Research ArticleCancer Drug Development

Pathway-Based Identification of Biomarkers for Targeted Therapeutics: Personalized Oncology with PI3K Pathway Inhibitors

See allHide authors and affiliations

Science Translational Medicine  04 Aug 2010:
Vol. 2, Issue 43, pp. 43ra55
DOI: 10.1126/scitranslmed.3001065

Abstract

Although we have made great progress in understanding the complex genetic alterations that underlie human cancer, it has proven difficult to identify which molecularly targeted therapeutics will benefit which patients. Drug-specific modulation of oncogenic signaling pathways in specific patient subpopulations can predict responsiveness to targeted therapy. Here, we report a pathway-based phosphoprofiling approach to identify and quantify clinically relevant, drug-specific biomarkers for phosphatidylinositol 3-kinase (PI3K) pathway inhibitors that target AKT, phosphoinositide-dependent kinase 1 (PDK1), and PI3K–mammalian target of rapamycin (mTOR). We quantified 375 nonredundant PI3K pathway–relevant phosphopeptides, all containing AKT, PDK1, or mitogen-activated protein kinase substrate recognition motifs. Of these phosphopeptides, 71 were drug-regulated, 11 of them by all three inhibitors. Drug-modulated phosphoproteins were enriched for involvement in cytoskeletal reorganization (filamin, stathmin, dynamin, PAK4, and PTPN14), vesicle transport (LARP1, VPS13D, and SLC20A1), and protein translation (S6RP and PRAS40). We then generated phosphospecific antibodies against selected, drug-regulated phosphorylation sites that would be suitable as biomarker tools for PI3K pathway inhibitors. As proof of concept, we show clinical translation feasibility for an antibody against phospho-PRAS40Thr246. Evaluation of binding of this antibody in human cancer cell lines, a PTEN (phosphatase and tensin homolog deleted from chromosome 10)–deficient mouse prostate tumor model, and triple-negative breast tumor tissues showed that phospho-PRAS40Thr246 positively correlates with PI3K pathway activation and predicts AKT inhibitor sensitivity. In contrast to phosphorylation of AKTThr308, the phospho-PRAS40Thr246 epitope is highly stable in tissue samples and thus is ideal for immunohistochemistry. In summary, our study illustrates a rational approach for discovery of drug-specific biomarkers toward development of patient-tailored treatments.

Introduction

The genetic alterations of human cancer often result in deregulation of signal transduction pathways that are sometimes directly implicated in the disease phenotype. This insight constitutes the main rationale for pathway context–based drug discovery, in which molecular-targeted therapeutics are developed against both mutant oncoproteins (for example, BCR-Abl) and wild-type effector molecules [for example, mammalian target of rapamycin (mTOR)] that reside upstream of or within deregulated oncogenic pathways (14). Nevertheless, despite the appealing nature of the concept of “oncopathway” addiction rather than oncogene addiction (5, 6), there has been only limited success in our treatment of major cancers. One key challenge for improving treatment outcomes remains the ability to predict which patient subpopulations will benefit from a given therapy (7). One can, at least in theory, approach this issue by matching molecular-targeted drugs with patients according to the makeup of their deregulated “addiction” pathways in the tumor. This redefinition of cancer by deregulated pathway defect rather than anatomical tissue of origin can enable a rational, disease pathway–based approach to drug-tailored biomarker discovery that can lead to personalized oncology management and clinical trial design (7, 8).

The phosphatidylinositol 3-kinase (PI3K) pathway is frequently activated in human tumors. Usually, this is caused by loss-of-function mutations in the tumor suppressor PTEN (phosphatase and tensin homolog deleted from chromosome 10) (9) or by activating gain-of-function mutations in the p110-α catalytic subunit of PI3K (10). Both of these events lead to increased AKT activity, which promotes tumor cell survival and metastasis (11). Growth factor binding to receptor tyrosine kinases (RTKs) also triggers AKT activation, and oncogenic RTK signaling frequently involves coactivation of the mitogen-activated protein kinase (MAPK) pathway (12). Several studies have concluded that cancer drugs such as Gleevec (imatinib), Tarceva (erlotinib), and Iressa (gefitinib), which target oncogenic protein tyrosine kinases, are most effective when causing a concomitant down-regulation of PI3K-MAPK signaling (1316). Accordingly, small-molecule inhibitors targeting intracellular components of the RAS-MAPK and PI3K-AKT-mTOR pathways are currently undergoing clinical development (7, 11). However, predicting the efficacy of these pathway-targeted agents in patients remains a challenge because the molecular mechanisms underlying response and the compensatory pathway alterations preventing response are largely unknown (3, 4, 17). Moreover, current clinical markers for measuring pathway activation are inadequate (7, 18), so there is an urgent need to correlate drug action with the modulation of signaling pathways to provide the basis for biomarker discovery and clinical assay development as a means to optimize dosing regimens and to stratify patients for effective individualized therapy.

The PI3K and MAPK pathways are governed primarily by serine-threonine protein phosphorylation and are subject to extensive cross-talk regulation (3, 17, 19). These signaling cascades are composed of both AGC kinases [exemplified by AKT, phosphoinositide-dependent kinase 1 (PDK1), S6 kinase (S6K), protein kinase C (PKC), and ribosomal S6 kinase (RSK)] and MAPK family members [exemplified by MAPK kinase (MEK) and extracellular signal–regulated kinase (ERK)]. Many of these enzymes phosphorylate their protein substrates in a context-specific manner depending on the amino acids flanking the phosphoacceptor site (2023). The motifs phosphorylated by AKT [RXRXX(s/t)] and MAPK [PX(s/t)P] kinases are classical examples of consensus kinase phosphorylation sites (20, 21). Phosphorylation of such motifs controls protein-protein interactions and signaling networks. For example, the PDK1-docking motif antibody recognizes the phosphorylated C-terminal hydrophobic motif (HM) found in most AGC kinases (24), which constitutes a binding site [(F/K)XX(F/Y)(s/t)(F/Y)] for PDK1 (23). This facilitates PDK1-mediated T-loop phosphorylation of the interacting kinase, with the notable exception of AKTThr308, which is phosphorylated by PDK1 in a pleckstrin homology domain–dependent manner (23). Accordingly, antibodies directed against AKT, PDK1, and MAPK phosphomotifs can be used to study phosphoproteins involved in PI3K and MAPK signaling.

