Understanding how oncogenic mutations rewire regulatory-protein networks is important for rationalizing the mechanisms of oncogenesis and for individualizing anticancer treatments. We report a chemical phosphoproteomics method to elucidate the topology of kinase-signaling networks in mammalian cells. We identified >6,000 protein phosphorylation sites that can be used to infer >1,500 kinase–kinase interactions and devised algorithms that can reconstruct kinase network topologies from these phosphoproteomics data. Application of our methods to primary acute myeloid leukemia and breast cancer tumors quantified the relationship between kinase expression and activity, and enabled the identification of hitherto unknown kinase network topologies associated with drug-resistant phenotypes or specific genetic mutations. Using orthogonal methods we validated that PIK3CA wild-type cells adopt MAPK-dependent circuitries in breast cancer cells and that the kinase TTK is important in acute myeloid leukemia. Our phosphoproteomic signatures of network circuitry can identify kinase topologies associated with both phenotypes and genotypes of cancer cells.
Subscribe to Journal
Get full journal access for 1 year
only $20.83 per issue
All prices are NET prices.
VAT will be added later in the checkout.
Rent or Buy article
Get time limited or full article access on ReadCube.
All prices are NET prices.
Code is available from the GitHub repository: https://github.com/CutillasLab/ebdt
Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: the next generation. Cell 144, 646–674 (2011).
Kanehisa, M., Furumichi, M., Tanabe, M., Sato, Y. & Morishima, K. KEGG: new perspectives on genomes, pathways, diseases and drugs. Nucleic Acids Res. 45, D353–D361 (2017).
Fabregat, A. et al. Reactome pathway analysis: a high-performance in-memory approach. BMC Bioinformatics 18, 142 (2017).
Sacco, F., Perfetto, L. & Cesareni, G. Combining phosphoproteomics datasets and literature information to reveal the functional connections in a cell phosphorylation network. Proteomics 18, 1700311 (2018).
Tape, C. J. et al. Oncogenic KRAS regulates tumor cell signaling via stromal reciprocation. Cell 165, 910–920 (2016).
Terfve, C. D., Wilkes, E. H., Casado, P., Cutillas, P. R. & Saez-Rodriguez, J. Large-scale models of signal propagation in human cells derived from discovery phosphoproteomic data. Nat. Commun. 6, 8033 (2015).
Invergo, B. M. & Beltrao, P. Reconstructing phosphorylation signalling networks from quantitative phosphoproteomic data. Essays Biochem. 62, 525–534 (2018).
Donnella, H. J. et al. Kinome rewiring reveals AURKA limits PI3K-pathway inhibitor efficacy in breast cancer. Nat. Chem. Biol. 14, 768–777 (2018).
Lachmann, A. & Ma’ayan, A. KEA: kinase enrichment analysis. Bioinformatics 25, 684–686 (2009).
Johnson, H. et al. Molecular characterization of EGFR and EGFRvIII signaling networks in human glioblastoma tumor xenografts. Mol. Cell. Proteomics 11, 1724–1740 (2012).
Ficarro, S. B. et al. Phosphoproteome analysis by mass spectrometry and its application to Saccharomyces cerevisiae. Nat. Biotechnol. 20, 301–305 (2002).
Casado, P. et al. Kinase-substrate enrichment analysis provides insights into the heterogeneity of signaling pathway activation in leukemia cells. Sci. Signal. 6, rs6 (2013).
Ochoa, D. et al. An atlas of human kinase regulation. Mol. Sys. Biol. 12, 888 (2016).
Drake, J. M. et al. Phosphoproteome integration reveals patient-specific networks in prostate cancer. Cell 166, 1041–1054 (2016).
Casado, P. et al. Proteomic and genomic integration identifies kinase and differentiation determinants of kinase inhibitor sensitivity in leukemia cells. Leukemia 32, 1818–1822 (2018).
Singh, R., Xu, J. & Berger, B. Global alignment of multiple protein interaction networks with application to functional orthology detection. Proc. Natl Acad. Sci. USA 105, 12763–12768 (2008).
Sharan, R. & Ideker, T. Modeling cellular machinery through biological network comparison. Nat. Biotechnol. 24, 427–433 (2006).
Creixell, P. et al. Kinome-wide decoding of network-attacking mutations rewiring cancer signaling. Cell 163, 202–217 (2015).
