Research ResourceCancer

Kinase-Substrate Enrichment Analysis Provides Insights into the Heterogeneity of Signaling Pathway Activation in Leukemia Cells

See allHide authors and affiliations

Science Signaling  26 Mar 2013:
Vol. 6, Issue 268, pp. rs6
DOI: 10.1126/scisignal.2003573

Abstract

Kinases determine the phenotypes of many cancer cells, but the frequency with which individual kinases are activated in primary tumors remains largely unknown. We used a computational approach, termed kinase-substrate enrichment analysis (KSEA), to systematically infer the activation of given kinase pathways from mass spectrometry–based phosphoproteomic analysis of acute myeloid leukemia (AML) cells. Experiments conducted in cell lines validated the approach and, furthermore, revealed that DNA-dependent protein kinase (DNA-PK) was activated as a result of inhibiting the phosphoinositide 3-kinase (PI3K)–mammalian target of rapamycin (mTOR) signaling pathway. Application of KSEA to primary AML cells identified PI3K, casein kinases (CKs), cyclin-dependent kinases (CDKs), and p21-activated kinases (PAKs) as the kinase substrate groups most frequently enriched in this cancer type. Substrates phosphorylated by extracellular signal–regulated kinase (ERK) and cell division cycle 7 (CDC7) were enriched in primary AML cells that were resistant to inhibition of PI3K-mTOR signaling, whereas substrates of the kinases Abl, Lck, Src, and CDK1 were increased in abundance in inhibitor-sensitive cells. Modeling based on the abundances of these substrate groups accurately predicted sensitivity to a dual PI3K and mTOR inhibitor in two independent sets of primary AML cells isolated from patients. Thus, our study demonstrates KSEA as an untargeted method for the systematic profiling of kinase pathway activities and for increasing our understanding of diseases caused by the dysregulation of signaling pathways.

Introduction

The signaling axis composed of phosphoinositide 3-kinase (PI3K), Akt, and the mammalian target of rapamycin (mTOR) is a paradigm of oncogenic signaling, and several inhibitors that target different nodes in this pathway are in clinical development. However, as with other oncogenic signaling events, dysregulation of PI3K signaling in cancer is heterogeneous; whereas components of this pathway may be essential for the survival of certain cell types, these may be redundant in others (18). Cancers are caused by genetic mutations and cytogenetic alterations that affect gene expression (911) and pathway activities; however, it is often difficult to uncover the functional consequences that these genetic changes have on the activities of the biochemical pathways that determine the cancer phenotype (12). The activities of kinase signaling pathways are inherently linked to the wiring of signaling networks (13, 14), which may differ between cell populations and for which our knowledge is still limited. As a consequence, the prediction of how genetic alterations may affect kinase pathway activity is substantially hampered.

Large-scale phosphoproteomics, now routine in many mass spectrometry (MS) laboratories, enables the quantification of thousands of phosphorylation sites in a single experiment (15, 16). Because, by definition, each phosphorylation site is the result of kinase activity (which is opposed by phosphatase activity), it should theoretically be possible to use phosphoproteomics data to obtain an estimate of proximal pathway activity downstream of each kinase present in the system under investigation (17). This would require the measurement of known kinase substrates (that is, specific phosphorylation sites) that could then be used as markers of kinase pathway activity. Databases of substrate-kinase relationships are publically available (1821), and, although not comprehensive, a subset of the sites quantifiable by large-scale phosphoproteomics is represented in such repositories. The challenge in using this information is the low specificity of kinase-substrate relationships because several different kinases may phosphorylate the same substrates, and proteins phosphorylated by a given kinase in one cell type may not be found or may be poor substrates of that kinase in other cell types. In addition, the stoichiometry of phosphorylation is difficult to measure routinely, and protein phosphorylation is a dynamic posttranslational modification that can quickly change during the course of an experiment. As a result, variables that are difficult to control, such as the circadian clock, cell confluence, and shear stress, can all affect protein kinase activity (2224). Such difficulties can contribute to the generation of substantial “noise” in the phosphoproteomics data. Therefore, known substrates of a given kinase may show inconsistent extents of phosphorylation across experimental conditions when considered independently.

Signal averaging is a well-known means to enhance the signal-to-noise ratio from inherently noisy data (25), and integration of large-scale biological results can lead to greater depth of biological insights than are generated when considering single variables (26). We therefore tested whether considering and averaging the amounts of several substrates of the same kinase (thus diluting the contribution of outliers) could be used to gain reliable information on the activity of a kinase pathway from MS-based phosphoproteomics. Our working hypothesis was that the results of this analysis, which we term kinase-substrate enrichment analysis (KSEA), may then be considered as a measure of global kinase pathway activation.

Therefore, the aim of KSEA is to systematically infer pathway activity from phosphoproteomics data. Phosphopeptides identified in large-scale phosphoproteomics experiments were arranged into substrate groups, defined as containing phosphorylation sites known to be substrates of specific kinases or as sharing specific phosphorylation motifs. We then calculated the extent and statistical significance of the enrichment of these substrate groups relative to the phosphoproteomics data set. Here, we first tested the approach in two acute myeloid leukemia (AML) cell lines with and without treatment with inhibitors of PI3K and mTOR. The results of KSEA correlated with those obtained by Western blotting analysis and demonstrated a dose-dependent inhibition of protein kinases known to be downstream of PI3K. Application of the approach to systematically profile kinase signaling in primary AML blasts isolated from patients identified kinase pathway activities present in this disease; some of which were associated with (and thus predicted) the sensitivity of primary AML to a PI3K inhibitor. Therefore, KSEA may represent a general approach to quantify kinase pathways and to stratify cancers based on their predicted responses to signaling inhibitors.

Results

Evaluation of KSEA in two AML cell lines of different sensitivities to kinase inhibitors

To test the accuracy of KSEA, we first used the approach to infer relative kinase pathway activities across the P31/Fuj and Kasumi-1 AML cell lines. P31/Fuj cells, which are deficient in PTEN (phosphatase and tensin homolog), are resistant to multiple drugs compared to Kasumi-1 cells, which have wild-type PTEN (17). We analyzed the phosphoproteomes of both cell lines with liquid chromatography–tandem MS (LC-MS/MS) methodology, as previously described (17, 27). We analyzed six independent replicates for each cell line in each experiment, and each experiment was performed three times, generating a total of 18 biological replicates per cell line. We quantified a total of 4129 unique phosphorylated peptide ions across the cell lines (Data set 1), with several hundreds of these being preferentially found in either P31/Fuj or Kasumi-1 cells. The actual numbers depended on the cutoff selected for the P value (Fig. 1A and fig. S1A). The overall abundances of most phosphorylated peptides were similar between the cell lines, with the log2 values of the fold differences centering ~0 (that is, showed no change) (Fig. 1A and fig. S1B).

