Research ResourceCancer

Integration of protein phosphorylation, acetylation, and methylation data sets to outline lung cancer signaling networks

See allHide authors and affiliations

Science Signaling  22 May 2018:
Vol. 11, Issue 531, eaaq1087
DOI: 10.1126/scisignal.aaq1087
  • Fig. 1 Data acquisition and analysis.

    (A) A CFN was created by using a machine learning, pattern recognition algorithm (t-SNE) to identify which PTMs clustered together, filtering PPIs (red edges) to retain only those between proteins whose PTMs coclustered. (B) Outline of TMT mass spectrometry coupled with immunoprecipitation (IP) using modification-specific antibodies. Phospho-Ser/Thr (pS/T) peptide IP was accomplished in multiple steps with AGC/PSD-family kinase substrate, AKT (v-akt murine thymoma viral oncogene homolog 1) substrate, AMP kinase substrate, and ATM/ATR (ataxia telangiectasia mutated/ATM related) substrate antibodies (see Materials and Methods). Phosphopeptides were also further purified on a TiO2 column (81). m/z, mass/charge ratio; LCMS/MS, liquid chromatography–mass spectrometry. (C) Bird’s eye view of the CCCN derived from PTMs from 15 independent experiments, each with six multiplex samples, from comparison of 45 lung cancer cell lines [12 derived from from small cell lung cancer (SCLC) and 33 from non–small cell lung cancer (NSCLC)] to normal lung tissue, and selected cell lines treated with anticancer drugs. This disconnected network includes threshold-filtered Spearman correlations among t-SNE–clustered PTMs (yellow edges are positive correlations; blue are negative correlations). Also shown are negative correlations among different modification types within the same protein, which are useful for revealing antagonistic relationships among PTMs. Node size and color reflect total of all PTM ratios in the data set (fold change key). This network is available for download as data file S1 and on the NDEx repository (https://doi.org/10.18119/N9F59Z). This network combined with the CFN that contains filtered PPI edges may be explored at https://cynetworkbrowser.umt.edu/.

  • Fig. 2 Networks derived from composite shortest paths from drug targets to drug-affected proteins.

    (A to C) Graphs showing the sum of shortest paths from each target to each protein whose PTMs were more than twofold affected by the MET and ALK inhibitor crizotinib (A) and the EGFR, ERBB2, and ERBB3 inhibitor gefitinib (Iressa) (B) in H3255 cells treated for 1 to 24 hours. A key defining node shape and border colors and edge colors is shown bottom right. Directed edges are shown with arrowheads; these indicate one protein acting on another protein (for example, kinases phosphorylating substrates). Undirected edges without arrowheads indicate various other types of interactions. Edge line thickness is proportional to the strength of interactions, as defined in PPI databases. Node size and color are proportional to the changes in PTMs for each protein in response to indicated drug treatments (see scale bar; yellow indicates positive change; blue, negative; green, no change). Many yellow nodes represent overall increases in acetylation in these graphs (C). Fold change for individual PTMs in response to indicated drugs is plotted using heat maps on a blue-yellow scale (see scale bar). Several proteins exhibited phosphorylation decreases (represented by blue on the heat map) and concomitant acetylation increases (represented by yellow) in response to gefitinib.

  • Fig. 3 Heat maps and network showing PTMs affected more than twofold by geldanamycin.

    (A to C) Heat maps of PTMs of HSPs (A), proteins involved in endocytosis (B), and proteins involved with the cytoskeleton (C) in each of two lung cancer cell lines cultured with serum and either untreated or treated with geldanamycin (100 μM for 15 or 24 hours) relative to serum-starved, untreated cultures. Scale bar (B) indicates fold change on a blue-yellow scale. (D) Network graph plotted as in Fig. 2 showing the sum of shortest paths from HSP90AA1 and HSP90AB1 (center) to each protein whose PTMs were more than twofold affected by geldanamycin in H2228 or H3122 cells. Node size and color indicate the sum of PTM changes in response to geldanamycin in both cell types (refer to the key in Fig. 2).

