. 2015 Sep 24;163(1):202-17.
Epub 2015 Sep 17.
Kinome-wide Decoding of Network-Attacking Mutations Rewiring Cancer Signaling
Free PMC article
Item in Clipboard
Kinome-wide Decoding of Network-Attacking Mutations Rewiring Cancer Signaling
Free PMC article
Cancer cells acquire pathological phenotypes through accumulation of mutations that perturb signaling networks. However, global analysis of these events is currently limited. Here, we identify six types of network-attacking mutations (NAMs), including changes in kinase and SH2 modulation, network rewiring, and the genesis and extinction of phosphorylation sites. We developed a computational platform (ReKINect) to identify NAMs and systematically interpreted the exomes and quantitative (phospho-)proteomes of five ovarian cancer cell lines and the global cancer genome repository. We identified and experimentally validated several NAMs, including PKCγ M501I and PKD1 D665N, which encode specificity switches analogous to the appearance of kinases de novo within the kinome. We discover mutant molecular logic gates, a drift toward phospho-threonine signaling, weakening of phosphorylation motifs, and kinase-inactivating hotspots in cancer. Our method pinpoints functional NAMs, scales with the complexity of cancer genomes and cell signaling, and may enhance our capability to therapeutically target tumor-specific networks.
Copyright © 2015 The Authors. Published by Elsevier Inc. All rights reserved.
Network-Attacking Mutations (A) Six distinct types of network-attacking mutations (NAMs) can be defined based on perturbations of signaling network dynamics, network structure, and dysregulation of phosphorylation sites. Cancer mutations could generate or destroy molecular logic gates, for example by creating new, or by removing existing, phosphorylation sites. Alternatively, mutant proteins could become activated by new upstream proteins (incoming edges) or start perturbing new downstream substrates (outgoing edges). Finally mutations could turn signaling proteins (e.g., protein kinases) constitutively “on” or “off.” The effect of these NAMs on the cue-signal-output flow of information is illustrated for each comparing the wild-type (WT) and mutant (Mut) cases. (B) After mapping mutations at the genomic and proteomic level, every NAM class defined in (A) is modeled on the different protein domains and motifs currently included in ReKINect following a distinct procedure: mutations on the essential residues of the kinase and SH2 domains are classified as node inactivating. Acidic mutations mimicking the phosphorylated/active state of kinases are classified as node activating. Mutations perturbing phosphorylation motifs and causing changes in the upstream kinase phosphorylating the target protein are classified as upstream rewiring. On the other hand, mutations in residues that determine specificity of the kinase or SH2 domains (Creixell et al., 2015) perturb domain specificity and are classified as downstream rewiring. Finally, our genome-specific MS experiments enable the identification of mutations generating phosphorylatable residues or the extinction of phosphorylation sites by mutating away from phosphorylatable residues. See also Supplemental Experimental Procedures.
Overview of NAMs in Cancer Cell Lines and in the Global Repository of Cancer Somatic Mutations as Predicted by ReKINect For each cell line and for the global repository of cancer somatic mutations we show the number of unique missense variants and how many of these variants fall within kinase proteins, SH2 proteins or phosphorylation sites (using a five-residue flanking region window surrounding the phosphorylation site). From these we then illustrate the fraction of variants falling within the respective domains and the fraction that can be interpreted by ReKINect. In the case of ES2, all of the 27 variants hitting an SH2 protein, hit outside SH2 domains, thus ReKINect could not make any predictions as to their effect (ghosted). It should be noted that the genesis of phosphorylation sites cannot be predicted from in silico analysis alone but require genome-specific-MS experiments. See also Figure S1.
NAMs Leading to Genesis and Extinction of Phosphorylation Sites (A) Two examples of network-attacking mutations generating new phosphorylation sites on HSF1 and TANC1, as evidenced by exome sequencing data and MS spectra matching the phosphorylated mutation. (B) Two examples of network-attacking mutations causing the extinction of known phosphorylation sites on RAB11FIP1 and TNKS1BP1, supported by exome-sequencing data and MS spectra matching the unphosphorylatable mutated residue. See also Figures S2 and S3.