Fig. 1

Principle and evaluation of KSEA. P31/Fuj and Kasumi-1 cells were seeded and harvested 24 hours later (12 replicates per condition). Proteins were extracted and digested with trypsin. Phosphorylated peptides were enriched with TiO2 and identified and quantified by LC-MS/MS. Average fold differences in abundances between both cell lines (P31/Fuj over Kasumi-1) and the statistical significance of these differences were calculated for each phosphorylated peptide. These data were then used for KSEA analysis. (A) Volcano plot showing the fold difference between the two cell lines in the abundance of each phosphorylated peptide of the respective P values. Fold difference is represented as the log2 (a positive value for the log2 of the fold difference indicates the increased abundance of a phosphorylated peptide in P31/Fuj cells, whereas a negative value indicates increased abundance in Kasumi-1 cells) and P value as log10 (a significance log10 P value <−1.3 corresponds to a linear P value of <0.05). (B) Illustrative examples of how KSEA calculates substrate group enrichment across the two cell lines for the kinases Akt1 and ERK1. Data points are the log2 of the fold difference between both cell lines of peptides containing phosphorylation sites listed in the PhosphoSite database as substrates of the indicated kinases. (C) Substrate group enrichment data of kinases represented with more than three entries in the substrate group. (D) Western blotting analysis showing the phosphorylation of proteins on sites previously reported to correlate with the activity of kinases indicated with an arrow in (C). Blots are representative of three independent experiments showing similar results. (E) KSEA applied to the data shown in (A) to calculate the enrichment of substrate groups defined by the named motif with the Delta counts method.

We provide examples of KSEA outputs in the distribution of fold differences in the amounts of phosphorylated peptides listed as substrates of Akt1 and extracellular signal–regulated kinase 1 (ERK1) by the PhosphoSite database (Fig. 1B). The specific peptides in these substrate groups were listed (fig. S2A), and we found that the results of three different tests of normality were consistent with the data being normally distributed (fig. S2B). The means of the log2 fold differences (the amounts of the phosphorylated peptides in P31/Fuj cells divided by those in Kasumi-1 cells) were 0.9 and −0.5 for the Akt1 and ERK1 substrate groups, respectively (Fig. 1B). Because a log2 fold change of 0 means that there is no difference between the samples being compared, a log2 fold change of 0.9 indicates that, on average, the abundances of the phosphorylated peptides that are Akt1 substrates were greater in P31/Fuj cells than in Kasumi-1 cells, whereas a log2 fold change of −0.5 indicates that known ERK1 substrates were more abundant in Kasumi-1 cells than in P31/Fuj cells. Therefore, we used average substrate abundances to calculate a value of enrichment, defined here as the ratio of the mean fold difference in the abundances of phosphorylated peptides known to be substrates of a particular kinase (that is, in the substrate groups indicated in Fig. 1B), relative to the mean fold difference in the abundances of all of the phosphorylated peptides quantified in the experiment (indicated in Fig. 1A). The results of this enrichment analysis were then presented for groups found with substrates in both P31/Fuj and Kasumi-1 cell lines (Fig. 1C and fig. S3A). The statistical significance of such enrichment was calculated with a z test (28). We also explored two other methods for calculating substrate group enrichment and statistical significance; these involved determining the log2 of the fold difference in the mean abundance of each substrate group across the conditions being compared (fig. S3B) or subtracting the number of phosphorylated peptides in a substrate group that were statistically significantly increased in abundance from those that were statistically significantly decreased in abundance across experimental conditions (fig. S3C). These two values of enrichment are referred to as the Fold and Delta counts, respectively. A paired t test of the means or the hypergeometric test was then used to infer the statistical significance of enrichment. Although they differed in overall magnitude, the results of these three measures of enrichment and statistical significance were consistent with each other (fig. S3), illustrating the robustness of the approach.

Phosphorylated substrate groups that we found to be enriched in Kasumi-1 cells included those for mitogen-activated or extracellular signal–regulated protein kinase kinase (MEK), ERK, and casein kinase 2 (CK2), whereas substrate groups for Akt, p21-activated kinase 1 (PAK1), protein kinase C (PKC) isoforms, ribosomal S6 kinases (RSKs), spleen tyrosine kinase (Syk), and RhoA-binding kinase (ROCK1) were enriched in P31/Fuj cells (Fig. 1C), suggesting that these kinases were more active in the respective cell lines. To test whether the results obtained by KSEA were an accurate reflection of kinase pathway activity, we compared these results with those obtained by Western blotting analysis of phosphorylation sites that correlate with the activity of selected kinases (based on antibody availability) (Fig. 1D). Kinases validated by Western blotting are indicated with an arrow (Fig. 1C). The extent of phosphorylation of ERK at Thr202 and Tyr204, which correlates with ERK activity and is a marker of MEK activation, was increased in Kasumi-1 cells compared to that in P31/Fuj cells, whereas the extents of phosphorylation of Ser473 of Akt, Ser380 of p90RSK, Thr505 of PKCδ, and Thr538 of PKCθ [all of which correlate with the activity state of these kinases (2932)] were more prominent in P31/Fuj cells than in Kasumi-1 cells (Fig. 1D). The extents of phosphorylation of glycogen synthase kinase 3β (GSK-3β) at Ser9, ribosomal S6 at Ser235/236, and myristoylated alanine-rich C-kinase substrate (MARCKS) at Ser152/156, which are catalyzed by Akt, RSK, and PKC, respectively (33, 34), were also increased in P31/Fuj cells compared to Kasumi-1 cells (Fig. 1D). Thus, comparison of results from KSEA (Fig. 1C) with those from Western blotting analysis (Fig. 1D) indicated that the substrate group enrichment values obtained by KSEA were a faithful reflection of pathway activity.

To complement KSEA based on knowledge of kinase-substrate relationships, we also applied the approach of taking phosphorylation motifs as the source of substrate groups. With this approach, we found that phosphorylated acidic motifs were enriched in Kasumi-1 cells, whereas phosphorylated basic motifs were enriched in P31/Fuj cells (fig. S4). The combined abundances of phosphorylated peptides with the motif SDxExE [associated with CK2 (35)] were more prominent in Kasumi-1 cells than in P31/Fuj cells (Fig. 1E). In contrast, phosphorylated peptides with the RxRxxS and KxRxxS motifs, which are associated with the activity of Akt and RSK isoforms (36), were enriched in P31/Fuj cells (Fig. 1E). These results are consistent with those obtained from KSEA based on databases of kinase-substrate relationships (Fig. 1C).