To enable the generation of drug-specific biomarker tools for PI3K pathway inhibitors, we have applied a differential phosphoproteome approach to identify and quantify drug-regulated serine-threonine phosphorylation events. Although immunoaffinity enrichment of phosphoproteins has been used to study tyrosine phosphorylation–based signal transduction (25), and global analyses of proteome and phosphoproteome dynamics were recently reported (2629), proteomic studies of drug-regulated serine-threonine phosphorylation events have been lacking, partly as a result of technical challenges. Herein, by taking advantage of phosphomotif antibodies to enrich for PI3K and MAPK pathway components, we identified phosphorylation sites specifically regulated by small-molecule PI3K pathway inhibitors to serve as unique and shared biomarkers. This approach also allows the generation of phosphorylation state–specific antibodies potentially useful as drug-specific biomarkers to predict anticancer efficacy in the clinic.

Results

Experimental strategy identifies phosphoprotein changes in response to drug treatment

To analyze the effects of AKT and PDK inhibition in a relevant, deregulated pathway context, we performed quantitative mass spectrometry (MS) on immunoprecipitates obtained with antibodies to motifs ([PX(s/t)P] and [RXXs/t]) from the PTEN-null PC-3 prostatic cancer cell line, which displays PI3K pathway activation and depends on PTEN loss for anchorage-independent growth. SILAC (stable isotope labeling by amino acids in cell culture)–labeled PC-3 cells were treated with a PDK1 inhibitor (compound 2) (30); an allosteric, highly selective AKT inhibitor (compound 17) (31); or the dual PI3K-mTOR inhibitor PI-103 (32). Western blot analysis confirmed that drug treatment induced differential regulation of phosphoproteins as judged by the phosphorylation of p90RSK, AKT, p70S6K, and S6RP (Fig. 1A). Consistent with pharmacological and genetic studies, phospho-p90RSKSer221 was selectively modulated by the PDK1 inhibitor (33). In contrast, all three compounds modulated the phosphorylation of AKTThr308, p70S6KThr389, and S6RPSer235/236, albeit to varying degrees. Whereas the AKT and PI-103 inhibitors were highly effective at inhibiting phosphorylation of both AKTSer473 and AKTThr308, the PDK1 inhibitor did not modulate the AKTSer473 site, consistent with genetic PDK1 deletion studies in embryonic stem cells (33). The PI-103 compound completely inhibited phosphorylation of S6RPSer235/236 as expected from its potent mTOR-inhibitory activity. Altogether, we concluded that PC-3 cells were well-suited for further profiling studies to discover both shared and uniquely modulated phosphoprotein biomarkers for PI3K pathway–targeted drugs.

Fig. 1

Western blot analysis of SILAC-labeled PC-3 cells and overview of experimental design. (A) SILAC-labeled PC-3 cells were incubated for 18 hours with either vehicle (dimethyl sulfoxide, DMSO) or PI3K pathway inhibitors (5 μM) targeting PDK1 (PDK1i), AKT (AKTi), or PI3K-mTOR (PI-103) and analyzed by Western blot with the indicated phosphorylation state–specific antibodies. (B) Schematic of stable isotope labeling with amino acid in cell culture and IAP for MS analysis (SILAC-MS-IAP). PC-3 cells grown either in light (12C-Arg, 12C-Lys) or in heavy (13C-Arg, 13C-Lys) medium were lysed and combined at a 1:1 ratio. Lysates were split into two aliquots: trypsin digests for MAPK and PDK1 motif IAP, and endoproteinase Glu-C for AKT motif IAP. Eluates were analyzed by liquid chromatography–Fourier transform ion cyclotron resonance, and ions were quantified by the Elucidator software 3.1. To reduce the false-positive discovery rate, we repeated each experiment by a reverse labeling approach. (C) Forward and reverse labeling allows confident annotation of quantitatively changing peptides in response to drug treatment. Example of ion mass/charge ratio (m/z) 826.674 (light peptide isoform) and m/z 835.451 (heavy peptide isoform) at the selected retention time (107.6 to 108.4 min) and the corresponding area under the curve used to calculate the change.

Specifically, to enable a quantitative assessment of phosphoprotein changes induced by pharmacological inhibitors, we grew cells either in “light” medium that contains the radio-neutral form of the natural amino acids lysine (12C6-Lys) and arginine (12C6-Arg) or in “heavy” medium that is fully substituted with 13C6-Lys and 13C6-Arg before treatment with either vehicle control or inhibitor for 18 hours, respectively. To ensure that the measured phosphorylation changes were robust and reproducible, we repeated the experiment with the labeling reversed (Fig. 1, B and C). Using phosphomotif-specific antibodies, we enriched each treatment condition for peptides containing the AKT substrate [RXX(s/t)], PDK1-docking [(F/K)XX(F/Y)(s/t)(F/Y)], and MAPK substrate [PX(s/t)P] motif before MS profiling.

