Characterizing the Clinical Relevance of an Embryonic Stem ...Human Cancer Biology Characterizing...

Published OnlineFirst December 8, 2009; DOI: 10.1158/1078-0432.CCR-09-1939 Published OnlineFirst December 8, 2009; DOI: 10.1158/1078-0432.CCR-09-1939 Published OnlineFirst December 8, 2009; DOI: 10.1158/1078-0432.CCR-09-1939

Human Cancer Biology

Characterizing the Clinical Relevance of an Embryonic Stem Cell

Phenotype in Lung Adenocarcinoma

Marvaretta Stevenson,1 William Mostertz,2 Chaitanya Acharya,2 William Kim,2 Kelli Walters,2

William Barry,3 Kristin Higgins,4 Sascha A. Tuchman,1 Jeffrey Crawford,1

Gordana Vlahovic,1 Neal Ready,1 Mark Onaitis,5 and Anil Potti1,2

Abstract Purpose: Cancer cells possess traits reminiscent of those ascribed to normal stem cells.

Auth2Inst3Com5Surg

Recei

Gran

datio

The c

page

accor

Note

Rese

geno

Thes

Clinic

Requ

Duke

919-6

F

do

ww

It is unclear whether these phenotypic similarities are the result of a common biological

phenotype, such as regulatory pathways.

Experimental Design: Lung cancer cell lineswith corresponding gene expression data and

genes associatedwith an embryonic stem cell identity were used to develop a signature of

embryonic stemness (ES) activity specific to lung adenocarcinoma. Biological character-

istics were elucidated as a function of cancer biology/oncogenic pathway dysregulation.

The ES signature was applied to three independent early-stage (I-IIIa) lung adenocarcino-

ma data sets with clinically annotated gene expression data. The relationship between the

ES phenotype and cisplatin (current standard of care) sensitivity was evaluated.

Results: Pathway analysis identified specific regulatory networks [Ras (P = 0.0005), Myc

(P = 0.0224), wound healing (P < 0.0001), chromosomal instability (P < 0.0001), and in-

vasiveness (P < 0.0001)] associated with the ES phenotype. The prognostic relevance of

the ES signature, as related to patient survival, was characterized in three cohorts

[CALGB 9761 (n = 82; P = 0.0001), National Cancer Institute Director's Challenge Con-

sortium (n = 442; P = 0.0002), and Duke (n = 45; P = 0.06)]. The ES signature was not

prognostic in prostate, breast, or ovarian adenocarcinomas. Lung tumors (n = 569) and

adenocarcinoma cell lines (n = 31) expressing the ES phenotype were more likely to be

resistant to cisplatin (P < 0.0001 and P = 0.006, respectively).

Conclusions: Lung adenocarcinomas that share a common gene expression pattern

with normal human embryonic stem cells were associated with decreased survival, in-

creased biological complexity, and increased likelihood of resistance to cisplatin. This

indicates the aggressiveness of these tumors. (Clin Cancer Res 2009;15(24):7553–61)D Februa

ry 21

, 201

2

Lung cancer remains the leading cause of cancer-related deathsin theUnited States with only a 15% five-year overall survival rate(1, 2). In addition, many patients that are diagnosed early andtreated with standard cisplatin-based therapy still go on to haveCTE

ors' Affiliations: 1Division of Oncology, Department of Medicine,

itute for Genome Sciences and Policy, and Departments of

putational Biology and Bioinformatics, 4Radiation Oncology, and

ery, Duke University, Durham, North Carolina

ved7/23/09; revised9/22/09; accepted9/26/09; publishedOnlineFirst 12/8/09.

t support: Emilene Brown Cancer Research Fund, The Jimmy V Foun-

n, American Cancer Society, and National Cancer Institute.

osts of publication of this article were defrayed in part by the payment of

charges. This article must therefore be herebymarked advertisement indance with 18 U.S.C. Section 1734 solely to indicate this fact.

: Supplementary data for this article are available at Clinical Cancer

arch Online (http://clincancerres.aacrjournals.org/) and at http://data.

me.duke.edu/Stemness.

e data have been presented previously at the American Society of

al Oncology annual conference on June 1, 2009.

ests for reprints: Anil Potti, Institute for Genome Sciences and Policy,

University, Box 3382, 101 Science Drive, Durham, NC 27708. Phone:

84-6567; Fax: 919-668-4777; E-mail: [email protected].

