Single cell profiling of potentiated phospho-protein networks in cancer cells

Cell, Vol. 118, 1–20, July 23, 2004, Copyright 2004 by Cell Press

Single Cell Profiling of PotentiatedPhospho-Protein Networks in Cancer Cells

sion include mutations to key signaling proteins as wellas epigenetic changes to gene expression patterns (Ha-nahan and Weinberg, 2000). Cancer genesis occurs in

Jonathan M. Irish,1 Randi Hovland,2,3

Peter O. Krutzik,1 Omar D. Perez,1

Øystein Bruserud,3,4 Bjørn T. Gjertsen,3,4

a stepwise progression, has underlying stochastic ele-and Garry P. Nolan1,*ments, and is reflective of the genetic selection of the1Department of Microbiology & Immunologycancer in the face of both immune system action andBaxter Laboratory of Genetic Pharmacologythe environmental requirements of cancer cells. The sig-Stanford Universitynaling profile of any given cancer cell is therefore theStanford, California 94305sum of numerous influences: epigenetic, genetic, and2 Center for Medical Genetics and Molecularmicroenvironmental. The current molecular understand-Medicineing of cancer signaling rests largely on extrapolationsHaukeland University Hospital and Proteomic Unitfrom studies of cell lines and as such is not adequately(PROBE)representative of the signaling phenotypes of a complexUniversity of Bergenpopulation of cancer cells in the body. In contrast, theBergenheterogeneity of cancer cell responses to therapy canNorwaybe thought of as mirroring the signaling differences that3 Department of Internal Medicinehave arisen during evolution of the cancer cell popula-Haematology Sectiontion in the body. Until now, this cell by cell informationHaukeland University Hospitalon cancer cell populations—required to model signalingBergennetwork pathologies that relate to cancer cell subsetsNorwayand disease progression—has not been available for4 Institute of Medicineanalysis.Haematology Section

Phospho-protein members of signaling cascades,University of Bergenand the kinases and phosphatases that interact withBergenthem, are required to initiate and regulate proliferativeNorwaysignals within cells. It might be predicted that geneticchanges common in cancers, such as receptor tyrosinekinase mutations and other signaling-related cytoge-Summarynetic alterations (Spiekermann et al., 2002; Wheatley etal., 1999), would change the potential of pre-existingAltered growth factor responses in phospho-protein-signaling networks to respond to external stimuli anddriven signaling networks are crucial to cancer celllead to identifiable patterns of signal transduction asso-survival and pathology. Profiles of cancer cell signalingciated with gene mutation. For instance, acute myeloidnetworks might therefore identify mechanisms by whichleukemia (AML) is a cancer wherein dysregulated growthsuch cells interpret environmental cues for continuedand inhibition of apoptosis lead to the accumulation ofgrowth. Using multiparameter flow cytometry, we mon-immature myeloid progenitor cells and oncogenic pro-itored phospho-protein responses to environmentalgression (Lowenberg et al., 1999). Two key parallel signalcues in acute myeloid leukemia at the single cell level.transduction networks active in cells that are consideredBy exposing cancer cell signaling networks to potenti-progenitors of AML (Reya et al., 2001) are the STATating inputs, rather than relying upon the basal levelspathway (Coffer et al., 2000; Smithgall et al., 2000) andof protein phosphorylation alone, we could discernthe Ras/MAPK pathway (Platanias, 2003). Several re-

unique cancer network profiles that correlated withports suggest that STATs, such as Stat3 and Stat5, are

genetics and disease outcome. Strikingly, individual can- constitutively activated in AML (Benekli et al., 2002; Birk-cers manifested multiple cell subsets with unique net- enkamp et al., 2001; Turkson and Jove, 2000; Xia et al.,work profiles, reflecting cancer heterogeneity at the 1998). But, a causal link between basal STAT phosphory-level of signaling response. The results revealed a dra- lation and leukemogenesis in primary patient materialmatic remodeling of signaling networks in cancer has not been demonstrated, despite significant evi-cells. Thus, single cell measurements of phospho-pro- dence implicating these proteins in oncogenic pro-tein responses reveal shifts in signaling potential of a cesses (Benekli et al., 2003; Bowman et al., 2000; Buet-phospho-protein network, allowing for categorizing of tner et al., 2002; Calo et al., 2003; Nieborowska-Skorskacell network phenotypes by multidimensional molecu- et al., 1999). Thought to act upstream of these pathways,lar profiles of signaling. abnormalities of the Flt3 (fms-like tyrosine kinase 3 )

receptor tyrosine kinase are detected in approximatelyIntroduction 30% of AML patients and are well-established as a nega-

tive prognostic indicator in AML (Gilliland and Griffin,Intracellular signaling and interpretation of environmen- 2002; Kottaridis et al., 2001; Thiede et al., 2002). Expres-tal cues play central roles in cancer cell initiation and sion of mutant, activated Flt3 in cell lines has been ob-maintenance. Actions that lead to cancer cell progres- served to activate STAT and Ras/MAPK signaling (Haya-

kawa et al., 2000; Mizuki et al., 2000). However, basallevels of Stat5 phosphorylation have been reported to*Correspondence: [email protected]

Cell2

be statistically indistinguishable in Flt3 mutant and wild- as well as attenuated Stat1 phosphorylation followingtype patient-derived AML blasts (Pallis et al., 2003). The cytokine stimulation. This is consistent with a modeldifferences observed between cell line studies and pri- where AML arises from proliferative signaling in the ab-mary patient samples raises the question of why, in sence of differentiation and apoptosis. Therefore, po-primary cells, there is not a strong linkage between acti- tentiated signaling can reveal relationships in phospho-vating mutations in signaling proteins, increased basal protein networks that are reflective—either causal orphosphorylation of putative targets, and disease state. reactive—of the mutations thought to initiate or sustain

Upon consideration, it could be proposed that it is the growth of a particular cancer phenotype and mightnot the absolute level of the signaling proteins them- be applicable to mapping of signaling changes duringselves (or their basal phosphorylation status) that will normal development.be most informative of cancer signaling processes, butthe manner in which such proteins perform in signalingnetworks. A potentially informative approach to illumi- Resultsnate the structure of signaling networks would be toimpose a signaling input on cancer cells and monitor Phospho-Protein Response Panel Distinguisheschanges in the level of phosphorylation on multiple key Leukemic Signal Transduction Networksnodal proteins. As such, one could measure the effects A cytokine response panel, composed of 36 phospho-of known (and unknown) mutations on signaling events protein states (6 basal states and 30 cytokine re-in cancer cell subtypes. Furthermore, one could detect sponses), was designed to survey altered signal trans-the influence of such mutations in sensitizing or potenti- duction of cancer cells (Figure 1A). Each square in theating the response of the cancer cell to other inputs grid approximates a multidimensional flow cytometryor environmental cues. By providing multiple discrete file that contained at least 30,000 cell events. The cyto-inputs to the signaling system, one is theoretically polar- kine responses of each phospho-protein node wereizing signaling networks to reveal their underlying struc- compared to the basal state (shown in this profile setture(s). Thus, by simultaneously measuring multiple as normalized to black) and represented by calculatingphosphorylation events at the single cell level across a the log2 fold difference in median fluorescence intensitycomplex population of cancer cells, it would be possible (MFI) of stimulated sample divided by unstimulated sam-to build and distinguish pathway anomalies in different ple. Although the data are visually simplified in a re-cell subsets. This could allow for resolution, in different sponse panel, the multiparameter flow cytometry dataindividuals, of those signaling profiles that differentiate are available for further inspection of single cell phos-cancer subtypes and relate to disease progression. We pho-protein responses of interesting subsets, as exam-describe this as interrogating the potentiation of signal- ined in detail later (Figure 5).ing pathways—that is, stimulating a network to reveal The cytokine response panel included detection ofits signaling potential. phosphorylated Stat1, Stat3, Stat5, Stat6, p38, and

A cohort of adult AML patients with characterized Erk1/2. We collected data on unstimulated cells andmutations in Flt3 and known cytogenetic abnormalities cells stimulated for 15 min with Flt3 ligand (FL), GM-(Bruserud et al., 2003) provided us with an opportunity CSF, G-CSF, IL-3, or IFN�. This panel was applied toto compare gene mutations with signal transduction the U937 histiocytic lymphoma cell line (Figure 1A), theprofiles in primary cancer samples. To meet a standard HL-60 AML cell line, and multiple samples of a CD33�

of measuring signaling events simultaneously in a con- subset from normal peripheral blood (Figure 1B). CD33�

trolled context we used intracellular phospho-specificcells represent differentiated cells of myeloid lineage

flow cytometry (Krutzik and Nolan, 2003; Perez and No-and were used to assess variability in cytokine re-

lan, 2002) to detect multiple phosphorylated, activatedsponses among samples from blood donors. In general,signaling molecules in primary leukemic cells drawnthe cytokine responses of normal lymphocyte subsetsfrom this cohort of well-characterized AML patients.differed very little between normal donors (�2 � 0.1,Flow cytometry is useful in a clinical setting as relativelyn � 6).small sample sizes—as few as 10,000 cells collected—

In contrast, the cytokine responses of AML cells var-can produce a considerable amount of statistically trac-ied in numerous ways between patient samples. Thetable multidimensional signaling data and reveal keycytokine response panels for 6/30 representative AMLcell subsets responsible for a phenotype (Krutzik et al.,patient samples are shown (Figure 1C, AML-P012004). We tested the hypothesis that genetic alterationsthrough -P06). For example, although AML-P02 andto signaling genes would cause leukemic cells to react inAML-P03 displayed many similarities, AML-P02 re-an inappropriate or sensitized manner to environmentalsponded to GM-CSF and G-CSF with phosphorylationinputs and that this differential signaling can be readof Stat5 while AML-P03 responded only to GM-CSF atout by single cell flow cytometry.Stat5 (Figure 1B). Additionally, although most patientHere, we apply potentiated signaling to interrogatesamples and controls displayed potent phosphorylationmultiple key nodal phospho-proteins in cancer net-of Stat1 following IFN�, AML-P05 lacked this response.works, map signaling mechanisms that unite cancerRepeat measurements of cell lines, primary cell sam-subtypes, and demonstrate an intimate association be-ples, and AML samples were collected (n � 3) and thetween oncogene mutation, remodeling of signaling net-technique and monoclonal antibodies displayed a highworks, and pathology. The profiles of signal transduc-level of reproducibility similar to previous studies (Krut-tion that correlated with poor response to chemotherapy

showed potentiated Stat5 and Stat3 phosphorylations zik and Nolan, 2003).