Electron transfer dissociation MS is essential for efficient sequencing of basophilic AKT substrates

Initially, we performed a data-dependent shotgun analysis of each immunoprecipitated sample to get a qualitative measure of the ability to derive amino acid sequence information from each isolated subpopulation of phosphopeptides. The analysis yielded 102, 158, 70, and 45 phosphorylated peptides that harbored the expected motifs [RXX(s/t)], [PX(s)P], [PX(t)P], and [(F/K)XX(F/Y)(s/t)(F/Y)], respectively (table S1). Peptides that were isolated by the AKT substrate motif antibody ([RXX(s/t)]) are inherently basophilic and thus were sequenced by both electron transfer dissociation (ETD) MS and collision-activated dissociation (CAD). As expected, CAD-acquired tandem mass spectrometry (MS/MS) spectra exhibited a strong neutral loss of the phosphate group and yielded poorly interpretable spectra, whereas ETD provided information-rich MS/MS spectra, including peptide sequences of known AKT substrates such as S6K, FOXO3, and Bad. Peptides isolated with the PDK1-docking motif antibody were readily identifiable by CAD MS/MS, and protein identifications were not significantly improved by ETD MS/MS. As expected, several RSK family members were identified in these eluants, confirming enrichment of PDK1 substrates on the basis of HM phosphorylation.

Seventy-one phosphoproteins are modulated by PI3K pathway inhibitors

Of the >500 unique phosphopeptides quantified in this study, 377 proteins were readily annotated into biological processes by the gene ontology format (Fig. 2A and table S2). To investigate whether the identified phosphoproteins exhibited a disproportional enrichment for any particular biological processes or pathways, we evaluated the potential enrichment for all categories with a hypergeometric test. Briefly, the test compares the proteins identified in this study with the background distribution of each category. The gene ontology distribution did not show a bias toward any protein function (Fig. 2A). However, performing the same hypergeometric analysis for annotated pathways (Panther) revealed significant enrichment for the PI3K-AKT, MAPK, and PTEN signaling pathways, as expected (Fig. 2B).

Fig. 2

High-level summary of SILAC-IAP-MS analyses. (A) Gene ontology analysis of 377 manually validated phosphopeptides containing the expected phosphorylated motifs (PX(s/t)P, RXX(s/t), or (F/K)XX(F/Y)(s/t)(F/Y). Only categories with 15 or more assigned proteins are shown. (B) Ingenuity analysis of these proteins for enriched pathways (right-tailed Fisher’s exact test). (C) Venn diagram showing the number of unique and shared phosphopeptides inhibited >1.4-fold. (D) The panels show the distribution of identified peptides that are modulated by each respective inhibitor. Binning at intervals of 1 (1.5 to 2, 2 to 3, 3 to 4, etc.).

Of the identified phosphopeptides, most showed a 1:1 ratio between control and treatment groups. However, 96 phosphorylated peptides showed >1.4-fold change in response to drug treatment in two independent experiments, representing 71 differentially phosphorylated proteins (Table 1). Specifically, we detected 36 peptides whose phosphorylation was regulated by the AKT inhibitor, 53 peptides by the PDK1 inhibitor, and 28 peptides by PI-103 (Table 1). The AKT and PDK1 compounds inhibited the phosphorylation of a common subset of 19 proteins (Fig. 2C). Among these phosphoproteins, 11 were also regulated by PI-103, suggesting that these phosphorylation sites are involved in core PI3K pathway signaling.

Table 1

Summary of phosphorylated proteins that show >1.4-fold reduction in response to PI3K pathway inhibitor treatment. ND, not determined. Abbreviations for the amino acids are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr.

View this table:

Bioinformatic analysis reveals PI3K pathway nodes that are modulated by inhibitors

We analyzed the fold change distribution for the AKT, PDK1, and PI-103 inhibitor–treated samples, looking at protein phosphorylation changes >1.4-fold in both the forward and the reverse labeling experiment (Fig. 2D). This analysis revealed that the phosphorylation of most regulated proteins was modulated two- to threefold, whereas a few proteins were more profoundly modulated. As indicated, among the identified phosphoproteins, phosphorylation of S6RP (RPS6) was most strongly regulated by PDK1, AKT, and PI3K inhibition (>50-fold for each of the three treatments). For AKT and PI-103 inhibitors, AKT1 substrate 1 (or PRAS40) stood out as the most highly regulated phosphoprotein. Its phosphorylation was regulated sixfold by these inhibitors compared to threefold with the PDK1 inhibitor. For the PDK1 inhibitor, the two most dynamically regulated phosphoproteins were p90RSK (RPS6KA6) (>10-fold) and PANK4 (10-fold). In general, the most significant changes were observed for known substrates of AKT (for example, PRAS40), PDK1 (for example, p90RSK), and mTOR (for example, RPS6). The MS analyses also confirmed the lack of modulation of phospho-p90RSKSer221 by inhibitors of AKT and PI3K-mTOR, consistent with the initial Western blot analysis (Fig. 1A). This result further validates phospho-p90RSKSer221 as a specific pharmacodynamic marker of PDK1 inhibition useful for lead optimization and drug development (34). Collectively, the above bioinformatic analyses validated the performance and robustness of the approach and support the use of immunoaffinity precipitation (IAP)–MS as a profiling approach to analyze compound-specific modulation of the PI3K-MAPK pathways.

Most regulated proteins are two-hop nodes away from the targeted protein kinases

To assess the extent to which the modulated peptides relate to known AKT, PDK1, or PI3K biology, we next mined the Ingenuity Pathway Analysis (knowledge network database) for one-hop neighborhoods for AKT, PDK1, and PI3K-mTOR, respectively. Additionally, we asked how closely the hit proteins relate to the PI3K pathway by connecting all known one-link protein-protein interactions. We found that 35 proteins map within two hops of the PI3K pathway (Fig. 3). The fact that we identified many distal nodes could be due to higher abundance or to signal amplification, because the signal is transmitted further downstream in the pathways from the inhibited kinase.

Fig. 3

Regulated phosphorylation networks of PC-3 cells in response to pharmacological inhibitors. Connectivity maps of phosphoproteins regulated by AKTi, PDK1i, and PI-103, respectively, projected on a two-hop network. The gene symbols for the corresponding drug-inhibited kinases are highlighted (PDPK1 encodes PDK1 and FRAP1 encodes mTOR). Phosphorylation changes >1.4-fold are highlighted in red. Proteins are defined by their gene symbols (nodes) and biological relationship by edges (lines). Node shapes are defined as follows: triangle down, kinase; triangle up, phosphatase; diamond (vertical), enzyme; hexagon, translation regulator; oval (vertical), transmembrane receptor; oval (horizontal), transcription regulator; circle, other.