Application of KSEA to characterize kinases involved in the cellular response to PI3K and mTOR inhibitors

We next applied KSEA to analyze the phosphoproteomes of P31/Fuj and Kasumi-1 cells treated with the structurally unrelated PI3K and mTOR inhibitors AZ12321046 (hereinafter named AZ123) and PI-103 (37, 38) and the mTOR complex 1 (mTORC1) and mTORC2 inhibitor Ku-0063794 (hereinafter named Ku-794) (39). Both of the PI3K and mTOR inhibitors (AZ123 and PI-103) blocked the phosphorylation of several markers of the PI3K signaling pathway in P31/Fuj and Kasumi-1 cells, whereas the mTOR-specific inhibitor Ku-794 did not substantially inhibit the phosphorylation of Akt in Kasumi-1 cells or of GSK-3β in both cell lines (fig. S5A). As expected, the proliferation of Kasumi-1 cells was more substantially inhibited by all of the inhibitors than was the proliferation of P31/Fuj cells (fig. S5B). The latter cell line is resistant to multiple kinase inhibitors, but this does not appear to be a result of the increased activity of membrane transporters; drug targets in this cell line were inhibited to the same extent as those in sensitive cells (fig. S5A) (40).

We used LC-MS/MS to compare the phosphoproteomes of P31/Fuj and Kasumi-1 cells after treatment with Ku-794, AZ123, or PI-103, relative to those of cells treated with dimethyl sulfoxide (DMSO) as a control. Three biological replicates were analyzed in each experiment, and the experiments were performed twice, resulting in six biological replicates for each condition. We quantified a total of 4129 phosphopeptides across 12 experimental conditions, which led to a total of ~310,000 quantitative data points (Data set 1). Tens to hundreds of phosphopeptides (depending on the selected cutoff for the fold differences and P values) were regulated by these inhibitors in each condition (fig. S6). Although phosphorylated peptide abundances predominantly decreased after treatment with inhibitors, there were also a statistically significant number of phosphorylated peptides whose abundances increased (fig. S6). Principal component analysis (PCA) of these data enabled us to separate vehicle- and AZ123-treated samples in P31/Fuj and Kasumi-1 cell lines, with some level of separation also observed between vehicle- and Ku-794–treated samples of both cell lines and between vehicle- and PI-103–treated samples of P31/Fuj cells (fig. S7).

KSEA of the phosphoproteomes of cells treated with the inhibitors (Fig. 2) showed a decrease in the abundances of the p70S6K, p90RSK, PKC, mTOR, Akt, and cyclin-dependent kinase (CDK) substrate groups upon inhibitor treatment (Fig. 2, A and C), consistent with these kinases being regulated by PI3K, mTOR, or both (31, 4143) and with our Western blotting analysis (fig. S5A). The changes in phosphorylation, as measured by KSEA, were inhibitor dose–dependent (Fig. 2C), indicating that the information on kinase activity provided by KSEA was at least semiquantitative. Treatment of cells with either AZ123 or Ku-794 induced a marked increase in the abundances of phosphorylated substrates of DNA-PK (Fig. 2, A and C). The analysis also showed that although the peptide motifs whose phosphorylation was reduced upon treatment with AZ123 or Ku-794 included those with basic sequences (Fig. 2, B and C, and fig. S8), the extent of phosphorylation of the SQ motif, which is recognized by DNA-PK and related kinases (44), increased (Fig. 2, B and D, and fig. S8B). Specifically, 132 phosphopeptides containing the SQ motif were present in our phosphoproteomics data. Of these, the abundances of 11 and 25 SQ motif–containing phosphorylated peptides were statistically significantly increased in P31/Fuj and Kasumi-1 cells, respectively, when cells were treated with AZ123. The extent of phosphorylation of the SQ motif was particularly increased by treatment with AZ123 or Ku-794 in Kasumi-1 cells (Fig. 2D). The phosphorylation of SQ motifs was mostly decreased after treatment of P31/Fuj cells with PI-103 (Fig. 2D and fig. S8B), consistent with the inhibition of DNA-PK by PI-103 as an off-target effect (45). Together, the KSEA data indicated that the activation of DNA-PK was an unexpected consequence of inhibiting PI3K and mTOR signaling in cells.

Fig. 2

KSEA results are consistent with the expected outcome of inhibiting PI3K and mTOR signaling and reveal the activation of DNA-PK downstream of PI3K. P31/Fuj and Kasumi-1 cells (six biological replicates per condition) were treated with vehicle (DMSO), AZ123 (0.1 or 1 μM), Ku-794 (0.1 or 1 μM), or PI-103 (1 μM) for 2 hours. Phosphorylated peptides were enriched with TiO2 and identified and quantified by LC-MS/MS. Average fold changes in the abundances of phosphorylated peptides in inhibitor-treated cells compared to those in DMSO-treated cells (control) in the six replicates and the statistical significance of these changes were calculated. These data were then used for KSEA analysis. (A) Heatmap showing the enrichment of substrate groups for the different kinases calculated by the KSEA algorithm with the PhosphoSite database and the “Fold” method of calculating enrichment. The extent of enrichment was calculated as the abundance of a phosphorylated peptide in the inhibitor-treated samples divided by its abundance in the DMSO-treated cells. m, number of substrates in the indicated kinase substrate group. (B) Heatmap showing enrichment of substrate groups based on common phosphorylation motifs. (C) Examples of substrate group enrichment as a function of treatment with PI3K and mTOR inhibitors. (D) As described for (C), but substrate groups were defined by their phosphorylation motif rather than by their upstream kinase. In each case, enrichment was calculated with the Fold method, that is, by averaging the normalized abundances of the phosphorylated peptides in the indicated substrate groups that were statistically significantly changed in the inhibitor-treated cells compared to those in control cells.

To validate these conclusions, we performed follow-up experiments with different molecular biology and biochemical techniques, which indicated that DNA-PK was activated as a result of the apoptotic process induced by inhibiting the PI3K-mTOR pathway (Supplementary Text and fig. S9). The effect of the DNA-PK pathway activity on the phosphoproteome of AML cells during the apoptotic process that we observed (fig. S9) is consistent with DNA-PK regulating different biological processes during apoptosis. These include DNA fragmentation and the nonhomologous end joining repair (NHEJR) system (46), both of which are implicated in the appearance of leukemogenic translocations in cells exposed to chemotherapy agents (47, 48).