Single Cell-Potentiated Networks in Cancer3

Figure 1. A Cytokine Response Panel Reveals Potentiated Signal Transduction Nodes in Primary Acute Myeloid Leukemias

(A) Stimulation states, shown in rows, included unstimulated or 20 ng/ml of FL, GM-CSF, G-CSF, IL-3, or IFN�. Target phosphorylations weredetected using phospho-specific antibodies for Stat1, Stat3, Stat5, Stat6, p38, and Erk1/2, shown in columns. Each square in the grid representsthe response of one phosphorylation site to one condition. The relationship between the grid and the flow cytometry data on which it is basedis diagrammed for U937 cells.(B) Representative cytokine response panels of the HL-60 AML cell line, normal CD33� leukocytes, and six AML patient samples. Repeatexperiments using these AML blasts yielded similar results (n � 3), and variation among normal, healthy donors was minimal (n � 6). Theresponse to stimulation at each signaling node is calculated as log2 (MFI stimulated / MFI unstimulated).

Basal Phosphorylation and Cytokine Responses phosphorylation states, we identified 7 of the 30 cyto-kine response states that displayed significant varianceof STAT and Ras-MAPK Phospho-Proteins Vary

Significantly among Primary AML Patient Samples across AML patient samples (�2 � 0.1, Figure 2B). Thesewere (1) phosphorylation of Stat3 following G-CSF, (2–5)Phospho-protein responses in many of the primary AML

blasts showed considerable induction and variance of phosphorylation of Stat5 following GM-CSF, G-CSF, IL-3,and IFN�, (6) Stat1 phosphorylation following IFN�, andSTAT and RAS/MAP-K pathway member phosphoryla-

tion (Figure 2A). To illustrate the differences in the basal (7) phosphorylation of Erk1/2 following FL. The re-maining cancer cytokine responses displayed variationstate across the AML samples, the first row of each

basal state has been represented according to a heat at or less than the variance in normal, differentiatedCD33� cells (Figure 1B). Including the two closest nodemap relative to the minimum among the 30 AML patient

blasts tested. As can be seen, there is considerable states to the significance threshold did not affect subse-quent patient groupings (Supplemental Figure S1 atvariation across all basal nodes. However, some signal-

ing nodes remain activatable, and, following stimulation, http://www.cell.com/cgi/content/full/118/2/��/DC1).Removing further nodes did affect the clustering, sug-we observed phosphorylation of associated signaling

proteins above the level observed in the basal state. In gesting that this collection of nodes represents the mini-mum set of node states from this analysis for an accurateaddition to the considerable variation of the six basal

Cell4

Figure 2. Basal and Potentiated Signaling Nodes that Varied among Cancer Samples Were Used to Define an AML Biosignature

(A) The cytokine response panel of 30 total AML patient samples. The first row of each has been colored to show the variation in the basalphosphorylation (relative to the minimum among the AML blasts). Of 900 cytokine responses assayed, 93 (10.3%) displayed a detectablephosphorylation increase following stimulation (greater than 0.55-fold on a log2 scale).(B) Significant cytokine responses were restricted primarily to the 7/30 cytokine response nodes with a variance across cancers greater than0.1 (yellow circles).(C) A graph of the absolute median plotted against the variance for each node state indicates the signal-to-noise threshold.

signaling profile. A graph of the absolute median against sis and then subsequently tested for correlation to clini-cal parameters—as opposed to pre-grouping patientsthe variance indicates the signal-to-noise threshold andby clinical outcome and then testing each node statethe relationship of variance to the median basal stateas a hypotheses—the statistical perils of multiple t testsor cytokine response (Figure 2C).can be minimized or avoided (Tilstone, 2003). Thismethod also allows for disease subgroups to be identi-

AML Patients with Similar Signal Transduction fied even when normal control cells are rare or difficultProfiles Display Similar Chemotherapy to obtain in bulk, as is the case with myeloid progenitorResponses, Flt3 Mutations, and cells (Reya et al., 2001), because differences amongCytogenetic Alterations experimental samples are used to define groups ratherWe composed a signal transduction based classification than comparison of each sample to a proposed nor-for AML using unsupervised clustering of phospho-pro- mal control.tein biosignatures (Figure 3A). This was a first step in Unsupervised clustering identified four main groupsdetermining whether underlying networks showed mecha- of AML patients (Figure 3A). Each of these groups wasnistic similarities to each other that corresponded with termed a signaling cluster (SC) and was referred toresponsiveness to apoptotic induction by chemotherapy. based on the primary signaling profile of the cluster: (1)Samples were grouped using the complete linkage hierar- SC-nonpotentiated (NP) had varied basal phosphoryla-chical clustering algorithm available in Multiple Experi- tion but displayed little or no potentiated phosphoryla-ment Viewer (MeV) (http://www.tigr.org/software/tm4/ tion in response to cytokine induction and a low Stat1mev.html). There are several advantages to unsuper- response to IFN� (italic letters denote the cognate acro-vised clustering based on potentiated signaling. When nym), (2) SC-potentiated-basal (PB) displayed both a

potentiated phosphorylation response to cytokine andpatients are clustered according to a biological hypothe-


Figure 3. AML Patients Grouped by Signal Transduction Biosignature Form Four Groups that Exhibit Significant Correlations to ClinicalPrognostic Markers

(A) The 13-parameter biosignatures of differentiated CD33� myeloid cells from six normal blood donors (D01 – D06), U937 and HL-60 cancercell lines, and 30 AML patient samples were grouped according to similarity using hierarchical clustering. The heat map for AML and cancercell line cytokine responses was scaled by subtracting donor sample medians to provide a dynamic color range. As shown previously, basalresponses are relative to the minimum among AML samples.(B) Four main groups of AML patients were identified based on the similarity of their signal transduction biosignatures. We designated thesegroups with signaling cluster (SC) nomenclature based on the signaling that defined them and mapped several clinical markers within theidentified patient groups.

high basal phospho-protein states, (3) SC-potentiated-1 significantly correlated with SC-P2 and Flt3 mutation(p � 0.03 and p � 0.001, respectively, Figure 3B).(P1) displayed intense potentiated signaling, lower basal

phosphorylation, and a p-Stat1 response to IFN�, and In patients with an SC-NP profile, cytogenetic alter-ations were grouped into a closely related branch (Figure(4) SC-potentiated-2 (P2) displayed many potentiated

cytokine responses in the context of low basal phos- 3, hierarchy branch AML-P23 through AML-P30) con-taining two patients with a 9;11 translocation (AML-P27phorylation and a low Stat1 response to IFN�.

Several clinical parameters were nonrandomly distrib- and AML-P24) and two patients with a loss of chromo-some 5 (AML-P23 and AML-P17). Patients with no de-uted among the SCs, and four of these are aligned below

the heat map (Figure 3B). A full chart of known AML tected cytogenetic alterations formed a closely relatedbranch of SC-P2 (Figure 3A, hierarchy branch AML-P13patient characteristics, ordered by SC, is provided for

further analysis (Supplemental Table S1 on the Cell web- through AML-P20). This division of altered or diploidcytogenetics among branches of SC-NP and SC-P2 wassite). The �2 statistical test was used to determine the

significance of the observed correlations between clini- statistically significant (p � 0.001).The signaling profiles and the activities representedcal factors previously determined to be prognostic for

AML and the biosignature-derived SCs. can be viewed in a more conventional network map (seeFigure 6) in an attempt to relate mechanism to signalingResistance to course one chemotherapy correlated

significantly with clustered signaling groups and oc- outcome. With the SC-P2 cluster, a composite pathwayshows the action of multiple basal state activators (incurred primarily in patients with an SC-P2 profile (p �

0.002, Figure 3B). Patients with an SC-P2 profile consti- orange) that induce a “basal high’ status for the targetedphospho-proteins (for SC-P2, only p-Stat5). In addition,tuted 33% of this cohort (10/30) and were primarily de-

fined by potentiated Stat5 and Stat3 myeloid signaling this map shows that one or more of several upstreamactivators can induce higher levels of phosphorylationin the absence of Stat1 phosphorylation following IFN�.

AML patient samples that displayed high myeloid cyto- above the basal phosphorylation level (that nodes are acti-vatable). In contrast, with SC-NP profiles, high basal phos-kine responses usually contained an internal tandem

duplication of Flt3—whereas the samples with a nonpo- phorylation at signaling nodes was observed (p-Erk1/2,p-Stat3, and p-Stat6), but additional phosphorylationtentiated SC profile lacked these mutations (p � 0.02).

