252 proteins [37], and Kim et al. used a single peptide match as a minimum criteria to identify 518 protein matches [35]. In the latter study, a single unique peptide spectral was detected for about 100 proteins, which has a higher FDR compared with proteins detected based on at least two unique spectral-peptide matches. Moreover, the Gao et al. study use individual samples whereas the Kim et al. study used pooled samples and both nondepleted and immunoaffinity-depleted preparations. Thus, comparisons of protein lists from different studies should take into account both protein identification criteria and sample preparation.

Spectral data provides multiple options for both relative and absolute quantification of protein levels. The most widely used method for vitreous proteomics has been based on label-free measurements of spectral-peptide matches, using either the number of unique [36] or total spectral matches [37] for a given protein. Addition label-free options the use of multiple reaction monitoring [44] and analyses of ion intensity and spectral peak area [45]. The use of high mass accuracy and resolution mass spectrometers not only improves the sensitivity of these label-free methods but also creates more robust quantitative options that involve isotope-labeling techniques [46]. Quantitative proteomic methods are of central importance to characterizing the changing in proteins in diabetes and diabetic retinopathy, and the topic of quantitative proteomics has been extensively reviewed elsewhere [47, 48].

Data Analysis

Vitreous proteomics from multiple laboratories has generated lists of proteins detected in vitreous fluid along with quantitative data used for comparisons of protein levels among patients with or without diabetic retinopathy. As describe above, the parameters used to collect these data differ at multiple levels. Thus, while these studies provide different perspectives of the vitreous proteome, the assimilation of data from different reports is complex and often relies on manual techniques. The in-depth comprehension and comparison of proteomic dataset from different groups will likely require integration of these data with emerging bioinformatics tools and strategies [49].

In contrast, there are multiple options available for data analysis within a given proteomic database. Vitreous proteomic databases have been used for quantitative comparisons of protein abundance among groups of subjects, analysis of amino acid modifications and protein fragments, and grouping of proteins according to gene ontology and functional networks [37]. One important limitation of this bioinformatics approach in further understanding the vitreous proteome is that many of the proteins that have been identified in this fluid are not well characterized. Moreover, the functions of these proteins, as well as other more full-characterized proteins, in the vitreous compartment are largely unknown. Thus, in addition to the organization of vitreous proteome using computer algorithms and databases, it is likely that functional studies will be needed to assess the actions of individual proteins within the vitreous milieu.

THE VITREOUS PROTEOME

Two main proteomic approaches, based on 2-D and 1-D gel pre-fractionation, have been used to characterize protein composition of the human vitreous and identify changes associated with diabetic retinopathy. Although differences in experimental methods

(as described above) complicate the comparison of these studies and data, a number of findings from vitreous proteomics have emerged.

2-DE-Based Proteomics

The earliest comparative proteomic studies were performed using vitreous samples separated by two-dimensional electrophoresis (2-DE). Nakanishi et al. [38] compared silver-stained proteins separated by 2-D electrophoresis of vitreous obtained from subjects with MH and diabetic retinopathy. This study analyzed proteins from 412 spots separated by 2-DE of diabetic retinopathy vitreous and identified proteins in 113 of these spots, which represented 50 different proteins. Comparison of vitreous was normalized to 100 mg of dialyzed protein, and the authors reported that Ig, a1-antitrypsin, a2-HS glycoprotein, and complement factor 4, and pigmented epithelial-derived factor (PEDF) were elevated in vitreous from diabetic retinopathy. While this study, and others that visualized vitreous proteins by 2-DE, detected several hundred spots of protein staining, these include a large fraction of spots corresponding to protein isoforms separated along the IEF gradient.

A report by Yamane et al. [29] using 2-DE detected more than 400 silver-stains spots and identified 78 proteins in vitreous from patients with MH and 600 spots and identified 141 in vitreous from patients with PDR. This study showed that vitreous (both MH and PDR) and plasma displayed similar patterns of proteins, and most proteins that were identified to be increased in PDR compared with MH were also found in serum. Comparisons of vitreous were normalized to 40 mL of undiluted vitreous volume. The authors concluded that the increases in proteins in the PDR vitreous were the result of increased RVP and hemorrhage. Four proteins, including PEDF, prostaglandin-D2- synthase, plasma glutathione peroxidase, and IRBP were identified in MH vitreous but not in serum, suggesting that these proteins are locally produced in the eye [29]. An analysis of relative protein-staining intensity among gel spots indicates that the most highly abundant proteins in the vitreous include serum albumin, PEDF, a1-antitrypsin, prostaglandin-D2-synthase, apolipoprotein A1, and transthyretin. Ouchi et al. detected over 200 spots using SYPRO Ruby staining of vitreous and identified proteins in 72 spots from vitreous from non-proliferative diabetic retinopathy (NPDR) with DME and 64 spots from vitreous from subjects with NPDR without DME [40]. Comparisons were normalized to 15 mg of total protein. ApoH was detected in non-DME vitreous but not in DME vitreous. PEDF, plasma retinol-binding protein (PRBP), apo A4, apo A1, Trip-11, and vitamin D–binding protein were reported to be elevated in DME vitreous [40].