Identification of substrate groups downstream of PI3K and mTOR

In addition to monitoring phosphorylation sites on predefined substrate groups, we also identified several other phosphorylation sites specifically affected (either inhibited or enhanced) as a result of treating cells with the PI3K and mTOR inhibitors (Fig. 3 and fig. S10). PI3K and mTOR inhibitors had a greater effect on the phosphoproteome of P31/Fuj cells than on that of Kasumi-1 cells (Fig. 3), consistent with this pathway being more active in the former cell line (Fig. 1 and fig. S5A). Sites dephosphorylated by one PI3K or mTOR inhibitor but not by the other (fig. S10) were regarded as being examples of off-target effects, whereas sites dephosphorylated by both PI-103 and AZ123 were regarded as being specifically downstream of the PI3K-mTOR pathway (Fig. 3). Sites regulated by different inhibitors were further classified into those dephosphorylated by the PI3K and mTOR inhibitors but not by the mTOR inhibitor (group 1), those whose phosphorylation increased in the presence of these inhibitors (group 2), and those affected by all three inhibitors (groups 3 and 4). We propose that phosphorylation sites in substrates of groups 1 and 2 (Fig. 3, A and C) are preferentially downstream of PI3K, but not mTOR, whereas sites in substrates of groups 3 and 4 (Fig. 3, B and D) are downstream of both PI3K and mTOR. We then applied KSEA on P31/Fuj and Kasumi-1 cell lines, taking these new substrate groups as the basis for the analysis. The results show that substrates of groups 1 and 3 were more abundant in P31/Fuj cells than in Kasumi-1 cells, whereas group 4 substrates were more abundant in Kasumi-1 cells (Fig. 3E), thus confirming our earlier data (Fig. 1) showing that the PI3K pathway is more active in P31/Fuj cells, and indicating that these newly defined substrate groups are readouts of the activity of the pathway.

Fig. 3

Definition of newly uncharacterized substrate groups as markers of activation of PI3K, mTOR, or both. The phosphoproteomes of P31/Fuj and Kasumi-1 cells (six replicates per condition) were treated and analyzed as outlined in Fig. 2. (A) Phosphorylation of peptides affected by the dual PI3K and mTOR inhibitors AZ123 and PI-103, but not by the mTOR-specific inhibitor KU-794, in P31/Fuj cells (P < 0.05 by one inhibitor and P < 0.1 by the other). (B) Phosphorylation of peptides affected by all inhibitors in P31/Fuj cells (P < 0.05 by one inhibitor and P < 0.1 by the others). (C) Phosphorylation of peptides affected by the dual PI3K and mTOR inhibitors AZ123 and PI-103, but not by the mTOR-specific inhibitor KU-794, in Kasumi-1 cells. (D) Phosphorylation of peptides affected by all inhibitors in Kasumi-1 cells. The phosphorylated peptides listed in (A) and (C) were regarded as having PI3K-dependent and mTOR-independent phosphorylation sites and were classified as group 1 or group 2 depending on whether they were decreased or increased in abundance, respectively, in response to inhibitor treatment. The phosphorylated peptides listed in (B) and (D) were regarded as having PI3K- and mTOR-dependent sites and were classified as group 3 or group 4. (E) The newly defined groups were incorporated into the KSEA algorithm, and their enrichment values were calculated for the P31/Fuj and Kasumi-1 cell lines.

Application of KSEA to identify kinase pathways in primary AML blasts

The substrate groups that we found with our chemical biological approach (Fig. 3) represent markers of PI3K-mTOR pathway activation in AML. To test whether the phosphorylation of these substrate groups could explain the sensitivity of cells to pathway inhibition, we incorporated them into a new database for KSEA. Values obtained by KSEA were then correlated with the data on the survival of primary AML cells (details of the samples can be seen in table S1) after exposure to the inhibitors (Fig. 4). As a control experiment and to assess whether previously frozen primary cells were metabolically active after thawing, we treated cells with the tyrosine kinase inhibitor sodium pervanadate before performing phosphoproteomic LC-MS/MS analysis. The results indicated that a functional kinase signaling network was preserved in essentially all primary AML cells after the freezing and thawing process (Data sets 2 and 3 and fig. S11).

Fig. 4

Application of KSEA to identify substrate groups differentially enriched in primary AML cells with different sensitivities to PI3K inhibition and to generate models to predict cell sensitivity. Frozen primary AML cells were thawed, cultured for 30 min, and then harvested. Phosphorylated peptides were enriched with TiO2 and identified and quantified by LC-MS/MS. Basal extents of phosphorylation were correlated with cell viability data after exposure of the same primary samples to AZ123 for 24 hours. Phosphorylated peptides whose abundance correlated with cell viability were used for KSEA. (A) Cell survival after exposure to 1 μM AZ123 for 24 hours was measured by MTS [3-(4,5-dimethyl-thiazol-2yl)-5-(3-carboxymethoxyphenyl)-2-(4-sulfophenyl)-2H-tetrazolium] assay in two groups of primary AML cell samples that were considered as training and testing sets. (B) Heatmap showing the enrichment values calculated by KSEA for the PI3K- and mTOR-dependent substrate groups defined in Fig. 3 and their correlation with the cell viability for the training set; R, Pearson’s correlation between substrate group enrichment and viability; m, number of phosphorylated peptides in the named substrate group; observed sensitivity, percentage viability of cells exposed to AZ123 [values are as in (A)]. (C) Heatmap showing the enrichment values calculated by KSEA for the substrate groups defined by the indicated kinase or tyrosine motif and their correlation with cell viability for the training set. (D) Heatmap of the enrichment values calculated by KSEA for the substrate groups defined by the indicated motifs and their correlation with cell viability for the training set. (E) Mathematical model to predict cell sensitivity to a PI3K and mTOR inhibitor based on substrate group enrichment data. The enrichment values shown in the top and middle heatmaps of (C) were separately summed for each sample. The ratio between both sums was calculated and correlated with the cell viability data shown in (A). (F and G) The regression equation obtained from (E) was rearranged as Predicted Sensitivity = (y + 4.8857)/0.1093, where y is the ratio of kinase substrate groups shown in (C). Predicted sensitivity was calculated in (F) the AML primary sample set (training set) and (G) an independent set of primary AML cells (the testing set) and plotted against observed cell sensitivity. (H) Mathematical model used to predict cell sensitivity to a PI3K and mTOR inhibitor based on the substrate group enrichment data shown in (D). (I and J) The regression equation obtained from (H) was rearranged as Predicted Sensitivity = (y + 2.2076)/0.0832, where y is the ratio of kinase substrate groups shown in (D). Predicted sensitivity was calculated in (I) the AML primary sample set (training set) and (J) an independent set of primary AML cells (the testing set).

