key: cord-0285300-s90bs7q6
authors: Mari, Tommaso; Mösbauer, Kirstin; Wyler, Emanuel; Landthaler, Markus; Drosten, Christian; Selbach, Matthias
title: In vitro Kinase-to-Phosphosite database (iKiP-DB) predicts kinase activity in phosphoproteomic datasets
date: 2022-01-14
journal: bioRxiv
DOI: 10.1101/2022.01.13.476159
sha: bc24c9125d373518d0a89fc47986cfaed20789e4
doc_id: 285300
cord_uid: s90bs7q6

Phosphoproteomics routinely quantifies changes in the levels of thousands of phosphorylation sites, but functional analysis of such data remains a major challenge. While databases like PhosphoSitePlus contain information about many phosphorylation sites, the vast majority of known sites are not assigned to any protein kinase. Assigning changes in the phosphoproteome to the activity of individual kinases therefore remains a key challenge.. A recent large-scale study systematically identified in vitro substrates for most human protein kinases. Here, we reprocessed and filtered these data to generate an in vitro Kinase-to-Phosphosite database (iKiP-DB). We show that iKiP-DB can accurately predict changes in kinase activity in published phosphoproteomic datasets for both well-studied and poorly characterized kinases. We apply iKiP-DB to a newly generated phosphoproteomic analysis of SARS-CoV-2 infected human lung epithelial cells and provide evidence for coronavirus-induced changes in host cell kinase activity. In summary, we show that iKiP-DB is widely applicable to facilitate the functional analysis of phosphoproteomic datasets.

Mass spectrometry (MS)-based proteomics provides extensive information about thousands of posttranslational modifications (PTMs) 1, 2 . Phosphorylation of serine, threonine or tyrosine residues is the most extensively studied PTM in humans 3, 4 , and over 200,000 phosphorylation sites have been described 5 . However, the vast majority of these sites have neither an annotated kinase nor a known biological function 6, 7 .

Therefore, while phosphoproteomics can now routinely quantify changes in thousands of phosphorylation sites 8, 9 , the functional interpretation of these data remains a major challenge.

Computational methods to functionally annotate phosphoproteomic data can be broadly divided into two categories. The first category consists of algorithms aiming at assigning kinases to unannotated phosphosites by integrating known kinase-substrate associations with other biological features (e.g. interaction networks or co-expression profiles) [10] [11] [12] [13] [14] . The output of such algorithms are scores representing the likelihood of a certain site to be phosphorylated by a specific kinase. The second category of methods seeks to annotate changes in kinase activity in quantitative phosphoproteomics datasets. To this end, observed changes in phosphopeptide levels across conditions are integrated with kinase-substrate annotations [15] [16] [17] . The output of such algorithms are kinase activity profiles across conditions. Both categories of algorithms share the need for a resource representing kinase-to-phosphosite associations. The most comprehensive resource for phosphosites function is PhosphoSitePlus (PSP) 5 .

However, fewer than 3% of known phosphosites have either a reported function or a known regulatory kinase in PSP 6 . Moreover, the distribution of kinase-to-phosphosite associations in PSP is skewed with a minority of well-studied kinases making up the majority of entries in the database. This bias remained essentially unchanged, despite tremendous technological developments in the field 18 .

Here, we address the challenge of predicting kinase activity in phosphoproteomic data from a different angle: instead of developing algorithms that rely on existing annotation, we expand on the knowledge of kinase-to-phosphosite annotation. To this end, we took advantage of recently published large-scale in vitro kinase data for over 300 human protein kinases 19 . We re-analyzed and filtered these data to compile it into an in vitro kinase-to-phosphosite database, or iKiP-DB. Using several published phosphoproteomics datasets of kinase activation/inhibition, we show that iKiP-DB outperforms PSP in its ability to detect changes in kinase activity. Finally, to apply iKiP-DB, we infected lung epithelial cells with SARS-CoV-2 and investigated changes in cellular kinase activity induced by the novel coronavirus.