Pan, C., Olsen, J. V., Daub, H. & Mann, M. Global effects of kinase inhibitors on signaling networks revealed by quantitative phosphoproteomics. Mol. Cell. Proteomics 8, 2796–2808 (2009).
Alcolea, M. P., Casado, P., Rodriguez-Prados, J. C., Vanhaesebroeck, B. & Cutillas, P. R. Phosphoproteomic analysis of leukemia cells under basal and drug-treated conditions identifies markers of kinase pathway activation and mechanisms of resistance. Mol. Cell. Proteomics 11, 453–466 (2012).
Klaeger, S. et al. The target landscape of clinical kinase drugs. Science 358, eaan4368 (2017).
Wilkes, E. & Cutillas, P. R. Label-free phosphoproteomic approach for kinase signaling analysis. Methods Mol. Biol. 1636, 199–217 (2017).
Wilkes, E. H., Terfve, C., Gribben, J. G., Saez-Rodriguez, J. & Cutillas, P. R. Empirical inference of circuitry and plasticity in a kinase signaling network. Proc. Natl Acad. Sci. USA 112, 7719–7724 (2015).
Alessi, D. R., Caudwell, F. B., Andjelkovic, M., Hemmings, B. A. & Cohen, P. Molecular basis for the substrate specificity of protein kinase B; comparison with MAPKAP kinase-1 and p70 S6 kinase. FEBS Lett. 399, 333–338 (1996).
Ubersax, J. A. & Ferrell, J. E. Jr. Mechanisms of specificity in protein phosphorylation. Nat. Rev. Mol. Cell Biol. 8, 530–541 (2007).
Hernandez-Armenta, C., Ochoa, D., Goncalves, E., Saez-Rodriguez, J. & Beltrao, P. Benchmarking substrate-based kinase activity inference using phosphoproteomic data. Bioinformatics 33, 1845–1851 (2017).
Barabási, A.-L. & Pósfai, M. Network Science (Cambridge Univ., 2016).
Jeong, H., Mason, S. P., Barabasi, A. L. & Oltvai, Z. N. Lethality and centrality in protein networks. Nature 411, 41–42 (2001).
Ciaccio, M. F., Wagner, J. P., Chuu, C. P., Lauffenburger, D. A. & Jones, R. B. Systems analysis of EGF receptor signaling dynamics with microwestern arrays. Nat. Methods 7, 148–155 (2010).
Vincent, A. M. & Feldman, E. L. Control of cell survival by IGF signaling pathways. Growth Horm. IGF Res. 12, 193–197 (2002).
Mertins, P. et al. Proteogenomics connects somatic mutations to signalling in breast cancer. Nature 534, 55–62 (2016).
Zhang, H. et al. Integrated proteogenomic characterization of human high-grade serous ovarian Cancer. Cell 166, 755–765 (2016).
Mason, J. M. et al. Functional characterization of CFI-402257, a potent and selective Mps1/TTK kinase inhibitor, for the treatment of cancer. Proc. Natl Acad. Sci. USA 114, 3127–3132 (2017).
Bachman, K. E. et al. The PIK3CA gene is mutated with high frequency in human breast cancers. Cancer Biol. Ther. 3, 772–775 (2004).
Samuels, Y. et al. Mutant PIK3CA promotes cell growth and invasion of human cancer cells. Cancer Cell 7, 561–573 (2005).
Zhao, L. & Vogt, P. K. Helical domain and kinase domain mutations in p110α of phosphatidylinositol 3-kinase induce gain of function by different mechanisms. Proc. Natl Acad. Sci. USA 105, 2652–2657 (2008).
Barbareschi, M. et al. Different prognostic roles of mutations in the helical and kinase domains of the PIK3CA gene in breast carcinomas. Clin. Cancer Res. 13, 6064–6069 (2007).
Davies, S. P., Reddy, H., Caivano, M. & Cohen, P. Specificity and mechanism of action of some commonly used protein kinase inhibitors. Biochem. J. 351, 95–105 (2000).
Hornbeck, P. V. et al. PhosphoSitePlus, 2014: mutations, PTMs and recalibrations. Nucleic Acids Res. 43, D512–D520 (2015).