We measured cell viability in two sets of AML primary blasts treated with 1 μM AZ123 (Fig. 4A). Phosphoproteome quantification followed by KSEA showed that group 1 substrates, which are downstream of PI3K (Fig. 3), were enriched in cells sensitive to AZ123 (R = −0.70; Fig. 4B), suggesting that the PI3K pathway may be more active in AZ123-sensitive cells. However, the difference in enrichment for this group between the most sensitive and the most resistant cells was less than twofold. In addition to the newly defined substrate groups (Fig. 3), we also monitored the enrichment of substrate groups taken from PhosphoSite, Phospho.ELM, and PhosphoPOINT alongside our collection of phosphorylation motifs (Fig. 4, C and D, and Data set 3). We found that substrate groups phosphorylated by CDC7 (cell division cycle 7), PDK1 (phosphoinositide-dependent protein kinase 1), and ERK1 were enriched in the AZ123-resistant cells (Fig. 4C), whereas substrates of CDK1 and of several tyrosine kinases, such as Src, Lck, and Abl, were increased in the AZ123-sensitive cells (Fig. 4C). This is consistent with the data regarding tyrosine motif enrichment; indeed, the average abundances of 99 peptides phosphorylated at tyrosines also had a negative correlation with cell viability (Fig. 4C). These data indicate that although markers of activation of the PI3K-mTOR pathway and substrates of tyrosine kinases were more enriched in cells sensitive to PI3K inhibition, resistant cells seemed to activate pathways parallel to PI3K-mTOR, including ERK1 and CDC7. KSEA based on phosphorylation motifs showed that the phosphorylation on several acidic motifs positively correlated with cell viability (Fig. 4D), whereas the enrichment of several basic and SP motifs (Fig. 4D) was negatively correlated.

Because KSEA values (the fold enrichment of the substrate groups) correlated with the sensitivity of a cell to a PI3K and mTOR inhibitor (Fig. 4, C and D), we hypothesized that these values might be used to predict the responses of a cell to an inhibitor. We generated a model based on linear regression by correlating the ratios of substrate group enrichment (Fig. 4C) with cell viability data (Fig. 4E). The regression function was rearranged and used to calculate a predicted viability (Fig. 4F and table S2). As measures of the accuracy of the model, we calculated the root mean square deviation (RMSD), a parameter that measures how well the predicted values fit with the observed values (49), and the Pearson’s correlation between observed and predicted sensitivity. The RMSD was 6.9 (table S2), and the Pearson’s correlation coefficient was 0.89 (Fig. 4F). Once the model was validated in this initial training set, we applied it to a new group of primary AML cells that we considered as the testing set (Fig. 4G). The RMSD between the predicted and observed viabilities was 9.2 for this testing set (table S2), and the correlation coefficient was 0.89 (Fig. 4G). We then applied the same modeling algorithms to the motif group enrichment data (Fig. 4D). Modeling based on the regression equation generated in this model (Fig. 4H) predicted viability after drug treatment (table S2) with good accuracy and linearity in both the training (RMSD = 7.6, R = 0.92; table S2 and Fig. 4I) and testing (RMSD = 7.3, R = 0.95; table S2 and Fig. 4J) sets.

In addition to phosphorylated peptides in specific substrate groups, other peptides showed a correlation with cell survival as a function of exposure to AZ123 (Fig. 5, A and B). We incorporated these phosphorylated peptides into two newly defined groups that were used as the basis of KSEA. As for our earlier analysis (Fig. 4), modeling based on the ratio between the enrichment values of these phosphorylated peptide groups (Fig. 5C, left panel) predicted sensitivity to the PI3K inhibitor AZ123 in two independent sets of primary AML cells. RMSD values obtained for the training and testing sets (15.3 and 12.0, respectively) were similar but slightly higher than those obtained by the analysis based on kinase and motif substrate groups (table S3). Together, these data (Figs. 4 and 5) illustrate the application of KSEA data to generate mathematical models that can be used to predict the responses of primary cancer cells to a particular drug.

Fig. 5

Definition of newly characterized KSEA groups of phosphorylated peptides that correlate with the sensitivity of primary AML cells to a PI3K and mTOR inhibitor and their application to generate models that predict sensitivity. The phosphoproteomics and cell viability data in Fig. 4 were mined to identify individual phosphorylated peptides that showed the greatest positive and negative associations with cell viability. These were incorporated into new groups for KSEA. (A) Heatmap showing phosphorylated peptide abundances and cell viability for the phosphorylated peptides included in the group of positive correlation. (B) Heatmap showing the correlation of phosphorylated peptide abundance and cell viability for the phosphorylated peptides included in the group of negative correlation. (C) Mathematical model to predict cell sensitivity to a PI3K and mTOR inhibitor based on the newly characterized groups of phosphorylated peptide. Enrichment values for the new correlated substrate groups defined in (A) and (B) were calculated with KSEA. Ratio values between both groups were calculated for each sample in the training test and correlated with cell viability data (left panel). Linear regression was used to predict cell viability data as a function of ratio values in the training and testing sets as shown in Fig. 4. Predicted and observed cell viability data were correlated for the training and testing sets (middle and right panels, respectively).

Heterogeneity of substrate group enrichment across primary AMLs

To further characterize kinase signaling heterogeneity in primary AML, we measured substrate group enrichment across blasts from 20 different AML patients relative to untransformed peripheral blood cells from five donors. We pooled the enrichment data based on five different sources of kinase-substrate relationships, namely, Phospho.ELM, PhosphoPOINT, PhosphoSite, our collection of phosphorylation motifs, and the new PI3K- and mTOR-dependent substrate groups that we defined (Fig. 3).

To investigate the relationship between substrate groups taken from the different databases of kinase-substrate relationships and motifs, we classified these groups by unsupervised hierarchical clustering of the enrichment analysis (of basal, untreated phosphoproteomes) after normalizing the data across all tumor samples. The results showed 16 main clusters (at a cutoff height of 1 in the cluster dendrogram; fig. S12A). Although the kinase-substrate databases use inconsistent nomenclature to name protein kinases, close inspection of the substrate enrichment clusters indicated that the enrichment data were generally consistent across primary AMLs irrespective of the source of the substrate groups used for KSEA. For example, the substrate groups named CK2-A1, CK2_alpha, and CSNK2A1 in Phospho.ELM, PhosphoPOINT, and PhosphoSite, respectively, all of which consist of substrates of CK2α, grouped in clusters 1 to 3 together with substrates of other CK2 isoforms and with acidic motifs (the preferred substrate motifs for CKs) (fig. S13). Cluster 5 contained substrates of Abl (PhosphoSite), Abl (Phospho.ELM), and ABL-25 (PhosphoPOINT), all of which correspond to the tyrosine kinase Abl and were grouped with substrates of the tyrosine kinase Btk and with the tyrosine phosphorylation motif. The enrichment of several substrate groups defined by basic motifs clustered with those defined by protein kinase C (cluster 11), consistent with the known substrate specificities of PKC.

