A total of 27 patients who undertook fundus surgery at the Department of Ophthalmology in Shanghai General Hospital were enrolled. Vitreous fluid was obtained from 12 patients during four different surgeries (each group, n = 3) for DIA proteomic analysis: PPV for RRD (RD group), SiO removal after successful retinal reattachment (SO group), cataract surgery after successful retinal reattachment with sterilized air tamponade (FA group), and PPV for epiretinal membrane (ERM group). The ERM group was regarded as control. For enzyme-linked immunosorbent assay (ELISA) analysis, vitreous fluid was obtained from 15 other patients (15 eyes: five FA eyes during cataract surgery for the ELISA-FA group and 10 SO eyes during SiO removal for the ELISA-SO group) to confirm the differentially expressed protein transferrin. Patients with metabolic diseases including hypertension and diabetes, ocular trauma, or any associated concurrent eye condition, such as vitreous hemorrhage, macular degeneration, high myopia, glaucoma, uveitis, or grade C to D PVR were excluded. The PVR was graded according to the updated classification of the Retina Society Terminology Committee (1991) (Machemer et al., 1991). Data concerning prior surgeries, duration of retinal detachment, date of primary retinal repair surgery, and severity of PVR were recorded as clinical characteristics.

Before intravitreal operation, undiluted samples of vitreous humor, retro-oil fluid, or vitreous cavity fluid were obtained under visual control by aspirating liquefied vitreous from the vitreous cavity with a 1-ml syringe connected to a hose flute needle before the infusion. The volume of samples is approximately 200–300 μl. Undiluted samples were divided into 1.5-ml tubes (Eppendorf, Freemont, CA, United States), frozen at −270°C immediately, and kept at −80°C until analysis.

The samples were digested, and the peptides were desalted and fractioned. The samples were mixed in 8 M urea, 100 mM Tris-HCl, pH 8.5, and 5 mM DTT, then 1 M iodoacetamide (15 mM IAA in UA buffer) was added to block reduced cysteine residues, and the samples were incubated for 40 min in darkness. Finally, 0.5 µg Lys-C (1:100) was added and set for 4 h, then 0.5 µg trypsin was added and set for 16 h at 37°C, and the resulting peptides were collected as a filtrate. The peptides of each sample were desalted and concentrated by vacuum centrifugation and reconstituted in 10 µl of 0.1% (v/v) formic acid. Furthermore, 2-μg sample peptides were mixed with some standard iRT peptides for further data-dependent acquisition (DDA) and DIA quantification analysis.

LC–MS/MS DIA analyses were performed in cooperation with Applied Protein Technology (APTBIO, Shanghai). Shotgun data-dependent mass spectrometry and assay library generation were completed using 2 ug peptides that are label-free, mixed with iRT peptides, and analyzed using a Knauer Smartline HPLC system, Easy nLC-1200 (Fortis Technologies, Neston, United Kingdom), according to instructions, operating in shotgun mode for 2 h. Peptide fractionation was operated at a flow rate of 250 nl/min with 95% buffer A, followed by a linear increase to 30% buffer B across 97 min, to 100% B across 13 min, and was sustained at 100% for 10 min. The top 20 most intense MS1 precursors with charge states between 2 and 5 were used for MS2 fragmentation, with a 15-s exclusion window. Fragment MS2 ions were collected for 120 ms across a 300–1,800-m/z range. Peptide sequences from raw profile mode wiff files were assigned using parallel searches against UniProtKB databases of human proteins with Maxquant software (1.5.3.17), appended with common contaminants and decoy sequences. Precursor ion mass errors were set to 30–75 ppm.

DIA was analyzed. Similarly, samples and iRT peptides were label-free and analyzed using a Knauer Smartline HPLC system, Easy nLC-1200 (Fortis Technologies, Neston, United Kingdom), according to instructions, operating in shotgun mode for 2 h. Peptide fractionation was operated at a flow rate of 250 nl/min with 95% buffer A, followed by a linear increase to 30% buffer B across 97 min, to 100% B across 13 min, and was sustained at 100% for 10 min. Each DIA cycle contained one full MS–SIM scan, and 30 DIA scans covered a mass range of 350–1,650 m/z with the following settings: SIM full-scan resolution was 120,000 at 200 m/z; AGC 3e6; maximum IT 50ms; profile mode; DIA scans were set at a resolution of 30,000; AGC target 3e6; Max IT auto; and normalized collision energy was 30 eV.

For DDA library data, database was searched by using MaxQuant 1.5.3.17 software. The database was downloaded at the website http://www.uniprot.org. The iRT peptide sequence was added into the database (>Biognosys|iRTKit|Sequence_fusionLGGNEQVTRYILAGVENSKGTFIID PGGVIRGTFIIDPAAVIRGAGSSEPVT GLDAKTPVISGGPYEYRVEATFGVDES NAKTPVITGAPYEYRDGLDAASYYAPVRA DVTPADFSEWSKLFLQFGAQGSPFLK). The parameters were set as follows: enzyme, trypsin; maximum missed cleavage site, 2; fixed modification, carbamidomethyl (C); and dynamic modification, oxidation (M) and acetyl (Protein N-term). All reported data were based on 99% confidence for protein identification as determined by false discovery rate {FDR = N (decoy) * 2 / [N (decoy)+ N (target)]} ≤ 1%. A spectral library was constructed by importing the original raw files and DDA searching results into Spectronaut Pulsar X_12.0.20491.4 (Biognosys).