NAMs Causing Downstream Rewiring (A) Three positions in direct contact with the substrate peptide, named αD1, HRD+4, and DFG+1, and likely involved in determining specificity for substrate positions P−3, P−2, and P0 (i.e., the phospho-acceptor site), respectively, harbor several cancer somatic mutations, three of which were selected for experimental validation. (B) Experimental validation by position scanning peptide library (PSPL) array of the specificity drift caused by downstream rewiring NAMs. Heat maps show normalized, averaged data from two independent experiments illustrating the specificity drift for the cancer variants PKD1 D665N and PKCγ D484G and M501I in substrate positions P−3, P−2, and P0, respectively. The results are also shown in logo form plotting the normalized information content in the wild-type and mutant specificity switch position (logos generated using Seq2Logo [Thomsen and Nielsen, 2012]). (C) The P0 specificity switch of the PKCγ variant M501I was subsequently confirmed by quantifying the phosphorylation rate of identical peptide substrates containing either Ser or Thr at the phosphorylation site position (RRRRR
SWYFGG and RRRRR TWYFGG) by mutant and wild-type kinase variants. The graph shows the mean ± SD (n = 4). (D) PKCγ expression levels are markedly increased in the tumor sample harboring the PKCγ M501I downstream rewiring mutation. (E) Comparison of the differences in substrate specificity typically observed between wild-type human kinases (gray histogram) and those mutant kinases reported here (black arrows). As evident from the plot, in two out of the three cases, the magnitude of the specificity drift caused by the cancer mutations is comparable to the specificity difference existing between different wild-type kinases. See also Figure S4.
NAMs Causing Upstream Rewiring (A) Upstream rewiring mutations will cause a new kinase (from K1 to K2) to phosphorylate the mutant protein (S, red). By plotting the probability of both kinases to phosphorylate the wild-type and mutant variants of the protein, we can visualize, quantify, and compare different upstream rewiring mutations. (B) The rewiring power and the rewiring angle can be computed by considering the necessary trajectory that the mutation causes (from the “origin” right-bottom triangle to the “destination” left-top triangle). The rewiring power is equivalent to the magnitude of the vector and measures the rewiring capacity of the mutation. The rewiring angle is the angle of the vector from the diagonal and distinguishes whether the rewiring effect is mainly driven by kinase resignation (i.e., a loss of phosphorylating ability of the wild-type kinase, angle >45°), depicted in blue, or by kinase take-over (i.e., an increase of phosphorylation ability of a new kinase, angle <45°), depicted in green. The three examples illustrate how three different mutations (A–C) can lead to different outcomes, such as the same rewiring power but different main driving force (A and B) or the same driving force but different magnitude (B and C). (C) Illustration of the two main driving processes that cause upstream rewiring, namely the reduced ability of the original kinase to phosphorylate the new mutant substrate variant (resignation) and the increased ability of a second kinase to phosphorylate the mutant substrate protein (take-over). (D) Representation of all the upstream rewiring mutations identified in the global repository of somatic mutations at different distances relative to the phosphorylation site (from five residues before a phosphorylation site, P−5, to five residues after a phosphorylation site, P+5). Rewiring events mainly driven by resignation are shown in blue and those mainly driven by take-over are shown in green. (E) Quantification of the percentage of mutations leading to upstream rewiring depending on their position relative to the phosphorylation site. (F) Assessment of the median magnitude of rewiring for mutations based on their position relative to the phosphorylation site. (G) The median rewiring angle (orange and yellow bars) and the ratio of take-over over resignation rewiring mutations (gray line) conditioned on the position of the mutation relative to the phosphorylation site. See also Figure S5.
Constitutive Activation and Inactivation of Kinases by NAMs (A) ReKINect identified ES2 cells as containing the constitutively activating BRAF V600E mutation. (B) An immunoblot and associated quantification, illustrating the phosphorylation of BRAF substrate MEK in the mutant cell line ES2 (in red) compared to the wild-type cell lines (in black), using total MEK and β-tubulin for normalization. (C) ReKINect identified several cancer mutations in catalytically essential residues of kinase domains. (D) A quantification of all mutations from the global repository of cancer somatic mutations predicted to inactivate kinases and the catalytically essential positions they hit. Mutations leading to kinase domain catalytic inactivation are enriched (χ
2 test, p = 1.69 × 10 −16) in cancer somatic mutations (with particular preference for the aspartate, D, and glycine, G, in the DFG motif).