KSEA of basal (untreated) primary AML showed that the substrate groups most frequently enriched in primary AML were those for CK2, CDKs, and PAKs, which were statistically significantly increased in 40 to 80% of the tumors (Fig. 6, top panels, and Data set 3). We also observed a general enrichment in the phosphorylation of substrates of other kinases, including cyclic adenosine monophosphate–dependent protein kinase (PKA), PKC, mitogen-activated protein kinases (MAPKs), and tyrosine kinases, although those increases were not statistically significant in most cases (Fig. 6, middle panels). The group of phosphorylated peptides linked to PI3K activity (Fig. 3, group 1) was statistically significantly enriched in 55% of the tumors relative to untransformed cells (Fig. 6, bottom panels), thus indicating that PI3K signaling may be particularly active in these AML cases.

Fig. 6

Kinase substrate groups most frequently enriched in primary AML cells. Primary cells from 20 AML patients and GMPB (granulocyte colony-stimulating factor–mobilized peripheral blood) cells from five healthy donors were thawed, cultured for 30 min, and harvested. Phosphopeptides were enriched with TiO2 and identified and quantified by LC-MS/MS. Average fold differences in the abundances of phosphorylated peptides between cells from AML patients and those of healthy donors and the statistical significance of these changes were calculated. These data were then used for KSEA analysis. Left panels show heatmaps indicating the extent of enrichment (calculated as the fold increase over the average value in the GMPB samples) for the substrate groups defined by the indicated upstream kinases for each AML sample. Right panels show heatmaps indicating the statistical significance of the enrichment for each sample at each kinase substrate group. m, number of substrates in the substrate group for the indicated kinase; n, number of cases with statistically significant enrichment of the group for the indicated kinase; percentage refers to the proportion of AML cases with statistically significant enrichment of kinase substrates.

Discussion

Cancers are caused by genetic mutations and cytogenetic alterations that affect gene expression (911) and signaling pathway activities, but it is often difficult to uncover the functional consequences that these genetic changes have on the activity of the biochemical pathways that drive the cancer phenotype (12). The activities of kinases are linked to the wiring of signaling networks (13), which may be different in different cell populations and for which our knowledge is still limited, which hampers the prediction of how genetic alterations may affect kinase pathway activity.

Previous attempts to derive information about kinase activities from phosphoproteomics data involved mining the known specificities of kinases for the phosphorylation of substrates in the context of linear motif sequences (50, 51). This work led to the development of several tools to link phosphorylation data to upstream kinases based on phosphorylation motifs (50, 51). Here, we investigated a systematic approach to infer kinase pathway activation based on values of substrate group enrichment obtained from previous knowledge of kinase-substrate relationships. We found that the information obtained with this approach was a reflection of kinase pathway activity, as it agreed with data obtained by targeted immunochemical methods (Fig. 1) and was consistent with the results expected by inhibiting PI3K signaling (Fig. 2). The results from our substrate-enrichment approach were also consistent with each other regardless of the databases of kinase-substrate relationships or statistical tests that we used. However, the comprehensiveness and fidelity of databases of kinase-substrate relationships that are currently available limit the analytical depth by which KSEA can derive information about kinase activation. To address this limitation, we also used phosphorylation motifs as the basis of KSEA. Both approaches were complementary in this study and pointed to the activation of DNA-PK as a result of inhibiting PI3K (Fig. 2), a hypothesis that was later validated by other methods that demonstrated that DNA-PK was activated as a result of the apoptotic process triggered by inhibition of PI3K and mTOR (fig. S9). An additional means by which the limitation on knowledge of kinase-substrate relationships may be addressed involves defining new substrate groups consisting of phosphopeptides readily measurable by MS and being bona fide readouts of pathway activities. We demonstrated this here with the identification of substrates for kinases downstream of PI3K and mTOR (Fig. 3). Although these substrate groups may not be linked to single kinases, collectively these provide readouts of activities linked to specific pathways. In addition, chemical genetic approaches (52, 53) could also be used in the future to extend databases of kinase-substrate relationships.

Consistent with the dependency of cancer cells on kinase pathway activation, we found that kinase substrate groups were in general more enriched in AML primary blasts than in untransformed cells. In line with the frequent activation of the PI3K pathway in AML (54), we also observed a statistically significant enrichment of phosphorylated substrate groups downstream of PI3K in 55% of cases. In addition, our analysis revealed that phosphorylated substrates of CK2α were significantly enriched in ~40% of the tumors (Fig. 6). This percentage is similar to the 33% of AML cases that were found to have a high abundance of this kinase (55). Phosphorylated substrates of the kinases CDK and PAK were also significantly enriched in at least 40% of the AML cases that we studied; however, although these kinases are important for the biology of several cancer types (56, 57), their role in AML is not well understood.

Because susceptibility to therapies that target kinase signaling may depend on the combined activation of different pathways within the network, we hypothesized that cells showing distinct responses to a given inhibitor would exhibit enrichment of different kinase-substrate groups. We therefore assessed whether KSEA could be used to identify kinase pathways enriched in AML primary blasts that show different responses to the PI3K-mTOR inhibitor AZ123. As a result, we found that cells that were resistant to AZ123 had increased amounts of phosphorylated substrates of ERK1, 3-phosphoinositide–dependent protein kinase-1 (PDPK1), and CDC7, whereas those cells that were sensitive to the inhibitor had increased amounts of phosphorylated substrates of CDK1 and tyrosine kinases such as Src and Abl. We confirmed the relationship between the abundances of these substrate groups and cell viability with a model based on linear regression, which showed that it was possible to predict the responses of two independent sets of primary AML cells to the PI3K and mTOR inhibitor AZ123 with good accuracy. These data illustrate the use of KSEA to generate input data for mathematical models that can be used to predict the responses of cancer cells to drugs that target kinase signaling. Our results also indicate that measuring the activity of the pathway being targeted and the activities of parallel pathways may be beneficial in predicting responses to inhibitors that target kinase signaling.

Materials and Methods

Reagents

The P31/Fuj (JCRB 0091) and Kasumi-1 (JCRB1003) cell lines were obtained from the Japanese Collection of Research Bioresources. AZ12321046-011 (AZ123) and Ku-006794 (Ku-794) were provided by S.C.C. (Astra Zeneca).

Cell culture

P31/Fuj and Kasumi-1 cells were grown in RPMI 1640 supplemented with 10% fetal bovine serum, penicillin and streptomycin (each at 100 U/ml), and 50 μM β-mercaptoethanol at 37°C in a humidified atmosphere at 5% CO2. Cells were maintained at a confluency of 0.5 × 106 to 2 × 106 cells/ml.

