Tear proteome analysis in ocular surface diseases using label-free LC-MS/MS and multiplexed-microarray biomarker validation

Abstract

We analyzed the tear film proteome of patients with dry eye (DE), meibomian gland dysfunction (MGD), and normal volunteers (CT). Tear samples were collected from 70 individuals. Of these, 37 samples were analyzed using spectral-counting-based LC-MS/MS label-free quantitation, and 33 samples were evaluated in the validation of candidate biomarkers employing customized antibody microarray assays. Comparative analysis of tear protein profiles revealed differences in the expression levels of 26 proteins, including protein S100A6, annexin A1, cystatin-S, thioredoxin, phospholipase A2, antileukoproteinase, and lactoperoxidase. Antibody microarray validation of CST4, S100A6, and MMP9 confirmed the accuracy of previously reported ELISA assays, with an area under ROC curve (AUC) of 87.5%. Clinical endpoint analysis showed a good correlation between biomarker concentrations and clinical parameters. In conclusion, different sets of proteins differentiate between the groups. Apolipoprotein D, S100A6, S100A8, and ceruloplasmin discriminate best between the DE and CT groups. The differences between antileukoproteinase, phospholipase A2, and lactoperoxidase levels allow the distinction between MGD and DE, and the changes in the levels of annexin A1, clusterin, and alpha-1-acid glycoprotein 1, between MGD and CT groups. The functional network analysis revealed the main biological processes that should be examined to identify new candidate biomarkers and therapeutic targets.

Figure 1

Workflow overview. A general workflow outlining the steps used in the study for the identification of proteins deregulated in DE and MGD tear samples, using LC-MS/MS label-free quantitative proteomics, candidate biomarkers validation, and clinical correlation strategy.

Figure 2

Canonical Discriminant Analysis results showing the separation between the samples as a function of the values for the 26 discriminant proteins obtained by stepwise discriminant analysis. Each of the points represents a sample from each group. Good separation between the groups is apparent. The MGD group was closer to the control group (CT) than the DE group. Key: squares, CT group; triangles, MGD group; circles, DE group.

Figure 3

The functional interaction network obtained using the Functional Interaction Reactome Cytoscape plugin. Proteins in the interaction network are represented as nodes (red dashed-line circles), while the interaction between any two proteins is represented by a line. These interactions can be direct (physical) and/or indirect (functional) in nature. Node colors indicate different modules obtained by topological clustering of the network using the functional interaction clustering method.

Figure 4

Antibody microarrays. (A) Spotting pattern for the 12-subarray format. (B) Representative image of the arrays showing the distribution of the standard calibration curve and samples. Each microarray slide contains a standard calibration curve (left column) and the fluorescence acquisition for 5 tear samples. Fluorescence scans of the microarray multiplex assays were acquired at 633 nm.

Figure 5

Comparison of the concentrations of candidate biomarkers in control (CT) and dry eye (DE) groups, measured using the customized antibody microarrays. Concentration is expressed in ng/mL. Green circles represent the mean concentration for the group. (A) Whisker plot of protein S100A6 concentrations. (B) Whisker plot of protein MMP9 concentrations. (C) Whisker plot of protein CST4 concentrations. (D) ROC curve analysis obtained by logistic regression using the three candidate biomarkers.