Evidence and Model for Functional Mutations and Tumor-Specific Network Medicine (A) The functional mutations found in this study are clear examples of single amino acid mutations that can severely perturb signaling networks. (B) Our study shows how non-recurrent cancer mutations on non-conserved residues can be functionally important and that functional recurrent (orange) and non-recurrent (red) NAMs can converge at the signaling network level. We also identified a case where a functional mutation in a low-abundant protein is accompanied by its overexpression. (C) The deployment of tools like ReKINect should enable the proposition of more refined signaling mechanisms underlying cellular cancer phenotypes and identification of driver and therapeutically relevant mutations.
Interpreting Functional Cancer Somatic Mutations in Repositories of Cancer Genome Data and Cell Lines, Related to Figures 1 and 2 (A) The gap between the number of unique cancer somatic mutations reported by global sequencing efforts and the ones for which the community has been able to attribute a driving role in cancer (list of genes in the cancer gene census [Forbes et al., 2011]) has been growing drastically in the last years. (B) A more comprehensive understanding of signaling networks and how mutations perturb them would help close the interpretation gap described in (A). (C and D) Data summary of the provenance (C) and different experimental observations (D) concerning mutations, phosphorylation sites, and proteins found using exome sequencing (NGS) and (phospho)proteomics-based mass spectrometry. Stringent filters were applied to ensure data quality, including 95% of sequence space covered by 10X sequencing reads for the NGS data, standard filters applied in subsequent steps (see Experimental Procedures), and high MaxQuant and localization scores for the MS data (see Experimental Procedures). (E) Using the global repository of somatic cancer mutations, we quantified the enrichment of mutations in functional residues covered by ReKINect, and to what extent different protein domains are affected by somatic missense mutations. As one would expect, and can be observed in (E), the mutation frequency generally depends on the fraction of the genome that a given domain covers (genome coverage), as shown in the scatter plot. However, several signaling (triangles) and non-signaling (circles) domains harbor many more mutations than it would be expected by random chance or genome coverage alone (darker blue denoting lower P-value) and are mutated in a wider range of cancers (data point size). These include signaling domains like serine-threonine kinase domains, S_TKc, tyrosine kinase domains, TyrKc, and SH2 domains. (F) The results on (B) show the enrichment in cancer mutations on specific residues, calculated as the fraction of functional residues mutated and not mutated with respect to the fraction of the proteome covered by functional and non-functional residues. Odds-ratios and P-values were computed using a Fisher’s Exact Test with Multiple-Test Correction.
Genesis of Phosphorylation Sites by Cancer Mutations Experimentally Confirmed by Mass Spectrometry, Related to Figure 3 (A) Samples are processed following Spike-In SILAC standard procedures, where a mix of all the samples, in our case the five ovarian cancer cell lines, is used as an internal reference, so that peptides from different samples can be compared (Geiger et al., 2011; Monetti et al., 2011). (B) Additional examples of NGS and MS-annotated NAMs generating phosphorylation sites, including KIAA1279 G66S, ZBED4 N105S, AHNAK P4206S and MDN1 T4534S, showing the mapping of NGS and MS data, where the mutation toward phosphorylatable residues are confirmed as phosphorylation sites by the MS data using genome-specific searches.