DIA data was analyzed with Spectronaut Pulsar X_12.0.20491.4 searching the above-mentioned constructed spectral library. The main software parameters were set as follows: retention time prediction type, dynamic iRT; interference on MS2 level correction, enabled; cross-run normalization, enabled. All results were filtered based on a Q value cutoff of 0.01 (equivalent to FDR <1%).

Differentially expressed proteins were clustered by hierarchical cluster method, and data was presented as heat maps. Gene Ontology (GO) terms were analyzed, and the main process of GO analysis included blast, GO mapping, GO annotation, and interproScan for annotation augmentation. The differences of proteins were calculated using Fisher’s exact test. GO enrichment of differential proteins was studied for biological process, cellular component, and molecular function using Blast2Go (Software Tool, https://www.blast2go.com). Pathway enrichment of protein clusters was performed by using Kyoto Encyclopedia of Genes and Genomes (KEGG, http://www.kegg.jp/) pathway database. The differences of enrichment pathways were calculated using Fisher’s exact test. To assess functional associations between differential proteins, the online tool STRING tool (Search Tool for the Retrieval of Interacting Genes/Proteins) was applied with high confidence (0.70) and organism (Homo sapiens).

The demographics and clinical characteristics of patients enrolled in the study and the description of the corresponding vitreous samples are summarized in Tables 1. The study cohorts consisted of 27 patients—14 women and 13 men—with ages between 52 and 77 years. The sample cohort for DIA comprised 58.3% men, and there was no significant difference in age and sex between groups (p > 0.05). Fifteen patients were selected for the validation of transferrin by ELISA with similar age and sex. Detached retina extended two to three quadrants without the involvement of macula in all RD patients. The most common ocular comorbidity were lattice degeneration and grade A to B PVR, which were balanced between groups.

The DDA analysis of fractionated peptides by HPLC system Easy nLC-1200 identified 1,016 protein groups and 6,768 peptides at a detection threshold of 0.05. The proteins (false discovery rate, FDR <0.01) and peptides were added to the spectral library for LC–MS/MS data identification and extraction. Each sample was then separately prepared, fractioned, and analyzed for DIA data extraction. With data imported into Sepctronaut Plusar X, different protein groups were identified and quantified following retention time alignment. Protein expression patterns were different between groups in protein distribution and expression content. It was found that there were 817, 905, and 910 protein groups for the three individuals in the RD group; 856, 919, and 850 protein groups for SO; 680, 762, and 774 for FA; and 673, 695, and 792 protein groups for ERM that acted as control. The number of protein groups of RD and SO samples was slightly larger than that of ERM or FA (Figure 1A). A heat map of clustered fold changes for the main differentially expressed proteins was presented (Figure 1B).

The average data points per peak, peak capacity, and iRT stability were reproducible across the samples (Figures 2A–C). The principal component analysis plots, protein FDR, and coefficient of variance of the quality control samples using the top 50 differentially abundant proteins showed stability and reproducibility across all the samples (Figures 2D–F).

The differential abundance of proteins was analyzed using hierarchical cluster. From the identified proteins, 43 were found to be differentially expressed (10 upregulated and 33 downregulated proteins) in the SO group compared to the FA group (SO vs. FA), 120 (20 upregulated and 100 downregulated) in the SO group compared to the ERM group (SO vs. ERM), 100 (32 upregulated and 68 downregulated) in the RD group compared to the ERM group (RD vs. ERM); and 41 (five upregulated and 36 downregulated) in the FA group compared to the ERM group (FA vs. ERM) (fold change >1.5 or <−1.5 and p-value <0.05) (Table 2). Of these proteins, a heat map of clustered fold changes for the majority of differentially abundant proteins was summarized (Figures 3A,B). The overlapping changes between four groups of RD, ERM, FA, and SO were then shown in a Venn diagram (Figure 3C). Among the overlapping proteins, four protein groups were differentially abundant in all the groups, including profilin-1, coagulation factor X, immunoglobulin kappa variable 4-1, and immunoglobulin lambda-like polypeptide 1. Importantly, several differentially abundant proteins could be related to SiO treatment in analyzing SO vs. FA and SO vs. ERM profiling. The protein categories in the FA group were similar to the control.

