Combinatorial Approach for Large-scale Identification of Linked Peptides from Tandem Mass Spectrometry Spectra

The combination of chemical cross-linking and mass spectrometry has recently been shown to constitute a powerful tool for studying protein–protein interactions and elucidating the structure of large protein complexes. However, computational methods for interpreting the complex MS/MS spectra from linked peptides are still in their infancy, making the high-throughput application of this approach largely impractical. Because of the lack of large annotated datasets, most current approaches do not capture the specific fragmentation patterns of linked peptides and therefore are not optimal for the identification of cross-linked peptides. Here we propose a generic approach to address this problem and demonstrate it using disulfide-bridged peptide libraries to (i) efficiently generate large mass spectral reference data for linked peptides at a low cost and (ii) automatically train an algorithm that can efficiently and accurately identify linked peptides from MS/MS spectra. We show that using this approach we were able to identify thousands of MS/MS spectra from disulfide-bridged peptides through comparison with proteome-scale sequence databases and significantly improve the sensitivity of cross-linked peptide identification. This allowed us to identify 60% more direct pairwise interactions between the protein subunits in the 20S proteasome complex than existing tools on cross-linking studies of the proteasome complexes. The basic framework of this approach and the MS/MS reference dataset generated should be valuable resources for the future development of new tools for the identification of linked peptides.The study of protein–protein interactions is crucial to understanding how cellular systems function because proteins act in concert through a highly organized set of interactions. Most cellular processes are carried out by large macromolecular assemblies and regulated through complex cascades of transient protein–protein interactions (1). In the past several years numerous high-throughput studies have pioneered the systematic characterization of protein–protein interactions in model organisms (2–4). Such studies mainly utilize two techniques: the yeast two-hybrid system, which aims at identifying binary interactions (5), and affinity purification combined with tandem mass spectrometry analysis for the identification of multi-protein assemblies (6–8). Together these led to a rapid expansion of known protein–protein interactions in human and other model organisms. Patche and Aloy recently estimated that there are more than one million interactions catalogued to date (9).But despite rapid progress, most current techniques allow one to determine only whether proteins interact, which is only the first step toward understanding how proteins interact. A more complete picture comes from characterizing the three-dimensional structures of protein complexes, which provide mechanistic insights that govern how interactions occur and the high specificity observed inside the cell. Traditionally the gold-standard methods used to solve protein structures are x-ray crystallography and NMR, and there have been several efforts similar to structural genomics (10) aiming to comprehensively solve the structures of protein complexes (11, 12). Although there has been accelerated growth of structures for protein monomers in the Protein Data Bank in recent years (11), the growth of structures for protein complexes has remained relatively small (9). Many factors, including their large size, transient nature, and dynamics of interactions, have prevented many complexes from being solved via traditional approaches in structural biology. Thus, the development of complementary analytical techniques with which to probe the structure of large protein complexes continues to evolve (13–18).Recent developments have advanced the analysis of protein structures and interaction by combining cross-linking and tandem mass spectrometry (17, 19–24). The basic idea behind this technique is to capture and identify pairs of amino acid residues that are spatially close to each other. When these linked pairs of residues are from the same protein (intraprotein cross-links), they provide distance constraints that help one infer the possible conformations of protein structures. Conversely, when pairs of residues come from different proteins (interprotein cross-links), they provide information about how proteins interact with one another. Although cross-linking strategies date back almost a decade (25, 26), difficulty in analyzing the complex MS/MS spectrum generated from linked peptides made this approach challenging, and therefore it was not widely used. With recent advances in mass spectrometry instrumentation, there has been renewed interest in employing this strategy to determine protein structures and identify protein–protein interactions. However, most studies thus far have been focused on purified protein complexes. With today''s mass spectrometers being capable of analyzing tens of thousands of spectra in a single experiment, it is now potentially feasible to extend this approach to the analysis of complex biological samples. Researchers have tried to realize this goal using both experimental and computational approaches. Indeed, a plethora of chemical cross-linking reagents are now available for stabilizing these complexes, and some are designed to allow for easier peptide identification when employed in concert with MS analysis (20, 27, 28). There have also been several recent efforts to develop computational methods for the automatic identification of linked peptides from MS/MS spectra (29–36). However, because of the lack of large annotated training data, most approaches to date either borrow fragmentation models learned from unlinked, linear peptides or learn the fragmentation statistics from training data of limited size (30, 37), which might not generalize well across different samples. In some cases it is possible to generate relatively large training data, but it is often very labor intensive and involves hundreds of separate LC-MS/MS runs (36). Here, employing disulfide-bridged peptides as an example, we propose a novel method that uses a combinatorial peptide library to (a) efficiently generate a large mass spectral reference dataset for linked peptides and (b) use these data to automatically train our new algorithm, MXDB, which can efficiently and accurately identify linked peptides from MS/MS spectra.