proteomicanalysisofbreastcancermolecularsubtypes ...€¦ · between phosphorylation by kinases,...

Proteomic analysis of breast cancer molecular subtypesand biomarkers of response to targeted kinase inhibitorsusing reverse-phase protein microarrays

Zachary S. Boyd,1 Qun Jenny Wu,2 Carol O’Brien,1

Jill Spoerke,1 Heidi Savage,1 Paul J. Fielder,2

Lukas Amler,1 Yibing Yan,2 and Mark R. Lackner1

1Department of Development Oncology Diagnostics and2Department of Pharmacodynamic Biomarkers, Genentech, Inc.,South San Francisco, California

AbstractAlthough breast cancer molecular subtypes have beenextensively defined by means of gene expression profilingover the past decade, little is known, at the proteomiclevel, as to how signaling pathways are differentiallyactivated and serve to control proliferation in differentbreast cancer subtypes. We used reverse-phase proteinarrays to examine phosphorylation status of 100 proteinsin a panel of 30 breast cancer cell lines and showeddistinct pathway activation differences between differentsubtypes that are not obvious from previous geneexpression studies. We also show that basal levels ofphosphorylation of key signaling nodes may have diag-nostic utility in predicting response to selective inhibitorsof phosphatidylinositol 3-kinase and mitogen-activatedprotein kinase/extracellular signal-regulated kinase kinase.Finally, we show that reverse-phase protein arrays allowthe parallel analysis of multiple pharmacodynamic bio-markers of response to targeted kinase inhibitors and thatinhibitors of epidermal growth factor receptor and mito-gen-activated protein kinase/extracellular signal-regulatedkinase kinase result in compensatory up-regulation ofthe phosphatidylinositol 3-kinase/Akt signaling pathway.[Mol Cancer Ther 2008;7(12):3695–706]

IntroductionBreast cancer is a heterogeneous disease with distinctmolecular subtypes that vary in prognosis and showdifferential response to both targeted and chemotherapeu-tic agents (1, 2). Over the past 30 years, the clinical

management of breast cancer has become increasinglypersonalized with routine use of diagnostic tests todetermine whether patients should receive targeted thera-pies, such as Tamoxifen for hormone receptor–positivedisease (3) and Trastuzumab for HER2-positive disease (4).Several trends suggest that this personalization is likely tocontinue and, indeed, perhaps result in even finerstratification of patients in the future. First, an explosionof gene expression profiling and classification studies hasrevealed additional complexity and substructure within theoverall framework of breast cancer (2). That is, whereasmost expression profiling studies recognize at least threemain subtypes, termed luminal, basal-like, and HER2amplified, certain studies have uncovered gene sets thatrefine the classification breast of tumors into as many assix distinct subtypes (5). Assuming that these subtypesharbor distinct pathological alterations and mechanisms ofsignaling pathway dysregulation, it seems likely that noveltherapeutics targeting particular signaling pathways mayhave activity in specific subsets of patients. Second, thereare currently over 1,400 phase II or phase III clinical trialsunder way in breast cancer,3 a number of which examinethe role of new targeted therapies in disease management,suggesting an increasing necessity of additional diagnosticstrategies to target novel agents to appropriate patientpopulations.Although gene expression studies have been very fruitful

at identifying breast cancer subtypes and predictingdisease recurrence (6), there is nevertheless a need forcomplementary diagnostic technologies that can assess theoverall status of signaling networks and provide informa-tion that is not apparent at the transcriptome level. Thereasons for this are severalfold. First, transcriptionalchanges in gene expression are a distal output of signaltransduction pathways (7) and may not provide a real-timereflection of the activation state of a given pathway.Second, the high dimensionality of gene expression datacreates a false discovery problem that makes it difficult torelate specific transcriptional changes to the activation ofa particular pathway (8). Thirdly, in past studies wheregene and protein expression were directly compared, amajority of the comparisons showed poor concordancebetween gene and protein expression (9–12); thus, it is nota certainty that gene expression can predict protein expres-sion levels. Finally, the functional components of signaltransduction pathways at a cellular phenotypic level areposttranslational in nature and involve a complex interplay

Received 8/28/08; accepted 9/30/08.

The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby markedadvertisement in accordance with 18 U.S.C. Section 1734 solely toindicate this fact.

Note: Z. Boyd and J. Wu contributed equally to this work.

Requests for reprints: Mark R. Lackner, Genentech, Inc., 1 DNA Way,South San Francisco, CA 94080. Phone: 650-467-1846;Fax: 650-225-7571. E-mail: [email protected]

Copyright C 2008 American Association for Cancer Research.

doi:10.1158/1535-7163.MCT-08-0810 3 www.clinicaltrials.gov

3695

Mol Cancer Ther 2008;7(12). December 2008

Published Online First on December 3, 2008 as 10.1158/1535-7163.MCT-08-0810

on July 9, 2020. © 2008 American Association for Cancer Research. mct.aacrjournals.org Downloaded from

Published OnlineFirst December 3, 2008; DOI: 10.1158/1535-7163.MCT-08-0810

http://mct.aacrjournals.org/

between phosphorylation by kinases, dephosphorylationby phosphatases, and protein degradation through ubiq-uitination. These rapid and dynamic changes in pathwaystatus are inherently difficult to capture in real timethrough gene expression surrogates. Also, because manyinhibitors currently in development are specific kinaseinhibitors that impair components of these signal trans-duction cascades, readouts of pathway activation stateswould be of great utility to both the drug and companiondiagnostic development process. For instance, assays forpretreatment phosphorylation levels could be used todetermine whether a signaling pathway was activated ina patient’s tumor and, hence, whether the patient mightbenefit from a specific inhibitor that targeted a key nodein that pathway (i.e., predictive biomarkers; ref. 13). Inaddition, assays for phosphorylation status of kinasesubstrates that could be conducted on small amounts ofpatient material pretreatment and posttreatment couldserve as reliable indicators that a drug was effectivelymodulating target activity at a particular dose and schedule(i.e., pharmacodynamic biomarkers).In light of these considerations, we have investigated the

