Identification of Protein Complexes by Integrating Protein Abundance and Interaction Features Using a Deep Learning Strategy.

Many essential cellular functions are carried out by multi-protein complexes that can be characterized by their protein-protein interactions. The interactions between protein subunits are critically dependent on the strengths of their interactions and their cellular abundances, both of which span orders of magnitude. Despite many efforts devoted to the global discovery of protein complexes by integrating large-scale protein abundance and interaction features, there is still room for improvement. Here, we integrated >7000 quantitative proteomic samples with three published affinity purification/co-fractionation mass spectrometry datasets into a deep learning framework to predict protein-protein interactions (PPIs), followed by the identification of protein complexes using a two-stage clustering strategy. Our deep-learning-technique-based classifier significantly outperformed recently published machine learning prediction models and in the process captured 5010 complexes containing over 9000 unique proteins. The vast majority of proteins in our predicted complexes exhibited low or no tissue specificity, which is an indication that the observed complexes tend to be ubiquitously expressed throughout all cell types and tissues. Interestingly, our combined approach increased the model sensitivity for low abundant proteins, which amongst other things allowed us to detect the interaction of MCM10, which connects to the replicative helicase complex via the MCM6 protein. The integration of protein abundances and their interaction features using a deep learning approach provided a comprehensive map of protein-protein interactions and a unique perspective on possible novel protein complexes.

Proteogenomic Features of the Highly Polymorphic Histidine-rich Glycoprotein Arose Late in Evolution.

Histidine-rich glycoprotein (HRG) is a liver-produced protein circulating in human serum at high concentrations of around 125 µg/ml. HRG belongs to the family of type-3 cystatins and has been implicated in a plethora of biological processes, albeit that its precise function is still not well understood. Human HRG is a highly polymorphic protein, with at least five variants with minor allele frequencies of more than 10%, variable in populations from different parts of the world. Considering these five mutations we can theoretically expect 35 = 243 possible genetic HRG variants in the population. Here, we purified HRG from serum of 44 individual donors and investigated by proteomics the occurrence of different allotypes, each being either homozygote or heterozygote for each of the five mutation sites. We observed that some mutational combinations in HRG were highly favored, while others were apparently missing, although they ought to be present based on the independent assembly of these five mutation sites. To further explore this behavior, we extracted data from the 1000 genome project (n ∼ 2500 genomes) and assessed the frequency of different HRG mutants in this larger dataset, observing a prevailing agreement with our proteomics data. From all the proteogenomic data we conclude that the five different mutation sites in HRG are not occurring independently, but several mutations at different sites are fully mutually exclusive, whereas others are highly intwined. Specific mutations do also affect HRG glycosylation. As the levels of HRG have been suggested as a protein biomarker in a variety of biological processes (e.g., aging, COVID-19 severity, severity of bacterial infections), we here conclude that the highly polymorphic nature of the protein needs to be considered in such proteomics evaluations, as these mutations may affect HRG's abundance, structure, posttranslational modifications, and function.

Schistosoma mansoni infection affects the proteome and lipidome of circulating extracellular vesicles in the host.

Eggs, schistosomula and adult Schistosoma worms are known to release extracellular vesicles (EV) during in vitro incubations and these EVs are postulated to affect the host responses. So far only those EVs released during in vitro incubations of schistosomes have been studied and it is unknown whether in blood of infected hosts the schistosomal EVs can be detected amidst all the circulating EVs of the host itself. In this study we analyzed the protein as well as the phospholipid composition of EVs circulating in blood plasma of S. mansoni infected hamsters and compared those with the EVs circulating in blood of non-infected hamsters. Although neither proteins nor lipids specific for schistosomes could be detected in the circulating EVs of the infected hamsters, the infection with schistosomes had a marked effect on the circulating EVs of the host, as the protein as well as the lipid composition of EVs circulating in infected hamsters were different from the EVs of uninfected hamsters. The observed changes in the EV lipid and protein content suggest that more EVs are released by the diseased liver, the affected erythrocytes and activated immune cells.