Perfetto, L. et al. SIGNOR: a database of causal relationships between biological entities. Nucleic Acids Res. 44, D548–D554 (2016).
Carlson, S. M. et al. Large-scale discovery of ERK2 substrates identifies ERK-mediated transcriptional regulation by ETV3. Sci. Signal. 4, rs11 (2011).
Shah, K., Liu, Y., Deirmengian, C. & Shokat, K. M. Engineering unnatural nucleotide specificity for Rous sarcoma virus tyrosine kinase to uniquely label its direct substrates. Proc. Natl Acad. Sci. USA 94, 3565–3570 (1997).
Xue, L., Geahlen, R. L. & Tao, W. A. Identification of direct tyrosine kinase substrates based on protein kinase assay-linked phosphoproteomics. Mol. Cell. Proteomics 12, 2969–2980 (2013).
Cohen, P. & Knebel, A. KESTREL: a powerful method for identifying the physiological substrates of protein kinases. Biochem. J. 393, 1–6 (2006).
Ferguson, F. M. & Gray, N. S. Kinase inhibitors: the road ahead. Nat. Rev. Drug Discov. 17, 353–377 (2018).
Schwanhausser, B. et al. Global quantification of mammalian gene expression control. Nature 473, 337–342 (2011).
Adams, J. A. Activation loop phosphorylation and catalysis in protein kinases: is there functional evidence for the autoinhibitor model? Biochemistry 42, 601–607 (2003).
Juric, D. et al. Phosphatidylinositol 3-kinase α-selective inhibition with alpelisib (BYL719) in PIK3CA-altered solid tumors: results from the first-in-human study. J. Clin. Oncol. 36, 1291–1299 (2018).
Sarker, D. et al. First-in-human phase I study of pictilisib (GDC-0941), a potent pan-class I phosphatidylinositol-3-kinase (PI3K) inhibitor, in patients with advanced solid tumors. Clin. Cancer Res. 21, 77–86 (2015).
Rajeeve, V., Pearce, W., Cascante, M., Vanhaesebroeck, B. & Cutillas, P. R. Polyamine production is downstream and upstream of oncogenic PI3K signalling and contributes to tumour cell growth. Biochem. J. 450, 619–628 (2013).
Gruhler, A. et al. Quantitative phosphoproteomics applied to the yeast pheromone signaling pathway. Mol. Cell. Proteomics 4, 310–327 (2005).
Larsen, M. R., Thingholm, T. E., Jensen, O. N., Roepstorff, P. & Jorgensen, T. J. Highly selective enrichment of phosphorylated peptides from peptide mixtures using titanium dioxide microcolumns. Mol. Cell. Proteomics 4, 873–886 (2005).
Montoya, A., Beltran, L., Casado, P., Rodriguez-Prados, J. C. & Cutillas, P. R. Characterization of a TiO2 enrichment method for label-free quantitative phosphoproteomics. Methods 54, 370–378 (2011).
Vizcaino, J. A. et al. 2016 update of the PRIDE database and its related tools. Nucleic Acids Res. 44, D447–D456 (2016).
Perkins, D. N., Pappin, D. J., Creasy, D. M. & Cottrell, J. S. Probability-based protein identification by searching sequence databases using mass spectrometry data. Electrophoresis 20, 3551–3567 (1999).
Cutillas, P. R. & Vanhaesebroeck, B. Quantitative profile of five murine core proteomes using label-free functional proteomics. Mol. Cell. Proteomics 6, 1560–1573 (2007).
Tsou, C. C. et al. IDEAL-Q, an automated tool for label-free quantitation analysis using an efficient peptide alignment approach and spectral data validation. Mol. Cell. Proteomics 9, 131–144 (2010).
Bateman, N. W. et al. Maximizing peptide identification events in proteomic workflows using data-dependent acquisition (DDA). Mol. Cell. Proteomics 13, 329–338 (2014).
Lawrence, R. T., Searle, B. C., Llovet, A. & Villen, J. Plug-and-play analysis of the human phosphoproteome by targeted high-resolution mass spectrometry. Nat. Methods 13, 431–434 (2016).
Fabian, M. A. et al. A small molecule–kinase interaction map for clinical kinase inhibitors. Nat. Biotechnol. 23, 329–336 (2005).
Elkins, J. M. et al. Comprehensive characterization of the published kinase inhibitor set. Nat. Biotechnol. 34, 95–103 (2015).