In testing the correlation of six cell surface antigens with was not usually observed following stimulation. Thesepathway maps are based on observations of signalingSC groups or Flt3, we noted that loss of CD15/Lewis X

Cell6

Figure 4. Flt3 Mutation in Primary AMLs IsAssociated with Potentiated Myeloid SignalTransduction Nodes

Basal and cytokine response node states ofpatient samples with wild-type or mutant Flt3are shown for the 30 AML samples assayed.Each circle represents the level of STATphosphorylation detected in an individualAML patient sample, grouped according towild-type Flt3 (black) or detected mutant Flt3(yellow). (A) To assess basal phosphorylation,samples were compared to the minimum ob-served among cancers. (B–E) The phosphory-lation of Stat5 detected following GM-CSF,G-CSF, and IL-3 and of Stat3 following G-CSFis shown as a fold increase above basal. (F)Cumulative myeloid cytokine responses, cal-culated by summing individual responses(B–E), were compared in patients with andwithout Flt3 mutations.

by flow cytometry (Figure 5) and are built to represent phorylation of Stat5 following IL-3 correlated with Flt3status, but this was not statistically significant (p � 0.07,signaling clusters (Figure 3). Thus, while individual AML

samples showed differences from the composite repre- Figure 4D). Stat3 phosphorylation following G-CSF, theonly Stat3 biosignature response state displaying signif-sentations (Figure 7 and see Discussion), signaling pro-

file correlations can be clearly identified for groups (Fig- icant variance among AMLs, correlated significantly withFlt3 mutation (p � 0.01, Figure 4E). Other cytokine re-ure 6), related to mechanisms, and used to form new

hypotheses. sponses were not observed to differ in patients with andwithout Flt3 mutation. To represent a combined effectof Flt3 mutation on myeloid signal transduction nodes,AML Blast Cells from Patients with Flt3 Mutations

Display Increased Stat5 and Stat3 we summed these four cytokine response node states.This correlation between myeloid cytokine responsesPhosphorylation in Response

to GM-CSF and G-CSF and Flt3 mutation in AML patient samples was signifi-cant (p � 0.005), and all samples with detectable Flt3We used a t test to assess the significance of the ob-

served correlation between Flt3 mutation and poten- mutations displayed positive myeloid cytokine re-sponses (Figure 4F). Thus, there is a strong linkage be-tiated myeloid signal transduction. The cytokine re-

sponses of patient samples with Flt3 mutations were tween the responsiveness of the cells to this set ofmyeloid stimuli and oncogenic mutation of Flt3, sug-compared with patient samples that had no detectable

receptor tyrosine kinase mutation (Figure 4). We ob- gesting that activating mutations of Flt3 potentiate cer-tain elements of the AML response beyond those whichserved no significant difference in the basal phosphory-

lation of Stat5 in either normal or Flt3 mutant AML sam- are canonically observed.ples (Figure 4A), as had been previously observed inpatient samples (Pallis et al., 2003). Because expression Multidimensional Single Cell Profiling Reveals

Cancer Cell Subsets with Distinctof mutant Flt3 had been demonstrated to have a potenteffect on Stat5 phosphorylation in cell lines, we asked Signal Transduction Profiles

Flow cytometry has the ability to simultaneously mea-whether there was a connection between Flt3 mutationsand potentiated Stat5 and Stat3 phospho-protein sig- sure multiple events per cell and, as such, has allowed

the delineation of phenotypic distinctions of immunenaling.Four of the p-Stat5 and p-Stat3 myeloid cytokine re- and other cell subsets. We aimed to determine whether

data simultaneously collected on two phospho-proteinsponses were identified as displaying significant vari-ance across cancers (Figure 2; p-Stat3 following G-CSF, states (Krutzik et al., 2004) in these AML samples

showed signaling differences that suggested cancersand p-Stat5 following GM-CSF, G-CSF, and IL-3), so wecompared these responses in samples with and without were composed of different subsets of cells with distinct

signaling modes. Two-dimensional contour plots for theFlt3 mutation. Phosphorylation of Stat5 following GM-CSF and G-CSF was significantly potentiated in patient G-CSF response of Stat5 and Stat3 are shown (Figure

5) grouped according to the signaling cluster hierarchysamples with Flt3 mutations (p � 0.04 and 0.02, respec-tively; Figures 4B and 4C). Sample calculations of t val- (which is correlated with the Flt3 mutational status)—for

cognate AML patient samples (see Supplemental Figureues are provided (Supplemental Table S2 online). Phos-


Figure 5. Representative 2D Flow Cytometry Plots of Stat5 and Stat3 Phosphorylation following G-CSF Stimulation in AML Patient Sampleswith Wild-Type Flt3 and Mutant Flt3 (ITD)

Two-dimensional contour plot representations of Stat5 and Stat3 phosphorylation (y and x axis, respectively) in patient samples from SC-NPand SC-P2. Both the level of basal phosphorylation and the response to G-CSF are shown.

S2 for a complete two-dimensional flow cytometry data- relapsed following chemotherapy. In this cohort, detec-tion of an SC-P2-potentiated signaling profile providedset). G-CSF stimulated both Stat3 and Stat5 phosphory-

lation and, to a large extent, this dual signaling took a better prediction of chemotherapy response than thatobtained through identification of Flt3 mutation alone.place simultaneously in individual cells. The linked po-

tentiation of p-Stat5/G-CSF and p-Stat3/G-CSF in indi- Therefore, the signaling profile of samples with detect-able mutations may indicate whether these mutationsvidual primary AML blast cells was most common in

samples with detectable Flt3 mutations (Figure 5B). successfully elicit potentiated signaling, which thesedata suggest would be predictive of aggressive pathol-However, we noted the reproducible presence of single

positive subpopulations in some patients (note AML- ogy. Higher dimensional resolution with additional stim-uli, or groups of stimuli, and downstream phospho-read-P13 and P20 as compared to P11). Thus, significant

cancer heterogeneity at the level of signaling heteroge- outs could provide the additional data that wouldprovide a mechanistic insight to answer this question,neity can be observed.

Four of the SC-NP profiles revealed the presence of as discussed below.a small, induced, p-Stat5/p-Stat3 double-positive sub-population (AML-P08, P23, P26, P30, Figure 5A). This Discussionraises the interesting possibility that the presence of adouble positive, G-CSF-inducible p-Stat5/p-Stat3 sub- We observe that the varied genetic and clinical pheno-

types of leukemic cancers are profoundly linked to dis-population, independent of obvious Flt3 mutational sta-tus, is predictive of an aggressive form of AML that is cernible patterns in phospho-protein signal transduction

networks. This allowed us to map signal transductionnot subject to DNA damage induction of apoptosis. Insome of these patients with a double positive p-Stat5/ mechanisms of cell subsets within individual patient

samples, cluster patient samples based on signalingp-Stat3 response to G-CSF, we did not observe Flt3mutation, though it is possible that additional mutations profiles, and ultimately observe that certain leukemic

signaling profiles of phospho-protein targets were sig-in other regulators phenocopy the effect of the Flt3 mu-tation. Of chemotherapy-treated patients with detect- nificantly linked with mutation of Flt3, cytogenetic

changes, expression of the cell surface marker CD15,able Flt3-ITD mutation and a non-SC-P2 profile (4treated, 3 were too sick to be treated), 4/4 went into and response to chemotherapy. Critical to successful

clustering was the inclusion of the response of signalingremission following course 1 chemotherapy. In contrast,7/7 patients with an SC-P2 profile and Flt3-ITD mutation networks to environmental cues, along with the basal

Cell8

levels of phosphorylation. Such profiling can be used relate with resistance to chemotherapy-induced celldeath. We observed that the response to course oneto (1) understand signaling mechanisms in malignantAML chemotherapy was a relapse in 9/9 treated patientsprogression, (2) describe cancers by molecular pheno-whose samples displayed an SC-P2 profile (Figure 3), atype, and (3) translate this information for clinical appli-pattern characterized by multiple potentiated myeloidcations. Thus, the signaling phenotype associated withsignaling states (p-Stat5, p-Stat3, p-Erk1/2) found ina particular cell context (differentiation state, activationcombination with a low p-Stat1/IFN� response and highor proliferation capacity, mutation, or class of mutations)basal Stat5 phosphorylation (Figure 3). This demon-can be revealed in the pattern of phosphorylations de-strates that signal transduction-based classification oftected when that network is queried to process stimuli.human cancer, a priori of knowledge of clinical outcome,can produce a patient classification that is predictive ofPotentiated Signaling Reflects Underlyingresponse to therapy.Network Differences

