Uncovering the Molecular Secrets of Inﬂammatory Breast ...Uncovering the Molecular Secrets of Inﬂammatory Breast Cancer Biology: An Integrated Analysis of Three Distinct Affymetrix

Human Cancer Biology

Uncovering the Molecular Secrets of Inflammatory BreastCancer Biology: An Integrated Analysis of Three DistinctAffymetrix Gene Expression Datasets

Steven J. Van Laere1,2, Naoto T. Ueno3, Pascal Finetti5, Peter Vermeulen1, Anthony Lucci3,Fredika M. Robertson4, Melike Marsan1,2, Takayuki Iwamoto3, Savitri Krishnamurthy3, Hiroko Masuda3,Peter van Dam1, Wendy A. Woodward3, Patrice Viens5, Massimo Cristofanilli6, Daniel Birnbaum5,Luc Dirix1, James M. Reuben3, and Francois Bertucci5

AbstractBackground: Inflammatory breast cancer (IBC) is a poorly characterized formof breast cancer. So far, the

results of expression profiling in IBC are inconclusive due to various reasons including limited sample size.

Here, we present the integration of three Affymetrix expression datasets collected through the World IBC

Consortium allowing us to interrogate the molecular profile of IBC using the largest series of IBC samples

ever reported.

ExperimentalDesign:Affymetrix profiles (HGU133-series) from137 patients with IBC and 252patients

with non-IBC (nIBC) were analyzed using unsupervised and supervised techniques. Samples were classified

according to themolecular subtypes using the PAM50-algorithm. Regressionmodels were used to delineate

IBC-specific and molecular subtype-independent changes in gene expression, pathway, and transcription

factor activation.

Results: Four robust IBC-sample clusterswere identified, associatedwith thedifferentmolecular subtypes

(P < 0.001), all of which were identified in IBCwith a similar prevalence as in nIBC, except for the luminal A

subtype (19% vs. 42%; P < 0.001) and the HER2-enriched subtype (22% vs. 9%; P < 0.001). Supervised

analysis identified and validated an IBC-specific, molecular subtype-independent 79-gene signature, which

held independent prognostic value in a series of 871 nIBCs. Functional analysis revealed attenuated TGF-bsignaling in IBC.

Conclusion: We show that IBC is transcriptionally heterogeneous and that all molecular subtypes

described in nIBC are detectable in IBC, albeit with a different frequency. The molecular profile of IBC,

bearing molecular traits of aggressive breast tumor biology, shows attenuation of TGF-b signaling,

potentially explaining the metastatic potential of IBC tumor cells in an unexpected manner. Clin Cancer

Res; 19(17); 4685–96. �2013 AACR.

IntroductionInflammatory breast cancer (IBC) is an aggressive form of

locally advanced breast cancer. At the time of diagnosis,virtually all patients have lymph node metastases and one

third of the patients have metastases in distant organs.Consequently, patient prognosis is poor (1). The rate ofpathologic complete response (pCR) after primary anthra-cycline-based chemotherapy ranges from 15% to 30% and

Authors' Affiliations: 1Translational Cancer Research Unit, OncologyCenter, General Hospital Sint-Augustinus, Antwerp, Wilrijk; 2DepartmentOncology, KU Leuven, Leuven, Belgium; 3Morgan Welch InflammatoryBreast Cancer Program and Clinic, and 4Department of ExperimentalTherapeutics, The University of Texas MD Anderson Cancer Center,Houston, Texas; 5D�epartement d'Oncologie Mol�eculaire, Centre deRecherche en Canc�erologie de Marseille, UMR891 Institut National de laSant�e et de la Recherche Medicale (INSERM), Institut Paoli-Calmettes(IPC), Marseille, France; and 6Department of Medical Oncology, G. MorrisDorrance Jr. Endowed Chair in Medical Oncology, Fox Chase CancerCenter, Philadelphia, Pennsylvania

Note: Supplementary data for this article are available at Clinical CancerResearch Online (http://clincancerres.aacrjournals.org/).

S.J. Van Laere, N.T. Ueno, P. Finetti, and F. Bertucci contributed equally tothis work.

Corresponding Authors: Steven J. Van Laere, Translational CancerResearch Unit, Department Medical Oncology, Sint-Augustinus, Oos-terveldlaan 24, 2610 Wilrijk, Belgium. Phone: 32-3-443-36-37; Fax: 32-3-443-30-36; E-mail: [email protected]; Francois Bertucci,Institut Paoli-Calmettes, Department of Medical Oncology, 232, bdSte-Marguerite, 13273 Marseille Cedex 09, France. Phone: 33-4-91-22-35-37; Fax: 33-4-91-22-36-70; E-mail: [email protected];and Naoto T. Ueno, Morgan Welch Inflammatory Breast Cancer Pro-gram and Clinic, Section of the Translational Breast Cancer Research,Department of Breast Medical Oncology, The University of Texas MDAnderson Cancer Center, 1515 Holcombe, Unit 1354, Houston, TX77030. Phone: 713-792-8754; Fax: 713-794-4385; E-mail:[email protected]

doi: 10.1158/1078-0432.CCR-12-2549

�2013 American Association for Cancer Research.

ClinicalCancer

Research

www.aacrjournals.org 4685

on February 11, 2021. © 2013 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from

Published OnlineFirst February 8, 2013; DOI: 10.1158/1078-0432.CCR-12-2549

http://clincancerres.aacrjournals.org/

the 5-year survival remains around 40%. Currently, IBCremains a poorly characterized disease lacking specificmolecular targets for therapy, although ErbB2 is oftenamplified.

The low frequency of occurrence (approximately 5% ofall breast tumors) combined with the small size of diag-nostic samples, in addition to the fact that patients with IBCare treated with neo-adjuvant chemotherapy, are the factorsthat have hampered past molecular studies, specificallythose that were directed at deciphering the molecular biol-ogy of this disease by genome-wide approaches (2). Themajor stumbling block has been the poor statistical powerwhen multiple genes (typically a multitude of 10,000) aretested on a small number of samples. High genes/samplesratios, even after removing noisy and noninformative data,result in dramatic downscaling of the significance level (i.e.,multiple testing correction) leading to loss of statisticalpower (3).

Despite the issues outlined earlier, several studies aimedat defining the molecular signature of IBC using genome-wide gene expression profiling. Most of these studies haveshown that amolecular signature of IBC is definable (4–10).Nevertheless, when comparing the crude gene lists, only alimited number of genes were commonly identified acrossall studies (7, 9). Also, when translating the gene lists intobiologic processes or signal transduction pathways, therewas significant ambiguity. Potential reasons for this obser-vation have been discussed previously and include: (i)

distinct case definitions of IBC used across different studies,(ii) variability of the non-IBC (nIBC) control group, (iii)differences in the characteristics of the IBC and nIBC groupsbetween different studies, notably with respect to the hor-mone receptor status, and (iv) interstudy technologic dif-ferences, that is, use of different platforms and subsequentlydifferent input gene lists (2).

In 2008, the World IBC Consortium (www.ibcconsor-tium.org) was founded with the goal of fostering colla-borations between international research groups whofocus on this rare but aggressive form of breast cancer.The research described in the current article is the firstproject spearheaded by the World IBC Consortium andaims at redefining the molecular profile of IBC by takinginto account the points of criticism raised with respect tothe earlier studies described earlier. Here, we used auniform case definition of IBC that has been put forwardby an international expert panel (11). Using gene expres-sion profiles from Affymetrix (HGU133-series) derivedfrom 3 different sites involved in IBC research, we wereable to obtain an unprecedented number of IBC samples(N ¼ 137), which allows us to resolve both the samplesize-related and the platform-related issues identified inthe previous studies.