Kuhn, M. Building predictive models in R using the caret package. J. Stat. Softw. 28, 1–26 (2008).
Kim, S. Y. & Volsky, D. J. PAGE: parametric analysis of gene set enrichment. BMC Bioinformatics 6, 144 (2005).
Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl Acad. Sci. USA 102, 15545–15550 (2005).
Horn, H. et al. KinomeXplorer: an integrated platform for kinome biology studies. Nat. Methods 11, 603–604 (2014).
This work was primarily funded by the BBSRC (BB/M006174/1). Barts and The London Charity (297/2249), CRUK (C15966/A24375 and C16420/A18066) and the QMUL Life Science Initiative also contributed to funding.
P.R.C. is a co-founder of Kinomica Ltd. The other authors declare no competing financial interests.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Integrated supplementary information
(a) Comparison of kinase inhibitor selectivity as determined by a Discover-X assay (this study) and Klaeger et al. (Science 358, 2017) for compounds present in both datasets. (b) Coefficients of variation of phosphopeptide peak areas in basal (DMSO controls) samples as a function of peak area cut-off values (n = 4 independent experiments). (c) Number of phosphopeptides quantified as a function of peak area intensity cut-off values. (d) Number of phosphorylation sites with CV<50% in basal (untreated) cells. (e) Number of unique phosphorylation sites found to be putative downstream targets (PDTs) of at least one kinase. (f) Number of kinases with at least five PDTs.
The ChemPhoPro allows the community to browse our experimental results and identified PDTs. The front page (a) provides a search box into which the name of protein or perturbagen (e.g. inhibitor) can be entered. If the name entered is recognized as a kinase, the kinase view (b) is displayed. In this view, the perturbagens panel provides information about perturbagens reported to affect the activity of the kinase, together with the corresponding results of our own DiscoverX studies. The following panel details putative downstream targets (PDTs) of the kinase identified by our analysis (organized by cell line) and the underlying experimental data can be accessed by clicking a PDT’s graph icon. The final panel in this view lists known kinase-substrate relationships extracted from UniProt and provides links to heatmaps (c) showing the behavior of these substrates in our experimental data. The protein view (d) shows reported substrates and PDTs in the context of their protein location. Clicking the graph icon of a PDT reveals its behavior by cell line and inhibitor. Finally, the perturbagen view (e) provides basic information about the inhibitors that we have used in this study, and data about their activity (both previously reported, and determined by our DiscoverX assays). Throughout the interface, links allow easy navigation between kinase, protein and perturbagen pages. An application programming interface (API) is available for programmatic access to the database. The database may be accessed at http://chemphopro.org.
Phosphopeptides were matched to predefined phosphorylation motifs that are common in kinase substrates (a) or to predicted kinase-substrate (Netphorest) groups (b). Enrichment was calculated as Log2[(k/m)/(q/t)] for (a) or as (k/m)/(q/t) for (b), where k is number of phosphosites in PDT that match the named motif/Netphorest group, m is total number of phosphosites in PDTs, q is total number of phosphosites in motif/Netphorest group, and t is total number of phosphosites. Full sets of motif/Netphorest groups in different PDTs are shown in Supplementary Dataset 4.
Supplementary Figure 4 Kinase signaling analysis using PDTs and previous databases of kinase-phosphosite relationships.
(a) Number of downstream targets of kinases poorly annotated in previous databases of K-P relationships. (b) Number of K-P relationships and number of kinases present in different databases of K-P relationships. (c) Experimental design of phosphoproteomic experiment of P31/Fuj cells treated with the named kinase inhibitors, and summary of results. P-values were calculated using two-sided paired t-test of log transformed data and FDR was calculated by adjusting p-values for multiple testing using the Benjamini-Hochberg method. These statistics were applied to the mean phosphoproteomics data of two independent experiments performed in technical replicate. (d) Results of changes in kinase activities as a function of treatment with the named kinase inhibitors (as determined by KSEA) when using different databases of K-P relationships as input of the KSEA algorithm. Z-scores were calculated using the PAGE method as in (BMC Bioinformatics 6, 144, 2005) and (PNAS 112, 7719-7724, 2015) and p-values using a two-sided Kolmogorov-Smirnov test as in (PNAS 102, 15545-15550, 2005). FDR was calculated by adjusting p-values for multiple testing using the Benjamini-Hochberg method.
