UBC Faculty Research and Publications

Characterization of heterotypic interaction effects in vitro to deconvolute global gene expression profiles… Buess, Martin; Nuyten, Dimitry S; Hastie, Trevor; Nielsen, Torsten; Pesich, Robert; Brown, Patrick O Sep 14, 2007

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


52383-13059_2007_Article_1672.pdf [ 2.11MB ]
JSON: 52383-1.0215893.json
JSON-LD: 52383-1.0215893-ld.json
RDF/XML (Pretty): 52383-1.0215893-rdf.xml
RDF/JSON: 52383-1.0215893-rdf.json
Turtle: 52383-1.0215893-turtle.txt
N-Triples: 52383-1.0215893-rdf-ntriples.txt
Original Record: 52383-1.0215893-source.json
Full Text

Full Text

Open Access2007Buesset al.Volume 8, Issue 9, Article R191ResearchCharacterization of heterotypic interaction effects in vitro to deconvolute global gene expression profiles in cancerMartin Buess*, Dimitry SA Nuyten§, Trevor Hastie†, Torsten Nielsen¶, Robert Pesich* and Patrick O Brown*‡Addresses: *Department of Biochemistry, Stanford University School of Medicine, Stanford, CA 94305, USA. †Department of Statistics, Stanford University School of Medicine, Stanford, CA 94305, USA. ‡Howard Hughes Medical Institute, Stanford University School of Medicine, Stanford, CA 94305, USA. §Departments of Radiation Oncology and Diagnostic Oncology, Netherlands Cancer Institute, 1066 CX Amsterdam, The Netherlands. ¶Department of Pathology and Laboratory Medicine, University of British Columbia, Vancouver, BC, Canada, V5Z 1M9. Correspondence: Patrick O Brown. Email: pbrown@pmgm2.stanford.edu© 2007 Buess et al.; licensee BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.Deconvolution of cancer expression profiles<p>I  an eff rt to deconvolut  global gene-expression profiles, an interaction between some breast cancer cells and stromal fibroblasts was found t  induce an i t f ron response, which may be associated wi h  greater propensity fo  tumor progression.</p>AbstractBackground: Perturbations in cell-cell interactions are a key feature of cancer. However, little isknown about the systematic effects of cell-cell interaction on global gene expression in cancer.Results: We used an ex vivo model to simulate tumor-stroma interaction by systematically co-cultivating breast cancer cells with stromal fibroblasts and determined associated gene expressionchanges with cDNA microarrays. In the complex picture of epithelial-mesenchymal interactioneffects, a prominent characteristic was an induction of interferon-response genes (IRGs) in a subsetof cancer cells. In close proximity to these cancer cells, the fibroblasts secreted type I interferons,which, in turn, induced expression of the IRGs in the tumor cells. Paralleling this model,immunohistochemical analysis of human breast cancer tissues showed that STAT1, the keytranscriptional activator of the IRGs, and itself an IRG, was expressed in a subset of the cancers,with a striking pattern of elevated expression in the cancer cells in close proximity to the stroma.In vivo, expression of the IRGs was remarkably coherent, providing a basis for segregation of 295early-stage breast cancers into two groups. Tumors with high compared to low expression levelsof IRGs were associated with significantly shorter overall survival; 59% versus 80% at 10 years (log-rank p = 0.001).Conclusion: In an effort to deconvolute global gene expression profiles of breast cancer bysystematic characterization of heterotypic interaction effects in vitro, we found that an interactionbetween some breast cancer cells and stromal fibroblasts can induce an interferon-response, andthat this response may be associated with a greater propensity for tumor progression.Published: 14 September 2007Genome Biology 2007, 8:R191 (doi:10.1186/gb-2007-8-9-r191)Received: 26 March 2007Revised: 14 June 2007Accepted: 14 September 2007The electronic version of this article is the complete one and can be found online at http://genomebiology.com/2007/8/9/R191Genome Biology 2007, 8:R191BackgroundCommunication between different cell types is fundamentalfor the development and homeostasis of multi-cellular organ-isms. Cells of different origin communicate in a network ofinteractions via proteins, peptides, small molecular signals,the extracellular matrix and direct cell-cell contact. TheseR191.2 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. http://genomebiology.com/2007/8/9/R191heterotypic interactions provide information that is neces-sary for the regulation of the gene expression programs innormal development [1], differentiation [2], topologic organ-ization [3] and homeostasis [4] of complex tissue structures.Given the important physiological role of intercellular com-munication to maintain the delicate dynamic equilibrium of anormal tissue, it is not surprising that aberrant cell-cell inter-action signals have been implicated in cancer developmentand progression [5-10]. Although the characteristics androots of the heterotypic interaction effects are fundamentalaspects of normal physiology and disease, they have not beensystematically explored.In cancer biology, there is increasing