Although a common SC-P2 profile emerged, at theFor many years, researchers have attempted to link ac-individual sample level we observed differences intivities of protein(s) to phenotypic outcomes in cancersignaling mechanisms contributing to STAT and Ras-progression and response to chemotherapy. We ob-MAPK potentiation (Figures 7A–7C). For example, AML-served that by stimulating networks of phosphoproteinsP11, AML-P21, and AML-P29 were all profiled as SC-P2,in leukemic cells to respond to environmental cues, wehad detectable Flt3 mutations, and relapsed following“align” these proteins according to the network’s re-chemotherapy. In looking at their individual signalingsponse potential. Thus, in AML cancers with varyingprofiles, we see that AML-P11 had high basal Stat5pathology and disease origin, responses to environmen-phosphorylation in resting cells but also was able total cues grouped AML types according to the underlyingphosphorylate Stat5 following G-CSF, IL-3, and IFN�.performance boundaries of their signal transduction net-Cells from AML-P21 displayed responses in p-Stat5 toworks (whose response is driven in part by the mutationsG-CSF and IFN� but lacked high resting phosphorylationthat occurred during the history of that cancer). As anof Erk1/2 and Stat5. In AML-P29, we detected low basalexample, although AML-P14 and AML-P24 had rela-phosphorylation but observed that Stat5 was phosphor-tively similar patterns of basal STAT and Ras-MAPKylated following numerous stimuli. While Stat5 providedpathway phosphorylation, these samples exhibited verya common link in these signaling systems in that it wasdifferent phospho-protein patterns following cytokinecapable of being activated above basal levels, variationsstimulation (Figure 3A) and had differences in chemo-existed in other nodes (see Figure 7 for details). Whattherapy response, Flt3 mutational status, and cytoge-united this group was that cancers exhibiting a lack ofnetics (Figure 3B and Supplemental Table S1 online).response to chemotherapy (a DNA damage-inducingSo while a profile of basal signaling alone would haveregimen that induces cell death via apoptosis) werefound these two patient samples to be similar, by usingcomposed of cell subsets with common potential tothe response to environmental cues combined with arespond to one or more upstream cytokine activatorsclustering analysis to group similar molecular pheno-and could frequently be detected to have an activating

types, we identified key differences in the signaling net-mutation in a signaling gene (e.g., Flt3-ITD). In common

works of the cancer cells in these two samples.between SC-P2 and SC-NP was that both profiles por-

Each response node in the cluster can be mapped totray at least one upregulated basal state that likely con-

a potential for a phospho-protein to be activated by an tributes to driving cellular division, but the ability to acti-upstream signaling event. Or, the node can be mapped vate further signaling past basal levels was what enabledas a basal state that is activated above background correlation of signaling profiles with a lack of responselevels (shown in orange as an auxiliary input to the sig- to chemotherapy.naling system, Figures 6 and 7). Thus, we see there are In the signaling cluster groups, basal phosphorylationsignificant differences in signaling mechanisms defining differences helped distinguish between subgroups withcomposite groupings of SC-P2 and SC-NP (Figure 6). cytogenetic alterations (branches of SC-NP and SC-P2,The composite SC-P2 profile (which is characterized Figure 3). Patient samples with cytogenetic alterationsprimarily by response to cytokine stimulation above were distinguished from samples with diploid cytoge-basal levels) had high basal Stat5 and lower basal phos- netics in branches of the two main SCs (p � 0.001,phorylation at other nodes and could be potentiated for Figure 3). This suggests that the signal transductionStat5, Stat3, or Erk by one of many cytokines. In strong profile of leukemic blasts reflects the presence or ab-contrast, SC-NP was nonresponsive to upstream stimuli sence of karyotypic abnormalities and that cytogeneticbut had varying basal high levels of Stat3, Stat6, and changes at different loci can lead to similar effects (phe-Erk and was low basal for Stat5. Thus, the primary dis- nocopying) on the signal transduction profile of a cell.tinction between the two groups was that one group In addition, correlation of Flt3 mutation with low CD15/had low basal phosphorylation levels but responded to Lewis X antigen expression (p � 0.001, Figure 3) hadenvironmental cues, while the second group had high not previously been reported. A connection betweeninternal basal phosphorylations at some of these phos- Flt3 signaling and CD15 expression has been observedphoprotein nodes, but these nodes were nonresponsive in models of myeloid cell expansion where FL injectionto upstream environmental inputs such as cytokine stim- following transplantation of bone marrow cells to aulation. SCID-hu mouse resulted in increased relative numbers

Under this hypothesis of potentiated signaling, we of CD33�/CD15 cells (Namikawa et al., 1999).predicted that particular signal transduction profiles, As reported by others (Pallis et al., 2003), we observed

no significant difference in basal phosphorylation ofsuch as niche-specific, pro-survival patterns, would cor-


Figure 6. Cancer Biosignatures and a Potentiated Model of Cancer Cell Signaling

(A) A general method for identifying a cancer biosignature is shown using an example of STAT and Ras/MAPK signaling node states in AML.(B) Composite maps of cancer networks from profiles SC-NP and SC-P2 were built out of common signaling events observed in each cluster.Highlighted nodes were detected to be high basal or potentiated in most of the samples from a profile group.

Stat5 between normal and Flt3 mutant AML patients samples—compared with cell lines or primary cell sub-sets (Figure 1)—shows the existence of multiple unique(Figure 5). The results suggest that the presence of Flt3-

ITD is linked with potentiatiation or sensitization of the cancer subsets based on signaling potential within indi-vidual AML patients. Thus, the well-recognized limita-phosphorylation of Stat5, and possibly Stat3, in re-

sponse to the myeloid cytokines GM-CSF and G-CSF tion for understanding cancer microheterogeneity mightbe overcome using multiparameter approaches that dis-(Figures 4 and 5). The clustering of Flt3 mutational status

with biosignature profiles supports the hypothesis that tinguish cancer cell subset responses to input stimuli.For instance, although Stat3 and Stat5 are both thoughtgene mutations conferring proliferative advantages pro-

duce detectable signatures in the mitogenic signaling to be constitutively active in leukemias and share manystructural similarities, in SC-P2, the Flt3-correlated po-network of cancer cells (Figures 6 and 7).tentiation of the GM-CSF response only occurred withStat5 (data not shown, see Supplemental Figure S2);Cancer Heterogeneity at the Level

of Signaling Response Stat3 showed no response to GM-CSF in any of thetested AML patient samples (Figures 2 and 5). In con-A critical advantage of single cell based biosignatures

is that, as shown above, they approximate complex data trast, in a given cell in SC-P2, the phosphorylation ofboth Stat5 and Stat3 generally occurred simultaneouslywell enough to allow biologically significant clustering

while retaining the underlying flow cytometric data for following G-CSF stimulation (Figure 5). Interestingly,multiple cancer subpopulations could be observed atfurther examination. We had also examined the original

2D plots of the phospho-protein expression profiles and the level of the signaling response. For example, AML-P13 and AML-P20 each had a subpopulation of cellsobserved that higher dimensional representations of the

data identified subsets of cells with unique signaling fea- that were only p-Stat5 positive in response to G-CSF.However, AML-P11 contained a subpopulation of cellstures.

Despite being composed of �95% leukemic blasts, that were p-Stat3 positive in the basal state but unre-sponsive to the environmental cues applied in thesethe heterogeneity of cytokine responses among AML

Cell10

Figure 7. Three AML Patients Profiled as SC-P2 Showed Similarities and Differences in Potentiated Signaling Mechanisms

Pathway maps summarizing the signaling phenotype of individual patient samples are shown for three profiled as SC-P2, as per Figure 6.Differences and similarities are highlighted in the text.

experiments. Whether these subset differences repre- tems are directly linked to apoptotic response and otherkey mutations in signaling proteins such as Bcl-2 andsent simply nonresponsive cells, cancer “stem” cells

(Reya et al., 2001), or other stages of clonal evolution p53 (J.I. and G.P.N., unpublished data).within the cancer remains to be determined.

The linkage observed between dual responses and SummaryThis study of signaling mechanisms in primary cancerFlt3 status suggests strongly that Flt3 mutation can, in

significant subset(s) of cells as observed in two-dimen- samples indicates that an activated phospho-proteomecan be informative of cancer pathology. The finding thatsional flow cytometry, lead to unexpected coactivation

of important phospho-protein transcription factors. Does potentiated cytokine responses of these cancers canbe combined with information on basal phosphorylationthis represent a requirement during selective pressure

that these cells maintain a low basal state of phosphory- states and clustered to create a classification with clini-cal relevance helps explain mechanism of signalinglation and then respond to appropriate stimuli for contin-

ued cell division? Do the basal high states, combined functions and might inform the design of mechanism-based cancer therapies. In a broader context, the ap-with nonresponsiveness to cytokine induction, of the

SC-NP cancer cells represent a different “solution” that proach allows a distinction of cell types based on signal-ing profiles in the absence of other specific knowledgeallows for the maintenance of cell division of the cancer?

These two cancer escape strategies (distinguished by their about the differentiation state or pathology of the cell. Inthis regard, we have observed examples of potentiatednetwork profiles) are differentially responsive to chemo-

therapeutic induction—and might allow us to use this infor- signaling in HIV-1 infection (J. Fortin and G.P.N., unpub-lished data) and in autoimmune systemic lupus erythe-mation to design therapies to attack the aggressive cancer

types based on the signaling profile that they present. matosus (P. Krutzik, M. Hale, and G.P.N., unpublisheddata). While other approaches for measuring multipleExtraction of additional high-dimensional information

could be applied to further refine our understanding of events exist, they are usually carried out on “averaged”cell populations (gene chips or high throughput proteo-systems such as those presented here, especially when

additional phospho-proteins or other single cell events mics), whereas multiparameter flow cytometry mea-sures the immediate response of signaling networks atare measured simultaneously (Krutzik et al., 2004; Krut-

zik and Nolan, 2003). This could provide the additional the level of individual cells in complex populations. Mea-suring the resting state of signaling proteins, and howinformation required to understand whether these sys-


were compared by calculating the log2 ratio of each sample’s MFIsuch proteins align in signaling networks to single ordivided by the minimum among cancers. To determine the statisticalmultiple inputs, in correlated datasets at the single cellsignificance of observed versus expected distributions, we used alevel therefore provides a fuller understanding of how�2 test (Figure 3B), and to determine the significance of the difference

cells process information in normal and diseased states. in mean of two populations, we used a student’s t test ( � 0.05,Figure 4). Standard t tests were used to validate existing hypotheses

Experimental Procedures (Kiyoi et al., 2002; Mizuki et al., 2003).