Materials and MethodsPatients and samples

Tumor samples were obtained from patients with breastadenocarcinoma treated in our 3 institutions: the InstitutPaoli-Calmettes (IPC, Marseille, France; 71 IBC and 139nIBC), the MD Anderson Cancer Center (MDA, Houston,TX; 25 IBC and 58 nIBC), and the General Hospital Sint-Augustinus (TCRU, Antwerp, Belgium; 41 IBC and 55nIBC). Each patient gave written informed consent and thisstudywas approvedby the Institutional ReviewBoards of all3 participating centers. The present dataset includes 137samples from patients with IBC and 252 samples frompatients with nIBC. Patients with IBC were selected bystrictly adhering to the consensus diagnostic criteria pub-lished by Dawood and colleagues (11). IBC cases includedlocally advanced [American Joint Committee on Cancer(AJCC) stage III] andmetastatic cases (AJCC stage IV). nIBCcases includedboth early-stage disease (AJCC stages I and II)and advanced stage disease (locally advanced, AJCC stageIII; and metastatic, AJCC stage IV). Estrogen receptor (ER)and ErbB2 expression were defined using probe sets205225_at (ESR1) and 216836_s_at (ERBB2), as describedpreviously (12). As a control, we verified the correlationbetween mRNA status and immunohistochemistry (IHC)-based protein status as discrete variables. The positive ornegative mRNA status was defined using a 2-componentGaussian mixture distribution model as described previ-ously (13). Data are provided in the Supplementary Fig. S1and Supplementary Table S1. Tumor grade was determinedusing the genomic grade index (GGI; ref. 14). All patientswere treated using a multidisciplinary approach accordingto standard guidelines. Clinicopathologic data of our seriesare shown in Table 1 and are consistent with literature.

Translational RelevanceThe clinical relevance of the research described in this

article is manifold. First, due to the extent of our series,which includes by far the largest series of inflammatorybreast cancer (IBC) samples ever reported, we were ableto delineate the molecular profile of IBC with enhancedaccuracy, taking into account various points of criticismraisedwith respect to earlier studies.Our report is thefirstto analyze the expression profiles of IBC samples in lightof a uniform case definition for IBC put forward by aninternational expert panel. Also, the present study is thefirst to deal with the fact that the differential distributionpattern of themolecular subtypes between IBC and non-IBC (nIBC) affects its outcome. Furthermore, our datasuggest that differentially expressed genes between IBCand nIBC in a molecular subtype-independent mannerhold the fingerprints of aggressive breast tumor behav-ior. Therefore, our expression series not only suits IBCresearch, but might also be instrumental for scientistsfocused on breast cancer invasion and metastasis ingeneral. Finally, this study gives voice to the World IBCConsortium, which is founded to foster collaborationsbetween researchers active within the IBC community.We believe that our effort is an example of how colla-borations with a strong international backbone canprosper research well into the 21st century.

Van Laere et al.

Clin Cancer Res; 19(17) September 1, 2013 Clinical Cancer Research4686




Data processing and normalizationRNA extraction from the 389 samples and hybridization

onto Affymetrix GeneChips (HGU133-series) was conduct-ed as described before (6, 8, 9, 15). Expression data werenormalized by guanine cytosine robust multi-array analysis(16). For each of the 3 datasets separately, probe sets withexpression values above log2(100) in at least 1%of the arrayswere filtered in. Next, the list of common informative probesets (N ¼ 9,926) was identified. This list was used to mergethe distinct datasets. Therefore, we conducted regressionnormalization, using the Limma-package in BioConductor,to remove technical, laboratory-specific, variation in geneexpression between the distinct datasets. A principal com-ponent analysis (PCA)was done on themerged dataset priorand after the regression normalization to verify the accuracy

of the regression normalization in removing the laboratory-specific variation in gene expression.

Molecular subtypesWe classified each sample in the merged dataset accord-

ing to the molecular subtypes (luminal A, luminal B, basal-like, HER2-enriched, and normal-like) using the PAM50-methodology described by Parker and colleagues (17). Inaddition, samples predicted to belong to the claudin-lowsubtype by the 9-cell line claudin-low predictor (18) wereconsidered claudin-low. To further investigate the effect ofthe regression normalization on the biologic variation ingene expression, the molecular subtype classification wascompared with the expression of an ER activation signature(19).

Table 1. Histoclinical data of IBC and nIBC samples

IBC nIBC

Characteristic (n ¼ 137) (n ¼ 252) P value OR

Median age, y (range) 50 (24–82) 55 (24–89) 0.002 NAAJCC stageI 0 (0%) 65 (26%) <0.001 NAII 0 (0%) 97 (39%)III 83 (75%) 66 (27%)IV 28 (25%) 20 (8%)

Histologic typeDuctal 122 (90%) 203 (81%) 0.042 NALobular 7 (5%) 23 (9%)Other 6 (4%) 26 (10%)

pNNegative NR 58 (43%) NR NRPositive NR 77 (57%)

pTpT1 NR 46 (34%) NR NRpT2 NR 65 (49%)pT3 NR 23 (17%)

ER (mRNA level) <0.001 2.36Negative 66 (48%) 71 (28%) (1.50–3.74)Positive 71 (52%) 181 (72%)

ERBB2 (mRNA level) 0.001 2.38Negative 95 (69%) 212 (84%) (1.39–4.00)Positive 42 (31%) 40 (16%)

GGI 0.003 1.97High 98 (72%) 141 (56%) (1.24–3.18)Low 39 (28%) 111 (44%)

Systemic therapy (CT and/or HT and/or T)Yes 134 (98%) 219 (86%) NR NRNo 0 (0%) 33 (13%)Not available 3 (2%) 20 (1%)

Median follow-up, months (range)a 56 (2–165) 80 (6–182) <0.001 NRDMFS-5yb 53% (43–65) 82% (77–88) <0.001 NA

Abbreviations: CT, chemotherapy; HT, hormone therapy; NR, not relevant; NA, not applicable; T, trastuzumab.aCalculated only for M0 patients without any metastatic relapse udring follow-up, from the date of diagnosis to the date of last news.bNumbers in brackets represent CI95%.

Molecular Profiling of IBC

www.aacrjournals.org Clin Cancer Res; 19(17) September 1, 2013 4687




Unsupervised analysisTo investigate common themes in gene expressionwithin

each tumor phenotype (IBC or nIBC) separately as well asfor both tumor phenotypes combined, we adopted thefollowing strategy.We randomly selected 100 samples fromthe 3 groups of interest (group 1, IBC; group 2, nIBC; andgroup 3, combined IBC/nIBC). On this selection, unsuper-vised hierarchical cluster analysis (UHCA) usingManhattandistance as distance measure and Ward linkage as thedendrogram drawingmethodwas conducted only for thosegenes having a SD greater than 2. UHCA was followed by acluster robustness analysis using the silhouette algorithm.The silhouette algorithm divides the dendrogram into anincreasing number of clusters. We chose to analyze a max-imum number of 20 clusters. For each increase in thenumber of clusters, the algorithm generates a score, themagnitude of which is proportional to the robustness ofthe identified clusters. This analysis was repeated 10 times,resulting in a series of 10 silhouette scores for each numberof clusters. The number of clusters in each group of interestwas then determined by sequentially comparing the silhou-ette scores (silhouette scores for 2 clusters vs. silhouettescores for 3 clusters, silhouette scores for 3 clusters vs.silhouette scores for 4 clusters, and so on) and equals thehighest number of clusters for which the silhouette scoresignificantly drops when increasing the number of clusterswith one unit.

