2.5. MS/MS data curation

Data were analysed automatically using the SequestHT search engine with the Percolator algorithm for validation of protein identifications (Proteome Discoverer 1.4; Thermo Scientific). Data were searched against the human proteome obtained from the SwissProt database on 22/11/2015, selecting trypsin as the enzyme and allowing for up to two missed cleavage sites, a precursor ion mass tolerance of 10 p.p.m. and 0.6 Da for product ions. Carbamidomethylcysteine was selected as a fixed modification, while oxidation of methionine (+15.994u), modification of serine and threonine with HexNac (+203.08u), and/or HexNacNeuNac (STn) (+494.17u), considering the possibility of partially inefficient α‐neuraminidase treatment, and/or T (+365.13u) were defined as variable modifications. For whole tumour proteome analysis, only high confidence peptides were considered. In glycoproteomics studies, due to the high lability of the sugar moieties under CID conditions, and the consequent difficulty in identifying modified peptides, Sequest results of low confidence peptides were also considered. Protein grouping filters were thus set to consider glycosylations with low confidence and ΔCn better than 0.05. The strict maximum parsimony principle was applied. A protein filter counting peptides only on top‐scored proteins was also set. Peptides were filtered for Xcorr ≥ 1.0 and ΔCn ≤ 0.05. Cytoplasm membrane proteins with at least one annotated glycosylation site were selected and the modifications were validated manually. Membrane proteins were sorted using NetOGlyc version 4.0 (http://www.cbs.dtu.dk/services/NetOGlyc/) (Steentoft et al., 2013) to generate the final protein list. Protein molecular and biological functions were interpreted using Panther (Mi et al., 2016).