use of reverse-phase protein arrays (RPPA) to determinethe signaling pathway status in a large panel of breastcancer cell lines that have previously been characterizedin detail at the gene expression and copy number leveland shown to reflect many of the genomic alterationscommonly found in breast tumors (13, 14). Previous studiesusing RPPAs in cell lines have focused on panels of celllines derived from multiple tissue types (15, 16) and havenot sought to address signaling network heterogeneitywithin the context of a single tumor type. The overarchingquestions we sought to address were, first, whether breastcancer molecular subtypes have distinct patterns of path-way activation that can be discerned through this typeof network analysis and, second, whether high levels ofactivation of a particular pathway can predict response toselective inhibitors of key nodes of that pathway. RPPAswere chosen for this study because they offer the potentialto profile potentially hundreds of phosphorylation eventsin very small quantities of tumor materials, such as thosethat might be obtained from laser capture microdissectionfrom a biopsy (17, 18). The basis of the technology is toimmobilize small amounts of lysate from a cell line ortumor sample in serial dilution on a microarray slide.Multiple samples are, thus, arrayed on a slide and can beprobed with antibodies that detect a particular phosphor-ylated epitope (17). Using this technology, we profiled100 key signaling nodes representing a number of path-ways known to be dysregulated in cancer across the panelof 30 cell lines. The key findings of this study are thatdifferentially phosphorylated proteins cannot only classifycell lines fairly accurately into molecular subtypes but alsoprovide an indication that basal-like cell line models arecharacterized by high levels of SRC family and epidermalgrowth factor receptor (EGFR)/RAS/extracellular signal-regulated kinase (ERK) signaling, whereas luminal celllines show high levels of activation of phosphatidylinositol

3-kinase (PI3K)/mTOR pathway components. In addition,we find that basal levels of signaling through the mitogen-activated protein kinase/ERK kinase (MEK)/ERK andPI3K/AKT pathway are correlated with in vitro responseto inhibitors of MEK kinase and PI3K, respectively. Finally,we show that targeted inhibitors of MEK, PI3K, andEGFR have measurable effects on key signaling pathwaycomponents, suggesting that RPPAs could play a role inoptimizing dose and schedule for such agents in earlystage clinical trials.

Materials andMethodsCell LinesBreast cancer cell lines AU565 BT-474, BT-549, BT-483,

CAMA-1, HCC1143, HCC1419, HCC1428, HCC1500,HCC1569, HCC1937, HCC1954, Hs578T, KPL-1, MCF-7,MDA-MB-175-VII, MDA-MB-231, MDA-MB-361, MDA-MB-435S, MDA-MB-436, MDA-MB-453, MDA-MB-468,and ZR-75-1 were obtained from American Type CultureCollection. The cell lines CAL-120, CAL-51, CAL-85-1,EFM-19, EFM-192A, and EVSA-T were obtained fromthe Deutsche Sammlung von Mikroorganismen und Zell-kulturen GmbH. All cell lines were maintained in RPMI1640 or DMEM supplemented with 10% fetal bovineserum (Sigma), nonessential amino acids and 2 mmol/LL-glutamine. In vitro cell viability studies were conductedas described previously (13). Kinase inhibitors used in thisstudy include the MEK1/2 inhibitor PD0325901 (Pfizer;ref. 19), the PI3K/mTOR inhibitors PI-103 (20) and GDC-0941 (21), and the EGFR inhibitor Erlotinib (Tarceva,Genentech). PD0325901 is a selective MEK inhibitor thatacts through a novel noncompetitive allosteric mechanismof action involving binding to a site adjacent to the ATP-binding site and locking the kinase in a closed andcatalytically inactive conformation (22). PI103 is a potentand selective PI3K inhibitor with low IC50 values againstrecombinant PI3K isoforms p110a, p110h, p110y, p110g,mTOR, and DNA-PK (20). GDC-0941 is a potent andselective inhibitor of all four isoforms of the catalyticsubunit of PI3K that is currently in phase I clinical trials(21). Erlotinib (Tarceva) is a small molecule inhibitor ofEGFR that is approved for treatment of advanced non–small cell lung cancer and first line metastatic pancreaticcancer (23).

Western Blot AnalysesFor analyses of basal levels of phosphorylated Akt, 40,000

cells per well were plated in six-well plates in mediacontaining 10% fetal bovine serum and allowed to grow for48 h or until cells reached 60% to 80% confluence. Cellswere washed with cold PBS and processed directly inthe plate. DNase lysis buffer [80 AL; 10 mmol/L Tris-HCl(pH 8), 3 mmol/L MgCl2, 10 mmol/L NaCl, 1% TritonX-100 supplemented with protease inhibitor cocktail(Sigma P 8342), phosphatase inhibitor I cocktail (Sigma P2850), phosphatase inhibitor II cocktail (Sigma P 5726),phenylmethylsulfonyl fluoride, and DNase (Sigma D7291)]was added to each well and then incubated, shaking at

Reverse-Phase Protein Arrays in Breast Cancer Cells3696





37jC for 10 min. NuPAGE LDS 4� sample buffer(Invitrogen) and NuPAGE 10� sample reducing agent(Invitrogen) were added, and the samples were heated to70jC for 10 min and then loaded directly into NuPAGE4% to 12% Bis-Tris precast gel (Invitrogen). Primaryblotting antibodies used were total AKT and pAKT (S473;Cell Signaling Technology). Secondary blotting antibodiesused were polyclonal goat anti-mouse IgG horseradishperoxidase and polyclonal goat anti-rabbit IgG horseradishperoxidase (both from Dako, Glostrup, Denmark or CellSignaling Technology).

PTENSmall Interfering RNAExperimentsOnTARGET Plus small interfering RNA (siRNA) specific

to human PTEN (Dharmacon) or a control siRNA that doesnot target any sequence in the human genome (nontargetcontrol, Dharmacon) were used in transient transfectionexperiments. Optimal siRNA duplex and lipid concentra-tions were determined for each cell line. NontargetedsiRNA nucleotides (Dharmacon) were used as nonspecificcontrols in all experiments. Transfection efficiency ofPTEN siRNA was evaluated by Western blot for humanPTEN. MDA-MB231 cells were plated at 6,000 cells perwell in a 96-well plate with 0.125 AL of LipofectamineRNAiMAX (Invitrogen) and 50 nmol/L of siRNA per well.Cells were incubated for 3 d in siRNA then harvested andprocessed for RPA analysis, as described below.