2009 American Association for Cancer Research.

i:10.1158/1078-0432.CCR-09-1939

7553w.aacrjournals.org

RETRA

Researon Marcclincancerres.aacrjournals.org Downloaded from Researon Marcclincancerres.aacrjournals.org Downloaded from Researon Marcclincancerres.aacrjournals.org Downloaded from

disease relapses (3). Research efforts continue in lung cancertreatment to improve the clinical outcomes of these patients.Recently, attention has turned toward stem cells as a new

frontier in the battle against cancer. The hallmark traits of stemcells—self-renewal and differentiation capacity—are mirroredby the high proliferative ability and phenotypic plasticity ofcancer cells. There is also emerging evidence to suggest that sig-naling pathways that are critical to regulating the function ofnormal embryonic stem cells may also be active in tumors(4–7). It is unclear, however, whether these phenotypic similar-ities between normal stem cells and tumor cells are the result ofa common biological phenotype(s), that is, changes in specificmolecular pathways and regulatory networks. One approach toaddressing this important question is to use genome-wide ex-pression data to better dissect the biology driving the “stemness”of a tumor. Such studiesmay eventually lead to improved under-standing of molecular mechanisms of disease progression andnovel therapeutic options.Seminal work from two laboratories has recently described

specific gene sets representative of “embryonic stemness” (ES)in normal stem cells (8, 9). Ben-Porath et al. (8) identified genesets associated with an embryonic stem cell identity that wereoverexpressed in poorly differentiated tumors and associated

Clin Cancer Res 2009;15(24) December 15, 2009

ch. h 25, 2020. © 2009 American Association for Cancerch. h 25, 2020. © 2009 American Association for Cancerch. h 25, 2020. © 2009 American Association for Cancer

http://clincancerres.aacrjournals.org/



Translational Relevance

Cancer cells and normal stem cells share certain

qualities, such as self-renewal andmultilineage differ-

entiation, which would allow these cancer cells to be

able to maintain tumor growth. We developed a 100-

gene signature that would identify embryonic stem

cell characteristics in lung adenocarcinoma that has

prognostic and therapeutic implications. Lung adeno-

carcinomas that share a common gene expression

pattern with normal embryonic stem cells were asso-

ciatedwith decreased survival, increased activation of

oncogenic pathways, and increased likelihood of re-

sistance to cisplatin, indicating the aggressiveness of

these tumors. Future steps should include evaluating

normal stem cell and lung cancer stem cell cultures to

further validate and manipulate the cancer biology/

oncogenic pathways found to be activated in tumors

with embryonic stemness. This may lead to new and

effective therapeutic strategies.


e

Published OnlineFirst December 8, 2009; DOI: 10.1158/1078-0432.CCR-09-1939

with poor clinical outcome in breast cancer. Wong et al. (9) iden-tified genes differentially expressed between normal embryonicstem cells and adult tissue stem cells. Cancers overexpressing theembryonic stem cell–like gene expression pattern were associat-ed with poor clinical outcomes in breast cancer also. These find-ings suggest that biological properties associated with embryonicstem cells may be related to tumor proliferation and invasive-ness. Building on this observation, we report the developmentand validation of a robust, clinically relevant signature of ES thatis specific to adenocarcinoma of the lung, the most commonsubtype of non–small cell lung cancer (NSCLC).

Materials and Methods

Below is a brief description of the methods. Complete details areavailable in Supplementary Methods. Supplementary data files can befound online.6

NSCLC cell lines. Cells were grown according to media recommen-dations by the commercial vendor (American Type Culture Collection)with minimal modification. Total RNA was extracted from 56 cell linesusing Qiagen RNeasy kits. Gene expression was measured using Affy-metrix Human Genome U133A GeneChips.7

Signature development. Two thousand eighty-one genes were ex-tracted from the Wong et al. data (courtesy of Dr. Howard Chang) and7,602 genes were extracted from the Ben-Porath et al. data (courtesy ofDr. Ittai Ben-Porath). Duplicate genes were removed from each gene list.Gene expression values corresponding to the two sets of gene lists wereextracted separately from the microarray data of the NSCLC cell lines todevelop individual gene expression signatures. Gene filtering was doneto isolate genes suspected of having the most effect. Intensity filteringwas used to filter out genes with consistently low intensity values orlow variance across samples. Interquartile range filtering was used to de-crease the total number of genes by including genes with intensity valuesfrom the 25th percentile to the 75th percentile.

Briefly, the statistical approach in signature development first in-volves the identification of cell lines with two distinct classes/pheno-types (ES and “no ES”). Gene expression data are obtained from the

RETRACTED F

6 http://data.genome.duke.edu/Stemness7 http://www.affymetrix.com/products

7554Clin Cancer Res 2009;15(24) December 15, 2009

Researcon Marchclincancerres.aacrjournals.org Downloaded from

two classes of cell lines and using a Bayesian regression analysis, a pre-dictor of ES is generated. The Bayesian model uses singular value de-composition and probit regression to build a gene signature from atraining data set. First, the most dominant genes are selected, whichshow the greatest amount of correlation to the observed phenotype,as determined by Student's t tests, and “metagenes” are defined as thetop orthogonal factors (principal components) identified by singularvalue decomposition to capture the greatest amount of variance in geneexpression within the collection of samples. This identifies the finalgenes to be included in a signature, or metagene, model. A Bayesianprobit function is then applied to determine the probability that eachsample has the ES phenotype. The model is then validated under leave-one-out cross-validation where the differentially expressed features thatcontribute to the metagene are properly reselected before fitting theBayesian model. Subsequently, the final model is used to predict theprobability of the ES phenotype in an independent data set. Furtherdetails are available in Supplementary Methods.

