TMT LC-MS/MS Data Analysis

Spectra acquired from LC-MS/MS experiments for the TMT experiments were processed using a Sequest-based software pipeline. First, a modified version of ReAdW.exe converted spectra to the mzXML format. These files were then searched against a database that contained the human proteome (Uniprot Database Organism ID: 9606, downloaded May 26, 2018) and Merkel Cell Polyoma Virus small and large T antigens concatenated to a database of all protein sequences reversed. For xenograft studies, a mouse proteome was appended to the database (Uniprot Database Organism ID: 10090, downloaded May 26, 2018) before reversal and concatenation. Mouse and human proteins are distinguished by their peptide identifications. The Sequest search engine takes care of assignments of peptides to mass spectra⁶⁸. A precursor ion tolerance of 50ppm and a production tolerance of 0.9Da were used as search parameters. Static modifications for TMT tags (+229.163Da) on lysine residues and the peptide’s N termini and carbamidomethylation (+57.021 Da) on cysteine residues were used in conjunction with a variable modification for oxidation (+15.995 Da) on methionine. Peptide-spectrum matches (PSMs) were then filtered using the linear discriminant analysis to a false discovery rate (FDR) of 1% as described previously⁶⁹. XCorr, ΔCn, missed cleavages, peptide length, charge state, and precursor mass accuracy were used as parameters for the LDA. The false discovery rate was estimated by using the target-decoy method. Peptides were identified and collapsed using principles of parsimony to a final protein-level FDR of 1%. For quantitation, we extracted the signal-to-noise (S:N) ratio of the closest matching centroid to the expected mass of the TMT reporter ion for each TMT channel from MS3 scans triggered by MS2 scans. MS3 spectra were filtered for a minimum TMT reporter ion sum S:N of 200 and an isolation specificity of at least 0.5. Protein level fold changes were calculated by comparing the sum signal to noise ratios for all observed peptides of a given protein. Proteins were ranked by nominal p-values obtained from Tukey’s Honest Significance Test performed posthoc on-peptide level linear models for each protein. These p-values were adjusted for multiple testing by employing Benjamini-Hochberg at a condition pairwise level to adjust false discovery rates. Pattern matching was performed by converting a desired protein expression pattern into a vector of 8 values, corresponding to the TMT channels used in the experiment, with each value taking a number between 0 (no expression) and 100 (full expression). This vector was compared to a vector of the scaled TMT values for each protein by calculating a cosine similarity value with values closer to 1 indicating more similarity.