Reverse-Phase Protein ArraysCells were grown in standard media and 10% fetal

bovine serum until they reached 60% to 80% confluenceand then lysed directly in plates in buffer consisting of a2.5% solution of 2-mercaptoethanol in loading buffer/T-PER plus phosphatase and protease inhibitors. Forpharmacodynamic studies, cells were plated in six-wellplates and allowed to adhere overnight and either lysedimmediately or after 6 h of incubation in the presence of1 Amol/L of each compound. Samples were shipped ondry ice to Theranostics Health, LLC for the reverse-phaseprotein array (RPPA) analysis. All samples were diluted toa final concentration of 0.5 mg/mL and then 30 AL of eachsample, arrayed in a series of six fold dilutions, was printedin duplicate on slides. Lysates derived from HeLa cells orHeLa cells treated with pervanadate were also printed oneach slide as low and high phosphorylation controls,respectively. The slides were then subjected to immunos-taining with a panel of 100 antibodies primarily directedagainst specific phosphorylated or cleaved proteins. Eachof these antibodies had previously undergone extensivevalidation for both phosphorylation and protein specificityusing single band detection at the appropriate MW byWestern Blotting. The specific protein and phosphoproteinassays are listed in Supplementary Table S1.4 To allownormalization of total protein on printed arrays, one to twoslides in each print run were stained with Sypro Rubyprotein blot stain (Invitrogen) and the value of these

stained arrays used for normalization of all end pointvalues. The intensity value for each end point wasdetermined by identifying spots for each duplicate dilutioncurve for each sample that were within the linear dynamicrange of the staining after background subtraction witheach spot (within slide local background and also againsta slide stained with secondary antibody only). Singleintensity values were obtained by multiplying each spotin the linear range by its dilution factor and averagingcandidate linear points. Finally, each value was normalizedrelative to the total protein intensity value for that samplederived from the Sypro Ruby–stained slide.

Statistical AnalysesData analysis was done with Partek Genomics Suite

software version 6.3 (Partek, Inc.). RPPA data waspreprocessed by log2 transformation and further linearscaling (z-score conversion) to ensure data normality andlinearity. The log-transformed data were used for principlecomponent analysis (PCA) and other statistical analyses,and the z-scores were used in hierarchical cluster analysis.For unsupervised hierarchical clustering analysis, a max/min ratio was calculated for each end point based on thehighest and lowest intensity samples. The top 50 assaysbased on max/min ratio were log transformed and mediancentered and then subjected to average linkage clusteringusing a correlation similarity metric. Analysis and visual-ization of data were conducted using cluster and treeviewsoftware (24).PCA was used to reduce the dimensionality of the data

and detect the dominant sources of variation in the data.For the identification of differentially expressed or phos-phorylated markers we, used a one-way ANOVA with a0.05 P-value threshold. To further validate the selectedmarkers from ANOVA, unsupervised hierarchical clusteranalysis was done to test the clear segregation of differentmolecular subtypes. Hierarchical cluster analysis was doneusing distance matrix of Pearson correlation with averagelinkage method and PCA with correlation dispersionmatrix and normalized eigenvector scaling. A post hocanalysis of ANOVA involving pairwise comparison wasalso done to select markers that showed statisticallysignificant phosphorylation differences between the sub-types. Volcano plots were used to view markers that notonly satisfied the statistical cutoff but also showedbiological significance based on fold change (i.e., 2-fold).To detect association between compound sensitivity

and phosphorylation of particular end points, we first logtransformed the in vitro EC50 values for each compound(Supplementary Table S1)4 and then used Pearson correla-tion and analysis of covariance to identify statisticallysignificant correlations between EC50 values and proteinintensities.

ResultsTechnicalValidation of ArraysWe profiled protein lysates from 30 breast cancer cell

lines with a panel of 100 antibodies recognizing primarily4 Supplementary material for this article is available at Molecular CancerTherapeutics Online (http://mct.aacrjournals.org/).

Molecular Cancer Therapeutics 3697





phosphorylated epitopes. Identities of the phosphorylatedand total proteins profiled, as well as normalized signalintensities for each of these antibody end points in all celllines are provided in Supplementary Table S1.4 Also shownis the molecular subtype for each cell line, determinedas previously described (13), as well as notable geneticalterations in each cell line (e.g., KRAS, BRAF, PTEN,PIK3CA status).We first sought to determine reproducibility of the

data by analyzing independently derived lysates (passage1 versus passage 2) from MDA-MB-231 cells and foundsignificant correlation between the 100 end points fromthe independent samples (Pearson correlation, r = 0.94;Supplementary Fig. S1).4 Next, we sought to compare

results from the RPPA analysis with independent dataderived from the same cell lines using different methods.First, we compared protein intensity for EGFR determinedby RPPA with mRNA levels determined on AffymetrixHuman Genome U133 Plus 2.0 arrays and found thesemeasurements to be significantly correlated (Pearsoncorrelation, r = 0.80; Supplementary Fig. S2A).4 Next, wecompared phosphorylated PTEN (S380) levels with PTENlevels determined from independent lysates derived fromthe same cell lines and found that cell lines determined tobe PTEN null (no protein band) by Western blotting wereuniformly low for the pPTEN (S380) RPPA end point. Wealso found good concordance between phosphorylationof AKT at S473 and T308 determined by Western blot and

Figure 1. Endpoints that showed dif-ferential phosphorylation between PTENor control siRNA-treated MDAMB231cells. A, phosphorylation sites thatshowed a statistically significant differ-ence between control and PTEN-treatedcells were subjected to hierarchical clus-ter analysis and visualized by means of aheatmap (red, high phosphorylation;green, low phosphorylation). B, path-way diagram showing the relationshipbetween proteins that show increased(red) or decreased (green ) phosphoryla-tion upon PTEN knockdown. Theseresults suggest coordinate up-regulationof the mTOR pathway and translationalmachinery consistent with the knownrole of PTEN as a negative regulator ofthis pathway.






RPPA analysis (data not shown). Together, these resultssuggest that determination of phosphorylated proteinstatus by RPPA is reproducible and also that the endpoint measurements agree with independent methods ofdetermining target status. To determine whether disruptionof key signaling nodes results in coordinate changes inphosphorylation in signaling networks, we used siRNAknockdown to genetically ablate the product of the tumorsuppressor gene PTEN in MDA-MB-231 cells. PTENencodes a phosphatase that is frequently deleted in a widevariety of cancers (25) and whose growth regulatoryproperties are primarily mediated via its lipid phosphataseactivity, which specifically reduces the cellular levels ofphosphatidylinositol 3,4,5-trisphosphate and, thus, servesto blunt signaling through the PI3K/AKT pathway whenintact (26). We confirmed PTEN knockdown at the proteinlevel by Western blot (data not shown) and then used at-statistic to identify phosphorylated end points thatdiffered significantly between control siRNA-treated cellsand PTEN siRNA-treated cells (using a cutoff of P < 0.05;Fig. 1A). Of note, we found that a number of key regulatoryphosphorylation sites in the PI3K/AKT pathway arealtered upon PTEN knockdown, including sites on mTOR,LKB1, p70S6K, and 4EBP1 (Fig. 1B). Interestingly, wefound a decrease in the phosphorylation of the T37 and T46

sites on 4EBP1 and a concomitant increase in the T70 site.It has been shown previously that 4EBP1 phosphorylationby mTOR on T37 and T46 is a priming event for subsequentphosphorylation of the carboxy-terminal (S65 and S70;ref. 27), and that dual phosphorylation of S65 and S70 isrequired to abrogate binding of 4EBP1 to the translationfactor eIF4E and, thus, stimulate initiation of translation(28). The reasons for differential behavior of thesephosphorylated sites on 4EBP1 are unclear but may reflecta regulatory feedback loop whereby the priming sites aredephosphorylated in response to the activation of theprotein. Together, these results suggest that perturbation ofa key signaling node can have consequences on pathwayactivation that can be assessed through multiplexed read-outs on RPPAs.