An Augmented Multiple-Protease-Based Human Phosphopeptide Atlas.

Although mass-spectrometry-based screens enable thousands of protein phosphorylation sites to be monitored simultaneously, they often do not cover important regulatory sites. Here, we hypothesized that this is due to the fact that nearly all large-scale phosphoproteome studies are initiated by trypsin digestion. We tested this hypothesis using multiple proteases for protein digestion prior to Ti(4+)-IMAC-based enrichment. This approach increases the size of the detectable phosphoproteome substantially and confirms the considerable tryptic bias in public repositories. We define and make available a less biased human phosphopeptide atlas of 37,771 unique phosphopeptides, correlating to 18,430 unique phosphosites, of which fewer than 1/3 were identified in more than one protease data set. We demonstrate that each protein phosphorylation site can be linked to a preferred protease, enhancing its detection by mass spectrometry (MS). For specific sites, this approach increases their detectability by more than 1,000-fold.

Spatial Organization in Protein Kinase A Signaling Emerged at the Base of Animal Evolution.

In phosphorylation-directed signaling, spatial and temporal control is organized by complex interaction networks that diligently direct kinases toward distinct substrates to fine-tune specificity. How these protein networks originate and evolve into complex regulatory machineries are among the most fascinating research questions in biology. Here, spatiotemporal signaling is investigated by tracing the evolutionary dynamics of each functional domain of cAMP-dependent protein kinase (PKA) and its diverse set of A-kinase anchoring proteins (AKAPs). Homologues of the catalytic (PKA-C) and regulatory (PKA-R) domains of the (PKA-R)2-(PKA-C)2 holoenzyme were found throughout evolution. Most variation was observed in the RIIa of PKA-R, crucial for dimerization and docking to AKAPs. The RIIa domain was not observed in all PKA-R homologues. In the fungi and distinct protist lineages, the RIIa domain emerges within PKA-R, but it displays large sequence variation. These organisms do not harbor homologues of AKAPs, suggesting that efficient docking to direct spatiotemporal PKA activity evolved in multicellular eukaryotes. To test this in silico hypothesis, we experimentally screened organisms with increasing complexity by cAMP-based chemical proteomics to reveal that the occurrence of PKA-AKAP interactions indeed coincided and expanded within vertebrates, suggesting a crucial role for AKAPs in the advent of metazoan multicellularity.

ROCK1 is a potential combinatorial drug target for BRAF mutant melanoma.

Treatment of BRAF mutant melanomas with specific BRAF inhibitors leads to tumor remission. However, most patients eventually relapse due to drug resistance. Therefore, we designed an integrated strategy using (phospho)proteomic and functional genomic platforms to identify drug targets whose inhibition sensitizes melanoma cells to BRAF inhibition. We found many proteins to be induced upon PLX4720 (BRAF inhibitor) treatment that are known to be involved in BRAF inhibitor resistance, including FOXD3 and ErbB3. Several proteins were down-regulated, including Rnd3, a negative regulator of ROCK1 kinase. For our genomic approach, we performed two parallel shRNA screens using a kinome library to identify genes whose inhibition sensitizes to BRAF or ERK inhibitor treatment. By integrating our functional genomic and (phospho)proteomic data, we identified ROCK1 as a potential drug target for BRAF mutant melanoma. ROCK1 silencing increased melanoma cell elimination when combined with BRAF or ERK inhibitor treatment. Translating this to a preclinical setting, a ROCK inhibitor showed augmented melanoma cell death upon BRAF or ERK inhibition in vitro. These data merit exploration of ROCK1 as a target in combination with current BRAF mutant melanoma therapies.

Facilitating protein disulfide mapping by a combination of pepsin digestion, electron transfer higher energy dissociation (EThcD), and a dedicated search algorithm SlinkS.