  • Fig. 4 EP300 interactions with PTM-modifying enzymes.

    (A) Shown are links in the CFN between EP300 and kinases, phosphatases, acetyl- and methyltransferases, and the actin polymerization–governing GTPase (guanosine triphosphatases) RAC1 and CDC42. In cases where there were more than one type of interaction between two proteins in the PPI databases (see the key in Fig. 2), to simplify graphs, white edges represent the composite of these interactions with the edge weight summed. (B) EP300 interactions with CLTC and enzymes that modify clathrin and govern EGFR endocytosis. The graph shows PTMs linked by black edges to the parent protein, and correlation edges (yellow, positive; blue, negative) between PTMs in the PTM CCCN, which depicts PTMs that cocluster with edges representing Spearman correlation greater than the threshold of |0.543|, as defined in fig. S9. Node size and color (refer to the key in Fig. 2) indicate ratios of PTM changes in lung cancer cell lines to those in normal lung tissue; see fig. S11 for responses to drugs. The combined CFN/CCCN graphs show the individual PTM response and the sum of all PTMs represented in the parent (gene) node.

  • Fig. 5 Dually acetylated and phosphorylated proteins: Interactions with bromodomain proteins and kinases.

    (A) Correlation between different PTM sites on the same protein: different phosphorylation sites (p p); phosphorylation and acetylation (p ac); and different acetylation sites (ac ac). Homo-PTM correlations (p p and ac ac) are compared to heterologous (p ac) PTMs (both P < 2.2 × 10−16, Welch two-sample t test). Individual PTM-PTM correlations are plotted in gray under boxplots (top). Correlation density between phosphorylation and acetylation sites on the same proteins is also shown (bottom). Similar plots in which tyrosine and serine/threonine phosphorylation are compared to each other and each to acetylation individually are shown in fig. S13. Negative correlations selected for display as edges are highlighted in blue. (B) Combined CFN and PTM CCCN for selected proteins modified by both phosphorylation and acetylation. In addition to correlation edges, negative Spearman correlations less than −0.5 are shown as blue edges (plotted as in Fig. 4B, except edges connecting proteins to their PTMs are light gray). (C and D) Selected regions of (B) expanded for clarity. (E) CFN interactions between tyrosine kinases, geldanamycin-affected dually acetylated and phosphorylated endocytic and cytoskeletal proteins, and bromodomain proteins (colored light red). Node size and color (refer to the key in Fig. 2) are in response to geldanamycin for (B) to (D). (F) Comparison of the number of interactions of bromodomain-containing proteins with dually modified proteins both phosphorylated and acetylated (p ac); those with negative correlations less than −0.5 (p ac neg); all acetylated proteins (all ac); all proteins not acetylated (all except ac); all proteins except those both phosphorylated and acetylated (all except p ac); and proteins only modified by phosphorylation (p only) or methylation (me only). CFN interactions between bromodomain-containing proteins and these groups of proteins were retrieved, and the number of edges was divided by the number of proteins in each group to obtain bromodomain interactions per gene.

  • Fig. 6 Geldanamycin induces changes in lipid raft localization of dually phosphorylated/acetylated proteins.