Breast Cancer Molecular SubtypesIntensive efforts over the past decade have revealed that

breast cancers can be stratified into molecular subtypeswith distinct molecular genetic alterations and prognosticoutcomes using gene expression profiling (2). Luminalbreast cancers are typically estrogen receptor positive andcharacterized by a coordinate expression of a number ofepithelial specific genes, a relatively good prognosis, andgood response rates to targeted hormonal therapies (3).HER2-positive breast cancers are characterized by thehigh-level gene amplification of HER2 oncogene, relativelypoor prognosis if untreated, and significant clinical benefitfrom the HER2-targeting monoclonal antibody Trastuzu-mab (Herceptin, Genentech; ref. 29). Basal-like breastcancers typically lack expression of HER2, ER, andprogesterone receptor and, hence, overlap with tumorsimmunohistochemically defined as ‘‘triple negative’’tumors (30). Basal-like breast cancers have a poor prognosis

and currently have not been shown to respond to anytargeted therapy and, thus, represent a particularlypressing unmet medical need (31). To date, little efforthas been focused on identifying signaling pathwayactivation differences that characterize the distinct molec-ular subtypes at the proteomic level, although such effortsare currently under way using RPPA technology alongwith human tumor specimens (31). Such efforts could haveimplications both for the development of novel-targetedagents that may show efficacy in particular subtypes, aswell as more personalized administration of emergingtargeted therapies. The cell lines used in the study havepreviously been classified into the three major molecularsubtypes using a combination of gene expression dataand HER2 status (13). Briefly, cell lines were assigned toluminal or basal-like classes using gene expression data,and then cell HER2 amplification status was assignedby means of quantitative reverse transcription–PCR ongenomic DNA to identify cell lines with greater than fourcopies of the HER2 locus. Here, we sought to determinewhether phosphorylated protein status from RPPA arrayscould provide insights into the signaling pathways thatare active in the different subtypes. We first used PCA toreduce dimensionality of the data and find patterns thatmight be related to the differential activity of signalingpathways in particular subtypes of breast cancer. Theresulting PCA plot (Fig. 2A) shows that the globalproteomic signature determined by this method largelyseparates basal-like cell lines from HER2 amplified andluminal cell lines along the second principal component.Also, with the exception of the HER2-amplified line BT474,the majority of the luminal lines are separated from theHER2 lines. This analysis suggests that the phosphorylatedprotein end points in this analysis are significantlycorrelated because the first three principal componentscan account for 61% of the variance in the data and alsothat distinct pathways may be activated in the differentsubtypes.To explore in more detail which particular end points

or signaling modules are predominantly active in a givensubtype, we constructed a one way ANOVAmodel derivedfrom pairwise comparisons of the different subtypesand then used the top end points from that model toperform unsupervised hierarchical clustering on the celllines (Fig. 2B). This analysis recapitulated the clusteringresults derived from gene expression analysis becauseit partitioned the cell lines into distinct clusters withluminal cell lines (based on gene expression) clearlyseparated from basal-like lines. To a lesser extent, theHER2-amplified cell lines formed a distinct grouping,although several of the HER2-amplified lines were foundto cluster with luminal cell lines in this analysis. Thisfinding is consistent with gene expression clustering ofbreast cancer cell lines because such studies report thatbreast cancer cell lines resolve into only two main clustersregarded as luminal and basal-like and do not reveal aseparate HER2-amplified subtype (14). Moreover, thisanalysis suggests specific pathway activation events may






be present in the different molecular subtypes (Fig. 2B andC). In particular, basal-like lines were found to be distinctfrom luminal and Her2-amplified lines in having low levelsof pPTEN and high levels of total EGFR, Pyk2 (Y402), andPKC-a (S567). HER2-amplified cell lines were distinct fromthe other subtypes in having high levels of pERBB3, pFAK,

and pEGFR (Y1173), and luminal cell lines were distinct inhaving higher levels of phosphorylation of pp70S6K (S371)and pA-RAF (S299). We also used volcano plots to comparethe fold change in intensity with the statistical significancelevel for luminal-basal markers identified from ANOVAanalysis (Supplementary Fig. S3).4 This analysis suggested

Figure 2. Molecular classification of breast cancer cell lines based on RPPA proteomic data. A, cell lines were classified into molecular subtype based ongene expression (luminal or basal-like) and Her2 amplification status (Her2), as indicated in the key. PCA was done using all 100 protein end points and theresulting PCA map is shown. Each spot represents a single cell line. x, y, and z axes represent three major PCs. Basal-like cell lines (red) were largelyseparated from HER2 amplified (green and blue) and luminal cell lines (purple ) along the second principal component. Also, with the exception of theHER2-amplified line BT474, the majority of the luminal lines are segregated from the HER2 lines in the direction of P1/P3 diagonal. The results suggestgeneral separation of cell lines of different subtypes into distinct PCA clusters, consistent with distinct patterns of pathway activation in the differentsubtypes. B, hierarchical clustering of differentially phosphorylated end points identified in a one-way ANOVA model. Cell lines are indicated on the verticalaxis and protein end points along the horizontal axis. The color of each end point corresponds to the z score, which indicates the number of standarddeviations from the mean. Red, intensity level above mean; green, intensity level below the mean. C, examples of end points that show subtype-specificdifferences in phosphorylation, including Pyk2 and EGFR (higher in basal-like lines), p70S6K (higher in luminal and Her2 amplified lines), and ERBB3 (higherin Her2 amplified). Graphs show box and whisker plots with median intensity levels and interquartile ranges along with individual data points representingeach cell line. Cell lines are color coded by molecular subtype according to the key provided, and y axis shows z scores.