Supplementary Figure 5 Validation of groups of phosphorylation sites as markers of kinase network edges.
(a) Motif enrichment analysis in phosphorylation site groups that define the named network edges. Z-scores values are shown in y-axes and x-axes show the number of phosphopeptides with the labeled motif. For clarity not all data-points are labeled. Enrichment was calculated as in Supplementary Fig. 3. (b) Enrichment of kinase network axes as a function of treatments with EFG or IGF-I for the indicated times. (c) Enrichment of phosphopeptides in the named edges in predicted kinase-substrate (Netphorest) groups. Enrichment was calculated as in Supplementary Fig. 3. (d) Enrichment of edges in P31/Fuj as a function of treatments with the named kinase inhibitors; z-scores were calculated using the PAGE method as in (BMC Bioinformatics 6, 144, 2005) and (PNAS 112, 7719-7724, 2015) and p-values using a two-sided Kolmogorov-Smirnov test as in (PNAS 102, 15545-15550, 2005). FDR was calculated by adjusting p-values for multiple testing using the Benjamini-Hochberg method (n = 2 independent experiments performed in technical replicate).
(a) Comparison of PDT kinase enrichment across the named cell models. Z-scores were calculated using the PAGE method as in (BMC Bioinformatics 6, 144, 2005) and (PNAS 112, 7719-7724, 2015) and p-values using a two-sided Kolmogorov-Smirnov test as in (PNAS 102, 15545-15550, 2005). FDR was calculated by adjusting p-values for multiple testing using the Benjamin-Hochberg method. These statistics were applied to the mean phosphoproteomics data of four independent experiments. For clarify not all kinases are labeled. (b) Phosphopeptides ion intensities (summed peak area) of phosphopeptides containing the named phosphorylation site across the named cell models (n = 3 independent experiments performed in technical replicate). Boxplots show median and interquartile ranges. P-values were calculated by Annova using R (ggpubr).
Supplementary Figure 7 Accuracy of multivariate models for drug sensitivity prediction based on different readouts of kinase expression and network circuitry.
Accuracy was determined as the difference between calculated and actual drug responses (in Fig. 4 of main text) divided by actual response. Boxplots show median, interquartile and max-min ranges. RMSE, root mean square error (n=36 AML cases).
Association between edge enrichment and drug response by Spearman correlation. The kinase axes showing greater importance to the statistical learning models shown in Fig. 4 were plotted as a function of responses of 36 primary AML cells to the named kinase inhibitors (each data point represents the quantification of the named edge in a primary sample and thus, because of missing data points from Casado et al. (Leukemia 32, 1818-1822, 2018), sample size are edge specific).
Supplementary Figure 9 Kinase network topologies and kinase activities associated to kinase protein expression and phosphorylation.
(a) Kinase protein expression, phosphorylation extent (calculated as the sum of intensities of peptides or phosphopeptides, respectively, derived from a given kinase) or their ratio was correlated to the enrichment of edges defined by the same kinase (top graphs) or to the predicted kinase activity (bottom graphs). Numbers refer to positively correlated edges (Spearman r > 0, bottom row) or those with Spearman r >0.28 (p<0.01, top row). Network edge enrichment as well as protein and phosphoprotein expression were calculated for 30 primary AML cases and for 83 primary breast cancer samples from a meta-analysis of their proteomes and phosphoproteomes from Casado et al. (Leukemia 32, 1818-1822, 2018) and Mertins et al. (Nature 534, 55-62, 2016). Violin plots show the kernel density of the data, in which probability is related to the width of the plot. Boxplots within violin plots (top graphs) show the median, interquartile ranges, and max-min ranges. Statistical significance of differences between the spearman rank values for each analysis was calculated using a two-sided Kruskal-Wallis test. (b) Association between number of kinase downstream targets (PDTs) and the correlation of kinase activity with phosphokinase expression. (c) Relationship between number of phosphorylation sites detected in kinase with the association between kinase activity and phosphokinase expression. (d) Relationship between number of phosphorylation sites detected in kinase with the association between kinase activity and the expression of unmodified kinase. P-values were calculated using Pearson (n=69 ovarian samples).
