME)

Proteome analysis was performed on aqueous humor samples from patients with treatment-naïve DME and age-matched controls. A total of 891 proteins were successfully identified in the combined set of aqueous samples (Supporting Data Values S1). In total, 256 aqueous humor proteins were successfully identified and quantified in at least 70% of the samples in each group (Supporting Data Values S2), and statistical analysis was performed on these proteins.

In the PCA plot, DME samples could generally be separated from control samples based on their proteomes (Figure 1). After correction for multiple hypothesis testing, 101 proteins were statistically significantly regulated in the DME group compared to the controls (Table 3; Figure 2). Among the significantly regulated proteins, 55 proteins were increased in content in DME, and 46 proteins were decreased in content (Table 3; Figure 2).

Bioinformatics

The proteins regulated in DME were associated with complement and coagulation cascades, glycolysis/gluconeogenesis, cholesterol metabolism, pyruvate metabolism, ECM-receptor interaction, hypoxia inducible factor-1 (HIF-1) signaling, carbon metabolism, and glucagon signaling (Figure 3A,B). Among the regulated proteins, a large cluster of proteins was formed by proteins involved in complement and coagulation cascades (Figure 3B). STRING cluster analysis also identified a major cluster of proteins involved in complement and coagulation cascades (Figure 4A), which were all increased in content. Proteins involved in complement and coagulation cascades included complement factors (C2, C3, C4B, C5, C6, C7, C8A, C8G, C9, and CFH), fibrinogen chains (FGA, FGB, and FGG), vitronectin (VTN), alpha-1-antitrypsin (SERPINA1), antithrombin-III (SERPINC1), heparin cofactor 2 (SERPIND1), alpha-2-antiplasmin (SERPINF1), plasma kallikrein (KLKB1), kininogen-1 (KNG1), alpha-2-macroglobulin (A2M), and prothrombin (F2).

DME was associated with the regulation of a cluster of proteins involved in ECM–receptor interaction, including agrin, basement membrane-specific heparan sulfate proteoglycan core protein (HSPG2), dystroglycan, and fibronectin (Figure 4A). Fibronectin was upregulated in DME, whereas agrin, dystroglycan, and HSPG2 were downregulated.

Another cluster of regulated proteins contained proteins involved in cholesterol metabolism, including apolipoprotein A-I (APOA1), apolipoprotein C-III (APOC3), apolipoprotein A-IV (APOA4), low-density lipoprotein receptor-related protein 2 (LRP2), and phosphatidylcholine-sterol acyltransferase (LCAT; Figure 4A). All lipoproteins and phosphatidylcholine-sterol acyltransferase were increased in DME, while lipoprotein receptor-related protein 2 was downregulated. STRING cluster analysis discerned a cluster of regulated proteins that represented multiple KEGG pathways (Figure 4B–F). The proteins of this cluster were downregulated and included alpha-enolase (ENO1), fructose-bisphosphate aldolase A (ALDOA), fructose-bisphosphate aldolase C (ALDOC), L-lactate dehydrogenase A chain (LDHA), L-lactate dehydrogenase B chain (LDHB), phosphoglycerate kinase 1 (PGK1), and pyruvate kinase PKM (PKM). The regulation of these proteins indicates that DME was associated with the regulation of the carbon metabolism, the HIF-1 signaling pathway, the glucagon signaling pathway, the pyruvate metabolism, and glycolysis/ gluconeogenesis (Figure 4B–F).

Correlations with clinical parameters

Positive correlations with BCVA were observed for complement C2, complement C5, complement C8 alpha chain, fibronectin, hepatocyte growth factor-like protein, and zinc-alpha-2-glycoprotein. Negative correlations with BCVA were observed for 78 kDa glucose-regulated protein, amyloid-like protein 2, beta-crystallin S, low-density lipoprotein receptor-related protein 2, neurexin-3, and testican-1 (Figure 5).

Positive correlations with CST were observed for ceruloplasmin and the complement component C8 alpha chain (Figure 6). Negative correlations with CST were observed for desmocollin-1, desmoglein-1, and dipeptidyl peptidase 2 (Figure 6).

Validation by ELISA

Mass spectrometry identified an increased level of LBP in DME (fold change = 3.65; p = 1.3×10⁻⁸; Figure 7A). The upregulation of LBP in DME was confirmed by ELISA, thereby validating the proteomic analysis (fold change = 1.72; p = 0.048; Figure 7A). VEGF was significantly increased in DME (p = 0.015; Figure 7A). LBP correlated significantly with VEGF (r = 0.68; p = 0.0077; Figure 7B). LBP quantified with mass spectrometry correlated with CST (r = 0.59; p = 0.035). Nonetheless, the correlation was not confirmed by ELISA (r = 0.31; p = 0.27; Figure 7B).

Immunofluorescence

To illustrate the involvement of fibrinogen extravasation in DME, we performed immunostainings on human retinal tissue sections of DR patients and non-diabetic controls. As an indicator of blood–retinal barrier breakdown, we used the monoclonal antibody PAL-E [25], which recognizes plasmalemma vesicle-associated protein (PLVAP), which is crucially involved in DME [28]. The semiquantitative grading of PAL-E on the vascular endothelium revealed patchy to uniformly marked staining [24] in independent sections of the DR donor. The co-localization of PLVAP with the extravasation of fibrinogen was observed in DME, whereas fibrinogen in the non-diabetic control was more confined within blood vessels (Figure 8).