Figure 3. Results of unsupervised hierarchical clustering with the 50 most differentially phosphorylated proteins. Cell lines are indicated on thehorizontal axis, and protein end points are along the vertical axis. The molecular subtype based on gene expression and Her2 status is indicated in coloredboxes along the top row according to the key. Red, increased phosphorylation; green, phosphorylation. The heatmap on the left shows all 50 end points,whereas the three smaller heatmaps show specific clusters of phosphorylated proteins that are enriched in each molecular subtype.






that high levels of EGFR and phosphorylation of Pyk2and PKC-a are characteristics of basal-like cell lines,whereas phosphorylation of A-RAF, p70S6K, and PDK1are characteristic of luminal cell lines. In a parallelapproach, we ranked the end points based on the levelof differential phosphorylation (as described in Materialsand Methods) across the cell line panel and then used thetop 50 differentially phosphorylated end points in hierar-chical clustering analysis (Fig. 3). This analysis againshowed that differential phosphorylation of these signal-ing pathway proteins organizes the cell lines into twooverall clusters that are quite similar to those derived fromgene expression analyses. In addition, this analysisrevealed patterns of pathway activation that are notobvious from published gene expression analyses. Inparticular, basal-like cell lines were found to havehigh levels of phosphorylation of nonreceptor tyrosinekinases, such as c-Abl and Pyk2, and in addition showedgenerally high levels of Erk1/2 phosphorylation andhigh total EGFR expression. In contrast, HER2-amplifiedcell lines were found to have high levels of phosphoryla-tion of components of HER receptor signaling (e.g.,SHC, ERBB3, EGFR), as well as other receptor tyrosinekinases (e.g., c-MET). Finally, luminal cell lines that donot have apparent amplification of HER2 showed gener-ally higher levels of activation of downstream signalingpathway components in the Akt/mTOR pathway (e.g.,p70S6K).

Predicting Sensitivity toTargetedTherapeuticsA potentially important application of RPPA technology

is the more personalized administration of targetedtherapies based on the signaling status of a given patient’stumor. The assumption is that if a patient’s tumor isaddicted to the continued activation of a particularpathway for continued growth and survival (32), thenphosphorylation at key nodes in that pathway may serve ashallmarks, indicating the presence of an activated pathwayand the potential for therapeutic intervention with inhib-itors targeting that pathway. To address the preclinicalfeasibility of such an approach, we sought to determinewhether the pretreatment levels of specific phosphorylated

end points showed correlation with in vitro response topotent and selective PI3K and MEK inhibitors. Geneticand biochemical analyses of MEK function have suggestedthat MEK activity is necessary for the transforming andproliferative effects of this pathway, suggesting thattherapeutics that completely inhibit MEK function mayhave utility in the treatment of cancers driven by activationof the RAS/RAF axis (33). Similarly, PI3K is a keytransducer of growth factor signals from receptor tyrosinekinases, as well as a frequently mutated oncogene,suggesting that PI3K inhibitors might have beneficialeffects in treating cancers driven by pathologic alterationsof this pathway (34). As described in Materials andMethods, we used Pearson correlation and an analysis ofcovariance model (ANCOVA) to identify the phosphory-lated proteins that showed the highest correlation withdrug sensitivity for these agents. For the MEK inhibitor(Fig. 4), we found significant correlation (P < 0.05) betweensensitivity (i.e., low EC50) and expression of total EGFRand phosphorylated ERK1/2 (the direct targets of MEKphosphorylation), suggesting that constitutive signalingthrough the EGFR/RAS/MEK/ERK axis may constitutea positive predictive factor for MEK inhibition (Table 1;Fig. 4). We also note that phosphorylation of the non-receptor tyrosine kinases Pyk2 and Abl, as well as PKC-a,were positively correlated with response to the MEKinhibitor. Conversely, phosphorylation of components ofAkt and mTOR signaling, such as PDK1 and p70S6K, wasnegatively associated with response to the MEK inhibitor(Table 1). In contrast, we found that sensitivity to the PI3K/mTor inhibitor PI-103 was significantly correlated withelevated phosphorylation at key nodes in the PI3K/Akt/mTOR pathway, including pAkt (T308 and S473), PRAS40(T246), and FKHR (T24; Table 1), suggesting that highlevels of signaling through the pathway may be indicativeof pathway addiction and predictive of response to atargeted PI3K inhibitor.

Pharmacodynamic Effects of Targeted KinaseInhibitorsWe determined the quantitative in vitro pharmaco-

dynamic effects of small molecule kinase inhibitors of

Figure 4. Sensitivity to a MEK inhib-itor is associated with high levels ofEGFR/Pyk2 and Erk1/2 signaling. Log-transformed EC50 data for the MEKinhibitor and selected end points thatshowed significant association withMEK sensitivity based on Pearson cor-relation were used to perform unsuper-vised hierarchical clustering analysis.Sensitivity to the MEK inhibitor (green ;low EC50) is associated with high levelsof EGFR expression and high levels ofphosphorylation of Erk1/2 and Pyk2(red).






MEK1/2, PI3K, and EGFR on a subset of 24 signalingpathway components (Supplementary Table S2).4 Weevaluated phosphorylation of all 24 end points pretreat-ment and posttreatment (6-hour treatment) with eachcompound in the cell lines CAL85-1 and MDAMB231.CAL85-1 is a basal-like breast cancer cell line withsubmicromolar in vitro EC50 sensitivity to all threecompounds (0.33 Amol/L for GDC0941, 0.11 Amol/L forPD0325901, and 0.23 Amol/L for Erlotinib). MDAMB231 isa KRAS/BRAF mutant basal-like breast cancer cell linewith submicromolar in vitro EC50 sensitivity to GDC0941and PD0325901 (0.82 and 0.31 Amol/L, respectively), whichis relatively insensitive to Erlotinib (EC50 4.84 Amol/L).Figure 5 shows the results for the protein end points thatshowed at least a 2-fold change between the 0 and 6 hourtime points for at least one agent in each cell line. Treatmentwith the PI3K inhibitor GDC-0941 resulted in coordinatedown-regulation of multiple components of the PI3K/mTOR axis in both cell lines, including pAkt (S473),pp70S6K (T389), 4EBP1, and pS6 ribosomal protein. Incontrast, treatment with the MEK inhibitor resulted indecreased levels of pErk1/2 and, somewhat surprisingly,up-regulation of key nodes in the PI3K pathway, suchas pAkt and pS6. This finding suggests the existence ofa feedback loop whereby inhibition of the MEK/ERKaxis causes up-regulation of signaling in the PI3K/AKTpathway. Finally, treatment with the EGFR inhibitor hadlittle or no effect on downstream pathway components,such as pErk1/2, in the unresponsive KRAS mutant cellline MDAMB231 but did dramatically reduce pErk1/2levels in the sensitive cell line CAL85-1. In addition,

Erlotinib treatment also caused up-regulation of pAkt(S473) and pS6 in a manner similar to MEK inhibition inthis cell line.