Supervised analysesSupervised analysis, identifying differentially expressed

genes between IBC andnIBC, was conducted to evaluate theinfluence of alternative stage-matching approaches on theidentifiable biologic differences between IBC and nIBC.Therefore, global differences in gene expression betweenIBC and 3 alternatively composed nIBC control groups(stage I–IV, stage I–II, and stage III–IV) were investigatedusing the global test (20). In addition, for each comparison,we sought to identify lists of differentially expressed genesusing significance analysis of microarrays (SAM; ref. 21).The resulting gene lists were compared.

To further explore functional differences between IBCand nIBC, we repeated supervised analyses at the pathwaylevel. We applied 18 pathway gene expression signaturesoriginally described by Gatza and colleagues using theprocedures outlined in the original publication (19) and1 additional gene expression signature for VEGF-activation(22). In addition, for each of the previously defined com-parisons, we explored differences in patterns of transcrip-tion factor activation. For that purpose, we designed analgorithm to quantify the extent of transcription factoractivation in samples based on the gene expression data.Details are provided in the Supplementary Data and Sup-plementary Table S2.

Molecular subtype-independent supervised analysisBecause of the association of the tumor phenotype (IBC/

nIBC) with the molecular subtypes, the list of probe setsresulting from the IBC/nIBC comparison probably contains

many genes related to molecular subtype, and therefore wehypothesize that this gene list is not IBC-specific. To dissectIBC-specific variations in gene expression from molecularsubtype-specific variations in gene expression, supervisedanalysis (IBC vs. nIBC) was repeated by applying linearregression modeling (Limma-package). Before analysis, wedivided the dataset into a training set of 250 samples (84IBC samples and 166 nIBC samples) and an independentvalidation set of 139 samples (53 IBC samples and 86 nIBCsamples). The distributions of the molecular subtypes, thetumor stage, and the institution from which the sampleswere derived (IPC, MDA, and TCRU) were comparedbetween both datasets. The training dataset was analyzedusing linear regressionmodels incorporating the molecularsubtype classification and the IBC/nIBC classification toidentify probe sets with IBC-specific expression compo-nents. Using the global test (20), lists of probe sets wereanalyzed for global expression differences between IBC andnIBC in the validation set. Classifier models based on theidentified lists of probe sets with an IBC-specific expressioncomponent were constructed using the nearest shrunkencentroid algorithm implemented in the PAMR-package. Forthis purpose, 10-fold cross-validationon the training setwasconducted to select an appropriate d-value minimizing thecross-validated training error rate. The resultingmodelswereapplied on the validation set to estimate accuracy, sensitiv-ity, and specificity. In addition, we tested the performance ofthe classifier models to discriminate between samples frompatients with and without IBC in more homogenous sub-groups. For the sake of statistical power, we combinedluminal A and B samples into a luminal group and basal-like and claudin-low samples into a triple-negative breastcancer (TNBC) group. Finally, the strategy of dissecting IBC-specific effects frommolecular subtype-specific effects usinglinear regression models was repeated on the entire datasetfor the pathway and transcription factor activation data.

Analysis of publicly available nIBC dataGiven the higher metastatic risk of IBC compared with

nIBC, we tested the hypothesis that our 79-gene IBC/nIBCsignature, if biologically relevant, is associated with meta-static relapse in nIBC. Therefore, we collected expressiondata from 6 publicly available series of adjuvant chemo-therapy- and hormone therapy-na€�ve lymph node-negativepatients with early breast cancer. These series have beenused for similar purposes in earlier studies and follow-updata in termsofmetastatic relapse are available. Anoverviewof included datasets is provided in Supplementary Table S3.Redundancy related to patient samples included inmultipledatasets was eliminated, yielding 871 samples available foranalysis. Before analysis, we mapped hybridization probesfor the differentially expressed genes across the 2 oligonu-cleotide-based platforms used across the series. When mul-tiple probes weremapped to the sameGeneID (EntrezGeneidentification number), the onewith the highest variance ina particular dataset was selected to represent the GeneID.Analysis of each dataset was done separately to guarantee alarger number of genes common with our gene list. After

Van Laere et al.





having identified the common genes, we used distanceweighted discrimination (DWD; ref. 23) to normalize eachpublic dataset to be comparable with our present pooledAffymetrix dataset. The samples were classified as "IBC-like"or "nIBC-like" using a support vector machine (SVM)function based on the list of 79 differentially expressedgenes. The initial outcome of interest for this SVM classifierwas the separation of IBC from nIBC. The SVM model wasestablished in our original training set (84 IBC and 166nIBC samples) with a linear kernel as the parameter.

Statistical analysesIn the results section, only nominal P values are reported;

however, in case of multiple comparisons, P values werecorrected and were considered significant only if the falsediscovery rate (FDR)was smaller than 0.05 (9). To comparethedistribution according to categorical variables,x2 testingor the Fisher exact test were used when appropriate. Tocompare continuous data between 2 or more groups,respectively the Mann–Whitney U test and the Kruskal–Wallis test were used. To compare continuous data with 1reference value, the Wilcoxon signed rank test was used. Tocompare distributions of continuous variables, Spearmancorrelation analysis was conducted. The distant metastasis-free survival (DMFS) for patients with nIBC included in thepublic dataset was calculated from the date of diagnosisuntil the date of first metastatic relapse. The follow-up wascalculated from the date of diagnosis to the date of last newsfor event-free patients. Survival was calculated using theKaplan–Meier method and compared between groups withthe log-rank test. Univariate and multivariate analyses weredone using generalized linearmodels or Cox’s proportionalhazards in case of survival analysis. Patients who died fromother causes and did not experience a metastatic relapsebefore dying were censored at the time of death and werenot considered as an event for the estimation of DMFS.

ResultsUnsupervised analysis and heterogeneity of IBCFirst, we assessed the effect of the normalization strategy

on the technical and biologic variation in gene expression.Results of this analysis are presented in Supplementary Fig.S2. PCA shows that normalizationwas effective in removingtechnical, interlaboratory variation in gene expression. Afternormalization, ER activation scores were significantly dif-ferent between samples classified according to the variousmolecular subtypes (P < 0.001), suggesting that the nor-malization procedure did not remove biologic differences.Second, using UHCA followed by silhouette score anal-

ysis, we identified 4 robust clusters in our series of 137 IBCsamples (Fig. 1A). The details of the silhouette analysis areprovided in Supplementary Figs. S3 and S4. A significantassociationwas observed between these sample clusters andthe molecular subtypes (P < 0.001). In agreement with thisobservation, ER-, progesterone receptor (PR)-, and ERBB2-pathway activation scores were also significantly differentbetween the different clusters (P < 0.001). When a similaranalysis was conducted on the nIBC series (N ¼ 252), 6

sample clusters, again related to themolecular subtypes (P <0.001), were identified (Fig. 1B). Arguably, the increase innumber of clusters observed innIBC canbe explainedby theincrease in sample size (137 vs. 252). Therefore, we ana-lyzed 10 sets of 137 randomly selected nIBC samples. Themedian number of clusters in these series of samplesequaled 6 (range, 4–10), which was significantly differentfrom the number of sample clusters observed in the series ofIBC samples (P¼ 0.002), suggesting that IBC is indeed lessheterogeneous than nIBC.