Patients and Preparation of AML Blasts Derivation of Signaling Pathway MapsThe study was approved by the local Ethics Committee and samples Pathways considered important to the signaling profiles of SC-NPcollected after informed consent. Samples were selected from a and SC-P2 were identified and highlighted in a model of the probedlarge group of consecutive patients with de novo AML and high network. For an initial appraisal, the following rules were used. Basalperipheral blood blast counts (Bruserud et al., 2003). These patients nodes were considered “basal high” when more than half the pa-were admitted to the hospital from April 1999 to August 2003, the tients in a signaling cluster had signaling above the median formedian age was 60, and ages ranged from 29 to 84. As these patients cancers. Similarly, the responsiveness of a node state to an environ-were selected for high blast counts, enriched AML cell populations mental cue was highlighted when more than half of the patient’scontaining �95% cancer cells were prepared using a simple density node states in a profile group were above the median for cancers.gradient separation of peripheral blood samples (Ficoll-Hypaque; Other node states were dimmed. Upon examination, the Stat1 re-NyCoMed, Oslo, Norway; specific density 1.077) before safe, stan- sponse to IFN� was dimmed for both SC-NP and SC-P2, as theredardized cryopreservation according to previously developed tech- was an attenuation of this signaling response, relative to donor PBLniques (Bruserud et al., 2000). These patients represent the latter controls, in both of the main signaling groups. The basal state ofportion of a group studied previously for Flt3 signaling and mutation p38 and Stat1 were high in SC-NP, but were not observed abovein AML (Bruserud et al., 2003). The acute myeloid leukemia cell line the median frequently enough to highlight their activity. SignalingHL-60 and the monocytoid lymphoma cell line U-937 were from pathway maps were drawn based on multiple sources (McCubreyAmerican Type Culture Collection (www.atcc.org) and cultured in et al., 2000; Rane and Reddy, 2002; Shuai and Liu, 2003; StirewaltRPMI medium (Invitrogen, Carlsbad, California) � 10% Fetal Calf and Radich, 2003).Serum (HyClone, South Logan, Utah).

AcknowledgmentsStimulation of AML BlastsPeripheral blood containing �95% AML blasts was thawed into 5

We thank R. Tibshirani for insights on statistical approaches, O.ml Stem Span H3000 defined, serum-free medium (Stem Cell Tech-

Vintermyr and A. Molven for making available c-kit and N-ras muta-nologies, Vancouver, BC, Canada), counted, pelleted, and resus-

tional analysis, A. Eliassen for assistance on Flt3 mutational analysis,pended at 2 � 106 cells per ml. Six FACS tubes (Falcon 2052, BD-

N. Anensen for many useful discussions, R. Smith, R. Wolkowicz,Biosciences, San Jose, California) were then filled with 2 ml of each

R. Ihrie, and P. Yam for review of this manuscript, and K. Vang forleukemia sample and allowed to rest at 37�C for 2 hr. AML blasts

administrative assistance. This work was supported by Nationalwere resuspended gently to prevent aggregation and allowed to

Heart, Lung and Blood Proteomics Center Contract N01-HV-28183I.rest at 37�C for another 45 min. At this time, vehicle (deficient RPMI

We gratefully acknowledge the support of BD Biosciences-Phar-medium, Invitrogen) or 40 l of stimulus was added to each tube

mingen, and the Baxter Foundation. B.T.G. was supported by theto a final concentration of 20 ng/ml. Stimuli included human recom-

Norwegian Research Council Functional Genomics Program (FUGE)binant Flt3 ligand (FL), GM-CSF, G-CSF, IL-3, and IFN� (all cytokines

grant number 151859 and the Norwegian Cancer Society (DNK).from Peprotech, Inc., Rocky Hill, New Jersey). Samples were re-

R.H. and Ø.B. were supported by Helse Vest HF research grants.turned to the 37�C incubator for 15 min to allow signal transduction

P.O.K. was supported by a Howard Hughes Medical Institute Pre-and phosphorylation, after which 100 l of 32% para-formaldehyde

doctoral Fellowship. J.M.I. was supported by the James H. Clark(PFA, Electron Microscopy Services Fort Washington, Pennsylvania)

Stanford Graduate Fellowship and the J.G. Lieberman Fellowship.was added to each 2 ml tube of cells to a final concentration of1.6%. Cells were fixed for 15 min at room temperature, pelleted,

Received: January 10, 2004then permeabilized by resuspension in 2 ml ice cold methanol forRevised: May 28, 200410 min, and stored at 4�C until being stained for flow cytometry.Accepted: May 28, 2004Published: July 22, 2004Intracellular Phospho-Specific Flow Cytometry

PFA fixed, methanol permeabilized AML blasts were rehydrated byReferencesadding 2 ml phosphate-buffered saline (PBS), gentle resuspension,

and then centrifugation. The cell pellet was washed once with 2 mlBenekli, M., Xia, Z., Donohue, K.A., Ford, L.A., Pixley, L.A., Baer,PBS, resuspended in 150 l PBS � 0.1% BSA (Sigma, St. Louis,M.R., Baumann, H., and Wetzler, M. (2002). Constitutive activity ofMissouri), and split evenly into three new FACS tubes. Fifty microli-signal transducer and activator of transcription 3 protein in acuteters of an antibody mix containing 0.065 g primary conjugatedmyeloid leukemia blasts is associated with short disease-free sur-phospho-specific antibody per sample was added to each tube ofvival. Blood 99, 252–257.AML blasts and staining proceeded for 20 min at room temperature.

Alexa (Ax) dye (Molecular Probes, Eugene, Oregon) conjugated anti- Benekli, M., Baer, M.R., Baumann, H., and Wetzler, M. (2003). Signaltransducer and activator of transcription proteins in leukemias.bodies (all from BD-Pharmingen, San Diego, California) included

antibodies against phospho-Stat3(Y705)-Ax488, phospho-Stat5 Blood 101, 2940–2954.(Y694)-Ax647, phospho-Stat6(Y641)-Ax488, phospho-Stat1(Y701)- Birkenkamp, K.U., Geugien, M., Lemmink, H.H., Kruijer, W., andAx647, phospho-p38(T180/Y182)-Ax488, and phospho-Erk1/2(T202/ Vellenga, E. (2001). Regulation of constitutive STAT5 phosphoryla-Y204)-Ax647 and were applied in pairs (Stat5/Stat3, Stat1/Stat6, tion in acute myeloid leukemia blasts. Leukemia 15, 1923–1931.and Erk/p38) for subsequent detection of Ax488 and Ax647 on FL-1

Bowman, T., Garcia, R., Turkson, J., and Jove, R. (2000). STATs inand FL-4, respectively. Approximately 30,000 ungated events were

oncogenesis. Oncogene 19, 2474–2488.collected for each sample on a benchtop FACSCalibur dual-laser

Bruserud, O., Gjertsen, B.T., and von Volkman, H.L. (2000). In vitrocytometer (Becton Dickinson, Franklin Lakes, New Jersey).culture of human acute myelogenous leukemia (AML) cells in serum-free media: studies of native AML blasts and AML cell lines. J.Statistical AnalysisHematother. Stem Cell Res. 9, 923–932.Changes in phosphorylation of STAT and Ras-MAPK proteins fol-

lowing cytokine stimulation were approximated by calculating the Bruserud, O., Hovland, R., Wergeland, L., Huang, T.S., and Gjertsen,B.T. (2003). Flt3-mediated signaling in human acute myelogenouslog2 ratio of median fluorescence intensity (MFI) of stimulated versus

unstimulated cell populations. Differences in basal phosphorylation leukemia (AML) blasts: a functional characterization of Flt3-ligand

Cell12

effects in AML cell populations with and without genetic Flt3 abnor- multiple active kinase states using polychromatic flow cytometry.Nat. Biotechnol. 20, 155–162.malities. Haematologica 88, 416–428.

Platanias, L.C. (2003). Map kinase signaling pathways and hemato-Buettner, R., Mora, L.B., and Jove, R. (2002). Activated STAT signal-logic malignancies. Blood 101, 4667–4679.ing in human tumors provides novel molecular targets for therapeu-

tic intervention. Clin. Cancer Res. 8, 945–954. Rane, S.G., and Reddy, E.P. (2002). JAKs, STATs and Src kinasesin hematopoiesis. Oncogene 21, 3334–3358.Calo, V., Migliavacca, M., Bazan, V., Macaluso, M., Buscemi, M.,

Gebbia, N., and Russo, A. (2003). STAT proteins: from normal control Reya, T., Morrison, S.J., Clarke, M.F., and Weissman, I.L. (2001).of cellular events to tumorigenesis. J. Cell. Physiol. 197, 157–168. Stem cells, cancer, and cancer stem cells. Nature 414, 105–111.