After analyzing the 56 NSCLC cell lines, independent signatures weredeveloped, composed of data from 28 and 22 NSCLC cell lines for theWong et al. and Ben-Porath et al. data, respectively. These smallergroups of cell lines, identified through unsupervised hierarchical clus-tering, showed biphasic gene expression patterns that allowed the genesfrom the Ben-Porath et al. and Wong et al. data to be used to divide thecells lines into those that showed ES and “no ES.”

The resulting signatures were individually evaluated for their abilityto separate lung cancer data sets into patient samples that exhibited ESand “no ES”. After initial validation using a pilot cohort, CALGB 9761(n = 82; ref. 10) lung adenocarcinoma data set, these two signatureswere combined. Cell lines with discordant gene expression patternswere removed. The remaining cell lines from both signatures underwentunsupervised hierarchical clustering again to identify cell lines withgene expression patterns representing the extremes of the ES phenotype.This developed a final, more robust signature specific to lung adenocar-cinoma. The final signature was composed on 20 cell lines and 100dominant genes. The dominant 100 genes constituting the final modelare shown in Supplementary Table S1.

Patient cohorts. Several clinically annotated cancer data sets wereused to validate the signature. Three early-stage (I-IIIa) lung adenocar-cinoma data sets were used: National Cancer Institute (NCI) Director'sChallenge Consortium (n = 442; ref. 11), Duke (GSE3151; n = 45; ref. 12),and CALGB 9761 (GSE3593; n = 82; ref. 10). A small cohort of pa-tient samples from CALGB 9761 may overlap with samples fromthe NCI Director's Challenge Consortium. A data set including 25advanced-stage (IIIb/IV) lung adenocarcinomas was also evaluated.The majority of these lung cancer patients were smokers. Threenon-lung adenocarcinoma data sets were evaluated, including pros-tate adenocarcinoma (n = 79; ref. 13), breast adenocarcinoma (GSE2034;n = 286; ref. 14), and ovarian adenocarcinoma (n = 119; ref. 15). Normallung epithelium from smokers was evaluated from Spira et al. (GSE4115;n = 90; ref. 16). Duke data set samples were obtained from patientsenrolled in institutional review board–approved clinical trials throughthe Duke Lung Cancer Prognostic Laboratory.

The dominant 100 genes in the ES signature (the “metagene”) wereapplied to each validation data set using Bayesian regression methods(17, 18). This enabled the assignment of a probability value to individualsamples reflecting an estimation (range, 0-1) of a given sample's likelihoodof having an ES phenotype. A predicted probability of >0.5 was chosenas a predetermined cutoff to identify those samples with the ES pheno-type. The ES phenotype was then correlated with clinical survival datafor patients using Kaplan-Meier curves to determine clinical relevance.

Gene set analysis. Analysis of Sample Set Enrichment Scores andGene Set Enrichment Analysis were used to identify gene sets/pathwaysassociated specifically with the 100 genes constituting the final ES sig-nature. Analysis of Sample Set Enrichment Scores8 is a statistical meth-od used to measure the degree of oncogenic pathway variation or gene

bruary

21, 2

012

8 http://people.genome.duke.edu/∼jhg9/assess//index.shtml

www.aacrjournals.org

h. 25, 2020. © 2009 American Association for Cancer


Embryonic Stemness in Lung Adenocarcinoma


set enrichment for each sample in a gene expression data set (19, 20). Itexpands on Gene Set Enrichment Analysis9 (version 2.0), which is ananalytic tool for relating gene expression data to gene sets to identifyunifying biological themes (21, 22).

Oncogenic pathway analysis. Signatures of oncogenic pathway acti-vation and tumor biology/microenvironment status were applied toeach of the adenocarcinoma lung cancer tumor samples (10, 12, 23–26). Bayesian binary regression methodologies described previously(17, 18) were applied to estimate relative probabilities of pathway de-regulation/activation or expression.

Cisplatin sensitivity in lung tumors and cell lines. A cisplatin sensi-tivity predictor was applied to each of the adenocarcinoma lung cancertumor samples (17, 27). Bayesian binary regression methodologies de-scribed previously (17, 18) were applied to estimate relative probabil-ities of cisplatin sensitivity.

EC50 data for cisplatin was collected on 31 NSCLC cell lines (fromAmerican Type Culture Collection). Cells were grown in the presence ofcisplatin for 96 h, and sensitivity to cisplatin was determined by quan-

RETRAC

9 http://www.broad.mit.edu/gsea

7555www.aacrjournals.org

Researon Marcclincancerres.aacrjournals.org Downloaded from

tifying the percent reduction in growth (versus DMSO controls) at 96 husing a standard MTT colorimetric assay (CellTiter 96 Aqueous OneSolution Cell Proliferation Assay Kit; Promega; refs. 28, 29).