When the earlier outlined analysis was repeated on thecombined dataset of IBC and nIBC samples, 7 robustsample clusters were identified (Fig. 1C). Downstream ofthe first bifurcation, luminal A, luminal B, and normal-likesamples were segregated from basal-like, HER2-enrichedand claudin-low samples (P < 0.001). In addition, a signif-icant segregation of IBC and nIBC samples was observed (P< 0.001). Linear regression analysis using the cluster label asthe dependent variable and both the tumor phenotype andthe different molecular subtypes as independent variablesshowed that molecular subtypes, and not the tumor phe-notype, are independent predictors of the cluster label(tumor phenotype, P ¼ 0.104; molecular subtypes, P <0.050). These data suggest that differences in gene expres-sionbetween IBC andnIBCare dominated by themolecularsubtype-related differences in gene expression.

IBC/nIBC molecular comparison and influence of themolecular subtypes

We sought to determine differences in gene expressionbetween IBC and nIBC in a gene-by-gene supervisedmanner. Table 2 provides an overview of the results of thiscomparison. When comparing IBC to nIBC, we observedsignificant differences in gene expression, regardless of thetype of stage-matching conducted for the nIBC controlgroup. On initial review, differences between IBC andadvanced stage (III–IV) nIBC seemed smaller than betweenIBC and all nIBC samples, as suggested by the lowerQ value.However, when differences between IBC samples and non-stage-matched nIBC samples were analyzed in 10 equallylarge series or randomly selected samples, no significantdifferences forQ values were recorded (P¼ 0.345), suggest-ing that the determinant responsible for this difference issample size and is not due to the type of stage-matching(Table 2 and Fig. 2A). Similar analyses were conducted forthenumbers of differentially expressed genes at FDR<0.050and FDR < 0.100 (Table 2 and Fig. 2B and C). Lists ofdifferentially expressed genes (FDR < 0.100) identified inthe comparisons of IBC to alternatively composed nIBCcontrol groups (i.e., nonstagematched, advanced stage (III–IV) only, or low stage (I–II) only) were evaluated to assessthe effect of the type of stage-matching on the gene listcontent; results are presented by a Venn diagram (Fig. 2D).Of the genes identified in the smallest gene list (IBC vs. stageIII–IV nIBC), 98% and 91%of the genes were present in thegene lists comparing respectively IBC to all nIBC samplesand IBC to the stage I–II nIBC samples only. When com-paring IBCwith either all nIBC samples orwith the stage I–II






nIBC samples only, an average concordance of approxi-mately 87%was observed between both gene lists (data notshown). These results suggest that stage matching hadlimited influence, allowing us to retain all nIBC controlsamples for further analysis.

We then compared IBC and nIBC for several expressionprofiles, including molecular subtypes and pathway andtranscription factor activation signatures. Results are shownin Supplementary Table S4. About the molecular subtypes,similar frequencies for all subtypes were observed except forthe luminal A subtype (42% in nIBC vs. 19% in IBC; P <0.001) and the HER2-enriched subtype (9% in nIBC vs.22% in IBC; P < 0.001). Out of 19 tested pathways, 12(63%) were differentially activated between IBC and nIBC:8 pathwaysweremore activated in IBC (CTNB,HER2,MYC,RAS, IFN-a, IFN-g , TNF-a, and VEGF), whereas 4 pathways

were attenuated in IBC (ER, PR, P53, and TGF-b). Whenanalyzing the transcription factor activation profiles, weidentified 78of 234 (33%)differentially activated transcrip-tion factors, from which 38 (49%) were more activated inIBC and 40 (51%) were more activated in nIBC. Amongthe transcription factors hyperactivated in IBC, we identi-fied RELA, corroborating the results from previous studies(24, 25).

Molecular subtype-independent comparison of IBCand nIBC

Because of the strong correlation between the IBC/nIBCphenotype and the molecular subtypes, we hypothesizedthat the results presented earlier could be related to themolecular subtypes, and hence were not actually IBC-spe-cific. Thus, we reiterated the IBC/nIBC comparative analysis

HER2-ENRICHEDCLAUDIN-LOW

NORMAL-LIKELUMINAL BLUMINAL A

BASAL-LIKE

A



BASAL-LIKE

B



BASAL-LIKEIBC

C

Figure 1. UHCA. Clustering was done with Manhattan distance as similarity measure and Ward linkage as the dendrogram drawing method for probe setshaving a SD greater than 2 in our series of IBC (A, 298 probe sets) and nIBC (B, 251 probe sets) samples only, as well as in the combined dataset (C, 270 probesets). Median-centered gene expression data are represented in a matrix format with rows indicating genes and columns indicating samples.Overexpressed genes (compared with the median) are color-coded red and repressed genes are color-coded green. Color saturation indicates thelevel of overexpression or repression. The clusters that were considered stable using the silhouette algorithm are indicated in the dendrogram usingalternating gray and blue colors. The classification of the samples according to the molecular subtypes (basal-like, claudin-low, HER2-enriched,luminal A, luminal B, and normal-like) is represented underneath the heatmaps in matrix format. Tumor subtype classification (IBC vs. nIBC) is also indicatedunderneath the heatmap representing the combined dataset. Black vertical ticks indicate class membership. For convenience, the shading pattern ofthe dendrograms is repeated on the class-membership matrix.

Van Laere et al.





(gene expression, pathway activation, and transcriptionfactor activation) using linear regression models to identifyprobe sets that are differentially expressed between IBC andnIBC in a molecular subtype independent manner. To beable to evaluate the extent bywhich the identified probe setstruly reflect IBC biology, we divided our series into atraining and validation set. No differences between both

sets were observed with respect to the distribution of themolecular subtypes (P ¼ 0.955), the tumor stage (P ¼0.954) and the institution from which the samples werederived (P ¼ 0.761).

In the training set, we identified 491 probe sets,representing 443 unique and IBC-specific genes, ofwhich 231 (47%) and 260 (53%) were respectively up- or

A B C

D

25 30 35 40 45

0.00

0.02

0.04

0.06

0.08

Q values

Den

sity

0 500 1,000 1,500 2,000 2,500 3,000 3,500

0e+0

01e

-04

2e-0

43e

-04

4e-0

45e

-04

FDR<0.05

Den

sity

1,000 20,00 3,000 4,000

0e+0

01e

-04

2e-0

43e

-04

4e-0

45e

-04

FDR<0.10D

ensi

ty

STAGE I–IV STAGE III–IV

619 37 5

2,687403

0

273

5,902STAGE I–II

Figure 2. Supervised analysis between IBC and nIBC according to nIBC stage-matching. Comparison of parameters of differential expression [Q values (A),number of significant genes at FDR < 0.050 (B), and number of significant genes at FDR < 0.100 (C)] between IBC and nIBC based on a comparisonwith stage-matching (stage III–IV) and for the comparison with no stage-matching (stage I–IV). The blue vertical line indicates the value for the correspondingparameter for the stage-matched comparison. The black line indicates the distribution of the values of the same parameters when analyzing differential geneexpression in the nonstage-matched analysis using 10 random series of equal sample sizes, as was the case for the stage-matched comparison.The Venn diagram (D) shows the number of differentially expressed genes between IBC and nIBC samples when using stage-matched nIBC samples only(stage III–IV) as comparator or when using stage I–II nIBC only as comparator or when using all nIBC samples regardless of stage (stage I–IV) as comparator.Numbers indicate the number of genes in each section.