(a) Difference in kinase network edge enrichment between AML primary cells positive for NRAS mutation relative to wild-type cases. Fold difference was calculated as average in network edge weight in NRAS positive tumors minus the average in wild-type tumors (n = 7 and 20 for mutant and wild-type, respectively) P-values were calculated using the Wilcoxon test (two-sided). FDR was calculated by adjusting p-values for multiple testing using the Benjamini-Hochberg method. For clarity not all data points are labeled. (b) Values of node centrality were calculated from a network reconstructed using the network edges increased in NRAS mutant relative to wild-type cells. (c) A centrality index was calculated as the product between betweenness and degree values – as in (b) – for network nodes associated with the mutations shown. The phosphoproteomics data from which kinase network topologies were determined were obtained from Casado et al (Leukemia 32, 1818-1822, 2018). (d) Cell death was measured in the named AML models using a Guava assay as a function of treatment with the named inhibitors for 3 days. Values are mean ± SD (n=3 independent experiments). (e) Area under the curve (AUC) of the data in (d). Boxplots show median and interquartile ranges. Statistical significance was calculated in R by two sided paired t-test of log transformed data.
Supplementary Figure 11 Phosphoproteomic analysis of CRC cell lines isogenic for PIK3CA KD and HD mutations.
(a) Summary of results of phosphoproteomic experiments comparing CRC cell lines isogenic for PIK3CA KD and HD mutations. P-values were calculated using two-sided paired t-test of log transformed data without adjustments for multiple comparisons. These statistics were applied to the mean phosphoproteomics data of two independent experiments performed in technical replicate (b) Expression of selected markers of PI3K and MAPK pathway activity using LC-MS/MS (left panel) or Western blot (right panel). Western blot assays were performed in two independent experiments. *p< 0.05, **p< 0.01 and ***p< 0.005. Full scans of WBs are shown in Supplementary Fig. 13. (c) Correlation of circuitry markers of PI3K/MTOR and MAPK1/3 activities showing significant anti-correlation in primary breast cancers (n = 86) and CRC cell lines (n = 2 independent experiments performed in technical replicates). P-values were determined by Spearman rank test. (d) Cell viability of PIK3CA MUT isogenic cell lines as a function of treatments with the named inhibitors at the concentrations shown. Data points are mean ± SD (n = 4 technical replicates).
Supplementary Figure 12 PDTs present in PhosphoSitePlus as a function of EBDT parameter tuning and relationship between number of compounds in kinase inhibitor fingerprint and PDT size.
Number of phosphosites in PDTs matched that were present in PhosphoSitePlus for the same kinase as a function of tuning the EBDT algorithm with different probability thresholds (a) or site-kinase inhibition ratio thresholds (b). (c) Number of phosphosites found to be PDTs of given kinases as a function of number of compounds in kinase inhibitors fingerprints for the respective kinase. Total number of PDTs in dataset is 6,206. Boxplots show median, interquartile ranges, and ranges. Outliers are shown as individual data points.
Supplementary Figure 13 Western blot analysis of CRC cell lines isogenic for PIK3CA KD and HD mutations.
Full scans of all the blots shown in Supplementary Fig. 11. Named markers of PI3K/AKT and MAPK pathways were probed in two independent experiments confirming results of LC-MS analysis of the same markers in independent experiments. Click inside this box and insert image for Supplementary Fig. 14. For best results, use Insert menu to select a saved file; do not paste images. Source images must be JPEGs (no larger than 10 MB) saved in RBG color profile, at a resolution of 150–300 dpi. Optimize panel arrangement to a 2:3 height-to-width ratio; maximum online display is 600h x 900w pixels. Reduce empty space between panels and around image. Keep each image to a single page.
Supplementary Figs. 1–13 and Supplementary Notes.
PDTs identified in the study.
Results of motif enrichment analysis.
Comparison of KSEA using different K–P databases.
Markers of kinase–kinase relationships.
Impact of recurrent genetic mutations on the wiring of breast cancer and AML kinase networks.
Impact of mutations on CRC isogenic cell lines.
About this article
Cite this article
Hijazi, M., Smith, R., Rajeeve, V. et al. Reconstructing kinase network topologies from phosphoproteomics data reveals cancer-associated rewiring. Nat Biotechnol (2020). https://doi.org/10.1038/s41587-019-0391-9