Disulfide bond identification is important for a detailed understanding of protein structures, which directly affect their biological functions. Here we describe an integrated workflow for the fast and accurate identification of authentic protein disulfide bridges. This novel workflow incorporates acidic proteolytic digestion using pepsin to eliminate undesirable disulfide reshuffling during sample preparation and a novel search engine, SlinkS, to directly identify disulfide-bridged peptides isolated via electron transfer higher energy dissociation (EThcD). In EThcD fragmentation of disulfide-bridged peptides, electron transfer dissociation preferentially leads to the cleavage of the S-S bonds, generating two intense disulfide-cleaved peptides as primary fragment ions. Subsequently, higher energy collision dissociation primarily targets unreacted and charge-reduced precursor ions, inducing peptide backbone fragmentation. SlinkS is able to provide the accurate monoisotopic precursor masses of the two disulfide-cleaved peptides and the sequence of each linked peptide by matching the remaining EThcD product ions against a linear peptide database. The workflow was validated using a protein mixture containing six proteins rich in natural disulfide bridges. Using this pepsin-based workflow, we were able to efficiently and confidently identify a total of 31 unique Cys-Cys bonds (out of 43 disulfide bridges present), with no disulfide reshuffling products detected. Pepsin digestion not only outperformed trypsin digestion in terms of the number of detected authentic Cys-Cys bonds, but, more important, prevented the formation of artificially reshuffled disulfide bridges due to protein digestion under neutral pH. Our new workflow therefore provides a precise and generic approach for disulfide bridge mapping, which can be used to study protein folding, structure, and stability.

Proteome adaptation of Saccharomyces cerevisiae to severe calorie restriction in Retentostat cultures.

Stationary-phase, carbon-starved shake-flask cultures of Saccharomyces cerevisiae are popular models for studying eukaryotic chronological aging. However, their nutrient-starved physiological status differs substantially from that of postmitotic metazoan cells. Retentostat cultures offer an attractive alternative model system in which yeast cells, maintained under continuous calorie restriction, hardly divide but retain high metabolic activity and viability for prolonged periods of time. Using TMT labeling and UHPLC-MS/MS, the present study explores the proteome of yeast cultures during transition from exponential growth to near-zero growth in severely calorie-restricted retentostats. This transition elicited protein level changes in 20% of the yeast proteome. Increased abundance of heat shock-related proteins correlated with increased transcript levels of the corresponding genes and was consistent with a strongly increased heat shock tolerance of retentostat-grown cells. A sizable fraction (43%) of the proteins with increased abundance under calorie restriction was involved in oxidative phosphorylation and in various mitochondrial functions that, under the anaerobic, nongrowing conditions used, have a very limited role. Although it may seem surprising that yeast cells confronted with severe calorie restriction invest in the synthesis of proteins that, under those conditions, do not contribute to fitness, these responses may confer metabolic flexibility and thereby a selective advantage in fluctuating natural habitats.

qcML: an exchange format for quality control metrics from mass spectrometry experiments.

Quality control is increasingly recognized as a crucial aspect of mass spectrometry based proteomics. Several recent papers discuss relevant parameters for quality control and present applications to extract these from the instrumental raw data. What has been missing, however, is a standard data exchange format for reporting these performance metrics. We therefore developed the qcML format, an XML-based standard that follows the design principles of the related mzML, mzIdentML, mzQuantML, and TraML standards from the HUPO-PSI (Proteomics Standards Initiative). In addition to the XML format, we also provide tools for the calculation of a wide range of quality metrics as well as a database format and interconversion tools, so that existing LIMS systems can easily add relational storage of the quality control data to their existing schema. We here describe the qcML specification, along with possible use cases and an illustrative example of the subsequent analysis possibilities. All information about qcML is available at http://code.google.com/p/qcml.

Daily rhythms in the cyanobacterium synechococcus elongatus probed by high-resolution mass spectrometry-based proteomics reveals a small defined set of cyclic proteins.