Table 2. Results of supervised analyses between IBC and nIBC according to the nIBC stage

Comparison N IBC N NIBC N Genes Obs Q Exp Q SD Q P value FDR < 0.05 FDR < 0.10

IBC vs. stage I–IInIBC

137 162 9,926 48.589 10 2.493 <0.001 3,363 4,524

IBC vs. stage III–IVnIBC

137 86 9,926 30.176 10 2.393 <0.001 1,134 2,164

IBC vs. stage I–IVnIBC

137 252 9,926 53.574 10 2.385 <0.001 3,746 5,009

IBC vs. stage I–IVnIBC(Perm)

137 86 9,926 33.672 10 2.440 <0.001 1,930 3,037

Range (28,625–39,679) (2,395–2,483) (1.082–2.630) (2.132–3.731)






downregulated in IBC (Supplementary Table S5). For com-parison, 2,743 genes were differentially expressed (FDR <0.050) between IBC and nIBC in the training set whenmolecular subtype-dependent gene expression differenceswere not considered. Within the list of 491 probe sets, 79probe sets (16%) were uniquely IBC-specific, whereas theremaining probe sets (N ¼ 412) showed additional molec-ular subtype-specific gene expression variation. Both thecomplete list of probe sets with an IBC-specific expressioncomponent (N ¼ 491) and the list of probe sets that areuniquely IBC-specific (N ¼ 79) were used for further val-idation. Both lists are differentially expressed between IBCand nIBC (P < 0.001) in the independent validation set witha slightly greater expression difference for the list of unique-ly IBC-specific probe sets (observedQ¼51.677 vs. observedQ ¼ 50.893). Further details are provided in the Supple-mentary Fig. S5.

Using the nearest shrunken centroid method, classifiermodels were constructed on the training set and appliedonto the validation dataset. Cross-validated probabilities

for the training set and posterior probabilities for thevalidation set are provided for both models in Fig. 3. Forthe full list of probe sets with an IBC-specific expressioncomponent, we observed an overall test accuracy of 68%with a sensitivity of 66% and a specificity of 70% (OR,4.434;P<0.001).Using the list of probe setswith auniquelyIBC-specific expression pattern, we observed an overall testaccuracy of 71%with a sensitivity of 65%and a specificity of74% (OR, 5.136; P < 0.001). We next evaluated the perfor-mance of both models to discriminate between samplesfrom patients with and without IBC in more homogenoussubgroups. The model based on the list of uniquely IBC-specific probe sets (N¼ 79) conducted well in discriminat-ing between samples from patients with and without IBC inthe luminal (N¼ 70;OR, 3.909; P¼ 0.019) and TNBC (N¼40; OR, 4.839; P ¼ 0.027) series. Evaluating the modelbased on the full list of probe sets (N ¼ 491) in the earlierdescribed subgroups showed poor performance, probablydue to the presence of a molecular subtype-specific expres-sion component for themajority of the involved probe sets.

A

B

C

D

1.0

0.0

0

0 20 40 60 80 100 120 140

50 100 150 200 250

491 Probe set signature 79 Probe set signature

0.8

0.6

0.4

0.2

1.0

1.2

0.8

0.6

0.4

0.2

0.0

1.0

0.8

0.6

0.4

0.2

0.0

Cro

ss-v

alid

ated

pro

babi

litie

sP

oste

rior

prob

abili

ties

0 20 40 60 80 100 120 140

1.0

0.8

0.6

0.4

0.2

0.0

Pos

terio

r pr

obab

ilitie

s

1.0

0.0

0 50 100 150 200 250

0.8

0.6

0.4

0.2

1.0

1.2

0.8

0.6

0.4

0.2

0.0

Cro

ss-v

alid

ated

pro

babi

litie

s

Figure 3. Construction and assessment of the IBC/nIBC classification models. Using PAM, classification models to discriminate between IBC and nIBC wereconstructed. Thefirstmodelwasbasedon the full list of 491probesetswith an IBC-specificexpression component. The secondmodelwasbasedon the list of79 uniquely IBC-specific probe sets. Bothmodels were constructed using the training set and tested on the validation set. The leave-one-out cross-validatedprobabilities for all samples in the training set based on the full list of probe sets with an IBC-specific expression component (A) and the list ofuniquely IBC-specific probe sets (C) are provided in a dot plot format. Each sample is represented by 2 colored dots, with red corresponding to the cross-validated probability that the related sample belongs to the IBC phenotype and green corresponding to the cross-validated probability that the related samplebelongs to the nIBC phenotype. The real classification of the samples according to tumor phenotype is provided along the X-axis, with samples to theleft of the vertical dashed line being IBC samples and vice versa. One can clearly observe that themajority of the IBC and nIBC samples are correctly classifiedusing both models, when using a cross-validated probability of 0.500 as threshold for classification. Both models were subsequently evaluated in anindependent test set. The posterior probabilities to be classified as IBC for all samples in the validation set based on the full list of probe sets withan IBC-specific expression component (B) and the list of uniquely IBC-specific probe sets (D) are provided in a dot plot format. A red dot indicates an IBCsample; a green dot indicates a nIBC sample. The posterior probabilities are plotted in ascending order.Most nIBC samples have lowposterior probabilities tobe classified as IBC and vice versa. Again, when using a posterior probability of 0.500 as threshold for classification we obtain a classification accuracyof 68% and 71% for the model based on the full list of probe sets and the model based on the list of uniquely IBC-specific probe set.

Van Laere et al.





To identify relevant pathways, gene functions and regu-latory networks associated with IBC biology, we decided toanalyze the full list of probe sets with an IBC-specificexpression component. We reason that the presence of anIBC-specific expression component for each of the involvedprobe sets lends credit to the use of this list of probe sets as astarting point to decipher the molecular biology of IBC.Detailed results of the Ingenuity Pathway Analysis areprovided in Supplementary Tables S6 and S7 and Supple-mentary Figs. S6 and S7. We observed that inflammation-and immune-related processes, includingNF-kB–signaling,characterize the IBC tumor phenotype. We identified aregulatory gene network composed of genes repressed inIBC and centered on TGF-b (Supplementary Fig. S7A).Next, we focused on IBC-specific pathway and transcrip-

tion factor activation patterns. When molecular subtypeswere incorporated into the model, activation of the IFN-a–pathway and inhibition of the P53-, TGF-b-, and PR-pathways were associated with IBC (FDR < 0.050). Severalother pathways known to be associated with heterogeneityin breast cancer, including ER, HER2, and EGF receptor(EGFR), ranked among the lesser significant pathways (FDR¼ 0.153, 0.124, and 0.343, respectively), suggesting that theinfluence of the molecular subtypes on these results wasremoved. Upon reassessment of the transcription factoractivation profiles, we identified 18 IBC-specific transcrip-tion factors, 10 ofwhichhad attenuated activity (MZF1,GC,FOXO3/STAT5, BCL3, PITX2, MSX2, TCF3/TCF4, MYB,PEG10, and ID1) and 8 of which are activated (EGR, LXR,STAT3, HNF4A, THRA, SHH, FOXM1, andMKX2-2) in IBC(data not shown). Inhibition of SMAD in IBC was alsoobserved with a significant nominal P value although theFDRexceeded our thresholds for significance (FDR¼0.084;P ¼ 0.009; rank ¼ 23rd/234). As expected, ER rankedamong the lesser significant transcription factors (FDR ¼0.246; P ¼ 0.055; rank ¼ 52nd/234).