DiscussionWe show here that signaling pathway network analysisbased on RPPAs can efficiently classify breast cancer celllines into molecular subtype in a manner similar to geneexpression data. Moreover, this type of network analysisgives additional information as to the key pathways thatmay be dysregulated in particular disease subtypes andindividual cancers and also may be useful in monitoringpathway modulation in response to targeted therapiesand, thus, determining optimal biological dosing in clinicaltrials.The findings presented here are in broad agreement with

previous studies using RPPA analysis on laser capturemicrodissected cells obtained from surgically resectedprimary breast tumor materials from 25 patients thatshowed that tumors could be broadly clustered intosubgroups based on EGFR signaling, AKT/mTOR activa-tion, c-Kit/Abl activation, and Erk pathway activation (35).In addition, our studies extend these findings by showingthat activation of these pathway modules seems to occurin a subtype-specific manner and can provide the basis fortherapeutic intervention. In particular, we found that celllines defined as basal-like have high levels of EGFR,activated ERK1/2, and phosphorylation of SRC-activatedeffector kinases, such as c-Abl and Pyk2. This is intriguingbecause basal-like cell lines have been shown to be a highly

Table 1. Protein markers of kinase inhibitor sensitivity

Protein endpoint Correlation coefficient P Lower CI Upper CI

MEKi sensitivity pPyk2 Y402 �0.6004 0.0005 �0.7898 �0.3063pER S118 0.4760 0.0078 0.1397 0.7139EGFR �0.4715 0.0085 �0.7110 �0.1340

pc-Abl T735 �0.4655 0.0095 �0.7072 �0.1265pCofilin S3 0.4621 0.0101 0.1222 0.7050pPKC a S657 �0.4484 0.0130 �0.6962 �0.1051pp27 T187 �0.4061 0.0260 �0.6685 �0.0536pATF-2 T71 0.3986 0.0291 0.0447 0.6636

pErk1-2 T202 Y204 �0.3817 0.0374 �0.6523 �0.0249pEGFR Y992 0.3782 0.0393 0.0207 0.6499PEGFR Y1068 0.3764 0.0404 0.0186 0.6487pPDK1 S241 0.3758 0.0407 0.0180 0.6483pStat Y701 0.3694 0.0446 0.0105 0.6440pA-Raf S299 0.3620 0.0493 0.0019 0.6389

pp70S6Kinase S371 0.3559 0.0536 �0.0050 0.6348pc-Kit Y721 0.3556 0.0538 �0.0054 0.6346

PI3Ki-mTOR sensitivity pPRAS40 T246 �0.4964 0.0053 �0.7267 �0.1658pFKHRT24-FKHRL1 T32 �0.4927 0.0057 �0.7244 �0.1610

pHistone H3 S10 �0.3787 0.0390 �0.6503 �0.0214pAkt T308 �0.3785 0.0391 �0.6501 �0.0211pStat3 Y705 �0.3755 0.0409 �0.6481 �0.0176pAkt S473 �0.3518 0.0566 �0.6320 0.0097

Abbreviation: CI, confidence interval.






Figure 5. Pharmacodynamic modulation of signaling pathways by targeted kinase inhibitors. The heatmap shows end points with at least a 2-foldchange in intensity between pretreatment and 6-h treatment with at least one inhibitor in CAL85-1 cells (A) and MDAMB231 cells (B). Green, low levels ofphosphorylation; red, high levels of phosphorylation. Graphs to the right show normalized intensity values for selected end points that were modulated bycompound treatment.






invasive, mesenchymal subtype with high metastaticpotential (14, 36). Indeed, previous studies in a limitednumber of cell lines have shown that c-Abl acts down-stream of EGFR and Src and serves to promote invasion ofaggressive basal-like cell lines (37). In addition, studieshave shown that coexpression of c-Src and EGFR isrequired to trigger an oncogenic signaling program thatleads to the transformation of mammary epithelial cells(38). Our results suggest that this constellation of eventsmay occur endogenously specifically in basal-like breastcancer. Consistent with this notion, others have shownthat the Src inhibitor dasatinib specifically has antitumoractivity in basal-like cell lines (39, 40), and we show herethat activity as selective inhibitor of the downstream MEKkinase is correlated with the up-regulation of the c-Src axisand constitutive signaling through ERK1/2. Conversely,we show that luminal cell lines have relatively low levelsof activation of the EGFR/ERK/c-SRC axis but, incontrast, have elevated activation of downstream compo-nents of PI3K/mTOR signaling, such as PDK1 and p70S6kinase. A hypothesis emerging from this finding would bethat MEK inhibition would have little effect in luminaltumors but that a selective inhibitor of PI3K and mTORmight serve to inhibit tumor growth in these cancers andthat targeted combination therapy could provide apowerful therapeutic candidate for these tumors. Consis-tent with this, we find that high levels of phosphorylationof key members of the PI3K/mTOR pathways are amongthe top predictors of PI-103 sensitivity in our unbiasedanalysis. An important caveat is that cell lines grown intwo-dimensional culture in the presence of full serum areunlikely to fully recapitulate and mimic growth factorsignaling in three-dimensional tumors but the observationthat high levels of signaling through oncogenic pathwayspredicts in vitro response to targeted agents of thesepathway points to the utility of cell line models to generatepreclinical hypotheses that can be tested in the clinicalsetting.Our studies also highlight the potential utility of RPPAs

in confirming pathway modulation upon therapeuticintervention and applications in examining pharmacody-namic biomarkers of drug response. Previous studies haveshown that RPPAs could be used to monitor in vitro andin vivo pharmacodynamic response to the Akt inhibitorperifosine and that this inhibitor selectively inhibited thePI3K/AKT axis (41). We have shown here that an inhibitorof all isoforms of the class I catalytic subunit of PI3K,GDC-0941, results in potent and selective inhibition ofmultiple nodes in the PI3K/Akt pathway and, thus, thatRPPA arrays might have utility monitoring surrogatemarkers of compound activity. Conversely, we show that aselective MEK inhibitor results in potent down-regulationof pErk1/2 and actually increases signaling through thePI3K/Akt axis. This result highlights the fact thatsignaling pathways are dynamically linked networks andthat perturbations in one pathway may have unforeseenconsequences on interacting pathways that may affectresponse to therapeutic agents (42). Ideally, monitoring