Results

Development of an ES signature specific to lung adenocarcinoma.Genes were extracted from the Wong et al. (ref. 9; courtesy ofDr. Howard Chang) and the Ben-Porath et al. (ref. 8; courtesyof Dr. Ittai Ben-Porath) data to develop a signature specific tolung adenocarcinoma (Fig. 1). These gene sets were then usedto analyze 56 NSCLC (adenocarcinoma) cell lines (Fig. 2). First,we tested the robustness of the Wong et al. and Ben-Porath et al.gene sets individually; thus, gene expression signatures were de-veloped independently from each gene set (17, 18) that were ableto divide the lung cancer cell lines into two distinct groups: celllines with and without ES. Further details on the signature devel-opment are included in Materials and Methods. The prognosticability of the two signatures was evaluated in a pilot cohort of

Fig. 1. Development and validation of an ES signature specific to lung adenocarcinoma. IQR, interquartile range. *, NCI Director's Challenge Consortium).TED Feb

ruary

21, 2

012


ch. h 25, 2020. © 2009 American Association for Cancer




early-stage (I-IIIa) lung adenocarcinoma samples obtained froma multicenter cooperative group trial (CALGB 9761; n = 82;ref. 10). It is important to note that because our classifiers werebuilt from only lung adenocarcinoma cell lines, we tested theclinical relevance in a patient cohort that was exclusive to lungadenocarcinoma. It was found that patients with tumors withan ES designation had a poorer survival compared with thosewithout ES (Ben-Porath et al. survival P = 0.03 and Wong et al.survival P = 0.06; Supplementary Fig. S1).With an underlying hypothesis that optimizing the strengths

of the Wong et al. and the Ben-Porath et al. approaches into aunified signature would lead to a more robust clinically validrepresentation of the ES phenotype, we proceeded to generatea gene expression signature representative of ES using a meta-gene approach (10, 17). Importantly, the individual signatures

RETR



were combined after removing cell lines with discordantgene expression petterns (Fig. 2). A 100-gene ES signaturederived from 20 lung adenocarcinoma cell lines was created,and the intrinsic robustness was verified in a leave-one-outcross-validation. This final “hybrid” signature was used asthe classifier for further analysis.The ES signature is composed of several genes necessary for cell

cycle regulation, cell proliferation, angiogenesis, apoptosis regu-lation, and gene transcription (Supplementary Table S1). Geneshighly upregulated in the signature are known to be involved intumor invasion and cell proliferation, such as RAB25 andMAPK13. These genes were compared with the genes includedin other prognostic lung cancer signatures, such as Beer et al.(30), Chen et al. (31), Potti et al. (10), Bhattacharjee et al. (32),and Lu et al. (ref. 33; Supplementary Table S2). Although the

Fig. 2. Signature development. Gene sets from Ben-Porath et al. and Wong et al. were used to develop separate signatures for ES using NSCLC(adenocarcinoma) cell lines. The individual signatures were later combined (bottom left) after cell lines with discordant gene expression patterns wereremoved. Bottom right, leave-one-out cross-validation model for the combination signature.

ACTED Februa

ry 21

, 201

2






Fig. 3. A, Analysis of Sample Set Enrichment Scores and Gene Set Enrichment Analysis were used to identify gene sets/pathways associated with cells lineswith and without ES. Certain biologically relevant processes are highlighted. NES, no ES. B, oncogenic pathway deregulation shown as a function of ES inlung adenocarcinoma samples (n = 569). The distribution of samples from each data set is shown above the heat map. Red on the heat map indicatespathway deregulation/activation and blue on the heat map indicates normal activity. Bottom, probability of oncogenic pathway deregulation. WH, woundhealing; CIN, chromosomal instability; IGS, invasiveness; BCAT, β-catenin; TNF, tumor necrosis factor.

RETRACTED Februa

ry 21

, 201

2

7557 Clin Cancer Res 2009;15(24) December 15, 2009www.aacrjournals.org

Research. on March 25, 2020. © 2009 American Association for Cancerclincancerres.aacrjournals.org Downloaded from




individual genes are different among the listed signatures, thereis significant overlap at a functional/pathway level.Biology of ES in lung adenocarcinoma. To deepen our inves-

tigation of the biological underpinnings of the ES phenotype,we applied previously validated approaches, including GeneSet Enrichment Analysis (21, 22) and Analysis of Sample SetEnrichment Scores (19, 20), to cell lines that constitute our100-gene lung cancer–specific classifier. Elucidation of patternsof pathway activation specific to ES in lung adenocarcinoma re-vealed numerous biologically relevant features (Fig. 3A). For ex-ample, the most striking characteristics in the stemness cohortincluded gene sets/pathways associated with neoplastic trans-formation in normal tissues, cancer invasion, and metastasis(12, 34, 35). Specifically, the Ras pathway was associated withthe ES phenotype.In addition, previously described gene signatures that delin-

eate the activity of several cancer cell biology and oncogenicpathways (12, 23–26) were applied to 569 samples from pa-tients with early-stage lung adenocarcinoma. Oncogenic path-ways and cellular processes that were significantly associatedwith increased deregulation in adenocarcinomas with ES(Fig. 3B) included Ras (P = 0.0005), Myc (P = 0.0224), woundhealing (P < 0.0001), chromosomal instability (P < 0.0001),and invasiveness (P < 0.0001) gene signatures. Src, β-catenin,tumor necrosis factor, and E2F pathways were not significantlydifferent. These findings are not completely surprising because