In vitro kinase assay data from the work of Sugiyama and colleagues 19 deposited on ProteomeXchange (dataset identifier PXD011366) was downloaded and re-analyzed using MaxQuant v1.6.10.43 20 . Each kinase assay was assigned a specific experimental group in the experimental design of MaxQuant, and, when present, replicate experiments were combined in the same experimental group. The quantification method was set to di-methyl labeling, as described in the original publication. The MS scans were searched against the Human Uniprot database (2018-07) using the Andromeda search engine. FDR was calculated based on searches on a pseudo-reverse database and set to 0.05. The search included as fixed modifications carbamidomethylation of cysteine and as variable modifications methionine oxidation, N-terminal acetylation, asparagine and glutamine deamidation as well as Phospho (STY). Trypsin/P was set as protease for in-silico digestion of the reference proteome. Phosphosites were filtered for reverse hits, contaminants and phosphosites with a localization probability lower than 0.75. Additionally, we only considered singly phosphorylated sites for building the database. Phosphopeptide intensities were log2-transformed and corrected for the intensity measured in the light di-methyl label channel, corresponding to the control experiment. As in the original publication 19 , phosphosites were considered specifically phosphorylated by a recombinant kinase when having a log-transformed ratio treatment over control higher than 1. From these kinase-substrate lists we removed all sites that were not contained in a large-scale re-analysis of 110 human phosphoproteomics experiments 7 , which was downloaded from the PRIDE repository (dataset identifier PXD012174). As a second filter, we removed all redundant phosphosites assigned to 20 or more kinases; finally, we excluded all kinase set with less than 5 annotated phosphosites. We then extracted the ±7 amino acid sequence windows around the phosphorylated site to be used as the specific identifier for each site, and combined all kinase sets into a single GMT file to be used with the ssGSEA suite to perform PTM-set enrichment analysis (PTM-SEA) 15 . Since our database is exclusively composed of kinase-substrate associations, all sites were marked as "up" sites. The database is built to work with the flanking sequence centric mode of PTM-SEA. To compare our database with PSP, we extracted the phosphosites of PTMsigDB 15 annotated as "KINASE-PSP''.

All datasets used for benchmarking of iKiP-DB were downloaded from the supplementary tables containing information on phosphopeptide data of the respective publications [21] [22] [23] [24] [25] . When not already provided, we filtered for single-phosphorylated, high confidence phosphosites (localization probability > 0.75) and calculated log2-transformed ratios of treatment(s) over control(s). We then extracted the ±7 amino acid sequence windows around the phosphorylation site (±6 for Rosales et al.) and removed duplicate entries. For each phosphoproteomics dataset, resulting sites and ratio information were exported as GCT files 15 .

We calculated all our kinase enrichments using the ssGSEA2.0 suite 15,26 using the standard settings: rank sample normalization, weight for the Kolmogorov-Smirnov statistic of 0.75, area under the resulting curve as main statistic and normalized enrichment score (NES) as output score, minimum overlap between query and test set of 5 sites and 1000 permutation for p-value and NES calculations. All benchmarking datasets were analyzed using iKiP-DB and PTMsigDB 15 (v1.9.0) separately. Since PTMsigDB contains annotations for several experimental categories, we considered exclusively sets of kinase-substrate associations ("KINASE-PSP" group). Since Rosales and colleagues only provided ±6 sequence windows for their dataset, we adapted iKiP-DB and PTMsigDB by cutting the sequences contained in the databases to ±6 and removing all duplicate entries that resulted from this cut. Analyses outputs were further processed using in-house made scripts with the R programming language. KSEA scores were calculated using in-house R scripts based on the formula described in the original publication 16 .

Experiments were performed and generated in the same context as described previously by Wyler, Mösbauer, Franke and colleagues 27 

Proteomics samples were prepared combining SP3 sample preparation 28 and TMT labelling 29 using TMTpro reagents 30 . Briefly, lysates were pre-cleared via centrifugation, then reduced and alkylated with DTT and iodoacetamide, respectively.

Proteins were then incubated with a 1:1 mix of SeraMag beads A and B at a 10:1 weight:weight bead:protein ratio. Protein binding to the hydrophilic beads was induced by adding ACN and washed with 80% EtOH to remove contaminants. Protein digestion was performed on beads in 50mM HEPES pH8, with Trypsin and LysC at a 1:50 protein:enzyme ratio for approximately 16 hours. After digestion, peptides were quantified via BCA assay and directly labelled with TMTpro reagents (Thermo Fisher Scientific; product number A44520, lot number UL297970) following the manufacturer's protocol. Samples were randomly assigned to a TMT channel as follows: CoV2 4hrs repA -> 128C, CoV2 4hrs repB -> 129N, CoV2 4hrs repC -> 133N, CoV2 8hrs repA -> 133C, CoV2 8hrs repB -> 127N, CoV2 8hrs repC -> 131C, CoV2 12hrs repA -> 130C, 

Proteome and phosphoproteome fractions were online-fractionated on a EASY-nLC 1200 and acquired on an Exploris 480 mass spectrometer (Thermo Fisher Scientific) operated on profile-centroid mode, as previously described 32 . Peptide separation was achieved on a fused silica, 25 cm long column packed in-house with C18-AQ 1.9 µm beads kept at a temperature of 45°C. Mobile phase A consisted of 0.1% FA and 3% ACN in water, while mobile phase B consisted of 0.1% FA and 90% ACN.