The GO enrichment analysis of biological functions associated with the top 20 proteins which are differentially abundant in SO vs. FA showed that the SO group was enriched for proteins related to lipoprotein catabolic process, regulation of cell migration, synapse organization, morphogenesis of an epithelium, tube formation, associative learning, fat cell differentiation, and neural tube development (p-value <0.05). These proteins mainly functioned as cation transmembrane transporter, inorganic cation transmembrane transporter, iron transmembrane transporter, and in ATPase activity (p-value <0.05). They were mainly distributed in the extracellular region, synapse, glutamatergic synapse, and proton-transporting two-sector ATPase complex (p-value <0.05) (Figure 4A). The GO enrichment analysis of proteins differentially abundant in SO vs. REM showed that the SO group was enriched for proteins related to biological adhesion, cell adhesion, lipid metabolic process, negative regulation of response to stimulus, and negative regulation of Wnt signaling pathways (p-value <0.05). These proteins mainly play roles in iron binding, metal ion binding, cation binding, hyaluronic acid binding, transporter activity, and lipid transporter activity (p-value <0.05). The enriched proteins were mainly distributed in various components of the cell membrane, transport vesicle membrane, and apical part of cell (p-value <0.05) (Figure 4C).

The comparison analysis of RD vs. REM showed that the RD group was enriched for proteins related to cell recognition, immune response-regulating signaling pathway, immune response-activating signal transduction, antibacterial humoral response, and phagocytosis (p-value <0.05). These proteins mainly functioned in immunoglobulin receptor binding, glycosaminoglycan binding, signaling receptor binding, antigen binding, and hyaluronic acid binding (p-value <0.05). They were distributed in different cell components, such as membrane part, membrane, external side of the plasma membrane, cell surface, immunoglobulin complex, and circulation (p-value <0.05).

Using the KEGG pathway database, the differentially expressed proteins in the SO vs. FA and SO vs. ERM groups are shown in Figures 4B,D. Three pathways related to measles, fatty acid elongation, and hematopoietic cell lineage were significantly enriched in SO vs. FA. In addition, in the SO vs. ERM comparison, lysosomes and cell adhesion molecules were significantly enriched.

The KEGG analysis of the RD vs. ERM group revealed several main pathways associated with RD, including complement and coagulation cascades, cell adhesion molecules, platelet activation, glycolysis/gluconeogenesis, HIF-1 signaling, proteoglycans in cancer, antigen processing and presentation, microRNAs in cancer, and apoptosis. Among these, pathways related to cell adhesion molecules (CAMs), HIF-1 signaling pathway, platelet activation, propanoate metabolism, and glycolysis/gluconeogenesis were significantly enriched.

Protein–protein interaction network (PPI) analyses were then made by STRING database (https://string-db.org/) with high confidence (0.70) for four groups. The network is enriched in only two interactions between the total of 27 differential proteins in SO vs. FA with a PPI enrichment (p-value = 0.183). Furthermore, there were 78 interactions enriched between the 74 proteins found to be differentially expressed in SO vs. ERM with a PPI enrichment (p-value <1.0e-16). The PPI network was grouped into 11 relevant protein clusters using the Markov Cluster Algorithm (inflation parameter, 3) clustering option provided by STRING (Figure 5). These clusters share interactions among themselves, indicating that the molecules involved could play key roles in diverse pathways that contribute to SiO-related retinal cell pathology.

To infer the possible associated functions, clusters were classified according to Reactome pathways (Supplementary Tables S1, S2). According to the Reactome pathway analysis, cluster 1 (red) is larger and is associated with post-translational protein phosphorylation, regulation of insulin-like growth factor (IGF) transport, platelet degranulation, formation of fibrin clots, and metabolism of proteins. Cluster 1 (red) is composed of APLP2, CDH2, CHGA, CHGB, CPE, PRKCSH, SCG2, SCG3, SERPINC1, SPARCL1, and TF.

In addition, the PPI analysis of the RD vs. ERM group revealed 58 interactions to be enriched between the 54 proteins with a PPI enrichment (p-value <1.0e-16). Through the Reactome pathway analysis, one important and relevant protein cluster—cluster 1 (red)—is larger and associated with the regulation of post-translational protein phosphorylation, regulation of IGF transport, and glycosaminoglycan metabolism. In cluster 1 (red), the proteins included APLP2, GOLM1, HSP90B1, LUM, NCAN, PTPRZ1, SHISA5, SPARCL1, and VCAN.

After the discovery-based proteomics, transferrin was found to be an important node involved in SO vs. FA and SO vs. ERM group. Owing to the biological significance and also to the availability of the experiment, the protein was chosen for further testing using ELISA (YX-E201806H, Sinobestbio). The amount of transferrin was determined in 15 samples, including 10 ELISA-SO and five ELISA-FA. For the 10 eyes in SO, a level of 33.56 ± 1.292 mg/L transferrin was found, with a range of 31.52–35.14 mg/L. For the five eyes in ELISA-FA, a level of 36.63 ± 1.73 mg/L transferrin was found, with a range of 34.05–38.85 mg/L. For ELISA confirmation, transferrin was significantly downregulated in SO eyes relative to FA eyes (p < 0.05) (Figure 6).