the embryonic signature was found to be enriched with genesets involved in cancer invasiveness, metastasis, and the Raspathway. The findings from the Analysis of Sample Set Enrich-ment Scores/Gene Set Enrichment Analysis analyses and onco-genic pathway evaluation confirm the biological complexityassociated with ES in cancer at a pathway level.ES and prognosis in lung adenocarcinoma. As an initial mea-

sure of clinical relevance, the probability of having an ES phe-notype, as determined by the 100-gene signature, wasevaluated in smokers with normal lung epithelium (n = 90;ref. 16) compared with patients with early-stage (I, II, andIIIa; n = 569) and advanced-stage (IIIb and IV; n = 25) lungadenocarcinoma. This showed a significant difference inprobability of ES for smokers without cancer when comparedwith patients with early- or advanced-stage lung adenocar-cinoma in a stepwise fashion (P < 0.0001; Fig. 4A). In addi-tion, patients with poorly (n = 166), moderately (n = 209),and well (n = 60) differentiated adenocarcinomas alsoshowed significant differences in probability of ES (P <0.0001; Fig. 4A), with poorly differentiated tumors havingthe highest probability of ES. This result provides preliminaryevidence that ES may be associated with lung cancer invasionand progression.To further characterize the clinical relevance of the ES pheno-

type, we analyzed patients with early-stage lung adenocarci-noma. Using survival as a measurable endpoint, the probabilityry

21, 2

012

a

Fig. 4. A, probability of having ES in smokers with normal lung epithelium and the tumors of patients with early-stage (I-IIIa; n = 569) or advanced-stage(IIIb/IV; n = 25) lung adenocarcinomas (left). Right, comparison of patients with poorly, moderately, or well-differentiated tumor differentiation. B,Kaplan-Meier survival curves of early-stage lung adenocarcinoma cohorts were used for validation of the ES signature. Median survival (in months) is listed.

RETRACTED Febru





RETRAC


of ES in patient tumor samples from these three data sets [CALGB9761 (n = 82; ref. 10), NCI Director's Challenge Consortium (n =442; ref. 11), and Duke (n = 45; ref. 12)] was plotted as a functionof survival (Fig. 4B). These patients had not received adjuvant che-motherapy. Kaplan-Meier survival analysis showed that patientsamples designated as havingESwere associatedwith a poorer sur-vival compared with those without ES in two of the data sets(CALGB 9761 P = 0.0001 and NCI Director's Challenge Consor-

7559www.aacrjournals.org

Researon Marcclincancerres.aacrjournals.org Downloaded from

tium P = 0.0002), with a trend toward poorer survival in the thirddata set (Duke P = 0.06).Ben-Porath et al. and Wong et al. showed previously that the

genes they identified as associated with an ES phenotype werenot predominately related to cell proliferation. To show thatthe ES signature goes beyond just genes involved in cell cycleand cell proliferation regulation, these genes were removedand the signature was reanalyzed. Repeat evaluation of the

TED Februa

ry 21

, 201

2

Fig. 5. A, probability of sensitivity tocisplatin was analyzed as a function of ESin lung adenocarcinoma samples (n = 569).Red on the heat map indicates cisplatinsensitivity and blue on the heat mapindicates cisplatin resistance. B, cisplatinsensitivity for NSCLC cell lines as a functionof ES. Sensitivity to cisplatin wasdelineated from EC50 assays. Red on theheat map indicates cisplatin sensitivity andblue on the heat map indicates cisplatinresistance.


ch. h 25, 2020. © 2009 American Association for Cancer



e


NCI Director's Challenge Consortium continued to show asignificant difference in survival between patients with andwithout ES (P = 0.0009; Supplementary Fig. S2).In addition, Cox proportional hazard analysis was done on

the largest data set, the NCI Director's Challenge Consortium(n = 442; Supplementary Table S3). The results revealed thatprobability of ES (based on the 100-gene signature), along withage, gender, disease stage, and tumor size, were all statisticallysignificant independent predictors of patient survival in early-stage NSCLC.Although the signature of ES was developed from lung adeno-

carcinoma lines and thus likely would only be specific to lungadenocarcinoma, the signature was used to correlate ES and sur-vival in other forms of adenocarcinoma [prostate adenocarcino-ma (13), ovarian cancer (15), and breast adenocarcinoma (14)data sets] to confirm this hypothesis (Supplementary Fig. S3).Indeed, the signature was not found to be prognostic in theseother types of adenocarcinomas, confirming the tissue specificityof the signature for lung adenocarcinomas.ES and cisplatin sensitivity in lung cancer. Platinum-based

chemotherapy is a cornerstone of the treatment of lung cancer(36). Several efforts are currently under way exploring the abil-ity to predict a cancer's sensitivity to chemotherapy in an effortto individualize a patient's care (17, 27, 37–39). Sensitivity tocisplatin was evaluated by applying a previously validated che-mosensitivity signature (17) to the lung adenocarcinoma sam-ples to determine if there was any association between cisplatinsensitivity and ES. An analysis of 569 lung adenocarcinomas re-vealed that samples predicted to carry the ES phenotype weremore likely to be resistant to cisplatin (Fig. 5A) compared withthose without ES (P < 0.0001).In an effort to further validate this novel observation, in vitro