Shuai, K., and Liu, B. (2003). Regulation of JAK-STAT signalling inCoffer, P.J., Koenderman, L., and de Groot, R.P. (2000). The role ofthe immune system. Nat. Rev. Immunol. 3, 900–911.STATs in myeloid differentiation and leukemia. Oncogene 19, 2511–

2522. Smithgall, T.E., Briggs, S.D., Schreiner, S., Lerner, E.C., Cheng, H.,and Wilson, M.B. (2000). Control of myeloid differentiation and sur-Gilliland, D.G., and Griffin, J.D. (2002). The roles of FLT3 in hemato-vival by Stats. Oncogene 19, 2612–2618.poiesis and leukemia. Blood 100, 1532–1542.Spiekermann, K., Pau, M., Schwab, R., Schmieja, K., Franzrahe, S.,Hanahan, D., and Weinberg, R.A. (2000). The hallmarks of cancer.and Hiddemann, W. (2002). Constitutive activation of STAT3 andCell 100, 57–70.STAT5 is induced by leukemic fusion proteins with protein tyrosine

Hayakawa, F., Towatari, M., Kiyoi, H., Tanimoto, M., Kitamura, T.,kinase activity and is sufficient for transformation of hematopoietic

Saito, H., and Naoe, T. (2000). Tandem-duplicated Flt3 constitutivelyprecursor cells. Exp. Hematol. 30, 262–271.

activates STAT5 and MAP kinase and introduces autonomous cellStirewalt, D.L., and Radich, J.P. (2003). The role of FLT3 in haemato-growth in IL-3-dependent cell lines. Oncogene 19, 624–631.poietic malignancies. Nat. Rev. Cancer 3, 650–665.

Kiyoi, H., Ohno, R., Ueda, R., Saito, H., and Naoe, T. (2002). Mecha-Thiede, C., Steudel, C., Mohr, B., Schaich, M., Schakel, U., Platz-nism of constitutive activation of FLT3 with internal tandem duplica-becker, U., Wermke, M., Bornhauser, M., Ritter, M., Neubauer, A.,tion in the juxtamembrane domain. Oncogene 21, 2555–2563.et al. (2002). Analysis of FLT3-activating mutations in 979 patients

Kottaridis, P.D., Gale, R.E., Frew, M.E., Harrison, G., Langabeer, with acute myelogenous leukemia: association with FAB subtypesS.E., Belton, A.A., Walker, H., Wheatley, K., Bowen, D.T., Burnett, and identification of subgroups with poor prognosis. Blood 99,A.K., et al. (2001). The presence of a FLT3 internal tandem duplica- 4326–4335.tion in patients with acute myeloid leukemia (AML) adds important

Tilstone, C. (2003). DNA microarrays: vital statistics. Nature 424,prognostic information to cytogenetic risk group and response to

610–612.the first cycle of chemotherapy: analysis of 854 patients from the

Turkson, J., and Jove, R. (2000). STAT proteins: novel molecularUnited Kingdom Medical Research Council AML 10 and 12 trials.targets for cancer drug discovery. Oncogene 19, 6613–6626.Blood 98, 1752–1759.Wheatley, K., Burnett, A.K., Goldstone, A.H., Gray, R.G., Hann, I.M.,Krutzik, P.O., and Nolan, G.P. (2003). Intracellular phospho-proteinHarrison, C.J., Rees, J.K., Stevens, R.F., and Walker, H. (1999). Astaining techniques for flow cytometry: Monitoring single cell signal-simple, robust, validated and highly predictive index for the determi-ing events. Cytometry 55A, 61–70.nation of risk-directed therapy in acute myeloid leukaemia derived

Krutzik, P.O., Irish, J.M., Nolan, G.P., and Perez, O.D. (2004). Analysis from the MRC AML 10 trial. United Kingdom Medical Researchof protein phosphorylation and cellular signaling events by flow Council’s Adult and Childhood Leukaemia Working Parties. Br. J.cytometry: techniques and clinical applications. Clin. Immunol. Haematol. 107, 69–79.110, 206–221.

Xia, Z., Baer, M.R., Block, A.W., Baumann, H., and Wetzler, M. (1998).Lowenberg, B., Downing, J.R., and Burnett, A. (1999). Acute myeloid Expression of signal transducers and activators of transcription pro-leukemia. N. Engl. J. Med. 341, 1051–1062. teins in acute myeloid leukemia blasts. Cancer Res. 58, 3173–3180.McCubrey, J.A., May, W.S., Duronio, V., and Mufson, A. (2000). Ser-ine/threonine phosphorylation in cytokine signal transduction. Leu-kemia 14, 9–21.

Mizuki, M., Fenski, R., Halfter, H., Matsumura, I., Schmidt, R., Muller,C., Gruning, W., Kratz-Albers, K., Serve, S., Steur, C., et al. (2000).Flt3 mutations from patients with acute myeloid leukemia inducetransformation of 32D cells mediated by the Ras and STAT5 path-ways. Blood 96, 3907–3914.

Mizuki, M., Schwable, J., Steur, C., Choudhary, C., Agrawal, S.,Sargin, B., Steffen, B., Matsumura, I., Kanakura, Y., Bohmer, F.D.,et al. (2003). Suppression of myeloid transcription factors and induc-tion of STAT response genes by AML-specific Flt3 mutations. Blood101, 3164–3173.

Namikawa, R., Muench, M.O., Firpo, M.T., Humeau, L., Xu, Y.,Menon, S., and Roncarolo, M.G. (1999). Administration of Flk2/Flt3ligand induces expansion of human high-proliferative potential col-ony-forming cells in the SCID-hu mouse. Exp. Hematol. 27, 1029–1037.

Nieborowska-Skorska, M., Wasik, M.A., Slupianek, A., Salomoni, P.,Kitamura, T., Calabretta, B., and Skorski, T. (1999). Signal transducerand activator of transcription (STAT)5 activation by BCR/ABL isdependent on intact Src homology (SH)3 and SH2 domains of BCR/ABL and is required for leukemogenesis. J. Exp. Med. 189, 1229–1242.

Pallis, M., Seedhouse, C., Grundy, M., and Russell, N. (2003). Flowcytometric measurement of phosphorylated STAT5 in AML: lack ofspecific association with FLT3 internal tandem duplications. Leuk.Res. 27, 803–805.

Perez, O.D., and Nolan, G.P. (2002). Simultaneous measurement of

Supplemental Table 1: AML patient characteristics. A B C D E F P# Age Sex Prev.

Disease FAB CD11 CD14 CD15 CD33 CD34 FLT3 N-RAS C-KIT Cytogenetics Resist

C1 5 33 M M5 + - - + - ITD wt wt normal - 7 43 M M2 + - + + - wt wt wt normal NT

8 61 F Ca. mamm.

M4-5 + + + + - wt wt wt normal +

1 44 F Ca. ovarii M4 n.d. - - + - wt wt wt normal -

10 82 M M5 + + + + - wt wt wt normal + 23 77 F M2 + - - + + wt wt wt 42-46XX, -5, der(5) +

26 65 M M2 + - - - + wt wt wt t(13;15)(q10;q10), -17, +21, +22 NT

27 56 F M5a - - + + n.d. wt wt wt t(9;11)(p22;q23), +8 - 24 36 F NF M5 + - + + - wt wt wt t(9;11)(p21;q23) -

17 78 F M4 + + + + - wt wt wt 47-49XX, -4, -5, +der(8) NT

30 84 M MDS M5 + - - + + ITD exon1 wt n.d. NT 3 34 F M5a - - + + - KD wt wt t(9;11), (p22;q23) - 16 82 M M5 - - + + - wt wt wt 45X NT 6 61 M M4 - - + + - ITD exon1 wt normal - 2 82 M M1 - - n.d. + + wt wt wt 47XY, +19 NT 12 51 F M2 - + - + - ITD,wt- wt wt normal -

19 79 F M5 + - + + n.d. KD wt wt del(12)(p11), add(11)(p15) NT

9 69 M M2 + - n.d. + - wt wt exon8 inv(16) - 18 74 M M1 - - + + + wt wt wt 90-94XXYY NT 22 79 M M1 + - - - + ITD wt wt normal NT 13 58 F M2 + - - + - ITD wt wt normal + 14 63 F M4 + - - + + ITD wt wt normal +

11 29 M M4 + - - + + ITD, KD wt wt normal +

4 64 F M4 + - - + + ITD, wt- wt wt normal +

20 56 M M2 + - - + - ITD wt wt normal + 15 59 F M2 + - - + + wt wt wt -7 + 28 30 M M3 + - - + - ITD n.d n.d. t(15;17)(q22;q21) NT 25 36 M CML M2 + n.d. n.d. + + wt wt wt der(9)(ins22q12;q34) + 21 33 M M1 + - - + + ITD wt wt 47XX, +4 + 29 44 F M1 + - - + + ITD,wt- wt wt del(7)(q22) +

A, For membrane molecule expression, patients were regarded as positive when more than 20% of blasts stained positiveas judged by flow cytometric analysis. All AML populations were negative for T-lymphocyte (CD2, CD3) and B lymphocyte (CD19, CD20) markers; B, For FLT3 the samples were either considered wild type (wt) or tested positive for internal tandem duplications (ITD), loss of a wild type allele (wt-), or Asp835 point mutations in the kinase domain (KD). C, N-RAS was identified to have point mutations in exon 1 (Gly12 or Gly13) in two patients. D, c-kit was identified to have a deletion as well as a point mutation in exon 8 of AML-P09. E, Karyotypes are listed and were scored according to Wheatley for Figure 3. F, Not all patients were treated with course 1 of chemotherapy (C1). Non-treated (NT) patients were excluded from thestatistical analysis shown in Figure 3b. For those that received C1, remission is scored as a "-" and relapse is scored as a "+". For some parameters no data (n.d.) were available.