Circadian rhythms are self-sustained and adjustable cycles, typically entrained with light/dark and/or temperature cycles. These rhythms are present in animals, plants, fungi, and several bacteria. The central mechanism behind these "pacemakers" and the connection to the circadian regulated pathways are still poorly understood. The circadian rhythm of the cyanobacterium Synechococcus elongatus PCC 7942 (S. elongatus) is highly robust and controlled by only three proteins, KaiA, KaiB, and KaiC. This central clock system has been extensively studied functionally and structurally and can be reconstituted in vitro. These characteristics, together with a relatively small genome (2.7 Mbp), make S. elongatus an ideal model system for the study of circadian rhythms. Different approaches have been used to reveal the influence of the central S. elongatus clock on rhythmic gene expression, rhythmic mRNA abundance, rhythmic DNA topology changes, and cell division. However, a global analysis of its proteome dynamics has not been reported yet. To uncover the variation in protein abundances during 48 h under light and dark cycles (12:12 h), we used quantitative proteomics, with TMT 6-plex isobaric labeling. We queried the S. elongatus proteome at 10 different time points spanning a single 24-h period, leading to 20 time points over the full 48-h period. Employing multidimensional separation and high-resolution mass spectrometry, we were able to find evidence for a total of 82% of the S. elongatus proteome. Of the 1537 proteins quantified over the time course of the experiment, only 77 underwent significant cyclic variations. Interestingly, our data provide evidence for in- and out-of-phase correlation between mRNA and protein levels for a set of specific genes and proteins. As a range of cyclic proteins are functionally not well annotated, this work provides a resource for further studies to explore the role of these proteins in the cyanobacterial circadian rhythm.

Identification of enriched PTM crosstalk motifs from large-scale experimental data sets.

Post-translational modifications (PTMs) play an important role in the regulation of protein function. Mass spectrometry based proteomics experiments nowadays identify tens of thousands of PTMs in a single experiment. A wealth of data has therefore become publically available. Evidently the biological function of each PTM is the key question to be addressed; however, such analyses focus primarily on single PTM events. This ignores the fact that PTMs may act in concert in the regulation of protein function, a process termed PTM crosstalk. Relatively little is known on the frequency and functional relevance of crosstalk between PTM sites. In a bioinformatics approach, we extracted PTMs occurring in proximity in the protein sequence from publically available databases. These PTMs and their flanking sequences were subjected to stringent motif searches, including a scoring for evolutionary conservation. Our unprejudiced approach was able to detect a respectable set of motifs, of which about half were described previously. Among these we could add many new proteins harboring these motifs. We extracted also several novel motifs, which through their widespread appearance and high conservation may pinpoint at previously nonannotated concerted PTM actions. By employing network analyses on these proteins, we propose putative functional roles for these novel motifs with two PTM sites in close proximity.

Quantitative and qualitative proteome characteristics extracted from in-depth integrated genomics and proteomics analysis.

Quantitative and qualitative protein characteristics are regulated at genomic, transcriptomic, and posttranscriptional levels. Here, we integrated in-depth transcriptome and proteome analyses of liver tissues from two rat strains to unravel the interactions within and between these layers. We obtained peptide evidence for 26,463 rat liver proteins. We validated 1,195 gene predictions, 83 splice events, 126 proteins with nonsynonymous variants, and 20 isoforms with nonsynonymous RNA editing. Quantitative RNA sequencing and proteomics data correlate highly between strains but poorly among each other, indicating extensive nongenetic regulation. Our multilevel analysis identified a genomic variant in the promoter of the most differentially expressed gene Cyp17a1, a previously reported top hit in genome-wide association studies for human hypertension, as a potential contributor to the hypertension phenotype in SHR rats. These results demonstrate the power of and need for integrative analysis for understanding genetic control of molecular dynamics and phenotypic diversity in a system-wide manner.

Deep proteome profiling of circulating granulocytes reveals bactericidal/permeability-increasing protein as a biomarker for severe atherosclerotic coronary stenosis.