After column equilibration peptides resuspended in buffer A were separated with a 250 µl/min flow on a 110 minutes gradient:mobile phase B increased from 4% to 30% in the first 88 minutes, followed by an increase to 60% in the following 10 minutes, to then reach 90% in one minute, which was held for 5 minutes. The MS was operated in data dependent acquisition, with MS1 scans from 350 to 1500 m/z acquired at a resolution of 60,000, maximum injection time (IT) of 10 ms and an automatic gain control (AGC) target value of 3e6. The 20 most intense precursor ion peaks with charges from +2 to +6 were selected for fragmentation, unless present in the dynamic exclusion list (30 s). for . Phosphoproteomics data for PTM-SEA analysis was ( ) = − 1 < 0 prepared as described above for the benchmarking datasets. All statistical analysis was done with the R programming language (v3.6.6).

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE 33 partner repository with the dataset identifier PXD030395. Prior to publication, the data can be accessed with the following credentials: Username: reviewer_pxd030395@ebi.ac.uk; Password: nLLVTHZZ

To develop our database, we took advantage of a recent large-scale discovery study of substrates of the human kinome 19 . In this study, de-phosphorylated HeLa cell lysate was incubated with 385 recombinantly expressed human kinases to perform in vitro kinase assays. After proteolytic digestion and phosphopeptide enrichment, the sites phosphorylated by each kinase were identified via MS ( Figure 1A ). Intrigued by this study, we retrieved the raw mass spectrometry data (ProteomeXchange dataset identifier PXD011366) and re-analyzed it with MaxQuant 20 . In total, we obtained 159,618

kinase-to-phosphosites associations for 17,225 unique phosphosites, localized on 4,032 distinct protein groups ( Figure 1A ). In vitro kinase assays can induce phosphorylation of sites that are not phosphorylated under physiological conditions. Therefore, we first filtered the data using a catalogue of 112 manually curated datasets of phospho-enriched proteins from 104 different human cell types or tissues 7 . More than half of the in vitro sites were not observed in any cell or tissue dataset and therefore excluded ( Figure 1B) . As a second filter, we excluded all sites assigned to 20 or more kinases, since we reasoned that these highly redundant sites are not well suited to distinguish between different kinases ( Figure 1C ). Finally, we removed all kinases with less than five phosphosites since such a small number would not provide sufficient information to compute robust enrichment scores.

After filtering, we obtained an in vitro kinase-to-phosphosite database (or iKiP-DB) for 313 kinases (Supplemental Table S1 ). iKiP-DB provides kinase-substrate associations for 128 kinases not previously annotated in PSP ( Figure 1D ). When looking at the 185 kinases shared between both databases, PSP showed a strong bias towards few well-characterized kinases (e.g. CDK1, PRKCA, CK2A1) while most other kinases have only a small number of assigned sites. In contrast, kinase-substrate associations in iKiP-DB do not display such a bias and greatly extend the annotations for many kinases ( Figure 1E, upper panel) . Nevertheless, although PSP and IKiP-DB have been derived in a completely independent fashion, we observed a significant overlap between the number of phosphosites assigned to the same kinases in both databases ( Figure 1E , lower panel). 