pathway status in patients undergoing therapy could,thus, form the basis for subsequent intervention or evenrational combination of agents based on the particularbehavior of an individual’s cancer. We also observed thatpathway modulation could be associated with efficacy intwo EGFR inhibitor-treated cell lines because an erlotinibresponsive cell line showed dramatic reduction in pErk1/2whereas a KRAS mutant, nonresponsive cell line failedto show down-regulation of pErk1/2. This is intriguingbecause clinical studies in patients with metastatic non–small cell lung cancer have suggested that KRAS muta-tions are associated with lack of clinical benefit fromerlotinib (43, 44). A simple model for these observationsis that activation of KRAS through oncogenic mutationresults in constitutive signaling through Erk1/2, whichis not blocked by inhibition at the level of the receptor.This result again highlights the potential benefits of usingRPPAs to examine pathway behavior in response totherapeutic agents to make an early decision as to whethera particular patient will benefit from a given treatment.Another appealing facet of using RPPA technology for

molecular diagnostic applications suggested by our resultsis the ability to capture overall pathway activation that canarise from multiple independent activating inputs. Forinstance, activating mutations in p110a, loss of PTEN, andamplification of HER2 all occur in breast cancer (45–47)and all result in constitutive signaling through the PI3K/AKT/mTOR axis. In the absence of a global readout ofpathway activation, multiple independent assays (e.g.,PIK3CA mutations, PTEN IHC, HER2 FISH or IHC) wouldneed to be run on clinical specimens and might still notcapture novel inputs into the pathway. The multiplexnature of RPPAs allows key nodes of the pathway to beexamined in parallel and would greatly streamlinediagnostic analyses of patient material. Of course, anumber of hurdles need to be overcome before thismethod can be successfully used in clinical practice. Chiefamong these hurdles is the fact that tumor biopsies aremost commonly fixed in formalin, a process which resultsin extensive cross-linking of proteins and alters immuno-reactivity. Successful incorporation of RPPAs for analysisof clinical samples will require either collection of flashfrozen biopsy material or the development of novelfixation conditions that leave phosphoepitopes intact.Efforts directed at understanding preanalytic fluctuationsof phosphoproteins during tissue procurement (48) andthe development of novel fixatives have been reported(49) and should portend a bright future for the use ofRPPAs in the clinical setting.Taken together, our findings suggest that proteomic

analysis of cell line models of breast cancer can revealpathway activation portraits that are not apparent by othermeans and have the potential to form the basis forpersonalized administration of targeted agents.

Disclosure of Potential Conflicts of InterestAll authors are employees of Genentech, Inc. The authors disclosed noother potential conflicts of interest.






Acknowledgments

We thank David Dornan, Lisa Belmont, and Elaine Storm for helpfulcomments on the manuscript, Emmanuel Petricoin III and Maarten Leerkesfor useful discussions and advice, and Sedita Lakic for administrativeassistance.

References

1. Carey LA, Dees EC, Sawyer L, et al. The triple negative paradox:primary tumor chemosensitivity of breast cancer subtypes. Clin CancerRes 2007;13:2329–34.

2. Sorlie T, Tibshirani R, Parker J, et al. Repeated observation of breasttumor subtypes in independent gene expression data sets. Proc Natl AcadSci U S A 2003;100:8418–23.

3. Cigler T, Goss PE. Breast cancer adjuvant endocrine therapy. Cancer J2007;13:148–55.

4. Slamon DJ, Leyland-Jones B, Shak S, et al. Use of chemotherapy plus amonoclonal antibody against HER2 for metastatic breast cancer thatoverexpresses HER2. N Engl J Med 2001;344:783–92.

5. Hu Z, Fan C, Oh DS, et al. The molecular portraits of breast tumors areconserved across microarray platforms. BMC Genomics 2006;7:96.

6. Ross JS, Hatzis C, Symmans WF, Pusztai L, Hortobagyi GN.Commercialized multigene predictors of clinical outcome for breast cancer.Oncologist 2008;13:477–93.

7. Shaulian E, Karin M. AP-1 as a regulator of cell life and death. Nat CellBiol 2002;4:E131–6.

8. Cheng C, Pounds S. False discovery rate paradigms for statistical ana-lyses of microarray gene expression data. Bioinformation 2007;1:436–46.

9. Anderson L, Seilhamer J. A comparison of selected mRNA and proteinabundances in human liver. Electrophoresis 1997;18:533–7.

10. Celis JE, Kruhoffer M, Gromova I, et al. Gene expression profiling:monitoring transcription and translation products using DNA microarraysand proteomics. FEBS Lett 2000;480:2–16.

11. Gygi SP, Rist B, Gerber SA, Turecek F, Gelb MH, Aebersold R.Quantitative analysis of complex protein mixtures using isotope-codedaffinity tags. Nat Biotechnol 1999;17:994–9.

12. Nishizuka S, Charboneau L, Young L, et al. Proteomic profiling of theNCI-60 cancer cell lines using new high-density reverse-phase lysatemicroarrays. Proc Natl Acad Sci U S A 2003;100:14229–34.

13. O’Brien C, Cavet G, Pandita A, et al. Functional genomics identifiesABCC3 as a mediator of taxane resistance in HER2-amplified breastcancer. Cancer Res 2008;68:5380–9.

14. Neve RM, Chin K, Fridlyand J, et al. A collection of breast cancer celllines for the study of functionally distinct cancer subtypes. Cancer Cell2006;10:515–27.

15. Mendes KN, Nicorici D, Cogdell D, et al. Analysis of signalingpathways in 90 cancer cell lines by protein lysate array. J Proteome Res2007;6:2753–67.

16. Shankavaram UT, Reinhold WC, Nishizuka S, et al. Transcript andprotein expression profiles of the NCI-60 cancer cell panel: an integromicmicroarray study. Mol Cancer Ther 2007;6:820–32.

17. Espina V, Wulfkuhle J, Calvert VS, Liotta LA, Petricoin EF III. Reversephase protein microarrays for theranostics and patient-tailored therapy.Methods Mol Biol 2008;441:113–28.

18. Wulfkuhle JD, Edmiston KH, Liotta LA, Petricoin EF III. Technologyinsight: pharmacoproteomics for cancer-promises of patient-tailoredmedicine using protein microarrays. Nat Clin Pract 2006;3:256–68.

19. Brown AP, Carlson TC, Loi CM, Graziano MJ. Pharmacodynamic andtoxicokinetic evaluation of the novel MEK inhibitor, PD0325901, in the ratfollowing oral and intravenous administration. Cancer Chemother Pharma-col 2007;59:671–9.

20. Raynaud FI, Eccles S, Clarke PA, et al. Pharmacologic characterizationof a potent inhibitor of class I phosphatidylinositide 3-kinases. Cancer Res2007;67:5840–50.

21. Folkes AJ, Baker SJ, Chuckowree IS, et al. The discovery of GDC-0941: a potent, selective, orally bioavailable inhibitor of class I PI3 kinase forthe treatment of cancer AACR Annual Meeting. Poster. San Diego; 2008.