PI3K pathway inhibitors regulate the cytoskeletal protein machinery

To define common signaling modules, we analyzed the phosphoproteins modulated by both the AKT and the PDK1 inhibitors. The most prominent protein modules inhibited by compound treatment related to the cytoskeletal machinery responsible for processes such as cell polarity (PAK4) and cytoskeletal reorganization (filamins, stathmin, dynamin, and PTPN14) (Fig. 4A). These cytoskeletal proteins were also modulated by the PI3K-mTOR inhibitor, as were several transcription factors (STAT1 and EIF4EBP1) and coactivators (IWS-1) with known relevance to cancer (Table 1). Other prominent modules regulated by the AKT and PDK1 inhibitors relate to vesicle transport (LARP1, VPS13D, and SLC20A1) and protein translation (RPS6 and PRAS40) (Fig. 4A). This is not unexpected because cancer cell motility, migration, and invasion have been shown to be affected by these compounds (3, 11). Nevertheless, the clear grouping of this phenotypic trait in the cell functional cluster analysis makes these specific drug-regulated phosphoproteins good candidate pharmacodynamic biomarkers for PI3K-PDK1-AKT-mTOR drug development. Accordingly, follow-up validation by Western blotting with commercially available phosphoantibodies against stathminSer38, PLCγSer1248, and MARK2Thr595 verified these phosphoproteins as specific, drug-regulated biomarkers (Fig. 4B).

Fig. 4

Simplified pathway diagram highlighting the clustering of dynamically regulated phosphoproteins in distinct cellular processes. (A) Phosphoproteins regulated by both AKT and PDK1 inhibitor show enrichment of specific cellular processes including cytoskeletal reorganization (blue), vesicle transport (gray), and protein translation (yellow), consistent with the established role of the PI3K pathway. (B) Follow-up Western blot analysis of PC-3 cells treated with indicated PI3K pathway inhibitors (18 hours) with four commercially available phosphospecific antibodies confirms the quantitative results of SILAC-IAP-MS. UO126 was used as a reference MEK inhibitor.

Unique, drug-regulated biomarkers reflect drug-specific responses

We identified 14 and 24 phosphorylation sites that were uniquely regulated by the AKT and PDK1 inhibitor, respectively. Of these, several have not previously been described as PI3K-regulated phosphorylation sites. Inherent to our approach, these markers reflect true drug-specific biological effects, including potential off-target activities. Accordingly, when we compared PTEN-null Jurkat cells treated with either the AKT or the PDK1 inhibitor by flow cytometry, we noted significant differences in the ability of these compounds to induce apoptosis on the basis of annexin V staining (Fig. 5). The PDK1 inhibitor showed a strong time- and dose-dependent induction of apoptosis compared to the AKT inhibitor. Intriguingly, six of the regulated phosphoproteins unique to the PDK1 inhibitor (EIF4EBP1, FLNA, LMNB1, NFKB1, PKM2, and STAT1) are linked to apoptosis (table S2) and may explain drug-specific biological effects, thereby providing a scientific rationale for a differentiated drug development paradigm.

Fig. 5

Differential cell death response to the PDK1 and AKT inhibitors in PTEN-deficient Jurkat cells. Cells were treated with the indicated inhibitors and harvested at various time points. Live cells were stained for propidium iodide (PI)/annexin V (BD Biosciences) according to the manufacturer’s instructions and analyzed on a FACSCalibur (BD Biosciences). (A) Representative fluorescence-activated cell sorting plot indicating intensity of PI (y axis) and annexin V staining (x axis) plotted for control or drug-treated (10 μM) Jurkat cells. (B) Relative percentage of PI/annexin double-negative (live), PI-negative/annexin V positive (early apoptotic), or PI/annexin V double-positive (apoptotic/necrotic) cells were estimated from the above experiment and graphed. Similar results were obtained in two independent experiments.

Phospho-PRAS40Thr246 correlates with up-regulated PI3K-AKT signaling and predicts AKT inhibitor sensitivity in lung and breast cancer cell lines

Our proteomic study guided the generation of phosphorylation state–specific antibodies as candidate biomarker tools for up-regulated pathway nodes sensitive to our targeted inhibitors. As proof of principle, one such custom antibody directed against phospho-PRAS40Thr246 was evaluated in a panel of 96 lung and 67 breast cancer cell lines by reverse-phase protein array (RPPA) analyses (35). We observed a strong positive correlation between phospho-AKTSer473 and phospho-PRAS40Thr246 for both the breast cancer (R = 0.57, P = 8 × 10−7) and the lung cancer (R = 0.44, P = 2 × 10−6) cell lines (Fig. 6A). This is comparable to the correlation observed for phospho-AKTSer473 to phospho-GSK3Ser9/21 in the same cohort of breast (R = 0.62, P = 1 × 10−9) and lung (R = 0.37, P = 2 × 10−3) cell lines. Other known AKT effector molecules, such as phospho-S6RPSer235/236 and phospho-FOXO1Ser256, showed a less statistically significant correlation with phospho-AKTSer473 when assayed by RPPA analysis (R = 0.05, P = 1 × 10−1 and R = 0.2, P = 7 × 10−2, respectively). No statistically significant correlation between phospho-PRAS40Thr246 and PTEN protein expression status was observed, in accordance with PTEN activity depending on both expression and its mutational status. Notably, high basal phospho-PRAS40Thr246 in breast cancer cell lines was predictive of antiproliferative effects of the AKT inhibitor in accordance with its identification as an AKT inhibitor–regulated phosphorylation node (Fig. 6B). In contrast, high basal phospho-MEKSer217/221 and phospho-ERKThr202/Tyr204 expression were negatively correlated with response to the AKT inhibitor (Fig. 6C and fig. S2), suggesting that these lines are more dependent on the RAS-MAPK pathway and therefore not expected to respond to PI3K pathway inhibitors (3, 17, 19).