evidence for the impor-tance of the interaction between the malignant epithelial cellsand the surrounding stromal cells [7]. Tumors are not merelyaggregates of malignant cells but are in many respects organ-like structures, which include host stromal cells, such asfibroblasts, endothelial cells and so on, with which the malig-nant cells themselves intermingle and interact. Inductiveinteractions between these different cell lineages can play notonly a morphogenetic role but also an important mechanisticrole in the pathogenesis and progression of malignancy. Co-inoculation of stromal cells with pre-malignant or malignantepithelial cells can increase tumorigenicity and the capacity tometastasize for a variety of tumor types [11,12], includingbreast cancer [13]. On the molecular level, results from theknockout of single genes have demonstrated the importanceof specific signaling pathways in the tumor-stroma interac-tion. For example, conditional inactivation of the transform-ing growth factor (TGF)-β receptor type II in stromal cells ledto development of epithelial cancer of the prostate andforestomach in mice [14]. In the mammary gland, site-spe-cific knockout of TGF-β receptor type II in stromal fibroblastsled to defective mammary ductal development and increasedcarcinoma growth and metastasis [15]. Experiments explor-ing the interaction of tumor with stromal cells in vitro haverevealed changes in expression of several genes involved incancer [16-18]. These effects reveal the significance of onespecific signaling mechanism, but a more complete overviewof the molecular systems that mediate these cell-cell interac-tion effects remains to be revealed.Biopsy samples of human carcinoma frequently contain bothmalignant cells and stromal cells. Since gene expression pro-files of human cancer are generally derived from these mixedcell populations of grossly dissected tissues, the effects of het-erotypic interactions among the cells in the tumor tissue areexpected to leave their traces in the global gene expressionprofiles. Datasets representing expression profiles of thou-sands of genes in collections of benign and malignant tissuesfrom hundreds of patients have steadily grown in recent yearsand might be a rich latent source of insights into heterotypicbreast cancer, Allinen et al. [19] attempted to resolve thiscomplexity by fractionating the tissue using cell-surfacemarkers to separate different cell types. This led to the iden-tification of cell type specific gene expression profiles. As aresult of this analysis they suggested that a myofibroblastexpression of CXCL14 and CXCL12, which can bind to therespective receptor CXCR4 on the epithelial cells, is a specifictumor promoting mechanism leading to enhanced prolifera-tion, invasion and metastasis. In a different approach tosearch for the relevance of stromal signals in cancer data,West et al. [20] identified stromal-cell specific gene expres-sion signatures in breast cancer using gene expression datafrom fibroblastic tumors as in vivo models of homogenouspopulations of malignant mesenchymal cells. Based on stro-mal-cell specific signatures they were able to segregate breastcancer samples into two subgroups with distinct clinicaloutcome.A further layer of complexity, in addition to the simple addi-tive effects of the involved cells, might arise from the effectson gene expression profiles induced by heterotypic cell-cellinteractions. The deconvolution of these intercellular signal-ing effects poses an even greater challenge, since they resultin supra-additive non-linear behaviors, which are hard to dis-entangle and distinguish from the cell-intrinsic regulatoryprocesses. These cell interaction effects might account for asignificant proportion of the unrevealed information in thegene expression data from tissue specimens. Given the evi-dence that interactions between cells can play critical roles intumor progression, such data might be even more meaningfulthan prominent expression patterns that are driven by theproportional representation of a given cell type in a tissue[21].The primary aim of this work was to survey and characterizethe effects of cell-cell interaction in an attempt to disentanglethe complex network of intercellular signaling in a multi-cel-lular tissue and specifically in breast cancer. To extract theinformation about tumor-stroma interaction from globalgene expression profiles of cancer tissue, we applied anapproach based on in vitro modeling combined with subse-quent testing of the in vitro findings in published cancer data-sets. Observation of fundamental biological processes invitro, such as the cell cycle [22] and the reaction of fibroblaststo serum [23,24], or observation of the common response ofdifferent cell types to hypoxic conditions [25] has proven tobe a worthwhile approach to better understand complex bio-logical mechanisms underlying global gene expression pro-files in human cancer. Using a simple ex vivo co-culturesystem allowed us to address a few basic questions about het-erotypic cell-cell interactions. First, is global gene