Coronary atherosclerosis represents the major cause of death in Western societies. As atherosclerosis typically progresses over years without giving rise to clinical symptoms, biomarkers are urgently needed to identify patients at risk. Over the past decade, evidence has accumulated suggesting cross-talk between the diseased vasculature and cells of the innate immune system. We therefore employed proteomics to search for biomarkers associated with severe atherosclerotic coronary lumen stenosis in circulating leukocytes. In a two-phase approach, we first performed in-depth quantitative profiling of the granulocyte proteome on a small pooled cohort of patients suffering from chronic (sub)total coronary occlusion and matched control patients using stable isotope peptide labeling, two-dimensional LC-MS/MS and data-dependent decision tree fragmentation. Over 3000 proteins were quantified, among which 57 candidate biomarker proteins remained after stringent filtering. The most promising biomarker candidates were subsequently verified in the individual samples of the discovery cohort using label-free, single-run LC-MS/MS analysis, as well as in an independent verification cohort of 25 patients with total coronary occlusion (CTO) and 19 matched controls. Our data reveal bactericidal/permeability-increasing protein (BPI) as a promising biomarker for severe atherosclerotic coronary stenosis, being down-regulated in circulating granulocytes of CTO patients.

The PRoteomics IDEntification (PRIDE) Converter 2 framework: an improved suite of tools to facilitate data submission to the PRIDE database and the ProteomeXchange consortium.

The original PRIDE Converter tool greatly simplified the process of submitting mass spectrometry (MS)-based proteomics data to the PRIDE database. However, after much user feedback, it was noted that the tool had some limitations and could not handle several user requirements that were now becoming commonplace. This prompted us to design and implement a whole new suite of tools that would build on the successes of the original PRIDE Converter and allow users to generate submission-ready, well-annotated PRIDE XML files. The PRIDE Converter 2 tool suite allows users to convert search result files into PRIDE XML (the format needed for performing submissions to the PRIDE database), generate mzTab skeleton files that can be used as a basis to submit quantitative and gel-based MS data, and post-process PRIDE XML files by filtering out contaminants and empty spectra, or by merging several PRIDE XML files together. All the tools have both a graphical user interface that provides a dialog-based, user-friendly way to convert and prepare files for submission, as well as a command-line interface that can be used to integrate the tools into existing or novel pipelines, for batch processing and power users. The PRIDE Converter 2 tool suite will thus become a cornerstone in the submission process to PRIDE and, by extension, to the ProteomeXchange consortium of MS-proteomics data repositories.

Current challenges in software solutions for mass spectrometry-based quantitative proteomics.

Mass spectrometry-based proteomics has evolved as a high-throughput research field over the past decade. Significant advances in instrumentation, and the ability to produce huge volumes of data, have emphasized the need for adequate data analysis tools, which are nowadays often considered the main bottleneck for proteomics development. This review highlights important issues that directly impact the effectiveness of proteomic quantitation and educates software developers and end-users on available computational solutions to correct for the occurrence of these factors. Potential sources of errors specific for stable isotope-based methods or label-free approaches are explicitly outlined. The overall aim focuses on a generic proteomic workflow.

High confidence and sensitivity four-dimensional fractionation for human plasma proteome analysis.

Reducing the complexity of plasma proteome through complex multidimensional fractionation protocols is critical for the detection of low abundance proteins that have the potential to be the most specific disease biomarkers. Therefore, we examined a four dimension profiling method, which includes low abundance protein enrichment, tryptic digestion and peptide fractionation by IEF, SCX and RP-LC. The application of peptide pI filtering as an additional criterion for the validation of the identifications allows to minimize the false discovery rate and to optimize the best settings of the protein identification database search engine. This sequential approach allows for the identification of low abundance proteins, such as angiogenin (10(-9) g/L), pigment epithelium growth factor (10(-8) g/L), hepatocyte growth factor activator (10(-7) g/L) and thrombospondin-1 (10(-6) g/L), having concentrations similar to those of many other growth factors and cytokines involved in disease pathophysiology.

Database independent proteomics analysis of the ostrich and human proteome.