Fig. 6

Evaluation of PRAS40 as biomarker of AKT activation and inhibitor sensitivity. (A) A panel of 67 breast (top) and 96 lung (bottom) cancer cell lines was assayed for phospho-AKTSer473, phospho-PRAS40Thr246, and PTEN protein expression by reverse-phase protein microarrays (35). A strong correlation between phospho-AKTSer473 (x axis) and phospho-PRAS40Thr246 (y axis) was observed (lung: Spearman R = 0.57, P = 8 × 10−7; breast: Spearman R = 0.44, P = 6 × 10−7). Other AKT substrates, such as phospho-AKTSer473 and phospho-GSK3SerS9/21, showed a similar strong correlation (Spearman R = 0.37, P = 2 × 10−3; Spearman R = 0.62, P = 1 × 10−9). No statistically significant correlation between phospho-PRAS40Thr246 and PTEN protein expression was observed. GSK3α/β, glycogen synthase kinase 3α/β. (B) Correlation of phospho-PRAS40Thr246 protein concentrations (y axis) with drug sensitivity (x axis). Lower numbers on the x axis indicate increased sensitivity (that is, normalized antiproliferative effects). (C) Correlation of basal phospho-MEKSer217/221 levels (y axis) with sensitivity to AKT inhibition (x axis).

Phospho-PRAS40Thr246 shows distinct subcellular localization compared to phospho-AKTSer473 and superior phosphoepitope stability

Next, we used a conditional, PTEN-deficient prostate tumor mouse model (PTENfl/lflPB-cre) to investigate the relationship between phospho-PRAS40Thr246 and PI3K pathway activation in tumor tissue. As a result of probasin-Cre–mediated homozygotic PTEN deletion, these mice develop prostatic intraepithelial neoplasia and invasive adenocarcinomas (36). Consecutive sections of mouse prostate tissues, either expressing PTEN or with PTEN deleted, were stained for phospho-AKTSer473 and phospho-PRAS40Thr246 (Fig. 7A). Both markers were markedly up-regulated in the PTEN-deficient tissues. Whereas phospho-AKTSer473 exhibits strong plasma membrane staining (36), phospho-PRAS40Thr246 shows a strong cytoplasmic staining. The signal intensity was qualitatively similar for both antibodies. Because phospho-AKTSer473 is prone to dephosphorylation during tissue processing (18), we next tested the stability of the PRAS40Thr246 phosphoepitope. In nonfixed tissue (room temperature), phospho-PRAS40Thr246 is stable up to 1 hour, unlike the labile phospho-AKTSer473 epitope (fig. S2). Together, these data show that phospho-PRAS40Thr246 is a biomarker for PI3K-AKT pathway–activated tumors that is sensitive to down-regulation by PI3K pathway inhibitors. Moreover, the robust signal in immunohistochemical applications is a clear advantage over phospho-AKTSer473, which is subject to feedback regulation and is prone to cause false-negative results in clinical tumor biopsies.

Fig. 7

Immunohistochemistry of PRAS40Thr246 in PTEN-deficient mouse prostate cancer model and triple-negative human breast cancers. (A) Immunohistochemical micrographs from serial sections of mouse prostate tissue showing activation of phospho-AKTSer473 and phospho-PRAS40Thr246 in response to homozygotic deletion of PTEN compared to PTEN wild-type (wt) normal tissue. (B) Quantitative immunohistochemistry (H score shown) of PTEN, phospho-AKTSer473, and phospho-PRAS40Thr246 expression in a panel of 12 triple-negative human breast tumor samples (representative samples shown). All micrographs were obtained at ×200 magnification. (C) Correlation of PRAS40Thr246 with phospho-AKTSer473 by H score (Pearson R = 0.817632) from breast cancer samples in (B).

Phospho-PRAS40Thr246 correlates with AKT activation but not with PTEN protein expression in human breast tumors

To further assess the utility of our phospho-PRAS40Thr246 biomarker for clinical studies, we investigated the concentration of phospho-PRAS40Thr246 in triple-negative human breast cancer samples and its relationship with phospho-AKTSer473 and PTEN protein expression (Fig. 7, B and C). These markers were assessed by immunohistochemistry (H scores on serial sections from the same tissue block) (Fig. 7B). We found that phospho-PRAS40Thr246 correlated positively with phospho-AKTSer473 (Fig. 7C). By contrast, PTEN protein expression (as measured by immunohistochemistry) did not correlate with phospho-PRAS40Thr246. This observation is similar to our RPPA analysis of >100 cell lines and underscores that PTEN protein expression is an inadequate marker of PI3K pathway activation, in part because PTEN is often mutated (18). Another possible explanation for the lack of correlation of phospho-PRAS40Thr246 with PTEN protein expression is that there are multiple means of activating the PI3K pathway in addition to PTEN loss of function. For example, in breast cancer, it is well known that RTKs can activate the PI3K-AKT pathway and thus contribute to the lack of correlation between PTEN protein expression and phospho-PRAS40Thr246. However, as expected, and consistent with the idea of oncopathway addiction, we found that PIK3CA mutations closely correlated with PRAS40 phosphorylation on Thr246 (fig. S3). Thus, we conclude that phospho-PRAS40Thr246 is a general marker of PI3K pathway activation and not exclusive to PTEN-deficient tumors.

Discussion

Here, we have demonstrated a pathway-based proteomic approach for the identification of both unique and shared dynamically regulated biomarkers for molecularly targeted cancer drugs. The method is compound-centric and based on observations of specific drug-induced signaling pathway modulations in tumor cells in culture, which provides for several immediate advantages. First, it enables a rational, unbiased deduction of the compound’s effects on oncogenic signaling pathways due to both on-targets and (potentially unknown) off-targets for the compound. Second, it renders the approach universally applicable to various classes of agents. Finally, in cases where pretreatment tumor biopsies can be obtained, biomarkers identified through this approach can have value in predicting drug efficacy. Appropriate compounds can be selected for patients with tumor profiles that match the specific, drug-inhibited pathways. Phosphospecific antibodies developed against key drug-regulated phosphorylation sites provide an ideal tool for this approach because they can be used for immunohistochemistry on tumor biopsies and potentially on other sources of tumor cells, including circulating tumor cells, lymph node metastases, or ascitic fluid.