Garcia-Ramirez et al. [30] compared vitreous proteomes from PDR and MH subjects using fluorescence-based labeling differences in 2-DE. Vitreous samples were subjected to affinity depletion to removed albumin and IgG, and comparisons were normalized to 2-mg/mL protein eluate. This study reported that levels of eight proteins were increased in PDR vitreous, including zinc a2-glycoprotein, apo A1 and apoH, fibrinogen A, complement proteins C3, C4b, C9, and factor B. In addition, three proteins were identified to be decreased in PDR vitreous, including PEDF, IRBP, and inter-a-trypsin inhibitor heavy chain. Subsequent studies from this group further characterized the decrease in IRBP [50] and increased in apo A1 and apoH [51] in diabetic retinopathy.

Kim et al. [42] compare vitreous from subjects with MH and PDR. In this study, compared with MH, prostaglandin-H2 d-isomerase and PEDF were elevated, and a1- antitrypsin and beta V spectrin were reduced in PDR. Shitama et al. [31] compared the relative abundance of 105 proteins among approximately 400 spots visualized by 2-DE of vitreous samples collected from control subjects or patients with NPDR, PDR, RRD, or proliferative vitreoretinopathy. This study identified about ten proteins that were elevated in NPDR and PDR compared with control vitreous, including apo A4, complement C3, a1-B-glycoprotein, a1-antitrypsin, zinc a2-glycoprotein 1, vitamin D–binding protein, and fibrinogen g.

1-DE-Based Proteomics

Preparative 1-DE was also used in early studies to characterize the vitreous composition however comparative analyses of groups of samples required the development of databases and spectral-based quantitative methods. Koyama et al. [39] characterized the vitreous protein, separated by 1-DE, from a single subject with diabetic retinopathy. This report cataloged 84 different proteins in this vitreous sample.

Gao et al. [36] compared vitreous from three groups of subjects, including NDM, diabetes with no diabetic retinopathy (DM noDR), and PDR. This study identified 117 proteins, including 27 proteins that were elevated in vitreous from PDR compared with vitreous from NDM. This report revealed that PDR vitreous contains increased levels of a number of intracellular and plasma proteins, suggesting that retinal hemorrhage and increased RVP have a major impact on the composition of vitreous in diabetic retinopathy. A key observation generated from this work was that the effects of these newly discovered vitreous proteins on ocular functions were not readily apparent from previous descriptions of protein activities and subcellular locations. This report demonstrated that intravitreal injection of carbonic anhydrase I (CA-I) into rat vitreous increased RVP and retinal thickness via activation of the plasma kallikrein system [36]. The findings suggested a new pathway contributing to diabetic retinopathy which involved intraocular hemorrhage, lysis of erythrocytes to release intracellular CA-I, followed by activation of the kallikrein kinin system (Fig. 3). Moreover, beyond this specific pathway, this report demonstrated that the functions of proteins in the vitreous may not be readily inferred by previous descriptions of protein annotations, and that direct functional analyses of protein actions within the vitreous milieu may be needed to elucidate protein actions from the information generated by vitreous proteomics. Kim et al. [35] used both 2-DE and 1-DE fractionation methods to characterize both non-depleted and albumin/IgGdepleted vitreous from PDR and MH. Pooled samples were used, and comparisons of PDR and MH were normalized to 500 mg per lane for 1-DE. This study generated used multiple pre-fractionation methods and mass spectrometry platforms to generate the largest number of proteins identified from vitreous from diabetic retinopathy; however, the study was not designed to enable statistical comparisons among conditions.

Gao et al. [37] expanded the analyses of NDM, DM noDR, and PDR vitreous that was initiated previously [36]. This report identified 252 proteins in vitreous and used spectral-peptide counts to characterize the vitreous proteome. This analysis showed that albumin represents about 40% of the total soluble protein content (Fig. 4), and that the total spectral peptide content for albumin in PDR vitreous is increased by about

Fig. 3. Origins of vitreous proteins that been implicated in diabetic retinopathy progression. Diabetic retinopathy induces the release of active proteins into the vitreous by secretion (for example, VEGF), RVP (for example, plasma kallikrein), and retinal hemorrhages and cell lysis (for example, carbonic anhydrase I).

twoand fourfold compared with noDR and NDM vitreous, respectively. In addition to transport proteins, this analysis revealed that the protease inhibitor a1-antitrypsin, the anti-angiogenic factor PEDF, and complement C3 are highly abundant in PDR vitreous. This report also identified 56 proteins which differed in abundance in noDR and PDR compared with NDM. The majority of these changes were increases by twoto fourfold, which were comparable with increases in serum albumin (Fig. 5). For example, angiotensinogen (AGT) was show to be increased by twoto threefold in DM noDR and PDR vitreous. In addition, small subsets of proteins were increased by over tenfold or were decreased in noDR and PDR compared with NDM vitreous. As previously reported with CA-I, the functions of most of the vitreous proteins may require further study to evaluate their effects in the vitreous. This proteomic study also revealed that groups of proteins from the complement cascade, coagulation system, and kallikrein kinin system are present in the vitreous, suggesting that the vitreous proteome contains biochemical systems [37]. Further analyses revealed that a number of individual proteins existed as protein fragments, suggesting that the vitreous is proteolytically active, and certain protein functions may be associated with these fragments, as previously described for the anti-angiogenic factor endostatin, which is generated from the limited proteolysis of collagen XVIII [52].