cisplatin sensitivity was evaluated in a large cohort of lung ad-enocarcinoma cell lines (n = 31) that were originally used todevelop the ES signature. The EC50 was determined for thesecell lines and plotted against the ES of the lung cancer cell lines.Confirming the observation seen in the patient cohort (Fig. 5A),lung cancer cell lines with ES were more resistant to cisplatin(P = 0.006; Fig. 5B). This important finding that actual patientsamples and cancer cell lines with an ES phenotype are morelikely to be resistant to platinum-based therapy may furthercontribute to the poor prognosis in patients whose cancers havethis phenotype. The etiology of this resistance could possibly berelated to the association of stem cells with multidrug resistanceand enhanced DNA repair mechanisms (4).

Discussion

Numerous examples now can be found where gene expres-sion profiles have been able to dissect lung cancer subtypes(30–32). The use of gene expression signatures as surrogate

ETRACTED F


R


phenotypes has been particularly important, outlining the com-plexity of biological systems in a way that was not previouslyfeasible (10, 37, 40, 41). However, in lung cancer, studies ad-dressing the potential effect of specific stem cell regulatory net-works in defining clinically relevant cancer phenotypes havebeen limited.Cancer cells and normal stem cells share certain qualities,

such as self-renewal and multilineage differentiation, whichwould allow these cancer cells to be able to maintain tumorgrowth. This has sparked interest in exploring whether theseshared qualities are due to shared molecular pathways betweencancer cells and stem cells versus due solely to a small popula-tion of cancer stem cells that drive the behavior of a tumor. Fur-ther investigation is thus warranted to better understand thebiological and clinical relevance of “stem cell–like” characteris-tics in cancers, specifically lung cancer.Our findings describe a 100-gene signature representative of

ES specific to lung adenocarcinoma. Preliminary findings sug-gest that merely the presence of genes regulating cell prolifera-tion do not portend clinical relevance or predict a stem cellphenotype at the gene expression level (8, 9). In fact, in ouranalysis, removing proliferation-associated genes from the ESsignature did not affect its prognostic ability to identify patientsat risk for poor survival. The finding of biological complexity ata pathway level, as evidenced by activation of Ras, Myc, andchromosomal instability in the stemness phenotype, might beexpected, considering the fact that this phenotype also has ad-verse prognostic and therapeutic implications shown by its as-sociation with poor survival, poorly differentiated cancers, andresistance to cisplatin-based chemotherapy.It is, however, important to note here that the goals of our

current study are not to develop yet another gene signature thatpredicts lung cancer recurrence and/or sensitivity to chemother-apy. There are several other predictors that are likely more ro-bust in addressing these questions more directly (10, 11, 42,43). We used disease recurrence/survival and sensitivity to cis-platin merely to depict the clinical relevance of studying stemcell–associated phenotypes relevant to lung adenocarcinoma.Our results need to be considered in light of the fact that stemcells of all types (normal adult stem cells, embryonic stem cells,and cancer stem cells) are still poorly characterized. Our abilityto isolate these cells and ensure cell purity is limited, specifical-ly in lung cancer. As an alternative, the embryonic stem cellphenotype we describe here may carry meaningful clinical rel-evance and be representative of an “inherited” behavior in lungadenocarcinoma as it evolves in the process of differentiation.

bruary

21, 2

012

Disclosure of Potential Conflicts of Interest

No potential conflicts of interest were disclosed.

References
1. Jemal A, Siegel R, Ward E, et al. Cancer statis-tics, 2008. CA Cancer J Clin 2008;58:71–96.
2. In: Ries L, Melbert D, Krapcho M, et al., editors.SEER cancer statistics review, 1975-2005.Bethesda (MD): National Cancer Institute; 2008.

3. Wakelee H, Chhatwani L. Adjuvant chemother-apy for resected non-small cell lung cancer.Semin Thorac Cardiovasc Surg 2008;20:198–203.

4. Boman BM,WichaMS. Cancer stem cells: a steptoward the cure. J Clin Oncol 2008;26:2795–9.

5. Glinsky GV. “Stemness” genomics law governsclinical behavior of human cancer: implicationsfor decision making in disease management.J Clin Oncol 2008;26:2846–53.

6. Jordan CT, Guzman ML, Noble M. Cancer stemcells. N Engl J Med 2006;355:1253–61.

7. Clarke MF, Dick JE, Dirks PB, et al. Cancer stemcells—perspectives on current status and futuredirections: AACR Workshop on Cancer StemCells. Cancer Res 2006;66:9339–44.

h. 25, 2020. © 2009 A

8. Ben-Porath I, Thomson MW, Carey VJ, et al.An embryonic stem cell-like gene expressionsignature in poorly differentiated aggressive hu-man tumors. Nat Genet 2008;40:499–507.

9. Wong DJ, Liu H, Ridky TW, et al. Modulemap of stem cell genes guides creation of ep-ithelial cancer stem cells. Cell Stem Cell 2008;2:333–44.