Inhibition of deregulated oncogenic pathways in cancer cells will only be effective in cases where the tumor depends on such pathways for survival. Understanding this oncopathway addiction for various tumors is a major challenge, but there is emerging evidence of dominant, oncogenic pathways that are major drivers of the tumor phenotype in various cancer types (2, 3, 5, 6, 11, 37). Regardless, the direct linkage between the pathways that are modulated by the drug and the up-regulated, oncogenic pathways in a patient will provide a way to match the drug pathway profile with the patient tumor profile. This is an advance over more traditional approaches, including genetic analysis, in which pathway perturbations that result from mutations are inferred but not directly observed. Moreover, current drugs are provided to patients based only on known on-target effects, which are typically identified through in vitro assays and, hence, do not reflect all drug activities in vivo. The tool linking the drug pathway profile with the tumor pathway profile is the phosphoantibodies directed against the drug-regulated nodes. Hence, our approach could be useful for largely biomarker-driven clinical trials based on pretreatment biopsy sampling in amenable tumor types.

Specifically, here, we identified phosphoprotein biomarkers that are modulated by pharmacological inhibitors of AKT, PDK1, and PI3K-mTOR. To accurately quantify protein phosphorylation changes observed by MS, we used SILAC and imposed stringent criteria for calling hits (1.4-fold change, observed in at least two independent experiments). In contrast to traditional SILAC analyses, we applied a protein identification–independent analysis in which peptide pairs are first associated in high-resolution full-scan mass spectra and subsequently targeted, either by CAD for [PX(s/t)P]- and [(F/K)XX(F/Y)(s/t)(F/Y)]-containing peptides or by ETD for [RXX(s/t)]-containing peptides. Using our defined hit criteria, we quantified the changes in protein phosphorylation levels of 71 phosphoproteins, including key effector molecules of AKT (for example, PRAS40 and S6RP) and PDK1 (for example, PRAS40 and p90RSK). Using various knowledge databases (table S2), we were able to integrate the differentially regulated phosphorylation changes and reconstruct signaling modules that correlated well with nodes of the canonical PI3K pathway. Notably, most modulated phosphoproteins mapped to more distal nodes in the PI3K pathway located two hops downstream from the inhibited kinases (Fig. 3). These include proteins involved in cytoskeletal reorganization, motility, translation initiation, and tumor survival (Fig. 4A). Moreover, signaling nodes directly involved in cross talk between canonical pathways were identified (for example, S6RP), and proteins associated with either PI3K or MAPK pathway annotation in knowledge databases were often co-regulated by pharmacological AKT inhibition (Table 1), consistent with extensive pathway cross talk (3, 17, 19).

Several of the phosphoproteins identified in our study relate to cytoskeletal reorganization and are either newly or more recently described as PI3K pathway–regulated. One of these, filamin C (FLNc), originally identified as a physiological substrate of AKT (38), was inhibited 3.5-fold by the AKT inhibitor and fourfold by both the PDK1 and the PI3K-mTOR inhibitors. Accordingly, phosphorylation of FLNcSer2213 is severely impaired in PDK1−/− knockout tissues, whereas in PDK1 wild-type cells, it is increased in response to growth factor stimulation and inhibited by wortmannin (38). The filamin proteins (FLNa, FLNb, and FLNc) are thought to control cytoskeletal dynamics by stabilizing three-dimensional networks of actin and by acting as scaffolding molecules, tethering components of signaling pathways to the plasma membrane. Notably, we observed selective inhibition of phospho-FLNaSer2152 in response to the PDK1 inhibitor (4.3-fold) but not to the AKT inhibitor. This observation is consistent with phosho-FLNaSer2152 being a known phosphorylation site for p90RSK (39) and with its kinase activation loop (p90RSKSer221) being inhibited by the PDK1 inhibitor (~10-fold) (Table 1). It is unclear why the dual PI3K-mTOR inhibitor PI-103 also shows some, albeit modest, inhibition of the same FLNa phosphorylation site because this compound does not inhibit p90RSK (Fig. 1 and Table 1). Moreover, both the AKT and the PDK1 inhibitors regulated previously undescribed phosphorylation sites on FLNbThr2554 and PAK4Thr207, which we speculate are co-regulated events, analogous to the concerted role played by FLNa and PAK1 in controlling membrane ruffling and migration of melanoma cells (40). We also noted that the protein tyrosine phosphatase PTPN14, which is mutated in colorectal cancers (41), is regulated on phospho-PTPN14Ser486 in response to PDK1 and PI-103 inhibitor treatments. Because this enzyme plays a role in phosphorylation of adherens junctions that affects cell adhesion and migration, it may contribute to cancer cell invasion and metastasis (42). It will be interesting to test a possible role of both PAK4 and PTPN14 in controlling cancer cell motility.

Stathmin is another dynamically regulated phosphoprotein identified in this study that may be causally linked to cancer. The observed regulation of phospho-stathminSer38 by PI3K pathway inhibitors is potentially important because high concentrations of stathmin are observed in PTEN-deficient tumors and associated with a poor prognosis (18). The phosphorylation state of stathmin has been linked to the cell cycle–dependent efficacy of chemotherapeutic drugs (43). Whether phosphorylation of stathmin might in part predict sensitivity to PI3K-targeting agents is an important question.