Prognostic value of the IBC/nIBC signature in nIBCGiven the poor prognosis of IBC, we hypothesized that

our 79-gene list, if biologically relevant with respect to theIBC/nIBC distinction, might be associated with the occur-renceofmetastatic relapse innIBC ingeneral.We thus testedits prognostic value in a series of 871 clinically annotatedand chemotherapy- and hormone therapy-na€�ve node-neg-ative early breast cancers. The SVM model classified 83.6%of the samples with a "nIBC-like" profile (N ¼ 728) and16.4% with an "IBC-like" profile (N ¼ 143). We comparedthe DMFS of these 2 nIBC classes. With amedian follow-upof 110 months after diagnosis (range, 1–286), the "nIBC-like" class had a 5-year DMFS of 75%, which was signifi-cantly longer than the 5-year DMFS of 62% observed forsamples classified as "IBC-like" (P ¼ 0.002). Kaplan–Meierplots are shown in Supplementary Fig. S8. In univariateanalysis (Supplementary Table S8), the risk of distant met-astatic recurrence in the "IBC-like" class was significantlyelevated as comparedwith the "nIBC-like" class [HR, 1.530;95%confidence interval (CI), 1.160–2.020; P¼ 0.002]. Theother significant variables (P < 0.050) included the histo-

logic grade, the molecular subtypes, and the pathologictumor size. Patient age at diagnosis was not significant.Interestingly, our SVM model remained significant (HR,1.480; 95% CI, 1.040–2.100; P ¼ 0.029) in multivariateanalysis incorporating all variables significant in univariateanalysis (pT status, histologic grade, and molecular sub-types), suggesting that the 79-gene signature containsmolecular subtype-independent prognostic informationthatmight explain theworse prognosis of patients with IBC.

DiscussionThis whole transcriptome study of IBC, the largest one

reported so far, was made possible thanks to collaborationthrough the World IBC Consortium. The present studyaddressed several criticisms raised with respect to the earlierstudies: a uniform IBC definition put forward by an inter-national expert panel (11) was used, pretreatment sampleswere profiled using the sameAffymetrix platformand ahighnumber of IBC samples was included in addition to a highnumber of nIBC control samples encompassing differentdisease stages.

We have shown that IBC is transcriptionally heteroge-neous, as shown by the identification of 4 robust sampleclusters. The identification of all existing molecular sub-types in IBC corroborates this finding. Compared withnIBC, the transcriptional heterogeneity was less extensiveand the distribution pattern of the molecular subtypesbetween 2 tumor groups differed, particularly with respectto the luminal A andHER2-enriched samples. Overall, 75%of the IBC samples belong to the classically more aggressivesubtypes, basal-like, HER2-enriched, claudin-low, or lumi-nal B subtypes, whereas these subtypes account for 54% ofthe nIBC samples. The luminal A subtype represents 19%ofthe IBC samples, whereas in nIBC this subgroup represents42%. These data corroborate the results reported inpreviouspublications where the presence of all molecular subtypesin IBC, albeit with a different frequency, was reported(9, 26, 27). The major difference resides in the fact that allsubtypes, except for the luminal A and the HER2-enrichedtumor samples, exhibit a similar prevalence in IBC as innIBC, whereas in earlier studies IBC was reported to beenriched for the basal-like and HER2-enriched molecularsubtypes (9, 27).

One of the aims of this study was to redefine themolecular profile of IBC, taking into account the pointsof criticism raised in earlier studies. The issues related tosample size, the definition of IBC cases and the interstudydifferences in technical platform are, to the best of ourefforts, resolved by the study design but the objectionswith respect to the composition of the control groupremain. Our results clearly show that the differentialdistribution of the molecular subtypes between bothtumor groups needs to be considered. As shown byunsupervised hierarchical clustering, IBC and nIBC aresignificantly segregated downstream of the first bifurca-tion, suggesting that there are major differences in geneexpression between the IBC and nIBC tumor phenotypes.However, multivariate regression analysis revealed that






the presence of the different molecular subtypes is themain driver of the clustering pattern. Therefore, differ-ences in gene expression between IBC and nIBC aredominated by the differential distribution pattern of themolecular subtypes, which is corroborated by the fact thatonly 18% of the genes initially reported to be differen-tially expressed between IBC and nIBC remain differen-tially expressed after conducting the linear regressionanalysis to account for the influence of the molecularsubtypes. Of note, the number of genes with a uniquelyIBC-specific gene expression profile is even smaller andrepresents only 3% of the global expression differences.The differential pattern of the molecular subtypes, at leastin part, explains the higher proportion of samples with amolecular "poor-prognosis" profile in IBC than in nIBCwhen classified according to previously published prog-nostic gene signatures, as well as the differential activa-tion of biologic pathways (data not shown). On the otherhand, the influence of the composition of the controlgroup according to tumor stage seems to be limited.

We thus decided to compare IBC with nIBC, incorporat-ing all stages (stage I–IV) of breast cancer and to useregression modeling to discriminate between molecularsubtype-specific effects and tumor phenotype-specificeffects. Alternatively, we could have compared IBC with anonstage-matched but molecular subtype-matched nIBCcontrol group. However, this would lead to smaller samplesizes, and hence, considerable loss of statistical power, adisadvantage not encountered when using the regressionmodels on the entire dataset.

The identified IBC-specific molecular changes both con-firm results obtained in previous studies and reveal novelfindings. MARCKS, a gene that ranked among the top 10differentially expressed genes between IBC and nIBC, isinvolved in regulating cell motility via its function as aregulator of the actin cytoskeleton (28). Interestingly, Vander Auwera and colleagues reported that MARCKS was oneof the targets of miR-30b, a phosphoinositide 3-kinase(PI3K)-targeting miRNA specifically downregulated in IBC(29). Other examples of genes overexpressed in IBC andinvolved in the actin-based cell migration include RAC1,RHOF, and FNBP1. The regulation of cell motility andinvasion was one of the most consistent molecular aspectsassociated with IBC across different studies (4–6, 9). Also,the role of NF-kB in IBC biology reported in earlier studies(24, 25) is corroborated by the findings in the present study.

By far, the most important biologic observation madewithin the scope of this analysis relates to TGF-b signaling.Given the aggressiveness of IBC and the role of TGF-b intumor biology (30), the observation of IBC-specific TGF-b–repression is unexpected. Careful examination of theliterature provided several interesting aspects with respectto TGF-b–signaling in breast cancer. First, Giampieri andcolleagues showed that TGF-b–signaling switches the activ-ity of breast cancer cells from cohesive to single cellmotility.In addition, cells restricted to collective invasion werecapable of lymphatic invasion but not blood-borne metas-tasis (31, 32). Thus, the reduced TGF-b–signaling as a

signature of IBC implies that IBC tumor cells migratecollectively and invade primarily into the lymph vessels.This is consistent with the presentation of IBC, with thepresence of numerous tightly aggregated nests of tumorcells, defined as tumor emboli that undergo dermal lym-phatic invasion as the pathologic hallmark of IBC. On thebasis of the central role of TGF-b in inducing the process ofepithelial-to-mesenchymal transition (EMT; ref. 30), ourobservations suggest that EMT is not the primary means oftumor cell migration/invasion in IBC. This agrees with theoverexpression of E-cadherin in IBC (33–35), which is aprotein involved in retention of an epithelial phenotypethat is lost during EMT, during which there is a gain ofexpression of genes involved in mesenchymal cell differen-tiation. Second, recent studies illustrate the importance ofactive TGF-b–signaling in preventing the proliferation ofERþ tumor cells, and suggest that the loss of this inhibitoryinteraction may be important in early breast cancer pro-gression (36). Therefore, we hypothesize that the reducedTGF-b–signaling in IBC might lead to increased prolifera-tion of ERþ tumor cells, which might be one of the reasonsfor the low frequency of luminal A samples in IBC. Inter-estingly, we recently described thatCDKN1B, a TGF-b-targetresponsible for inhibition of cell proliferation, is repressedin IBC (36). In the current series, we observed a molecularsubtype independent repression of CDKN1B in IBC (datanot shown).