To assess if iKiP-DB can predict kinase activity in phosphoproteomic data, we first turned to epidermal growth factor receptor (EGFR) signalling 34 . To this end, we took advantage of a recent phosphoproteomic dataset 21 . In this study, the authors treated HeLa cells with EGF or TGFɑ over a time course, enriched for phosphopeptides via combined phospho-tyrosine and titanium-dioxide enrichment, and measured the resulting phosphoproteome by MS (Figure 2A) . This dataset provides an excellent benchmark for our database since i) EGFR/PI3K/AKT pathway is well studied and Table S2 ). The two databases produced very similar predictions for EGFR and the two downstream kinases AKT1 and RPS6KA1 (also known as p90RSK or MAPKAPK1) 21, 34 . In line with previous findings, our analysis reveals that EGFR is rapidly activated upon stimulation with both EGF and TGFɑ, followed by a decrease in activity caused by receptor internalization 21 . Both analyses indicate a stronger re-activation of EGFR upon TGFɑ treatment, which has been linked to the more efficient recycling of the receptor to the plasma membrane 21 . Of note, both PSP and iKiP-DB predicted EGFR at time point 1 min as the most activated kinase throughout the treatment, confirming the specificity of the predictions (Supplemental Table S2 ). The relevant phosphosites for EGFR from PSP and iKiP-DB showed similar trends across the dataset ( Figure 2C) . Additionally, the delayed and more sustained activation of AKT1 and RPS6KA1 is in line with the more downstream position of these kinases in the EGFR/PI3K/AKT pathway 35, 36 ( Figure 2B ). To ensure robustness of iKiP-DB across different algorithms to calculate kinase enrichments, we also used kinase-substrate enrichment analysis (KSEA) 16 and calculated KSEA scores and statistics, obtaining overall similar results ( Figure S1B , Supplemental Table S2 ). In summary, these data show that iKiP-DB can be used to predict the activation status of a well-studied kinase as well as a manually curated database. 

To further validate iKiP-DB as a useful tool to predict kinase activity, we searched the literature for phosphoproteomics datasets with the following characteristics: i) treatments with kinase inhibitors or knockdowns, as they would provide an easy ground truth to test our predictions; ii) studies focusing on a range of kinases, both well and less characterized; iii) datasets derived from samples other than HeLa cells, to test predictions across different model systems. According to these criteria, we selected four additional datasets for further benchmarking of iKiP-DB ( Figure 3) . Again, we used the PTM-SEA algorithm to compute kinase enrichment scores 15 . Figure S2B) . Similarly, analysis with iKiP confirms the knockdown of ALK in neuroblastoma NB1 cells, as well as a dose-dependent inhibition with lorlatinib ( Figure   3B , S2C and S2D).

The next two dataset we selected focused on the inhibition of ERK1/2 and CK2, which are well studied and annotated kinases in PSP ( Figure 1E ). ERK1/2 inhibitors are of particular interest due to their role in the clinic, for example for the treatment of melanoma, especially after the development of metastases 37 . In their work, Basken and colleagues tested two potent ERK1/2 inhibitors on WM239 metastatic melanoma cells and measured their effect on the phosphoproteome 24 . Kinase activity prediction with iKiP-DB correctly indicated the strong inhibition of both ERK1 and 2, while prediction with PSP could only detect the significant depletion of ERK1 ( Figure 3C , Supplemental Table S3 ). Next, we wanted to test whether our database could be used to unbiasedly identify the correct target of a kinase inhibitor. For this, we assessed which kinases were predicted as most inhibited by both drug treatments. Analysis with iKiP-DB correctly identified ERK2 as the most inhibited kinase by ravoxertinib and SCH772984

( Figure S2E ), which is well in line with the lower IC50 values of both compounds for ERK2 compared to ERK1 38, 39 . On the contrary, analysis with PSP did not indicate either ERK as the most inhibited kinase by either treatment ( Figure S2E , Supplemental Table   S3 ). Finally, in the last study we used for benchmarking, Rosales and colleagues treated two acute myeloid leukemia cell lines with a peptide-based kinase inhibitor targeting CK2 25 . While both PSP and iKiP-DB correctly predicted the inhibition of CK2A1, our database could also provide information regarding the inhibition of CK2A2, the other catalytic subunit of the kinase ( Figure 3D , Supplemental Table S3 ). Again, we assessed which kinases were predicted as most inhibited to evaluate the potential of our database for unbiased target discovery. Also here, iKiP-DB could be used to correctly assess the target of the inhibitor in all conditions ( Figure S2F , Supplemental Table S3 ).

In conclusion, reanalysis of published phosphoproteomic datasets highlights the ability of iKiP-DB to predict the activity of different kinases in various cell lines and with data generated via different quantification and enrichment strategies (SILAC, TMT, EasyPhos, TiO2, anti-p-Tyr antibodies). Hence, iKiP-DB can robustly predict kinase activity in diverse phosphoproteomic datasets.