As proof of principle for our approach, we evaluated an antibody against phospho-PRAS40Thr246 for clinical translation feasibility studies for several reasons. First, it was one of the most dynamically regulated markers in vitro (Table 1) and a direct substrate for AKT (3, 4, 44). Second, there is a great need for such an antibody because antibodies against phospho-AKTThr308 (the phosphorylation site that directly correlates with AKT activity) display low sensitivity and stability for immunohistochemistry, whereas those against phospho-AKTSer473, although widely used for immunohistochemistry, monitor a phosphorylation site that is subject to feedback regulation by mTORC2 and therefore does not always correlate with AKT activity (3, 4, 44). Moreover, most downstream substrates, although in principle feasible as biomarkers for AKT activity, are not optimal for immunohistochemistry, have restricted tissue expression, or are not exclusive PI3K-AKT pathway–regulated substrates. Finally, clinical studies examining rapamycin-induced AKT activation in PTEN-deficient glioblastoma patients have shown that induction of phospho-PRAS40 is significantly associated with shorter time to progression (44). These clinical findings, together with our data showing strong inhibition of phospho-PRAS40Thr246 by several PI3K pathway inhibitors, indicate the utility of this phosphoantibody for biomarker-driven clinical trials in patients with PI3K pathway–activated tumors. Indeed, our analysis of 67 breast cancer cell lines (Fig. 6B and fig. S3) and triple-negative breast cancer tumor tissues (Fig. 7, B and C) suggests that phospho-PRAS40Thr246 status, combined with mutational analysis of PI3K, could enable such a patient stratification approach. Our in vitro drug sensitivity data (Fig. 6C, fig. S1, and table S3) also imply that high levels of phospho-MEK (Fig. 6C) and phospho-ERK (fig. S1) could act as possible exclusion biomarkers for single-agent AKT inhibitor therapy independent of phospho-PRAS40Thr246 levels. Cancer patients showing both high phospho-RAS40Thr246 and phospho-ERK tumor staining may be more likely to benefit from dual PI3K-MAPK pathway inhibition achieved through combinations of drugs [for example, an AKT inhibitor with inhibitors of MEK or inhibitors of relevant RTKs such as insulin-like growth factor receptor (IGFR), epidermal growth factor receptor (EGFR), or cMET]. Accordingly, MEK and AKT inhibitor combination therapy is currently being tested in clinical trials for efficacy in tumors with deregulated core PI3K and MAPK pathway signaling.

Likewise, monitoring both phospho-PRAS40Thr246 (a biomarker of AKT enzyme activity) and phospho-AKTSer473 levels in response to PI3K pathway inhibitors may elucidate the role of feedback loops in determining the clinical efficacy of targeted therapies. For example, treatment of cancer cells with a catalytic inhibitor of AKT, A-443654, results in increased phosphorylation of AKTSer473 via a feedback mechanism that is independent of mTORC1 (45). By contrast, the allosteric AKT inhibitor used in this study does not result in feedback activation at the level of AKTSer473 because it blocks both kinase activity and AKT membrane translocation. By the same token, PI-103, which is a dual PI3K-mTOR catalytic inhibitor that inhibits both mTORC1 and mTORC2, can be differentiated from a catalytic AKT inhibitor by these antibody-based biomarkers: PI-103 inhibits phosphorylation of both AKTSer473 and AKTThr308 (Fig. 1A), as well as PRAS40Thr246 (Fig. 4B), indicating that it might provide a possible advantage over using a catalytic AKT inhibitor.

Together, the combination of technological advances with small-molecule kinase inhibitors described herein has allowed site-specific identification and quantification of hundreds to thousands of phosphorylation sites in an enriched tumor cell subproteome, thereby setting the stage for a paradigm shift in drug development. Most notably, the study highlights a general approach for the identification of uniquely regulated, yet simple, patterns of phosphopeptide alterations that are specific for targeted agents acting on different nodes within the same oncogenic pathway. This aspect of drug development is becoming a common challenge for many pharmaceutical and biotech companies, with their portfolio of targeted agents concentrated on one or a few core signaling pathways. Our approach can be used for informed substratification of cancer patients based on simple, drug-specific biomarker patterns into groups that are most likely to respond to pathway inhibition by individual agents. In addition, such a method can enable differentiation of small molecules developed against the same targets because different off-target effects are likely to be revealed through qualitatively and quantitatively different phosphopeptide alterations. This will also maximize the value of the drug pipeline and portfolio by avoiding exhaustion of less-defined, overlapping patient populations in clinical trials.

The pathway-based approach described here for deriving drug-specific biomarkers will also, in principle, enable repositioning of compounds on the basis of drug-regulated pathway effects due to (often unknown) off-target effects. For instance, an experimental or a marketed drug that is considered safe for humans and is used for a given tumor type on the basis of known on-target effects might have inhibitory effects on other signaling pathways from off-target interactions. These effects can be readily identified through our differential phosphoprofiling approach. To the extent that some of the identified pathways could be oncogenic pathways for certain tumors, this could guide biomarker-driven trials for such tumors containing up-regulation of the identified pathway. The ability to generate and use phosphorylation state–specific antibodies against identified, drug-specific phosphopeptide patterns will likely have wide-ranging clinical and practical implications through usage in biopsy-driven trials.

Materials and Methods

Antibodies and tissue culture reagents

All antibodies were from Cell Signaling Technology (CST) except for phospho-p90RSKSer221 (Invitrogen). The custom phospho-PRAS40Thr246 antibody (clone C77D7) is available from CST.

