Structure-based validation can drastically under-estimate error rate in proteome-wide cross-linking mass spectrometry studies

Abstract

Recent, rapid advances in cross-linking mass spectrometry (XL-MS) has enabled detection of novel protein-protein interactions and their structural dynamics at the proteome scale. Given the importance and scale of the novel interactions identified in these proteome-wide XL-MS studies, thorough quality assessment is critical. Almost all current XL-MS studies validate cross-links against known 3D structures of representative protein complexes. However, current structure validation approach only includes cross-links where both peptides mapped to the 3D structures.Here we provide theoretical and experimental evidence demonstrating this approach can drastically underestimate error rates for proteome-wide XL-MS datasets. Addressing current shortcomings, we propose and demonstrate a comprehensive set of four metrics, including orthogonal experimental validation to thoroughly assess quality of proteome-wide XL-MS datasets.3 Cross-linking mass spectrometry (XL-MS) is a powerful platform capable of unveiling protein interactions and capturing their structural dynamics 1 . Although XL-MS techniques were once limited to studying individual functional complexes at a time, the development of efficient MScleavable chemical cross-linkerssuch as disuccinimidyl sulfoxide (DSSO) 2 , disuccinimidyl dibutyric urea (DSBU) 3 and protein interaction reporters (PIRs) 4has broadened the applicability of XL-MS to proteome scale 5-8 . With the increased throughput of these techniques, the number of false positive cross-links and incorrect interactions derived from them can quickly add up with just one large-scale XL-MS experiment, if one is not careful. Therefore, thorough quality assessment methods have become critically important. Computationally defined false discovery rates (FDR) provide a quantitative metric allowing researchers to filter lists of cross-link identifications, until a theoretical quality level is achieved. In addition to FDR calculations, almost all proteome-wide XL-MS studies leverage available 3D structures of representative complexes to provide a means of validation and quality assessment 9, 10 . Here, we demonstrate fundamental flaws in this structure-based quality assessment approach that can drastically underestimate the error rates of large-scale XL-MS datasets.The maximum theoretical distance that a given chemical cross-linker can span (e.g. 30Å for DSSO 11 ) can assess agreement between existing 3D protein structures and XL-MS datasets.Particularly in small-scale studies handling intraprotein cross-links, the fraction of cross-linked residue pairs that satisfy this distance constraint may provide meaningful insights into protein flexibility or the quality of the cross-links detected. In proteome-wide XL-MS studies, researchers extend this concept and use representative, highly abundant complexes such as the ribosome and proteasome to estimate the quality of all interprotein cross-links reported. However, the majority of false positive identifications come from these interprotein crosslinks 12 . Moreover, true pos...

Structural analysis of 70S ribosomes by cross-linking/mass spectrometry reveals conformational plasticity

“…There is still development in the field regarding the distance measure that is best applicable for mapping cross-links, i.e., Euclidian versus surface-exposed distance 28 . Also, the usefulness of FDR calculations and their correlation to true positive protein–protein interactions and the corresponding structural models 29 as well as the choice of molecular models that are used for cross-link mapping are in some cases suboptimal. This is because (a) current studies have a bias for high-abundant proteins, but methods to address this issue are being implemented 30 and (b) few molecular models deposited in structure databases are being evaluated for cross-linking distances, and only one molecular model is selected for distance calculation per protein complex, despite the wealth of structural data.…”

Structural Analysis of 70S Ribosomes by Cross-Linking/Mass Spectrometry Reveals Conformational Plasticity

“…There is still development in the field regarding the distance measure that is best applicable for mapping cross-links, i.e., Euclidian versus surface-exposed distance [25]. Also, the usefulness of FDR calculations and their correlation to true positive protein-protein interactions and the corresponding structural models [26] as well as the choice of molecular models that are used for cross-link mapping are in some cases suboptimal. This is because (a) current studies have a bias for high-abundant proteins, but methods to address this issue are being implemented [27] and (b) few molecular models deposited in structure databases are being evaluated for cross-linking distances, and only one molecular model is selected for distance calculation per protein complex, despite the wealth of structural data.…”

A synthetic peptide library for benchmarking crosslinking-mass spectrometry search engines for proteins and protein complexes

“…One limitation of such methods to test bioinformatic approaches is that the formation of non-specific crosslinks during the experimental procedure cannot be fully eliminated, and such a crosslink would wrongly be defined as a false positive. Furthermore, such an approach often underestimates FDR since the crosslinks that are evaluated are limited to those that are formed between peptides contained in the same protein model 23 . This lack of a generally accepted method on how to control for falsely identified crosslinks may have severe implications that are far-reaching in biological research.…”