Computational protocol: Critical Assessment of Protein Cross-Linking and Molecular Docking: An Updated Model for the Interaction Between Photosystem II and Psb27

Similar protocols

Protocol publication

[…] The analysis of cross-linking MS (CX-MS) data is especially challenging owing to the fact that structurally meaningful interpeptide cross-links are obscured by the vastly greater number of non-informative intrapeptide cross-links, dead-end cross-links, and free peptides. Moreover, mass spectrometric analysis of PSII samples is even more difficult, as this protein complex consists of around 20, mostly hydrophobic, subunits per monomer. However, the major problem lies in the overwhelming number of theoretically possible peptides, in particular if post-translational modifications other than the cross-link itself are included in the analysis. Furthermore, the experimental data set may include spectra of copurifying impurities, which are not included in the database used for CX-MS analyses. Hence, careful inspection of the results for false-positive MS2 assignments is mandatory. Thus, we used the high-resolution crystal structure of PSII () to test and validate our CX-MS approach.Two PSII samples (inactive, monomeric Psb27-PSII and active, dimeric PSII) were incubated with the isotope-coded cross-linker bis(sulfosuccinimidyl)suberate (BS3, spacer length: 11.4 Å), then digested with trypsin and analyzed by LC-ESI tandem mass spectrometry. Data analysis was performed with StavroX v. 3.1.19 ().As the false discovery rate represents one of the main difficulties in the identification of cross-linked peptides, most software tools provide a scoring function that estimates the probability of a false positive assignment. StavroX calculates a score based on the identified fragment ions for each precursor mass (). To establish a correlation between the score and the false discovery rate, the dataset is searched analogously against a decoy database consisting of reversed protein sequences. Previous validation of the software based on model systems yielded a false discovery rate of 2% for scores >100 ().We first attempted to identify only H12-BS3 cross-links in the two PSII data sets, in order to test the reliability of the software. When scores ≤100 were considered, the number of hits in the decoy database equaled the number of possible cross-linked peptides identified with the native database. For scores >100 the number of peptide identifications for the native sequences clearly exceeded that for the decoy database, but StavoX still reported 9 and 10 decoy hits compared to 65 and 80 hits for the native database in the two datasets, respectively. In addition, the results obtained with the native database contained false-positive discoveries with scores >100, although these “cross-links” clearly contradict the PSII crystal structure. Thus, despite the fact that it is based on nearly complete assignment of the entire fragment ion series, the result shown in Figure suggests that the N-terminal portion of CP43 is cross-linked to PsbV, although these sequences are located on different sides of the membrane. Even though the number of false discoveries is clearly lower than that of true hits, it might nevertheless account for the contradictory and artefactual structural models, since each suggested assignment is the basis for a structural constraint.Moreover, assignments that are based only on the identification of H12-BS3 cross-linked peptides might be ambiguous, even though the score calculated for each identification is above the threshold. Figure shows an MS2 fragment pattern that was assigned to cross-linked peptides of PsbL and CP47 (score 136) and PsbL and PsbX (score 100). Most of the observed fragments were assigned to the α-peptide (PsbL), whereas the identity of the β-peptide is ambiguous. In the upper spectrum only two (b-)ions of the β-peptide (CP47, red arrows) were found, and in the lower spectrum none of the PsbX-specific fragments were identified. Although the identification of the PsbL/CP47 cross-link is more reasonable – also from the structural perspective – there is still considerable uncertainty about the correct assignment.Hence, to avoid ambiguous assignments and to reduce the false discovery rate further, we included the deuterated cross-linker (D12-BS3) in our analysis. For each peptide containing a H12-BS3 modification the dataset should then include a second peptide with the same amino acid sequence modified by D12-BS3. The mass difference of 12 Da between the two versions of the cross-linker should be matched by the corresponding precursor ions and, due to the nearly identical chemical properties of H12-BS3 and D12-BS3, both peptides are expected to exhibit a very similar retention time and the same pattern of fragment ions. So, we considered only those peptides that fulfilled the following three criteria: (1) identification in the H12-BS3 and D12-BS3 modified form, (2) simultaneous elution, and (3) similarity of fragment-ion patterns. Using this more stringent approach, we identified 10 intermolecular and five intramolecular cross-links (Supplementary Table and Supplementary Figures ). In addition to the similar fragment patterns of the peptides cross-linked with H12-BS3 and D12-BS3, the fragment masses in both spectra provide additional information which strengthens the validity of the assignment: While ions that do not include the cross-linking BS3 molecule exhibit identical masses in both spectra (e.g., y6α/β + 1), fragments containing the cross-linker show the predicted mass shift of 12 Da/z (e.g., b10α/β + 2, b12α/β + 2, and b13α/β + 2) as illustrated for the PsbU–PsbU cross-link (Figure ).Finally, it is noteworthy that the number of identified peptides exceeds the number of cross-links because, as a result of missed cleavages and methionine oxidation, some of the peptides included the same cross-link (e.g., a cross-link was identified in the oxidized and the reduced form of amethionine-containing peptide). Moreover, highly abundant peptides showed broad elution peaks and were identified several times. [...] The high-resolution crystal structure of mature PSII should facilitate an in silico approach to the prediction of potential cross-links in the protein complex. We used the Xwalk algorithm () to calculate solvent-accessible surface distances between lysine residues in PSII. Potential cross-links were limited to a maximum distance of 35 Å, as the distance constraint for BS3 is 11.4 Å and the size of two lysine side-chains is 13 Å. The additional 10 Å was added to allow for conformational dynamics. Interestingly, Xwalk did not predict the majority of cross-links that were identified by the MS approach. The reason for this unexpected result becomes obvious on closer inspection of the crystal structure of PSII (). Although the structure was solved with a high resolution of 1.9 Å, many of the protein termini are not resolved at all, due to their enhanced flexibility. On the other hand, in 14 of 15 cross-links a protein N-terminus provided at least one primary amine. This leads to the model of easy accessible and flexible cross-linking hotspots, like the N-termini of PsbO and PsbL (Figure ), that readily react with the cross-linker. As a consequence, the partly fixed cross-linker scans the surface within a 35-Å radius for a second reactive side-chain (preferably lysine, but also serine or threonine). Thus, we never observed cross-links between protein–protein interfaces that are hidden in the complex. Instead, surface-exposed, flexible parts of PSII are the preferred cross-linking sites. […]

Pipeline specifications

Software tools StavroX, Xwalk
Application MS-based untargeted proteomics
Chemicals Plastoquinone