Quantitative immunoprecipitation MS

PC-3 cells (2 × 108) were grown either in light (12C6-Lys, 12C6-Arg) or heavy (13C6-Lys, 13C6-Arg) Dulbecco’s modified Eagle’s medium (Invitrogen) supplemented with 10% fetal bovine serum (Hyclone) for 3 weeks before inhibitor treatment (5 μM, 18 hours). Light and heavy lysates were combined (1:1 ratio), and 20 mg was digested with either trypsin or Glu-C (Fig. 1B). Phosphopeptide enrichment was carried out as described (46) with the AKT substrate motif antibody for the Glu-C digest, and the PDK1-docking and MAPK substrate motif antibodies for the trypsin digest. Immunoprecipitations of the trypsin digests were performed serially. After desalting, samples were loaded onto a reverse-phase microcapillary column [360-μm outer diameter (OD) by 75-μm inside diameter (ID)] packed with 2.5 cm of Magic AQ C18 (Michrom Bioresources). After washing with 100% solvent A (0.1 M acetic acid in deionized H2O) for 5 min at 3 μl/min, bound peptides were eluted onto a spraying column (360-μm OD by 50-μm ID, Magic AQ C18 medium). A gradient (flow rate of 300 nl/min) was delivered by an Agilent 1100 Series binary high-performance liquid chromatography system comprising three distinct phases: (i) 5-45% B in 75 min, (ii) 45-100% B in 15 min, and (iii) 100-0% B in 5 min at 1 μl/min (A, 0.1 M acetic acid; B, 70% acetonitrile in 0.1 M acetic acid). Full-scan spectra were acquired (resolution of 50,000) on an ORBI trap mass spectrometer (Thermo Electron). Tandem mass spectra were simultaneously acquired for the seven most abundant ions in a data-dependent fashion (mass window, 3 daltons; dynamic exclusion, 45 s; repeat counts, 1). Targeted analysis was performed on a linear triple quadrupole–ETD mass spectrometer (Thermo Electron).

SILAC analyses

All binary SILAC analyses were carried out with the SILAC analyzer 3.1 (Elucidator, Rosetta Biosoftware). Briefly, all ions in a single-dimension liquid chromatography–Fourier transform MS run were grouped according to their respective isotope and charge clusters, after which all possible heavy-light SILAC pairs were evaluated. Retention time shifts for each potential pair were limited to 0.2 min, and the mass accuracy was limited to 5 parts per million.

Immunohistochemistry

Serial sections (5 μm) of formalin-fixed, paraffin-embedded samples were stained with the Ventana Discovery XT multistainer. Pictures were taken with a Zeiss Imager (Z1 microscope equipped with an AxioCam HRc camera). All micrographs were taken with ×200 magnification. H scores for each staining were derived by multiplying the fraction of positively stained tissue (percent) by the intensity of the staining (1+, 2+, or 3+) based on the mean of the subjective scores by two independent pathologists. The signal (H scores) was classified as negative (0 to 9), low (10 to 100), medium (101 to 200), or high (201 to 300).

Cell lines and proliferations assays

Breast and lung cancer lines were obtained from various commercial vendors (American Type Culture Collection, European Collection of Animal Cell Cultures, German Resource Centre for Biological Material, Health Science Research Resources Bank, and Immunobiological Laboratory) and cultured in vendor-recommended medium. Drug sensitivity was determined with CellTiter-Glo (Promega) at 24, 48, and 72 hours after inhibitor treatment. Normalized drug sensitivity values [X/X0] were calculated to correct for differences in basal growth rates (doubling times) across the various cell lines. The median effective concentration (EC50) was calculated after 72 hours of AKT inhibitor treatment with standard four-parameter sigmoidal curve-fitting algorithms.

Supplementary Material

www.sciencetranslationalmedicine.org/cgi/content/full/2/43/43ra55/DC1

Methods

Fig. S1. Correlation of basal phospho-ERKThr202/Tyr204 concentrations versus sensitivity to AKT inhibition.

Fig. S2. Immunohistochemical stability of phospho-AKTSer474 and phospho-PRAS40Thr246.

Fig. S3. Correlation analysis of phospho-PRAS40Thr246 with PI3K-pathway mutation status in breast cancer cell lines.

Table S1. Summary of identified peptides containing the [RXX(s/t)], [PX(s/t)P], and [(F/K)XX(F/Y)(s/t)(F/Y)] motif.

Table S2. Gene ontology summary for 377 identified phosphoproteins.

Table S3. Key information on breast cancer cell line panel.

Table S4. Key information on lung cancer cell line panel.

Footnotes

  • * These authors contributed equally to this work.

  • Present address: Fred Hutchinson Cancer Research Center, Seattle, WA 98109, USA.

  • Present address: Biogen Idec, Cambridge, MA 02432, USA.

  • || Present address: Daiichi Sankyo Research Institute, Daiichi Sankyo, Inc., Edison, NJ 08837, USA.

  • Citation: J. N. Andersen, S. Sathyanarayanan, A. Di Bacco, A. Chi, T. Zhang, A. H. Chen, B. Dolinski, M. Kraus, B. Roberts, W. Arthur, R. A. Klinghoffer, D. Gargano, L. Li, I. Feldman, B. Lynch, J. Rush, R. C. Hendrickson, P. Blume-Jensen, C. P. Paweletz, Pathway-based identification of biomarkers for targeted therapeutics: Personalized oncology with PI3K pathway inhibitors. Sci. Transl. Med. 2, 43ra55 (2010).

References and Notes

  1. Acknowledgments: We thank J. English, V. Richon, and G. Draetta for helpful guidance during the course of the study. We are grateful to our long-standing collaboration with scientists at CST. We thank B. Settlage, J. Conway, D. Spellman, and A. Bondarenko for developing SILAC Elucidator 3.1. We apologize for not being able to cite many relevant original papers, replaced by reviews, due to space limitation. Funding: Merck Research Laboratories. Author contributions: This study was conceived by P.B.-J. The experiments were designed by J.N.A., P.B.-J., R.C.H., and C.P.P. and performed by J.N.A., S.S., B.D., A.H.C., L.L., B.R., W.A., R.A.K., D.G., B.L., and C.P.P. Proteomic data analyses were carried out by A.C., J.R., and C.P.P. Immunohistochemical studies were performed by A.D.B. and D.G. Bioinformatic analyses were performed by T.Z., S.S., J.N.A., and I.F. The paper was written by J.N.A., P.B.-J., and C.P.P. All authors critically reviewed the manuscript. Competing interests: P.B.-J. is an inventor on a patent related to the results described here (8). All authors performed this study while employees at Merck Inc. The authors declare that they have no competing interests.
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article