10. Potti A, Mukherjee S, Petersen R, et al. A ge-nomic strategy to refine prognosis in early-stage


merican Association for Cancer




non-small-cell lung cancer. N Engl J Med 2006;355:570–80.

11. Shedden K, Taylor JM, Enkemann SA, et al.Gene expression-based survival prediction inlung adenocarcinoma: a multi-site, blinded vali-dation study. Nat Med 2008;14:822–7.

12. Bild AH, Yao G, Chang JT, et al. Oncogenicpathway signatures in human cancers as a guideto targeted therapies. Nature 2006;439:353–7.

13. Stephenson AJ, Smith A, Kattan MW, et al.Integration of gene expression profiling and clin-ical variables to predict prostate carcinomarecurrence after radical prostatectomy. Cancer2005;104:290–8.

14. Wang Y, Klijn JG, Zhang Y, et al. Gene-expres-sion profiles to predict distant metastasis oflymph-node-negative primary breast cancer.Lancet 2005;365:671–9.

15. Dressman HK, Berchuck A, Chan G, et al. Anintegrated genomic-based approach to individu-alized treatment of patients with advanced-stageovarian cancer. J Clin Oncol 2007;25:517–25.

16. Spira A, Beane JE, Shah V, et al. Airway epi-thelial gene expression in the diagnostic evalua-tion of smokers with suspect lung cancer. NatMed 2007;13:361–6.

17. Potti A, Dressman HK, Bild A, et al. Genomicsignatures to guide the use of chemotherapeu-tics. Nat Med 2006;12:1294–300.

18. West M, Blanchette C, Dressman H, et al. Pre-dicting the clinical status of human breast cancerby using gene expression profiles. Proc NatlAcad Sci U S A 2001;98:11462–7.

19. Edelman E, Porrello A, Guinney J, et al. Anal-ysis of Sample Set Enrichment Scores: assayingthe enrichment of sets of genes for individualsamples in genome-wide expression profiles.Bioinformatics 2006;22:e108–16.

20. Edelman EJ, Guinney J, Chi JT, Febbo PG,Mukherjee S. Modeling cancer progression viapathway dependencies. PLoS Comput Biol2008;4:e28.

21. Subramanian A, Tamayo P, Mootha VK, et al.Gene Set Enrichment Analysis: a knowledge-based approach for interpreting genome-wide


RETRAC

clincancerres.aDownloaded from

expression profiles. Proc Natl Acad Sci U S A2005;102:15545–50.

22. Mootha VK, Lindgren CM, Eriksson KF, et al.PGC-1α-responsive genes involved in oxidativephosphorylation are coordinately downregu-lated in human diabetes. Nat Genet 2003;34:267–73.

23. Bild AH, Potti A, Nevins JR. Linking oncogenicpathways with therapeutic opportunities. NatRev Cancer 2006;6:735–41.

24. Liu R, Wang X, Chen GY, et al. The prognosticrole of a gene signature from tumorigenicbreast-cancer cells. N Engl J Med 2007;356:217–26.

25. Chang HY, Nuyten DS, Sneddon JB, et al. Ro-bustness, scalability, and integration of awound-response gene expression signature inpredicting breast cancer survival. Proc Natl AcadSci U S A 2005;102:3738–43.

26. Carter SL, Eklund AC, Kohane IS, Harris LN,Szallasi Z. A signature of chromosomal instabil-ity inferred from gene expression profiles pre-dicts clinical outcome in multiple humancancers. Nat Genet 2006;38:1043–8.

27. Potti A, Nevins JR. Utilization of genomic sig-natures to direct use of primary chemotherapy.Curr Opin Genet Dev 2008;18:62–7.

28. Mosmann T. Rapid colorimetric assay for cel-lular growth and survival: application to prolifer-ation and cytotoxicity assays. J ImmunolMethods 1983;65:55–63.

29. Berridge MV, Tan AS. Characterization of thecellular reduction of 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT): subcel-lular localization, substrate dependence, andinvolvement of mitochondrial electron trans-port in MTT reduction. Arch Biochem Biophys1993;303:474–82.

30. Beer DG, Kardia SL, Huang CC, et al. Gene-expression profiles predict survival of patientswithlung adenocarcinoma. Nat Med 2002;8:816–24.

31. Chen HY, Yu SL, Chen CH, et al. A five-genesignature and clinical outcome in non-small-celllung cancer. N Engl J Med 2007;356:11–20.

32. Bhattacharjee A, Richards WG, Staunton J,

ebrua

r

7561 Clin Canc

TE F

Research. on March 25, 2020. © 2009 acrjournals.org

et al. Classification of human lung carcinomasby mRNA expression profiling reveals distinctadenocarcinoma subclasses. Proc Natl AcadSci U S A 2001;98:13790–5.

33. Lu Y, Lemon W, Liu PY, et al. A gene expres-sion signature predicts survival of patients withstage I non-small cell lung cancer. PLoS Med2006;3:e467.

34. Rhodes DR, Yu J, Shanker K, et al. Large-scalemeta-analysis of cancer microarray data identi-fies common transcriptional profiles of neo-plastic transformation and progression. ProcNatl Acad Sci U S A 2004;101:9309–14.