Mass spectrometry (MS)-based proteome analysis relies heavily on the presence of complete protein databases. Such a strategy is extremely powerful, albeit not adequate in the analysis of unpredicted postgenome events, such as posttranslational modifications, which exponentially increase the search space. Therefore, it is of interest to explore "database-free" approaches. Here, we sampled the ostrich and human proteomes with a method facilitating de novo sequencing, utilizing the protease Lys-N in combination with electron transfer dissociation. By implementing several validation steps, including the combined use of collision-induced dissociation/electron transfer dissociation data and a cross-validation with conventional database search strategies, we identified approximately 2,500 unique de novo peptide sequences from the ostrich sample with over 900 peptides generating full backbone sequence coverage. This dataset allowed the appropriate positioning of ostrich in the evolutionary tree. The described database-free sequencing approach is generically applicable and has great potential in important proteomics applications such as in the analysis of variable parts of endogenous antibodies or proteins modified by a plethora of complex posttranslational modifications.

Evaluating experimental bias and completeness in comparative phosphoproteomics analysis.

Unraveling the functional dynamics of phosphorylation networks is a crucial step in understanding the way in which biological networks form a living cell. Recently there has been an enormous increase in the number of measured phosphorylation events. Nevertheless, comparative and integrative analysis of phosphoproteomes is confounded by incomplete coverage and biases introduced by different experimental workflows. As a result, we cannot differentiate whether phosphosites indentified in only one or two samples are the result of condition or species specific phosphorylation, or reflect missing data. Here, we evaluate the impact of incomplete phosphoproteomics datasets on comparative analysis, and we present bioinformatics strategies to quantify the impact of different experimental workflows on measured phosphoproteomes. We show that plotting the saturation in observed phosphosites in replicates provides a reproducible picture of the extent of a particular phosphoproteome. Still, we are still far away from a complete picture of the total human phosphoproteome. The impact of different experimental techniques on the similarity between phosphoproteomes can be estimated by comparing datasets from different experimental pipelines to a common reference. Our results show that comparative analysis is most powerful when datasets have been generated using the same experimental workflow. We show this experimentally by measuring the tyrosine phosphoproteome from Caenorhabditis elegans and comparing it to the tyrosine phosphoproteome of HeLa cells, resulting in an overlap of about 4%. This overlap between very different organisms represents a three-fold increase when compared to dataset of older studies, wherein different workflows were used. The strategies we suggest enable an estimation of the impact of differences in experimental workflows on the overlap between datasets. This will allow us to perform comparative analyses not only on datasets specifically generated for this purpose, but also to extract insights through comparative analysis of the ever-increasing wealth of publically available phosphorylation data.

Improving SRM assay development: a global comparison between triple quadrupole, ion trap, and higher energy CID peptide fragmentation spectra.

In proteomics, selected reaction monitoring (SRM) is rapidly gaining importance for targeted protein quantification. The triple quadrupole mass analyzers used in SRM assays allow for levels of specificity and sensitivity hard to accomplish by more standard shotgun proteomics experiments. Often, an SRM assay is built by in silico prediction of transitions and/or extraction of peptide precursor and fragment ions from a spectral library. Spectral libraries are typically generated from nonideal ion trap based shotgun proteomics experiments or synthetic peptide libraries, consuming considerable time and effort. Here, we investigate the usability of beam type CID (or "higher energy CID" (HCD)) peptide fragmentation spectra, as acquired using an Orbitrap Velos, to facilitate SRM assay development. Therefore, peptide fragmentation spectra, obtained by ion-trap CID, triple-quadrupole CID (QqQ-CID) and Orbitrap HCD, originating from digested cellular lysates, were compared. Spectral comparison and a dedicated correlation algorithm indicated significantly higher similarity between QqQ-CID and HCD fragmentation spectra than between QqQ-CID and ion trap-CID spectra. SRM transitions generated using a constructed HCD spectral library increased SRM assay sensitivity up to 2-fold, when compared to the use of a library created from more conventionally used ion trap-CID spectra, showing that HCD spectra can assist SRM assay development.