Supplementary Table 2 - Statistical tests reported in Figure 4

Test Correlation of p-

Stat5/GM-CSF with Flt3 mutation

Test Correlation of p-Stat3 / G-CSF with Flt3 mutation Test

Correlation of p-Stat3/5/myeloid w. Flt3

mut. Test

Correlation between basal p-Stat5 and Flt3

mutation** Test Correlation of p-Stat5 /

G-CSF node w. Flt3 mut.

t-value (absolute) 2.129 t-value

(absolute) 2.931 t-value (absolute) 3.076 t-value

(absolute) 0.217 t-value (absolute) 2.563

p-value 0.0425648 p-value 0.0073015 p-value 0.0047619 p-value 0.7808521 p-value 0.0177399

Data Enumeration

Fold increase in p-Stat5 following GM-CSF

Data Enumeration

Fold increase in p-Stat3 following G-CSF

Data Enumeration

Summed p-Stat3/5 node responses**

Data Enumeration Basal p-Stat5 Data

Enumeration Fold increase in p-Stat5

following G-CSF

Patient Flt3 mutant

Flt3 wild type Patient Flt3

mutant Flt3 wild

type Patient Flt3 mutant

Flt3 wild type Patient Flt3

mutant Flt3 wild

type Patient Flt3 mutant

Flt3 wild type

AML-P01 -0.0248 AML-P01 0.0394 AML-P01 -0.1005 AML-P01 0.0772 AML-P01 -0.1286

AML-P02 0.6743 AML-P02 0.6750 AML-P02 2.3346 AML-P02 0.8568 AML-P02 0.6224



AML-P05 -0.0790 AML-P05 0.1945 AML-P05 0.1145 AML-P05 -0.1811 AML-P05 0.0780


AML-P07 -0.1040 AML-P07 -0.0127 AML-P07 -0.3249 AML-P07 0.8568 AML-P07 -0.0913


AML-P09* AML-P09* AML-P09* na AML-P09* AML-P09* na

AML-P10 -0.1175 AML-P10 -0.1814 AML-P10 -0.6249 AML-P10 -0.1811 AML-P10 -0.1959


AML-P12 0.2596 AML-P12 0.2465 AML-P12 1.1164 AML-P12 1.2066 AML-P12 -0.1294




AML-P16 0.4672 AML-P16 -0.0393 AML-P16 1.6480 AML-P16 0.5450 AML-P16 0.0390

AML-P17 0.2479 AML-P17 0.1821 AML-P17 0.6654 AML-P17 -0.3511 AML-P17 0.0524




AML-P21 -0.0516 AML-P21 1.7000 AML-P21 2.5326 AML-P21 -0.6104 AML-P21 1.0646

AML-P22 -0.0649 AML-P22 1.1431 AML-P22 1.5458 AML-P22 0.0649 AML-P22 0.4801

AML-P23 -0.1939 AML-P23 -0.1685 AML-P23 -0.8045 AML-P23 -0.2464 AML-P23 -0.2600

AML-P24 0.0778 AML-P24 0.0000 AML-P24 -0.1159 AML-P24 -0.0649 AML-P24 -0.1679

AML-P25 -0.5320 AML-P25 0.0383 AML-P25 -0.9353 AML-P25 1.7000 AML-P25 -0.3377

AML-P26 -0.1310 AML-P26 0.0515 AML-P26 -0.3398 AML-P26 -0.7394 AML-P26 0.0120

AML-P27 0.0764 AML-P27 -0.1428 AML-P27 0.1383 AML-P27 -0.9078 AML-P27 0.1804




Average 0.332 0.077 Average 0.726 0.179 Average 2.362 0.604 Average 0.230 0.180 Average 0.505 0.064

Variance 0.105 0.103 Variance 0.368 0.144 Variance 2.533 2.208 Variance 0.273 0.493 Variance 0.340 0.098

Num Samples 15 14 Num Samples 15 14 Num Samples 15 14 Num Samples 15 14 Num Samples 15 14

Std Error 0.120 Std Error 0.187 Std Error 0.572 Std Error 0.231 Std Error 0.172

df 27 df 27 df 27 df 27 df 27

* P09 excluded due to c-kit mutation * P09 excluded due to c-kit mutation ** p-Stat3/G-CSF, p-Stat5/GM-CSF/G-CSF/IL-3

** Hypothesis rejected due lack of significant correlation * P09 excluded due to c-kit mutation

* P09 excluded due to c-kit mutation * P09 excluded due to c-kit mutation

Supplementary Table 3 - Values of all 36 node states for all 30 AML patient samples.*

Name Description GWEIGH

T P01 P02 P03 P04 P05 P06 P07 P08 P09 P10 P11 P12 P13 P14 P15 P16 P17 P18 P19 P20 P21 P22 P23 P24 P25 P26 P27 P28 P29 P30

Rank

Variance EWEIGHT 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

1 1.716 p-Stat1-IFNg biosignat

ure 1 -

1.89 -

0.16 1.20 -

0.14 -

2.44 -

0.37 -

2.43 -

1.07 0.45 -

2.20 -

0.81 0.27 -

0.60 0.32 -

0.75 1.41 -

1.78 1.50 1.63 -

0.10 -

0.76 1.68 -

2.19 -

1.30 0.21 -

2.65 -

2.60 -

1.00 -

1.15 -

1.72

2 0.627 p-Stat5-IL-3 biosignat

ure 1 -

0.12 0.23 0.64 0.92 -

0.21 0.36 -

0.25 -

0.12 -

0.08 -

0.26 0.73 0.61 1.70 0.54 2.76 1.05 0.05 0.01 1.76 1.14 -

0.31 -

0.14 -

0.31 -

0.16 -

0.23 -

0.40 -

0.11 1.91 0.75 -

0.38

3 0.551 p-Stat6-basal biosignat

ure 1 1.31 1.08 2.57 0.79 2.58 1.39 3.44 0.70 0.79 0.76 0.95 1.35 0.53 1.38 1.09 2.11 1.53 0.95 0.69 0.95 0.00 0.38 1.66 1.50 0.66 1.63 2.19 1.79 0.39 1.82


ure 1 1.19 1.67 2.80 0.88 2.13 1.88 3.19 2.04 1.23 0.70 1.01 1.83 0.82 1.28 1.03 2.63 1.96 1.76 1.21 1.13 0.00 0.79 1.53 1.91 1.23 1.48 1.89 1.79 0.62 1.89

5 0.450 p-Erk1/2-

basal biosignat

ure 1 2.12 2.05 2.13 1.52 2.31 2.38 2.25 1.52 0.92 1.42 1.53 1.31 0.79 0.79 1.60 1.45 0.85 0.66 0.92 1.94 0.44 1.12 0.42 0.31 0.47 0.00 0.70 0.74 0.25 1.13

6 0.372 p-Stat3-G-

CSF biosignat

ure 1 0.03 0.66 0.03 0.64 0.18 0.13 -

0.03 0.43 1.88 -

0.19 0.74 0.23 1.45 0.56 0.45 -

0.05 0.17 1.14 1.84 0.36 1.69 1.13 -

0.18 -

0.01 0.03 0.04 -

0.16 0.21 1.23 0.27


ure 1 0.99 1.76 1.27 1.08 0.73 1.28 1.76 1.18 0.74 0.73 1.23 2.11 2.24 1.31 1.63 1.45 0.56 0.88 0.39 1.09 0.30 0.97 0.66 0.84 2.61 0.17 0.00 1.19 1.02 0.84

8 0.328 p-Stat5-IFNg biosignat

ure 1 -

0.30 0.07 0.38 0.45 -

0.06 0.84 -

0.33 0.02 0.33 -

0.11 0.23 -

0.11 0.15 0.23 0.73 1.07 0.07 1.47 2.17 0.37 0.23 1.49 -

0.33 0.46 0.08 -

0.32 0.02 0.56 0.53 0.20

9 0.289 p-Stat5-G-

CSF biosignat

ure 1 -

0.05 0.70 0.20 0.55 0.16 0.26 -

0.01 0.17 1.49 -

0.12 0.49 -

0.05 0.83 0.60 0.47 0.12 0.13 0.77 2.32 0.29 1.14 0.56 -

0.18 -

0.09 -

0.26 0.09 0.26 0.30 0.97 0.14

10 0.221 p-p38-basal biosignat

ure 1 0.97 1.52 1.21 0.97 1.18 1.60 1.91 0.65 1.14 0.31 0.86 1.31 0.51 0.78 0.93 1.66 1.93 0.87 0.77 1.21 0.00 0.51 0.69 1.06 1.21 0.57 0.77 1.61 0.53 1.69


ure 1 0.40 0.98 0.66 0.00 0.70 1.66 0.79 1.02 0.11 0.34 0.65 0.80 0.31 0.32 1.11 0.63 0.10 0.22 0.50 0.48 0.11 0.22 0.00 0.23 0.14 0.03 0.20 0.45 0.04 0.05

12 0.130 p-Erk1/2-FL biosignat

ure 1 0.20 0.34 0.69 0.32 0.12 1.00 0.14 0.03 0.24 0.23 0.06 0.10 0.51 0.16 0.17 1.52 0.11 1.04 0.77 -

