MaXLinker: Proteome-wide Cross-link Identifications with High Specificity and Sensitivity

Abstract

Protein-protein interactions play a vital role in nearly all cellular functions. Hence, understanding their interaction patterns and three-dimensional structural conformations can provide crucial insights about various biological processes and underlying molecular mechanisms for many disease phenotypes. Cross-linking mass spectrometry (XL-MS) has the unique capability to detect protein-protein interactions at a large scale along with spatial constraints between interaction partners. The inception of MS-cleavable cross-linkers enabled the MS2-MS3 XL-MS acquisition strategy that provides cross-link information from both MS2 and MS3 level. However, the current cross-link search algorithm available for MS2-MS3 strategy follows a “MS2-centric” approach and suffers from a high rate of mis-identified cross-links. We demonstrate the problem using two new quality assessment metrics [“fraction of mis-identifications” (FMI) and “fraction of interprotein cross-links from known interactions” (FKI)]. We then address this problem, by designing a novel “MS3-centric” approach for cross-link identification and implementing it as a search engine named MaXLinker. MaXLinker outperforms the currently popular search engine with a lower mis-identification rate, and higher sensitivity and specificity. Moreover, we performed human proteome-wide cross-linking mass spectrometry using K562 cells. Employing MaXLinker, we identified a comprehensive set of 9319 unique cross-links at 1% false discovery rate, comprising 8051 intraprotein and 1268 interprotein cross-links. Finally, we experimentally validated the quality of a large number of novel interactions identified in our study, providing a conclusive evidence for MaXLinker's robust performance.

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

“…We observed that there was no significant difference (all P >0.85) among the three sets in terms of their percentage of residue pairs satisfying the distance constraint ( Fig.1c ), even though the overall qualities of these three sets are drastically different by design. Additionally, we utilized our recently published search engine, MaXLinker 2 , to repeat the analysis and observed similar results, confirming that these findings are software-independent ( Extended Data Fig.1a , 1b ). We further re-analyzed raw files from two other publicly available studies representing different organisms ( E. coli 13 and Mouse 10 ) and cellular compartments (mitochondria 10 ) ( Fig.2a , b and Extended Data Fig.2a , b ).…”

“…Especially since the true FDR at the protein pair level can be significantly higher than the estimated FDR at the peptide pair level 7 , 15 , absolute precision will be critically important for confirming the quality of novel protein-protein interactions identified in a large-scale cross-linking study. Finally, we confirmed the usefulness and robustness of the three computational metrics (namely FSI, FMI and FKI) on the re-analyzed E. coli ( Fig.2c - e ) and Mouse mitochondrial ( Extended Data Fig.2c - e ) XL-MS datasets, and using the additional search engine MaXLinker 2 ( Extended Data Fig.1c - e ).…”

“…Cross-linking mass spectrometry (XL-MS) is a powerful platform capable of unveiling protein interactions and capturing their structural dynamics 1 . The wealth of information from proteome-wide XL-MS approaches facilitate large-sale identification of protein-protein interactions 2 , 3 , and high-throughput three dimensional structural modeling of functional protein complexes 4 - 6 . With the increased throughput of these techniques, the number of false positive cross-links and incorrect interactions can quickly add up with just one large-scale XL-MS experiment, if one is not careful.…”

“…To demonstrate our theory experimentally, we obtained a subset of 122 raw files from our recent proteome-wide human K562 XL-MS study 2 . Next, in order to generate three sets of cross-links with drastically different qualities, we ran the XlinkX search engine (Proteome Discoverer 2.2) using three criteria of increasing stringency (‘10% FDR’, ‘1% FDR’ and ‘1% FDR with ΔXlinkX score≥50’; Methods).…”

“…(ii) Fraction of mis-identifications (FMI) : Including the proteome of an unrelated organism in the search database as an internal negative control can be an efficient way to independently assess the underlying error rate of the cross-link search algorithm 2 , 14 , 15 (Methods).…”