Cell lysis and protein digestion

Cells were split at a confluency of 0.5 × 106 cells/ml, and after 24 hours of recovery, the cells were treated with vehicle (DMSO) or with 0.1 or 1.0 μM AZ123, Ku-794, or PI-103 for 2 hours. For each condition, three independent biological replicates were performed for each experiment, and all experiments were performed twice to give a total of at least six biological replicates for each condition. For each replicate, 10 × 106 cells were harvested by centrifugation, washed twice with cold phosphate-buffered saline supplemented with 1 mM Na3VO4 and 1 mM NaF, and lysed in 1 ml of urea buffer [8 M urea in 20 mM Hepes (pH 8.0), supplemented with 1 mM Na3VO4, 1 mM NaF, 1 mM Na4P2O7, 1 mM β-glycerophosphate, and 1 μM okadaic acid]. Cell lysates were further homogenized by sonication (with three 15-s pulses), and insoluble material was removed by centrifugation. Protein was quantified by the Bradford assay. For each replicate, 0.5 mg of protein was reduced and alkylated by sequential incubation with 4.1 mM dithiothreitol and 8.3 mM iodoacetamide for 15 min at room temperature in the dark. For protein digestion, the urea concentration was reduced to 2 M by the addition of 20 mM Hepes (pH 8.0). Immobilized tosyl-lysine chloromethyl ketone (TLCK)–trypsin [20 p-toluenesulfonyl-l-arginine methyl ester (TAME) units/mg] was then added, and samples were incubated overnight at 37°C. Digestion was stopped by adding a final concentration of 1% trifluoroacetic acid (TFA), and trypsin beads were removed by centrifugation. The resultant peptide solutions were desalted with C18-Oasis cartridges as indicated by the manufacturer with slight modifications. Briefly, Oasis cartridges were conditioned with 1 ml of acetonitrile (ACN) and equilibrated with 1.5 ml of wash solution (0.1% TFA, 2% ACN). Peptides were loaded in the cartridges and washed with 1 ml of wash solution. Finally, peptides were eluted with 0.5 ml of glycolic acid buffer (1 M glycolic acid, 5% TFA, 80% ACN).

Phosphorylated peptide enrichment

Enrichment of phosphorylated peptides was performed with TiO2 as previously described (27). Briefly, peptide eluents were normalized to 1 ml with glycolic acid buffer and incubated with 25 μl of TiO2 buffer (a 50% slurry in 1% TFA) for 5 min at room temperature. TiO2 beads were packed by centrifugation in C18 spin columns previously equilibrated with glycolic acid buffer. The columns were sequentially washed with 300 μl of glycolic acid, 50% ACN, and ammonium bicarbonate buffer [20 mM NH4HCO3 (pH 6.8) in 50% ACN]. For phosphopeptide elution, beads were incubated for 1 min at room temperature with 50 μl of 5% NH4OH in 50% ACN and centrifuged. This step was repeated three times. Eluents from the same sample were pooled and acidified with formic acid to a final concentration of 10%. Finally, samples were dried with a SpeedVac, and pellets were stored at −80°C.

LC-MS/MS analysis

LC-MS/MS was performed as previously described (17). Briefly, phosphopeptide pellets were resuspended in 20 μl of 0.1% TFA, and 4 μl was loaded into an LC-MS/MS system, which consists of a nanoflow ultrahigh pressure liquid chromatography (UPLC, nanoAccuity, Waters) coupled online to an Orbitrap XL mass spectrometer (Thermo Fisher Scientific). The top five most intense multiply charged ions were selected for collision-induced dissociation fragmentation in multistage activation mode. The resolution of MS1 was set to 60,000.

Data processing and statistical analysis

Peptide identification was performed by matching of the MS/MS data to the SwissProt database (downloaded March 2011) restricted to human entries (22,000 protein entries) with the Mascot search engine (58). Mass tolerances were set to 5 parts per million (ppm) and 600 millimass units for parent and fragment ions, respectively. Allowed variable modifications were phosphorylation on Ser, Thr, and Tyr; PyroGlu on N-terminal glutamine; and oxidation of methionine. Phosphopeptides with a Mascot expectancy of <0.05 (~2% false discovery rate) were included in a database of sites quantifiable by MS. Pescal software (17, 59) was then used to obtain peak heights of extracted ion chromatograms of phosphorylated peptides in this database across all the samples being compared. Pescal aligned retention times using those of peptides common to all samples as reference points along chromatograms. The extracted ion chromatogram (XIC) windows were 7 ppm and 2 min. XIC intensity values were normalized to the sum of all values in a sample and to the mean of phosphopeptide intensities across samples. The significance of the difference in the means of log2-transformed data across samples was assessed by Student’s t test followed by Benjamini Hochberg multiple testing correction.

Kinase-substrate enrichment analysis

Phosphopeptides with P < 0.05 (as assessed by t test analysis of log2-transformed data) were grouped into substrate sets. The common feature of phosphopeptides in these substrate groups was that they had sites known to be phosphorylated by a specific kinase or that the phosphorylated residue was present in the context of predefined phosphorylation motifs. The information on kinase-substrate relationships was obtained from publically available databases, namely, PhosphoSite (20), Phospho.ELM (21), and PhosphoPOINT (18), whereas the list of motifs was obtained from the literature and from an analysis of our data set with Motif-X (51). Three approaches were used to infer global differences in the abundance of substrate groups across samples. The first one involved counting the number of phosphopeptides in the substrate group whose abundances increased or decreased relative to those in control cells. The enrichment of kinase substrates was then calculated by a parameter termed “delta counts,” which was defined as the number of phosphorylated peptides in a substrate group that statistically significantly increased their abundance in inhibitor-treated cells relative to those in control cells minus the number of phosphorylated peptides whose abundance decreased. The advantage of this approach is that it does not involve mathematical division, and therefore, it is applicable to situations in which the number of substrates for a particular substrate group is zero in some of the samples being compared. The statistical significance of enrichment was then calculated with the hypergeometric test followed by Benjamini Hochberg multiple testing correction. The second approach investigated here to assess enrichment of substrate groups involved comparing the means (arithmetic average) of the abundances of all of the phosphorylated peptides in a given substrate group in a sample relative to those of the control. For this method, the abundances of phosphorylated peptides were first normalized to those of the control and then subjected to log2 transformation so that these abundances were all expressed relative to those of the control. In addition, to derive a value of the fold difference of a given substrate group across control and test samples, we also calculated P values with paired t tests followed by Benjamini Hochberg multiple testing correction. The third approach involved calculating the ratio of the means of the phosphorylated peptide abundances in the substrate groups relative to their abundances in the whole data set; that is, enrichment = mS/mP, where mS is the log2 of the mean abundances of the substrate group and mP is the log2 of the mean abundances of the whole data set. The statistical significance of enrichment was then calculated as previously described (28) with a z score = (mS − mP) * m1/2/δ, where m is the size of the substrate group and δ is the SD of the mean abundances of the whole data set. The z score was then converted to a P value in Excel 2007 as previously described (28).