As Table S4 ). At the latest time point, we identified several differentially regulated host proteins ( Figure 4C , Supplemental Table S4 ). Among these, many interferon-stimulated genes such as IRF7, IFIT1, IFIT2, IFIH1, OAS1 and MX2, were enriched upon infection. These proteins are well characterized for their antiviral function 40, 41 , also in the context of SARS-CoV-2 infection 27, 42, 43 . We also identified downregulation of ACE2, the most important entry factor of SARS-CoV-2 44 which is in line with the reported shedding of the receptor from infected cells 45 . To gain insight on how the virus changes the activity of host kinases, we analyzed the phosphoproteomic data using iKiP-DB ( Figure 4D , Supplemental Table   S4 ). This analysis divided kinases into three broad groups: early activated, late inhibited and late activated, with the last group being the largest ( Figure 3C ) 

Protein phosphorylation is a fundamental and ubiquitous process in biological systems 52 , and as such studying the phosphoproteome is highly informative for basic biology and disease 21, 53 . While novel technologies are rapidly increasing the depth and coverage of MS-derived phosphoproteomics data, their interpretation remains challenging, partially due to the scarcity of functional annotations of phosphosites 6 .

In this work, we described the re-analysis of published in vitro kinase assay data, which we compiled into a database of kinase-substrate associations we named iKiP-DB.

Compared to PSP as the current gold standard for phosphosite annotation 5 , our database increases the range of annotated kinases as well as the number of substrate sites annotated. Additionally, since our database is derived from experimental data, it is not biased towards the most well-studied portion of the kinome. We employed iKiP-DB in conjunction with PTM-SEA 15 to predict the activation or inhibition of kinases in published phosphoproteomics data. Direct comparison demonstrated an equal or higher degree of accuracy of iKiP-DB compared to the manually annotated data of PSP.

Interestingly, we also observed a superior prediction by iKiP-DB for CK2 and ERK1/2 inhibitor experiments, even though both CK2 and ERK1/2 are well annotated kinases in PSP. Finally, analysis of newly generated data of SARS-CoV-2 infected lung epithelial cells recapitulates the phenotypic effects of viral infections, as well as the known biology of the novel coronavirus.

One important limitation of iKiP-DB is that it is based on in vitro kinase experiments, which do not necessarily reflect the activity of kinases in vivo. We alleviate this problem by restricting the database to sites that have been observed in cells ( Figure   1B ). Nevertheless, individual phosphosites in iKiP-DB are not necessarily phosphorylated by the corresponding kinase under physiological conditions. Therefore, iKiP-DB is more useful for the large-scale analysis of kinase signatures, rather than at the level of individual phosphosites. It is actually surprising how well the in vitro kinase data alone (without any manual annotation) reflects cellular kinase activity (Figure 2 and   3 ).

In summary, we demonstrate that integrating phosphoproteomic datasets with in vitro kinase data via iKiP-DB greatly facilitates detection of altered kinase activity. We believe this tool will be broadly useful for phosphoproteomic data analysis. We provide iKiP-DB in GMT format, ready to use for PTM-SEA using the ssGSEA suite.

We would like to thank Dr. Evelyn Ramberger for the critical reading of the manuscript, as well as Mohamad Haji of the mass spectrometry core facility of the MDC for his invaluable help in sample preparation. Since this work would have not been possible without all the published data we used for the database and the benchmarking, we would like to express our gratitude to all the researchers involved in these studies that generated high-quality phosphoproteomics data we could use. We would like to especially mention Naoyuki Sugiyama, Yasushi Ishihama and Haruna Imamura, since the core of this work was based on their dataset. This work was supported by the German Ministry of Education and Research (BMBF) via the national research node for mass spectrometry in systems medicine MSTARS (031L0220B to M.S.).