35. van't Veer LJ, Dai H, van de Vijver MJ, et al.Gene expression profiling predicts clinical out-come of breast cancer. Nature 2002;415:530–6.

36. Non-Small Cell Lung Cancer CollaborativeGroup. Chemotherapy in non-small cell lungcancer: a meta-analysis using updated dataon individual patients from 52 randomised clini-cal trials. BMJ 1995;311:899–909.

37. Minna JD, Gazdar AF, Sprang SR, Herz J. Can-cer. A bull's eye for targeted lung cancer thera-py. Science 2004;304:1458–61.

38. Bepler G. Pharmacogenomics: a reality or stilla promise? Lung Cancer 2006;54 Suppl 2:S3–7.

39. Bepler G, Begum M, Simon GR. Molecularanalysis-based treatment strategies for non-small cell lung cancer. Cancer Control 2008;15:130–9.

40. Perou CM, Sorlie T, Eisen MB, et al. Molecularportraits of human breast tumours. Nature 2000;406:747–52.

41. Golub TR, Slonim DK, Tamayo P, et al. Molec-ular classification of cancer: class discovery andclass prediction by gene expression monitoring.Science 1999;286:531–7.

42. Sun Z, Wigle DA, Yang P. Non-overlappingand non-cell-type-specific gene expression sig-natures predict lung cancer survival. J Clin On-col 2008;26:877–83.

43. Lau SK, Boutros PC, Pintilie M, et al. Three-gene prognostic classifier for early-stage nonsmall-cell lung cancer. J Clin Oncol 2007;25:5562–9.

y 21,

2012

D

er Res 2009;15(24) December 15, 2009

American Association for Cancer


Retraction

Retraction: Characterizing the ClinicalRelevance of an Embryonic StemCell Phenotype in Lung Adenocarcinoma

The authors wish to retract the article titled "Characterizing the Clinical Rele-vance of an Embryonic Stem Cell Phenotype in Lung Adenocarcinoma," whichwas published in the December 15, 2009, issue of Clinical Cancer Research (1).

It has recently come to our attention that clinical information from a data setCALGB 9761 (GSE3593), available at the time of the signature development,was incorrect. Since survival data from this data set were integral to the creationof this signature, thismakes the findings of thismanuscript inaccurate. Drs. AnilPotti and Marvaretta Stevenson take full responsibility for this error. Therefore,we would like to formally retract the above paper.

Marvaretta StevensonWilliam MostertzChaitanya R. AcharyaWilliam KimKelli WaltersWilliam BarryKristin HigginsSascha A. TuchmanJeffrey CrawfordGordana VlahovicNeal ReadyMark OnaitisAnil Potti

Duke UniversityDurham, North Carolina

Reference1. Stevenson M, Mostertz W, Acharya C, Kim W, Walters K, Barry W, et al. Characterizing the

clinical relevance of an embryonic stem cell phenotype in lung adenocarcinoma. Clin CancerRes 2009;15:7553–61.

Published OnlineFirst February 21, 2012.doi: 10.1158/1078-0432.CCR-12-0337�2012 American Association for Cancer Research.

ClinicalCancer

Research

Clin Cancer Res; 18(6) March 15, 20121818

2009;15:7553-7561. Published OnlineFirst December 8, 2009.Clin Cancer Res Marvaretta Stevenson, William Mostertz, Chaitanya Acharya, et al. Cell Phenotype in Lung AdenocarcinomaCharacterizing the Clinical Relevance of an Embryonic Stem

Updated version

10.1158/1078-0432.CCR-09-1939doi:

Access the most recent version of this article at:

Material

Supplementary

http://clincancerres.aacrjournals.org/content/suppl/2009/12/21/1078-0432.CCR-09-1939.DC1Access the most recent supplemental material at:

Cited articles

http://clincancerres.aacrjournals.org/content/15/24/7553.full#ref-list-1

This article cites 42 articles, 14 of which you can access for free at:

Citing articles

http://clincancerres.aacrjournals.org/content/15/24/7553.full#related-urls

This article has been cited by 1 HighWire-hosted articles. Access the articles 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://clincancerres.aacrjournals.org/content/15/24/7553To request permission to re-use all or part of this article, use this link

Research. on March 25, 2020. © 2009 American Association for Cancerclincancerres.aacrjournals.org Downloaded from


http://clincancerres.aacrjournals.org/lookup/doi/10.1158/1078-0432.CCR-09-1939

http://clincancerres.aacrjournals.org/content/suppl/2009/12/21/1078-0432.CCR-09-1939.DC1

http://clincancerres.aacrjournals.org/content/15/24/7553.full#ref-list-1

http://clincancerres.aacrjournals.org/content/15/24/7553.full#related-urls

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

mailto:[email protected]

http://clincancerres.aacrjournals.org/content/15/24/7553


Characterizing the Clinical Relevance of an Embryonic Stem ...Human Cancer Biology Characterizing...

Documents

Transcript of Characterizing the Clinical Relevance of an Embryonic Stem ...Human Cancer Biology Characterizing...