Global Analysis of the Effect of Network Attacking Mutations on Cell Phenotypes, Related to Figure 3 (A) The effects of network topology, the presence of network attacking mutations and the classification of those mutations on the phenotypic impact of RNAi knockdown of kinase and SH2 domain containing genes were analyzed using a regression model (see Supplemental Experimental Procedures). (B) Regression variables determining model performance. For the NAM-model RNAi targets with NAMs in their close vicinity (d = 2) are more likely to cause phenotypic changes when knocked down. This likelihood is increased further if the classification of NAM is taken into account (Classified-model d = 1 and d = 2). In both models, the more proteins there are in the network vicinity of the RNAi target the less likely it is a phenotype will be observed (Any protein, d = 2) (see Supplemental Experimental Procedures). (C) The final regression result for the NAM model. Y model illustrates the regression model prediction expressed in terms of the number of nuclei, normalized to negative control, y data show the deviation from the model prediction expressed in terms of the experimental SD (see Supplemental Experimental Procedures). (D) RAB11FIP1 knockdown leads to a reduction in proliferation in all the RAB11FIP1
wt cell lines (ES2, OVAS, OVISE and TOV21). This response is lost in the cell line where a mutation has led to the extinction of a phosphorylation site on RAB11FIP1 (RAB11FIP1 T281M), KOC7C. Data shown is mean + SD from quadruplicate biological repeats. (E) TANC1 knockdown leads to a reduction in nuclear intensity in the OVAS cell line, in which a new phosphorylation site has appeared as a result of a mutation (TANC1 N251S). TANC1 wt cell lines (ES2, KOC7C, OVISE and TOV21) showed no decrease in intensity. Data shown is mean + SD from quadruplicate biological repeats. (F) The RAB11FIP1 mutated cell line KOC7C was the only cell line not to show a significant phenotype in FACS analysis of cell-cycle dynamics. Data shown is mean + SD from triplicate biological repeats.
High-Confidence Candidate NAMs Driving Downstream Rewiring and Downstream Rewiring Experimental Validation, Related to Figure 4 (A) Cancer mutants on position DFG+1 most likely causing downstream rewiring. Further focusing on alignment position DFG+1 reveals several additional cancer mutations with hydrophobic-to-β-branched aliphatic residues. (B) Cancer mutants on position αD1 most likely causing downstream rewiring. Further focusing on alignment position αD1 reveals several additional cancer mutations with D-to-N substitutions. (C) The enrichment of mutations in DFG+1 favoring Ser-to-Thr specificity switches (with nine mutants following this pattern and none in the opposite direction, as shown in the bottom figure) will most likely lead to a specificity drift from Serine toward Threonine phosphorylation preference. (D) Full PSSMs for the downstream rewiring cancer mutation in PKCγ driving changes in kinase substrate specificity. Two cancer mutations in PKCγ, D484G and M501I, switch specificity by perturbing the residue preference on the substrate position P-2 and P0 (i.e., the phospho-acceptor site) respectively. (E) Full PSSMs for the downstream rewiring cancer mutation in PKD1 driving changes in kinase substrate specificity. A cancer mutation in PKD1, D665N, causes a drift in substrate specificity on P-3 position.
Upstream Rewiring Graphs Using NetworKIN and Information Content in Various Phosphorylation Substrate Positions, Related to Figure 5 (A) As an extension to the graphs shown in Figure 5, here we show similar rewiring graphs computed using NetworKIN (Linding et al., 2007) instead of NetPhorest (Miller et al., 2008), and therefore including contextual information for improved accuracy. Please note that the extreme values in P-3 and P+1 graphs were added for completeness, but due to their outlier status are out of scale (values added and their numerical values can be seen in the figure); for further information and accurate numerical values please refer to Table S5 online. It is important to note that we only used top-scoring NetworKIN (Linding et al., 2007) and NetPhorest (Miller et al., 2008) predictions filtered to ensure maximum confidence, and that we reached the same conclusions using both algorithms. Because of this and the fact we have based our global observations on thousands of mutations, our conclusions drawn from Figure 5 and (B) are highly unlikely to have been affected by our choice of methods. (B) By analyzing PSSMs characterizing the peptide specificity of a large number of human protein kinases (from the NetPhorest repository), we could quantify how much each substrate position contributes to the kinase-substrate recognition process (from seven residues before the phosphorylation position, P-7, up to seven residues after, P+7). Similar to what we observed in Figure 5 and (A), in the case of mutations hitting different positions and their likelihood to lead to upstream rewiring, position P-1 contributes relatively little to the kinase-substrate recognition process.
All figures (13)
Unmasking determinants of specificity in the human kinome.
Cell. 2015 Sep 24;163(1):187-201. doi: 10.1016/j.cell.2015.08.057. Epub 2015 Sep 17.