    (A) Outline of cell fractionation experiments that separate organelles including different populations of endosomes (org1 to org4, top) and detergent-sensitive and detergent-resistant membranes including lipid rafts (raft1 to raft4, bottom). Org1 to org3 contain lysosomes and endosomes with different sedimentation velocity, respectively; org4 contains soluble, cytoplasmic proteins; raft1 contains detergent-sensitive proteins; raft2 to raft4 are detergent-resistant fractions of decreasing equilibrium density (10, 46, 47). (B to E) H3122 and H3255 cells treated with geldanamycin (“geld.”) or not treated (“C.”) were fractionated, and total protein amounts in each fraction [four organellar fractions (“org.”) and four raft fractions (“raft.”)] were determined in three separate experiments by mass spectrometry (B and C, H3122 cells) and, for a select few, by Western blotting (D, H3255 cells; E, cell type indicated). Mass spectrometry data [expressed as heat maps in (B)] and Western blot data (D, and fig. S14) were used to calculate the amount of each protein in each cell fraction as a proportion of the total in the whole cell, which is the sum of protein amounts in all cell fractions. The heat maps in (C) and (E) show fold change abundance in treated cells relative to control cells for each fraction for mass spectrometry and Western blot data, respectively. One hundred sixty-eight of 218 dually phosphorylated and acetylated proteins with negative correlations between these PTMs were detected in this experiment. Many exhibited changes in the raft2 fraction (P = 0.05394, Welch two-sample t test). Forty-five of 108 proteins whose PTMs changed significantly in response to geldanamycin (Fig. 3) were detected by mass spectrometry in this experiment; collectively, their amounts increased in the raft2 fraction (P = 0.01457). Of the six bromodomain-containing proteins detected, most decreased in the raft2 fraction (P = 0.04832 for all the bromodomain-containing proteins). Mass spectrometry data expressed as the amount of proteins in cell fractions are in table S2.

  • Fig. 7 Clusters of proteins dually modified by different PTMs.

    (A and C) Correlations between different PTM sites on the same protein plotted as in Fig. 5A. Homo-PTM correlations (me me; ac ac; p p) are significantly different from heterologous PTM correlations (me ac; p me; P < 2.2 × 10−16 in all cases). (B and D) Selected proteins modified by (B) acetylation and methylation, and (D) phosphorylation and methylation are shown with CFN links to PTM-modifying enzymes, as described in Fig. 4B. These figures highlight two cliques of PTM CCCN edges that coclustered and had a high Spearman correlation. Negative correlations between different PTMs on the same protein are indicated by blue edges. Node size and color (refer to the key in Fig. 2) reflect total of all PTM ratios in the data set.

  • Fig. 8 Pathways to SMARCA4 and NKX2-1.

    (A) Gene expression data from 29 NSCLC cell lines from the CCLE (49) plotted on a blue-red scale (key) (50). NKX2-1 and SMARCA4 (highlighted) are among the most dysregulated genes in lung cancer. (B) CFN and CCCN path from EP300 to NKX2-1 and SMARCA4, where node size and color (refer to the key in Fig. 2) indicate total ratio data for all experiments where lung cancer cell lines were compared to normal lung tissue. (C) Shortest paths from EGFR and MET to transcription factors in a CFN derived from the PPI data set with highly curated molecular interactions with a focus on direct interactions for which there is strong evidence (20, 85). Note that DDX5 and NCOA2 bind to NKX2-1 and SMARCA4 as in (B). Node size is CFN betweenness, and node color is normalized betweenness defined as CFN betweenness divided by the betweenness from all PPI data sets before filtering.

  • Fig. 9 Dual PTMs may function as dueling PTMs.

    (A) Venn diagram showing the overlap among proteins modified by more than one type of PTM. A similar Venn diagram in which tyrosine and serine/threonine phosphorylation are separated is shown in fig. S15A. (B) Schematic of the findings. Proteins in molecular signaling pathways modified by more than one PTM have different sets of interacting proteins (5, 6). PTM-driven interactions occur through recognition of specifically modified amino acid residues by protein domains listed under the PTM type in the figure. Acetylated (Ac) proteins interact with bromodomain and BET (bromodomain and extraterminal domain) proteins (green); methylated proteins interact with proteins containing tudor domains if methylated on arginine (RMe); or chromo, PWWP (’Pro-Trp-Trp-Pro’), and MBT (malignant brain tumor) domains if methylated on lysine (KMe) (purple). Phosphorylated proteins interact with several protein families (red): tyrosine phosphorylated proteins (pY) interact with SH2 (Src homology domain 2) and PTB (phosphotyrosine binding) domains; serine/threonine phosphorylated proteins (pS/T) interact with a variety of proteins including 14-3-3 protein family members and proteins containing the domains WW (domain with 2 conserved Trp residues), FHA (forkhead-associated domain), WD40 (WD or β-transducin repeats), and LRR (leucine-rich repeats). Our data suggest that where inverse correlations exist between different PTMs, these may function as exclusive “OR” (XOR) switches to direct alternative cellular outcomes. This principle applies to phosphorylation versus acetylation; phosphorylation versus methylation; and methylation versus acetylation.