22. Wang JY, Wilcoxen KM, Nomoto K, Wu S. Recent advances of MEKinhibitors and their clinical progress. Curr TopMedChem2007;7:1364–78.

23. BareschinoMA,SchettinoC, Troiani T,Martinelli E,Morgillo F,CiardielloF. Erlotinib in cancer treatment. Ann Oncol 2007;18 Suppl 6:vi35–41.

24. Eisen MB, Spellman PT, Brown PO, Botstein D. Cluster analysis anddisplay of genome-wide expression patterns. Proc Natl Acad Sci U S A1998;95:14863–8.

25. Steck PA, PershouseMA, Jasser SA, et al. Identification of a candidatetumour suppressor gene, MMAC1, at chromosome 10q23.3 that ismutated in multiple advanced cancers. Nat Genet 1997;15:356–62.

26. Goberdhan DC, Wilson C. PTEN: tumour suppressor, multifunctionalgrowth regulator andmore. HumMol Genet 2003;12 Spec No 2:R239–48.

27. Gingras AC, Gygi SP, Raught B, et al. Regulation of 4E-BP1phosphorylation: a novel two-step mechanism. Genes Dev 1999;13:1422–37.

28. Gingras AC, Raught B, Sonenberg N. Regulation of translationinitiation by FRAP/mTOR. Genes Dev 2001;15:807–26.

29. Shepard HM, Jin P, Slamon DJ, Pirot Z, Maneval DC. Herceptin.Handbook Exp Pharmacol 2008;181:183–219.

30. Bidard FC, Conforti R, Boulet T, Michiels S, Delaloge S, Andre F. Doestriple-negative phenotype accurately identify basal-like tumour? Animmunohistochemical analysis based on 143 ‘‘triple-negative’’ breastcancers. Ann Oncol 2007;18:1285–6.

31. Reis-Filho JS, Tutt AN. Triple negative tumours: a critical review.Histopathology 2008;52:108–18.

32. Weinstein, JA. Oncogene addiction. Cancer Res 2008;68:3077–80.

33. Kohno M, Pouyssegur J. Targeting the ERK signaling pathway incancer therapy. Ann Med 2006;38:200–11.

34. Workman P, Clarke PA, Guillard S, Raynaud FI. Drugging the PI3kinome. Nat Biotechnol 2006;24:794–6.

35. Wulfkuhle JD, Speer R, Pierobon M, et al. Multiplexed cell signalinganalysis of human breast cancer applications for personalized therapy. JProteome Res 2008;7:1508–17.

36. Sarrio D, Rodriguez-Pinilla SM, Hardisson D, Cano A, Moreno-BuenoG, Palacios J. Epithelial-mesenchymal transition in breast cancer relates tothe basal-like phenotype. Cancer Res 2008;68:989–97.

37. Lin J, Arlinghaus R. Activated c-Abl tyrosine kinase in malignant solidtumors. Oncogene 2008;27:4385–91.

38. Dimri M, Naramura M, Duan L, et al. Modeling breast cancer-associated c-Src and EGFR overexpression in human MECs: c-Src andEGFR cooperatively promote aberrant three-dimensional acinar structureand invasive behavior. Cancer Res 2007;67:4164–72.

39. Finn RS, Dering J, Ginther C, et al. Dasatinib, an orally active smallmolecule inhibitor of both the src and abl kinases, selectively inhibitsgrowth of basal-type/‘‘triple-negative’’ breast cancer cell lines growingin vitro . Breast Cancer Res Treat 2007;105:319–26.

40. Huang F, Reeves K, Han X, et al. Identification of candidate molecularmarkers predicting sensitivity in solid tumors to dasatinib: rationale forpatient selection. Cancer Res 2007;67:2226–38.

41. Hennessy BT, Lu Y, Poradosu E, et al. Pharmacodynamic markers ofperifosine efficacy. Clin Cancer Res 2007;13:7421–31.

42. Araujo RP, Liotta LA, Petricoin EF. Proteins, drug targets and themechanisms they control: the simple truth about complex networks. NatRev Drug Discov 2007;6:871–80.

43. Massarelli E, Varella-Garcia M, Tang X, et al. KRAS mutation is animportant predictor of resistance to therapy with epidermal growth factorreceptor tyrosine kinase inhibitors in non-small-cell lung cancer. ClinCancer Res 2007;13:2890–6.

44. Zhu CQ, da Cunha Santos G, Ding K, et al. Role of KRAS and EGFR AsBiomarkers of Response to Erlotinib in National Cancer Institute of CanadaClinical Trials Group Study BR.21. J Clin Oncol 2008;29:641–3.

45. Cantley LC. The role of phosphoinositide 3-kinase in human disease.Harvey Lect 2004;100:103–22.

46. Saal LH, Gruvberger-Saal SK, Persson C, et al. Recurrent grossmutations of the PTEN tumor suppressor gene in breast cancers withdeficient DSB repair. Nat Genet 2008;40:102–7.

47. Yarden Y, Sliwkowski MX. Untangling the ErbB signalling network.Nat Rev 2001;2:127–37.

48. Espina VA, Edmiston KH, Heiby M, et al. A portrait of tissuephosphoprotein stability in the clinical tissue procurement process. MolCell Proteomics 2008;7:1998–2018.

49. Vincek V, Nassiri M, Nadji M, Morales AR. A tissue fixative thatprotects macromolecules (DNA, RNA, and protein) and histomorphology inclinical samples. Lab Invest 2003;83:1427–35.






Published OnlineFirst December 3, 2008.Mol Cancer Ther reverse-phase protein microarraysbiomarkers of response to targeted kinase inhibitors using Proteomic analysis of breast cancer molecular subtypes and

Updated version

10.1158/1535-7163.MCT-08-0810doi:

Access the most recent version of this article at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

Subscriptions

Reprints and

[email protected] at

To order reprints of this article or to subscribe to the journal, contact the AACR Publications

Permissions

Rightslink site. (CCC)Click on "Request Permissions" which will take you to the Copyright Clearance Center's

.http://mct.aacrjournals.org/content/early/2005/11/01/1535-7163.MCT-08-0810.citationTo request permission to re-use all or part of this article, use this link



http://mct.aacrjournals.org/lookup/doi/10.1158/1535-7163.MCT-08-0810

http://mct.aacrjournals.org/cgi/alerts

mailto:[email protected]

http://mct.aacrjournals.org/content/early/2005/11/01/1535-7163.MCT-08-0810.citation


proteomicanalysisofbreastcancermolecularsubtypes ...€¦ · between phosphorylation by kinases,...

Documents