Verification by multiple reaction monitoring

Proteotypic peptide transitions for MRM-based targeted proteomics analysis were selected using Skyline v. 19.1.0.193 (RRID : SCR_014080). MRM assays were performed in an Eksigent nanoLC Ultra 1D plus system (AB SCIEX, Foster City, CA) coupled to a SCIEX 5500 QTRAP 5500 Mass Spectrometer (RRID : SCR_020517)via a Nanospray III source. A scheduled method was designed for the relative quantitation of 35 proteins, using 2-3 proteotypic peptides per protein and 3-4 transitions per peptide (332 transitions in total). The Homo sapiens UniProtKB reviewed database was used as background proteome. The selected enzyme was trypsin/P [KR |-] and peptide parameters were set to: a length range of 8 to 25 amino acids, 2+ and 3+ charged, no missed cleavages, and potentially modified residues such as methionine (Met, M) and cysteine (Cys, C). When possible, peptides were selected to cover distinct regions of the protein sequence. As described above, 10 µg of non-depleted vitreous samples were loaded on an SDS-PAGE gel and in-gel digested. Peptide concentration was determined using Thermo Fisher Qubit fluorimeter (RRID : SCR_018095), according to manufacturer’s instructions, and 1 µg of tryptic peptides was loaded onto a C18 Acclaim PepMapTM 100 column (Thermo Scientific, 300 µm I.D. × 5 cm, 5 µm particle diameter, 100 Å) using solvent A (2%B ACN, 0.1% formic acid in water) at 2 µL/min. After desalting, the trap column was switched online with a C18 BioSphere column (Nano-separations, 75 µm I.D. × 15 cm, 3 µm particle diameter, 120 Å) and peptides were fractionated in a 30 min gradient (4 to 90% of 100% ACN, 0.1% formic acid) at 300 nL/min, followed by 15 min of equilibration to initial conditions. The 5500 QTRAP system was operated in positive polarity and MRM scan mode, with an ion spray voltage of 2800 V, IHT of 150 °C, CUR of 20, GS1 of 25, medium collision gas, and DP of 80 V. Scheduled mode was enabled and detection window set at 300 sec. Collision energy and expected retention time for each transition were defined in Skyline. Beta-galactosidase standards and a pool of vitreous samples were injected alternately with the vitreous samples to monitor oscillations in the MS signal and in the retention time. Raw MS data were imported into Skyline and the automatically selected transition peaks were manually revised considering the retention time and the intensity distribution of the selected transitions. The total area of each protein was calculated by summing the area of the respective peptides (calculated as the sum of all peptide transitions). To correct the fluctuations in MS signal over time, the calculated total area of each protein was normalized by dividing it by the total area of digested beta-galactosidase (injected between each batch) and multiplying by the median. Statistical analysis by one-way ANOVA (Tukey’s HSD) and post-hoc tests and multivariate statistical analyses were performed using Metaboanalyst v5.0 (RRID : SCR_015539) (41). Partial least squares discriminant analysis (PLS-DA) was used to build a predictive model to define a panel of discriminatory biomarkers of vitreoretinal diseases. The predictive ability (Q2), R-Squared (R2), and accuracy of the model were calculated via cross-validation to define the optimal number of components for classification. Classical univariate and multivariate exploratory receiver operating characteristic (ROC) analyses were performed to evaluate the diagnostic potential of discriminatory proteins between the disease groups. ROC curves were generated in multivariate exploratory analyses by Monte-Carlo cross-validation using balanced sub-sampling, in which two-thirds of the samples are used to evaluate the feature importance. For model building, PLS-DA was defined as classification method and PLS-DA built-in as the feature ranking method, while the number of latent variables was defined to 2.