Supplementary Materials

  • www.sciencesignaling.org/cgi/content/full/11/531/eaaq1087/DC1

    Text S1. Data analysis considerations and network development.

    Fig. S1. Example t-SNE embeddings.

    Fig. S2. Initial evaluation of clusters derived from t-SNE embeddings.

    Fig. S3. Correlation between replicates.

    Fig. S4. Number of edges per gene in clusters.

    Fig. S5. Correlation between degree and betweenness.

    Fig. S6. CFN betweenness is not biased by overrepresentation in PPI databases.

    Fig. S7. The number of posttranslational modifications per protein weakly correlates with CFN betweenness.

    Fig. S8. The effect of applying PTM correlation threshold on CFNs.

    Fig. S9. Network density as a function of PTM correlation threshold defines a minimum for construction of a PTM CCCN.

    Fig. S10. Relative network density of drug-affected proteins.

    Fig. S11. EP300 interactions with PTM-modifying enzymes.

    Fig. S12. Combined EP300 CCCN and CFN.

    Fig. S13. Comparison of tyrosine phosphorylation, serine/threonine phosphorylation, and acetylation.

    Fig. S14. Western blot data from cell fractionation experiments.

    Fig. S15. Separation of tyrosine and serine/threonine phosphorylation and relative abundance compared to paxDB.

    Table S1. Proteins with PTMs that have negative correlations.

    Table S2. Amount of proteins in cell fractions.

    Data file S1. Cytoscape file: Complete cocluster correlation network.

    References (8991)

  • Supplementary Materials for:

    Integration of protein phosphorylation, acetylation, and methylation data sets to outline lung cancer signaling networks

    Mark Grimes,* Benjamin Hall, Lauren Foltz, Tyler Levy, Klarisa Rikova, Jeremiah Gaiser, William Cook, Ekaterina Smirnova, Travis Wheeler, Neil R. Clark, Alexander Lachmann, Bin Zhang, Peter Hornbeck, Avi Ma'ayan, Michael Comb

    *Corresponding author. Email: mark.grimes{at}mso.umt.edu

    This PDF file includes:

    • Text S1. Data analysis considerations and network development.
    • Fig. S1. Example t-SNE embeddings.
    • Fig. S2. Initial evaluation of clusters derived from t-SNE embeddings.
    • Fig. S3. Correlation between replicates.
    • Fig. S4. Number of edges per gene in clusters.
    • Fig. S5. Correlation between degree and betweenness.
    • Fig. S6. CFN betweenness is not biased by overrepresentation in PPI databases.
    • Fig. S7. The number of posttranslational modifications per protein weakly correlates with CFN betweenness.
    • Fig. S8. The effect of applying PTM correlation threshold on CFNs.
    • Fig. S9. Network density as a function of PTM correlation threshold defines a minimum for construction of a PTM CCCN.
    • Fig. S10. Relative network density of drug-affected proteins.
    • Fig. S11. EP300 interactions with PTM-modifying enzymes.
    • Fig. S12. Combined EP300 CCCN and CFN.
    • Fig. S13. Comparison of tyrosine phosphorylation, serine/threonine phosphorylation, and acetylation.
    • Fig. S14. Western blot data from cell fractionation experiments.
    • Fig. S15. Separation of tyrosine and serine/threonine phosphorylation and relative abundance compared to paxDB.
    • Legends for tables S1 and S2
    • Legend for data file S1
    • References (8991)

    [Download PDF]

    Other Supplementary Material for this manuscript includes the following:

    • Table S1 (Microsoft Excel format). Proteins with PTMs that have negative correlations.
    • Table S2 (Microsoft Excel format). Amount of proteins in cell fractions.
    • Data file S1 (.cys format). Cytoscape file: Complete cocluster correlation network.

    [Download Tables S1 and S2]

    [Download Data file S1]


    © 2018 American Association for the Advancement of Science

Navigate This Article