Our results also identified several transcription factorswith an IBC-specific activation profile. In line with theobservations made on repression of TGF-b signaling, weobserved reduced IBC-specific activity of the SMAD tran-scription factor, which acts downstream of TGF-b–signaling. In fact, correlation analysis using independentgene sets suggests that both observations are linked (datanot shown). Further corroborating our hypothesis that EMTis not the main mode of invasion observed in IBC is thefinding of IBC-specific attenuation of the TCF3/TCF4 tran-scription factor activity, which is involved in the inductionand maintenance of EMT through the repression of E-cadherin (37).

In summary, the results obtained within the scope ofthis study provide novel insights into the biology of IBCand open an entirely new avenue of research topics.Without doubt, the most remarkable identified featureof IBC is the attenuated TGF-b–signaling, and by exten-sion the potential effects thereof on IBC tumor cellmigration and invasion, which warrants further investi-gation. Our data also show that the differential distribu-tion pattern of the molecular subtypes between IBC andnIBC should be considered when doing IBC research(e.g., in the present study, up to 82% of the differentiallyexpressed genes between IBC and nIBC are not IBC-specific), not only in translational studies but also inbasic research. For example, the SUM149 cell line,although derived from a patient with IBC, remains abasal-like cell line, and therefore observations made usingthis model system might not be IBC-specific. In fact,application of the 79-gene classifier onto a collection of

Van Laere et al.





7 IBC cell lines obtained through the World IBC Con-sortium revealed that out of multiple replicates for eachcell line, only the SUM149, KPL4, and FC-IBC-01 celllines were reliably classified as IBC (Prof. Robertson;personal communication).

Disclosure of Potential Conflicts of InterestNo potential conflicts of interest were disclosed.

Authors' ContributionsConception and design: S.J. Van Laere, N.T. Ueno, P. Vermeulen, S.Krishnamurthy, P. van Dam, W.A. Woodward, P. Viens, L. Dirix, F. BertucciDevelopment of methodology: S.J. Van Laere, S. Krishnamurthy, L. Dirix,F. BertucciAcquisitionofdata (provided animals, acquired andmanagedpatients,provided facilities, etc.): S.J. Van Laere, N.T. Ueno, P. Finetti, F.M. Robert-son, S. Krishnamurthy, P. Viens, D. Birnbaum, L. Dirix, F. BertucciAnalysis and interpretation of data (e.g., statistical analysis, biosta-tistics, computational analysis): S.J. Van Laere, P. Finetti, T. Iwamoto,S. Krishnamurthy, L. Dirix, F. BertucciWriting, review, and/or revision of the manuscript: S.J. Van Laere, N.T.Ueno, P. Vermeulen, A. Lucci, F.M. Robertson, M. Marsan, T. Iwamoto,

H. Masuda, P. van Dam, W.A. Woodward, P. Viens, M. Cristofanilli,D. Birnbaum, L. Dirix, J.M. Reuben, F. BertucciAdministrative, technical, or material support (i.e., reporting or orga-nizing data, constructing databases): S.J. Van Laere, P. Finetti, M.Marsan,F. BertucciStudy supervision: S.J. Van Laere, N.T. Ueno, P. Vermeulen, A. Lucci, P. vanDam, P. Viens, D. Birnbaum, F. Bertucci

Grant SupportThe Institut National du Cancer (INCa) Translational Research Grant

2007 (to F. Bertucci); Translational Research Grant 2009 (D. Birnbaum); theLigue Nationale Contre le Cancer (label D. Birnbaum); the NIHR01CA138239-01 and R01 CA138239 ARRA supplement (to W.A. Wood-ward and J.M. Reuben); The State of Texas Grant for Rare and AggressiveCancers (to N.T. Ueno, J.M. Reuben, A. Lucci, and W.A. Woodward); TheAmerican Airlines Komen Foundation Promise Grant KGO81287 (to W.A.Woodward, J.M. Reuben, and A. Lucci); Komen Foundation GrantKG101478 (to W.A. Woodward); and Translational Research Grant 220-2008, Stichting tegen Kanker (L. Dirix).

The costs of publication of this articlewere defrayed in part by thepaymentof page charges. This article must therefore be herebymarked advertisement inaccordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Received August 1, 2012; revised January 18, 2013; accepted January 22,2013; published OnlineFirst February 8, 2013.

References1. Dirix LY, van Dam P, Prov�e A, Vermeulen PB. Inflammatory breast

cancer: current understanding. Curr Opin Oncol 2006;18:563–71.2. Bertucci F, Finetti P, BirnbaumD, Viens P.Gene expression profiling of

inflammatory breast cancer. Cancer 2010;116:2783–93.3. Clarke R, Ressom HW, Wang A, Xuan J, Liu MC, Gehan EA, et al. The

properties of high-dimensional data spaces: implications for exploringgene and protein expression data. Nat Rev Cancer 2008;8:37–49.

4. Van Laere S, Van der Auwera I, Van denEyndenGG, FoxSB, Bianchi F,Harris AL, et al. Distinct molecular signature of inflammatory breastcancer by cDNA microarray analysis. Breast Cancer Res Treat 2005;93:237–46.

5. Bertucci F, Finetti P, Rougemont J, Charafe-Jauffret E, Nasser V,LoriodB, et al.Geneexpressionprofiling formolecular characterizationof inflammatory breast cancer and prediction of response to chemo-therapy. Cancer Res 2004;64:8558–65.

6. Iwamoto T, Bianchini G, Qi Y, Cristofanilli M, Lucci A, Woodward WA,et al. Different gene expressions are associated with the differentmolecular subtypes of inflammatory breast cancer. Breast CancerRes Treat 2011;125:785–95.

7. Boersma BJ, Reimers M, Yi M, Ludwig JA, Luke BT, Stephens RM,et al. A stromal gene signature associated with inflammatory breastcancer. Int J Cancer 2008;122:1324–32.

8. Bekhouche I, Finetti P, Adelaïde J, Ferrari A, Tarpin C, Charafe-JauffretE, et al. High-resolution comparative genomic hybridization of inflam-matory breast cancer and identification of candidate genes. PLoSONE2011;6:e16950.

9. Van Laere S, Van der Auwera I, Van den Eynden G, Van Hummelen P,van Dam P, Van Marck E, et al. Distinct molecular phenotype ofinflammatory breast cancer compared to non-inflammatory breastcancer using Affymetrix-based genome-wide gene-expression anal-ysis. Br J Cancer 2007;97:1165–74.

10. Bi�eche I, Lerebours F, Tozlu S, Espie M, Marty M, Lidereau R. Molec-ular profiling of inflammatory breast cancer: identification of a poor-prognosis gene expression signature. Clin Cancer Res 2004;10:6789–95.

11. Dawood S, Merajver SD, Viens P, Vermeulen PB, Swain SM, BuchholzTA, et al. International expert panel on inflammatory breast cancer:consensus statement for standardized diagnosis and treatment. AnnOncol 2011;22:515–23.

12. Gong Y, Yan K, Lin F, Anderson K, Sotiriou C, Andre F, et al. Deter-mination of oestrogen-receptor status and ERBB2 status of breastcarcinoma: a gene-expression profiling study. Lancet Oncol 2007;8:203–11.