0.10 0.28 0.17 -

0.08 0.56 0.10 0.34 0.25 0.36 0.28 -

0.14

13 0.112 p-Stat5-GM-

CSF biosignat

ure 1 -

0.11 0.59 1.06 0.58 -

0.16 0.30 -

0.19 -

0.06 -

0.12 -

0.20 0.25 0.18 0.27 0.15 0.47 0.38 0.16 -

0.03 0.25 0.42 -

0.14 -

0.15 -

0.28 -

0.01 -

0.62 -

0.22 -

0.01 0.55 0.21 -

0.06

14 0.076 p-Erk1/2-IL-3 1 0.01 0.05 0.29 0.23 -

0.10 0.09 0.29 0.14 -

0.65 -

0.03 -

0.01 -

0.17 0.13 0.18 0.04 0.56 -

0.21 0.40 0.42 -

0.55 0.12 0.19 0.18 0.66 -

0.22 0.09 0.08 0.16 0.04 -

0.29

15 0.067 p-Stat1-G-

CSF 1 0.13 0.36 0.00 0.28 -

0.12 -

0.01 -

0.03 0.10 0.86 0.03 0.26 -

0.20 0.57 0.43 0.36 -

0.01 0.23 0.11 0.95 -

0.04 0.19 0.17 0.11 0.03 0.10 -

0.10 -

0.09 0.14 0.14 0.16

16 0.062 p-Stat3-IFNg 1 -

0.30 -

0.12 -

0.21 0.04 -

0.05 0.21 -

0.26 -

0.16 -

0.14 -

0.36 -

0.03 -

0.09 0.08 -

0.07 0.14 0.12 0.05 0.32 0.64 -

0.05 -

0.13 0.31 -

0.49 0.56 -

0.06 -

0.39 -

0.19 0.05 -

0.01 -

0.04

17 0.058 p-Erk1/2-IFNg 1 -

0.14 -

0.04 0.19 0.14 -

0.07 0.02 0.47 -

0.06 -

0.55 0.32 0.25 -

0.21 0.28 0.01 -

0.02 0.28 -

0.03 0.25 0.28 -

0.58 0.27 -

0.09 0.08 0.18 -

0.08 0.45 0.07 0.03 0.25 -

0.06

18 0.055 p-Stat5-FL 1 0.10 -

0.18 -

0.17 0.22 0.14 0.35 -

0.01 0.04 0.32 -

0.11 0.34 -

0.08 0.01 0.18 0.26 0.45 0.17 0.69 0.67 0.20 -

0.01 0.19 -

0.18 0.13 -

0.39 0.03 0.25 0.01 0.18 0.16

19 0.046 p-Erk1/2-GM-

CSF 1 0.01 0.12 0.44 0.49 -

0.04 0.35 0.29 -

0.04 -

0.12 0.18 -

0.10 -

0.26 0.17 0.17 0.06 0.13 -

0.03 0.01 0.12 -

0.61 0.14 0.01 -

0.10 0.10 -

0.26 0.11 -

0.01 0.11 0.19 -

0.18

20 0.041 p-Erk1/2-G-

CSF 1 0.01 0.20 0.11 0.08 -

0.01 0.07 0.31 -

0.19 -

0.12 0.18 0.01 -

0.34 0.24 0.14 -

0.10 0.18 -

0.21 0.24 0.40 -

0.47 0.22 -

0.18 -

0.21 0.02 -

0.13 0.28 -

0.19 0.02 0.14 0.10

21 0.035 p-p38-FL 1 0.24 0.46 0.81 0.37 0.12 0.59 0.28 0.29 0.55 0.18 0.19 -

0.05 0.25 0.33 0.55 0.54 0.15 0.49 0.32 0.06 0.47 0.32 0.17 0.58 0.58 0.16 0.23 0.31 0.24 0.32

22 0.030 p-Stat3-IL-3 1 0.11 0.01 -

0.15 0.24 -

0.11 0.19 -

0.05 0.01 0.05 -

0.24 0.16 0.14 0.20 0.11 0.36 -

0.12 -

0.12 0.37 0.21 0.06 -

0.06 0.23 -

0.33 0.07 0.02 -

0.19 0.15 0.12 0.24 -

0.15

23 0.028 p-p38-IL-3 1 0.14 0.18 0.48 0.42 0.16 0.29 0.22 0.25 -

0.10 0.04 0.10 -

0.05 0.22 0.16 0.53 0.39 0.21 0.21 0.54 -

0.02 0.15 0.21 0.01 0.32 0.45 0.21 0.10 0.25 0.49 0.30

24 0.024 p-p38-G-CSF 1 0.03 0.22 0.27 0.25 -

0.01 0.05 0.21 0.10 0.29 -

0.03 0.03 -

0.27 0.16 0.09 0.34 0.14 -

0.05 0.18 0.44 -

0.22 0.26 0.05 -

0.03 0.16 0.27 0.17 -

0.01 0.04 0.21 0.16

25 0.023 p-p38-IFNg 1 -

0.02 -

0.10 0.17 0.18 -

0.02 0.05 0.18 0.25 -

0.14 0.18 0.17 -

0.05 0.12 0.03 0.16 0.21 0.20 0.14 0.19 -

0.35 0.01 -

0.05 -

0.09 0.22 0.48 0.17 0.01 -

0.04 0.08 0.13

26 0.023 p-p38-GM-

CSF 1 0.18 0.18 0.55 0.47 0.08 0.27 0.19 0.09 0.01 0.16 0.08 -

0.12 0.08 0.09 0.14 0.22 0.14 0.17 0.21 -

0.16 0.15 0.20 -

0.03 0.17 0.49 0.10 0.01 0.17 0.23 0.23

27 0.019 p-Stat1-GM-

CSF 1 0.13 -

0.13 0.02 0.22 0.06 -

0.12 -

0.08 -

0.08 0.00 -

0.08 -

0.12 -

0.37 -

0.03 0.12 0.08 -

0.03 0.29 -

0.04 -

0.24 -

0.10 -

0.01 0.24 0.09 -

0.05 -

0.04 -

0.01 -

0.09 0.01 -

0.23 -

0.03

28 0.017 p-Stat6-FL 1 0.04 -

0.10 0.07 0.11 0.12 -

0.19 0.01 0.12 0.12 -

0.03 -

0.07 0.05 0.05 0.56 0.07 0.02 0.02 0.01 0.12 0.06 0.14 0.03 -

0.08 -

0.05 0.05 -

0.15 -

0.11 0.04 -

0.02 0.15

29 0.015 p-Stat6-IFNg 1 0.13 0.00 0.06 0.18 0.22 0.10 0.10 0.21 0.10 0.18 0.09 0.16 0.18 0.14 0.17 0.23 0.35 0.10 0.26 0.01 0.10 0.12 0.22 0.64 0.19 -

0.11 0.16 0.19 0.09 0.18

30 0.014 p-Stat1-FL 1 0.14 -

0.06 0.21 0.21 0.13 0.05 -

0.01 0.25 0.27 -

0.03 -

0.02 -

0.01 0.14 0.22 0.21 0.01 0.35 0.22 0.17 0.03 0.03 0.10 -

0.22 0.16 0.03 0.12 0.10 0.13 0.01 0.13

31 0.013 p-Stat6-G-

CSF 1 0.07 -

0.04 0.00 0.06 0.08 -

0.06 -

0.01 0.08 0.18 -

0.05 0.04 -

0.08 0.09 0.06 0.14 -

0.08 0.12 -

0.08 0.21 -

0.20 0.08 -

0.04 -

0.26 0.05 0.01 -

0.21 -

0.18 0.05 0.00 0.17

32 0.012 p-Stat3-GM-

CSF 1 -

0.05 -

0.07 -

0.33 0.08 -

0.10 0.06 -

0.08 -

0.07 -

0.05 -

0.16 0.01 -

0.02 0.02 0.03 0.03 -

0.19 -

0.01 -

0.01 0.04 -

0.06 -

0.06 0.19 -

0.28 0.03 -

0.14 -

0.06 -

0.16 -

0.16 0.06 0.12

33 0.012 p-Stat1-IL-3 1 0.05 -

0.10 -

0.01 0.21 -

0.08 -

0.10 -

0.08 -

0.15 -

0.01 -

0.11 -

0.15 -

0.21 -

0.15 0.08 0.07 -

0.29 0.02 -

0.10 -

0.03 -

0.06 -

0.06 0.03 0.10 0.01 -

0.28 -

0.01 -

0.04 0.06 -

0.15 -

0.16

34 0.012 p-Stat3-FL 1 0.10 -

0.03 -

0.25 0.07 0.08 0.02 0.05 -

0.05 0.07 -

0.03 0.12 0.03 -

0.01 0.02 0.08 0.01 -

0.02 -

0.15 0.20 0.03 0.05 0.08 -

0.21 -

0.02 -

0.12 -

0.06 -

0.21 -

0.08 0.09 0.16

35 0.009 p-Stat6-GM-

CSF 1 0.00 -

0.10 -

0.09 0.13 -

0.06 -

0.03 -

0.10 0.00 -

0.04 0.01 -

0.08 -

0.13 0.00 0.09 0.03 -

0.06 0.10 -

0.03 -

0.13 -

0.18 -

0.08 0.05 -

0.13 0.01 0.05 -

0.13 -

0.30 0.04 -

0.13 0.06

36 0.007 p-Stat6-IL-3 1 0.12 -

0.08 0.08 0.16 -

0.08 0.01 -

0.01 0.00 0.08 -

0.03 -

0.04 0.01 -

0.01 0.14 0.08 -

0.01 -

0.01 -

0.05 0.10 -

0.10 -

0.01 0.05 -

0.14 0.12 -

0.13 -

0.12 0.09 0.12 0.05 0.04

* Significance of node states to this system was determined by looking at the variance of node states across tumors. Rank was determined by variance of a node state across tumors. Complete linkage hierarchical clusterings are shown in Figure 3 and Supplementary Figure 1. The inner cells of this table can be copied and pasted as text to make a Stanford format file readable by the MeV software used for hierarchical clustering and produced the order indicated in Supplementary Table 1 . Clustering for this manuscript employed the significant node states flagged as "biosignature."

Documents

Single cell profiling of potentiated phospho-protein networks in cancer cells