Mass Spectrometry-Based Detection and Assignment of Protein Posttranslational Modifications

Status of Large-Scale Analysis of Post-Translational Modifications by Mass Spectrometry

The Coming of Age of Phosphoproteomics--from Large Data Sets to Inference of Protein Functions

Mutations, PTMs and Recalibrations

Illuminating the Dark Phosphoproteome

The Functional Landscape of the Human Phosphoproteome

Benchmarking Common Quantification Strategies for Large-Scale Phosphoproteomics

Prediction of Kinase-Specific Phosphorylation Sites through an Integrative Model of Protein Context and Sequence

KinomeXplorer: An Integrated Platform for Kinome Biology Studies

0: An Update on the Prediction of Kinase-Specific Phosphorylation Sites in Proteins

Accurate Prediction of Kinase-Substrate Networks Using Knowledge Graphs

KinPred: A Unified and Sustainable Approach for Harnessing Proteome-Level Human Kinase-Substrate Predictions

A Curated Resource for Phosphosite-Specific Signature Analysis

A Web-Based Tool for Kinase Activity Inference from Quantitative Phosphoproteomics

IKAP: A Heuristic Framework for Inference of Kinase Activities from Phosphoproteomics Data

Too Many Roads Not Taken

Large-Scale Discovery of Substrates of the Human Kinome

The MaxQuant Computational Platform for Mass Spectrometry-Based Shotgun Proteomics

Multilayered Proteomics Reveals Molecular Switches Dictating Ligand-Dependent EGFR Trafficking

Global Phosphoproteomics Reveals DYRK1A Regulates CDK1 Activity in Glioblastoma Cells

Integrated Proximal Proteomics Reveals IRS2 as a Determinant of Cell Survival in ALK-Driven Neuroblastoma

Kinase Pathway Inhibitors in Melanoma Cells

Targeting of Protein Kinase CK2 in Acute Myeloid Leukemia Cells Using the Clinical-Grade Synthetic-Peptide CIGB-300

Gene Set Enrichment Analysis: A Knowledge-Based Approach for Interpreting Genome-Wide Expression Profiles

Transcriptomic Profiling of SARS-CoV-2 Infected Human Cell Lines Identifies HSP90 as Target for COVID-19 Therapy. iScience 2021

Single-Pot, Solid-Phase-Enhanced Sample Preparation for Proteomics Experiments

Reproducible Workflow for Multiplexed Deep-Scale Proteome and Phosphoproteome Analysis of Tumor Tissues by Liquid Chromatography-Mass Spectrometry

TMTpro Reagents: A Set of Isobaric Labeling Mass Tags Enables Simultaneous Proteome-Wide Measurements across 16 Samples

Off-Line High-pH Reversed-Phase Fractionation for in-Depth Phosphoproteomics

SARS-CoV-2 Infection Triggers Profibrotic Macrophage Responses and Lung Fibrosis

The PRIDE Database and Related Tools and Resources in 2019: Improving Support for Quantification Data

Epidermal Growth Factor Receptor Cell Proliferation Signaling Pathways

Rsk1 Mediates a MEK-MAP Kinase Cell Survival Signal

Characterization of Regulatory Events Associated with Membrane Targeting of p90 Ribosomal S6 Kinase 1

Targeting the ERK Signaling Pathway in Melanoma

Discovery of (S)-1-(1-(4-Chloro-3-Fluorophenyl)-2-Hydroxyethyl)-4-(2-((1-Methyl-1H-Pyrazol-5-Yl )amino)pyrimidin-4-Yl)pyridin-2(1H)-One (GDC-0994), an Extracellular Signal-Regulated Kinase 1/2 (ERK1/2) Inhibitor in Early Clinical Development

Interferon-Stimulated Genes and Their Antiviral Effector Functions

Interferon-Stimulated Genes as Enhancers of Antiviral Innate Immune Signaling

Imbalanced Host Response to SARS-CoV-2 Drives Development of COVID-19

Interplay between SARS-CoV-2 and the Type I Interferon Response

SARS-CoV-2 Cell Entry Depends on ACE2 and TMPRSS2 and Is Blocked by a Clinically Proven Protease Inhibitor

Plasma ACE2 Activity Is Persistently Elevated Following SARS-CoV-2 Infection: Implications for COVID-19 Pathogenesis and Consequences

Regulation of RIG-I-like Receptor-Mediated Signaling: Interaction between Host and Viral Factors

Role of MAPK/MNK1 Signaling in Virus Replication

Identification of Regulatory Phosphorylation Sites in Mitogen-Activated Protein Kinase (MAPK)-Activated Protein Kinase-1a/p90rsk That Are Inducible by MAPK

Polo-like Kinases (plks), a Key Regulator of Cell Cycle and New Potential Target for Cancer Therapy

Cell Cycle Regulation by the NEK Family of Protein Kinases

Mass-Spectrometric Exploration of Proteome Structure and Function

Proteogenomics Connects Somatic Mutations to Signalling in Breast Cancer