13. Lehmann BD, Bauer JA, Chen X, Sanders ME, Chakravarthy AB, ShyrY, et al. Identification of human triple-negative breast cancer subtypesand preclinical models for selection of targeted therapies. J Clin Invest2011;121:2750–67.

14. Sotiriou C, Wirapati P, Loi S, Harris A, Fox S, Smeds J, et al. Geneexpression profiling in breast cancer: understanding the molecularbasis of histologic grade to improve prognosis. J Natl Cancer Inst2006;98:262–72.

15. Van der Auwera I, YuW, Suo L, VanNeste L, vanDamP, VanMarck EA,et al. Array-based DNAmethylation profiling for breast cancer subtypediscrimination. PLoS ONE 2010;5:e12616.

16. Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ,Scherf U, et al. Exploration, normalization, and summaries of highdensity oligonucleotide array probe level data. Biostatistics 2003;4:249–64.

17. Parker JS, Mullins M, Cheang MC, Leung S, Voduc D, Vickery T, et al.Supervised risk predictor of breast cancer based on intrinsic subtypes.J Clin Oncol 2009;27:1160–7.

18. Prat A, Parker JS, Karginova O, Fan C, Livasy C, Herschkowitz JI, et al.Phenotypic and molecular characterization of the claudin-low intrinsicsubtype of breast cancer. Breast Cancer Res 2010;12:R68.

19. GatzaML, Lucas JE, BarryWT, Kim JW,WangQ,CrawfordMD, et al. Apathway-based classification of human breast cancer. Proc Natl AcadSci U S A 2010;107:6994–9.

20. Goeman JJ, van de Geer SA, de Kort F, van Houwelingen HC. A globaltest for groups of genes: testing association with a clinical outcome.Bioinformatics 2004;20:93–9.

21. Tusher VG, Tibshirani R, Chu G. Significance analysis of microarraysapplied to the ionizing radiation response. Proc Natl Acad Sci U S A2001;98:5116–21.

22. Trinh XB, Tjalma WA, Dirix LY, Vermeulen PB, Peeters DJ, BachvarovD, et al. Microarray-based oncogenic pathway profiling in advancedserous papillary ovarian carcinoma. PLoS ONE 2011;6:e22469.

23. BenitoM,Parker J, DuQ,WuJ, XiangD,PerouCM, et al. Adjustment ofsystematic microarray data biases. Bioinformatics 2004;20:105–14.

24. Van Laere SJ, Vander Auwera I, Van den EyndenGG, Elst HJ,Weyler J,Harris AL, et al. Nuclear factor-kappaB signature of inflammatorybreast cancer by cDNA microarray validated by quantitative real-timereverse transcription-PCR, immunohistochemistry, and nuclear fac-tor-kappaB DNA-binding. Clin Cancer Res 2006;12:3249–56.

25. Lerebours F, Vacher S, Andrieu C, Espie M, Marty M, Lidereau R, et al.NF-kappa B genes have a major role in inflammatory breast cancer.BMC Cancer 2008;8:41.






26. Bertucci F, Finetti P, Rougemont J, Charafe-Jauffret E, Cervera N,Tarpin C, et al. Gene expression profiling identifies molecularsubtypes of inflammatory breast cancer. Cancer Res 2005;65:2170–8.

27. Van Laere SJ, Van den Eynden GG, Van der Auwera I, VandenbergheM, vanDamP,VanMarckEA, et al. Identification of cell-of-origin breasttumor subtypes in inflammatory breast cancer by gene expressionprofiling. Breast Cancer Res Treat 2006;95:243–55.

28. Chen X, Rotenberg SA. PhosphoMARCKS drives motility of mousemelanoma cells. Cell Signal 2010;22:1097–103.

29. Van der Auwera I, Limame R, van Dam P, Vermeulen PB, Dirix LY,Van Laere SJ. Integrated miRNA and mRNA expression profiling ofthe inflammatory breast cancer subtype. Br J Cancer 2010;103:532–41.

30. GotzmannJ,MikulaM,EgerA, Schulte-HermannR, FoisnerR,BeugH,et al.Molecular aspects of epithelial cell plasticity: implications for localtumor invasion and metastasis. Mutat Res 2004;566:9–20.

31. Giampieri S, Manning C, Hooper S, Jones L, Hill CS, Sahai E.Localized and reversible TGFbeta signalling switches breast cancercells from cohesive to single cell motility. Nat Cell Biol 2009;11:1287–96.

32. Giampieri S, Pinner S, Sahai E. Intravital imaging illuminates transform-ing growth factor beta signaling switches during metastasis. CancerRes 2010;70:3435–39.

33. Colpaert CG, Vermeulen PB, Benoy I, Soubry A, Van Roy F, van BeestP, et al. Inflammatory breast cancer shows angiogenesis with highendothelial proliferation rate and strong E-cadherin expression. Br JCancer 2003;88:718–25.

34. Kleer CG, van Golen KL, Braun T, Merajver SD. Persistent E-cadherinexpression in inflammatory breast cancer. Mod Pathol 2001;14:458–64.

35. Charafe-Jauffret E, Tarpin C, Bardou V-J, Bertucci F, Ginestier C,Braud A-C, et al. Immunophenotypic analysis of inflammatory breastcancers: identification of an "inflammatory signature." J Pathol 2004;202:265–73.

36. Ren Y, Wu L, Frost AR, Grizzle W, Cao X, Wan M. Dual effects of TGF-beta on ERalpha-mediated estrogenic transcriptional activity in breastcancer. Mol Cancer 2009;8:111.

37. HugoHJ, KokkinosMI, Blick T, AcklandML, ThompsonEW,NewgreenDF. Defining the E-cadherin repressor interactome in epithelial-mes-enchymal transition: the PMC42 model as a case study. Cells TissuesOrgans 2011;193:23–40.

Van Laere et al.





2013;19:4685-4696. Published OnlineFirst February 8, 2013.Clin Cancer Res Steven J. Van Laere, Naoto T. Ueno, Pascal Finetti, et al. Expression DatasetsBiology: An Integrated Analysis of Three Distinct Affymetrix Gene Uncovering the Molecular Secrets of Inflammatory Breast Cancer

Updated version

10.1158/1078-0432.CCR-12-2549doi:

Access the most recent version of this article at:

Material

Supplementary

http://clincancerres.aacrjournals.org/content/suppl/2013/02/08/1078-0432.CCR-12-2549.DC1

Access the most recent supplemental material at:

Cited articles

http://clincancerres.aacrjournals.org/content/19/17/4685.full#ref-list-1

This article cites 37 articles, 8 of which you can access for free at:

Citing articles

http://clincancerres.aacrjournals.org/content/19/17/4685.full#related-urls

This article has been cited by 9 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]

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

Permissions

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

.http://clincancerres.aacrjournals.org/content/19/17/4685To request permission to re-use all or part of this article, use this link



http://clincancerres.aacrjournals.org/lookup/doi/10.1158/1078-0432.CCR-12-2549

http://clincancerres.aacrjournals.org/content/suppl/2013/02/08/1078-0432.CCR-12-2549.DC1

http://clincancerres.aacrjournals.org/content/19/17/4685.full#ref-list-1

http://clincancerres.aacrjournals.org/content/19/17/4685.full#related-urls

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

mailto:[email protected]

http://clincancerres.aacrjournals.org/content/19/17/4685


Documents

Uncovering the Molecular Secrets of Inﬂammatory Breast ...Uncovering the Molecular Secrets of Inﬂammatory Breast Cancer Biology: An Integrated Analysis of Three Distinct Affymetrix