Bioinformatics and statistical analysis

A script was written in VBA (Microsoft Office 2007 version) to automate the application of KSEA algorithms outlined above. R (version 2.12.2) was used for hierarchical clustering and PCA. Tests of normality were performed in Prism v. 4.03.

Viability analysis

P31/Fuj and Kasumi-1 cells were seeded in 96-well plates at 10,000 cells per well. Twenty-four hours later, the cells were treated for 72 hours as indicated in the figure legends. Cell viability was then determined with the MTS assay. Each data point was assayed five times. Differences between treatments were subjected to t test analysis and considered statistically significant when P < 0.05.

Western blotting analysis

P31/Fuj and Kasumi-1 cells were lysed in base buffer [50 mM tris-HCl (pH 7.4), 150 mM NaCl, 1 mM EDTA, 1% Triton] supplemented with phophatase inhibitors (1 mM Na3VO4, 1 mM NaF, 2.5 mM Na4P2O7, 1 mM β-glycerol-phosphate) and protease inhibitors (protease inhibitor cocktail; Sigma-Aldrich). Seventy micrograms of protein was resolved by 6 or 10% SDS–polyacrylamide gel electrophoresis, as required, and transferred onto polyvinylidene difluoride membranes. Blocked membranes were incubated with primary and secondary antibodies and developed with SuperSignal West Pico Chemiluminescent Substrate (Thermo Scientific). Primary antibodies (see table S4) were used at 1:1000 dilution, whereas secondary antibodies were used at 1:5000 dilution.

Cell death assays

P31/Fuj and Kasumi-1 cells were seeded in six-well plates at 1 × 106 cells per well. Twenty-four hours later, cells were treated for 72 hours as indicated in the legends. Cells were harvested by centrifugation and stained with propidium iodide (PI) and annexin V (with the FITC Annexin V Apoptosis Detection kit), as indicated by the manufacturer. Fluorescence was measured in a BD LSRFortessa cell analyzer (Becton Dickinson), and data were analyzed with BD FACSDiva software v. 6.1.3 (Becton Dickinson). Cells were considered viable when they were negative for both annexin V and PI, whereas cells that were positive for annexin V and negative for PI were considered early apoptotic, cells that were positive for both annexin V and PI were considered late apoptotic, and cells that were negative for annexin V but were positive for PI were considered necrotic.

Primary AML cells

The characteristics of the AML patients are listed in table S1. All patients gave informed consent for the storage of their serum for research purposes. The study was conducted in accordance with the Local Research Ethics Committee, as described in a previous study (60). Primary AML blasts and GMPB cells were thawed following standard procedures. Cell viability was determined by trypan blue staining in a Vi-Cel XR Cell Viability Analyzer (Beckman Coulter). The samples were then split into two populations that were either treated with 1 mM sodium pervanadate or left untreated for 30 min. Samples were then harvested and processed for phosphoproteomic analysis.

Primary cell viability assays

Cell viability of primary cells after 24-hour treatment with the indicated compounds was determined with a Guava PCA cell analyzer (Guava Technologies Inc.) using the Guava ViaCount Reagent (Millipore) according to the manufacturer’s instructions.

Supplementary Materials

www.sciencesignaling.org/cgi/content/full/6/268/rs6/DC1

Supplementary Text

Fig. S1. Comparison of the phosphoproteomes of two AML cell lines.

Fig. S2. Examples of substrate group enrichment.

Fig. S3. Comparison of the different enrichment strategies.

Fig. S4. Results of KSEA based on substrate groups that were defined by a common motif.

Fig. S5. Characterization of the effects of PI3K inhibitors in P31/Fuj and Kasumi-1 cell lines.

Fig. S6. Overview of phosphopeptide quantification in P31/Fuj and Kasumi-1 cell lines treated with PI3K and mTOR inhibitors.

Fig. S7. PCA of phosphoproteomics data obtained from experiments in which P31/Fuj and Kasumi-1 cells were treated with AZ123, Ku-794, or PI-103.

Fig. S8. KSEA of the phosphoproteomes of inhibitor-treated cells based on substrate groups containing given phosphorylation motifs.

Fig. S9. DNA-PK is activated as a result of the initiation of apoptosis induced by PI3K and mTOR inhibitors.

Fig. S10. Differentiating between on-target and off-target effects of PI3K and mTOR inhibitors.

Fig. S11. Verification that the primary AML cells were metabolically active and had functional kinase signaling networks.

Fig. S12. Unsupervised hierarchical classification of primary AML cells based on enrichment data obtained with KSEA.

Fig. S13. Clusters of substrate groups found together in primary AML cells.

Table S1. Characteristics of AML primary samples.

Table S2. Results of linear regression model using the predefined substrate groups shown in Fig. 4.

Table S3. Results of linear regression model with the newly defined phosphorylated peptide groups shown in Fig. 5.

Table S4. Antibodies used for Western blotting experiments.

Data set 1. Phosphorylated peptides in cell lines.

Data set 2. Phosphorylated peptides in primary cells.

Data set 3. Details of KSEA in primary cells.

References and Notes

Acknowledgments: We thank A. Montoya for technical support, K. Ali for help with flow cytometry, and R. Petty and D. Taussig for providing primary cells. Funding: This work was funded by grants from the Medical Research Council (G0800914), BBSRC (BB/G0115023/1), Barts and The London Charity (297/298), and the Commission of the European Community (PIEF-GA-2009-254796). Author contributions: P.C. designed the study, performed research, analyzed data, prepared figures, and edited the paper; J.-C.R.-P. analyzed data; S.C.C. provided reagents and designed the study; S.G. provided reagents and designed the study; B.V. coordinated collaboration, provided reagents, and designed the study; S.J. designed the study and performed research; and P.R.C. conceived and wrote KSEA, designed the study, analyzed data, prepared figures, and wrote the paper. Competing interests: S.C.C. works for Novartis; S.J. works for Astra Zeneca; and B.V. and P.R.C. are consultants for Activiomics. Data availability: MS results and phosphorylated peptide spectra are available at the PRIDE database (http://www.ebi.ac.uk/pride): accession code PXD000185 and DOI: 10.6019/PXD000185.
View Abstract

Stay Connected to Science Signaling

Navigate This Article