26388442 Free PMC article.
Systematic analysis of somatic mutations in phosphorylation signaling predicts novel cancer drivers.
Mol Syst Biol. 2013;9:637. doi: 10.1038/msb.2012.68.
Mol Syst Biol. 2013.
23340843 Free PMC article.
Presence and utility of intrinsically disordered regions in kinases.
Mol Biosyst. 2014 Nov;10(11):2876-88. doi: 10.1039/c4mb00224e.
Mol Biosyst. 2014.
Kinomics: methods for deciphering the kinome.
Nat Methods. 2005 Jan;2(1):17-25. doi: 10.1038/nmeth731.
Nat Methods. 2005.
The resistance tetrad: amino acid hotspots for kinome-wide exploitation of drug-resistant protein kinase alleles.
Methods Enzymol. 2014;548:117-46. doi: 10.1016/B978-0-12-397918-6.00005-7.
Methods Enzymol. 2014.
Effect of naive and cancer-educated fibroblasts on colon cancer cell circadian growth rhythm.
Cell Death Dis. 2020 Apr 27;11(4):289. doi: 10.1038/s41419-020-2468-2.
Cell Death Dis. 2020.
32341349 Free PMC article.
MutaBind2: Predicting the Impacts of Single and Multiple Mutations on Protein-Protein Interactions.
iScience. 2020 Mar 27;23(3):100939. doi: 10.1016/j.isci.2020.100939. Epub 2020 Feb 27.
32169820 Free PMC article.
Profiling Cell Signaling Networks at Single-cell Resolution.
Mol Cell Proteomics. 2020 May;19(5):744-756. doi: 10.1074/mcp.R119.001790. Epub 2020 Mar 4.
Mol Cell Proteomics. 2020.
32132232 Free PMC article.
Lysine Methylation Regulators Moonlighting outside the Epigenome.
Mol Cell. 2019 Sep 19;75(6):1092-1101. doi: 10.1016/j.molcel.2019.08.026.
Mol Cell. 2019.
Systematic analysis of the intersection of disease mutations with protein modifications.
BMC Med Genomics. 2019 Jul 25;12(Suppl 6):109. doi: 10.1186/s12920-019-0543-2.
BMC Med Genomics. 2019.
31345222 Free PMC article.
Alexander J., Lim D., Joughin B.A., Hegemann B., Hutchins J.R.A., Ehrenberger T., Ivins F., Sessa F., Hudecz O., Nigg E.A. Spatial exclusivity combined with positive and negative selection of phosphorylation motifs is the basis for context-dependent mitotic signaling. Sci. Signal. 2011;4:ra42.
Anastassiadis T., Deacon S.W., Devarajan K., Ma H., Peterson J.R. Comprehensive assay of kinase catalytic activity reveals features of kinase inhibitor selectivity. Nat. Biotechnol. 2011;29:1039–1045.
Andreadi C., Cheung L.K., Giblett S., Patel B., Jin H., Mercer K., Kamata T., Lee P., Williams A., McMahon M. The intermediate-activity (L597V)BRAF mutant acts as an epistatic modifier of oncogenic RAS by enhancing signaling through the RAF/MEK/ERK pathway. Genes Dev. 2012;26:1945–1958.
Antal C.E., Hudson A.M., Kang E., Zanca C., Wirth C., Stephenson N.L., Trotter E.W., Gallegos L.L., Miller C.J., Furnari F.B. Cancer-associated protein kinase C mutations reveal kinase’s role as tumor suppressor. Cell. 2015;160:489–502.
Arnold R., Patzak I.M., Neuhaus B., Vancauwenbergh S., Veillette A., Van Lint J., Kiefer F. Activation of hematopoietic progenitor kinase 1 involves relocation, autophosphorylation, and transphosphorylation by protein kinase D1. Mol. Cell. Biol. 2005;25:2364–2383.
Research Support, N.I.H., Extramural
Research Support, Non-U.S. Gov't
Information Storage and Retrieval
Ovarian Neoplasms / metabolism*
Protein Kinases / chemistry
Protein Kinases / genetics*
Protein Kinases / metabolism*
LinkOut - more resources
Full Text Sources Medical Research Materials