expressionin a co-culture setting different from the expression in mono-culture and, if so, in which respect is it different? Second, howGenome Biology 2007, 8:R191interaction effects on global gene expression. The superposi-tion of the cell specific profiles, however, results in complexgene expression patterns that are difficult to interpret. Indo the responses to co-culture differ among different cellcombinations? Third, are the in vitro observations transfera-ble in vivo using published gene expression datasets from191.3http://genomebiology.com/2007/8/9/R191 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. Rhuman tissue specimens? We analyzed heterotypic interac-tion effects in stromal fibroblasts and a diverse set of benignand malignant breast epithelial cells in a mixed co-culturesetting by measuring changes in global gene expression usingDNA microarrays. The global view of the gene expressionresponses facilitated the identification of specific changes andpathways underlying these effects. Gene expression signa-tures paralleling a response to heterotypic interaction in thisex vivo model were shared by clinically distinct subgroups ofbreast cancer.ResultsIdentification of heterotypic interaction effectsAs a model for investigating the gene expression program inresponse to heterotypic cell-cell interaction in normal breastand in breast cancer, we examined cells representing thebenign and malignant epithelial cell compartment and themesenchymal cell compartment in an in vitro mixed co-cul-ture setting. The cells were co-cultivated for 48 h in low fetalbovine serum medium (0.2% FBS) to allow reciprocal signalexchange with minimal background from the influence ofundefined molecular signals inherent in FBS. We examinedthe effects of co-cultivation for each cell pair in at least twoindependent biological replicates. The gene expression pro-files of the co-cultures were compared to the expression pro-files of the corresponding cells kept in monoculture usingcDNA microarrays containing approximately 40,700 ele-ments, representing 24,472 unique Unigene clusters (buildnumber 173, released on 28 July 2004). To establish theexperimental approach, we first focused our experiments onthe breast cancer cell line MDA-MB231, the primary fibro-blast CCL-171 and the co-culture of these two cell types. Thedata were organized using unsupervised hierarchical cluster-ing of the replicate experiments to provide an overview of theeffects on global gene expression (Figure 1a). In the co-cul-ture, most genes displayed intermediate expression levels,which closely approximated the proportionally weightedaverage of their expression levels in the two cell types inmonoculture. However, one set of genes showed a consistent,significant increase in transcript abundance in the co-culturecompared to either monoculture, suggesting that induction ofthese genes was an effect of co-cultivation. Most of theseinduced genes were known to be interferon regulated (Figure1b). They included those encoding the myxovirus resistanceproteins 1 and 2 (MX1 and MX2), 2',5'-oligoadenylate syn-thetase 1 and 2 and 3 (OAS1, OAS2, OAS3) and interferon-induced protein with tetratricopeptide repeats 1 (IFIT1),phospholipid scramblase 1 (PLSCR1), eukaryotic translationinitiation factor 2-alpha kinase (EIF2AK2) and the signaltransducer and activator of transcription (STAT1). One ofthese genes, EPSTI1, had previously been reported to beinduced by co-cultivation of MDA-MB231 and a fibroblastlinked to interferon induction (for example, zinc finger pro-tein 187 (ZNF187), Homo sapiens peroxisomal proliferator-activated receptor A interacting complex 285 (PRIC285),hect domain and RLD 6 (HERC6)), we have confirmed thatthey are induced in MDA-MB231 cells by treatment with atype I interferon. As a more explicit approach to identifygenes with consistent changes in expression in response toco-culture we used significance analysis of microarrays(SAM) [26]. A set of 42 genes represented by 49 image cloneswere identified with a false discovery rate (FDR) of 0 (Addi-tional data file 1).To further validate the results obtained by cDNA microarrayanalysis OAS2 transcript levels were measured by quantita-tive real time PCR (Figure A in Additional data file 2). More-over, for STAT1 the increase in transcript levels in co-culture(2.8-fold) was paralleled by an increase in STAT1 protein asdetected by fluorescence assisted cell sorting (FACS) analysis(Figure B in Additional data file 2).Since breast cancer is a clinically and molecularly heterogene-ous disease, we selected a broad spectrum of different breastcancer cell lines to sample this heterogeneity and explored theeffects of heterotypic culture looking for subtype-specific andshared response patterns. We focused on epithelial-mesen-chymal interactions co-cultivating fibroblasts of different ori-gins (HTB125 (breast stromal fibroblast), HDF (fibroblastfrom breast skin) and CCL-171 (embryonic lung fibroblast)),in combination with normal breast epithelial cells (humanmammary epithelial cells (HMECs)) and seven widely usedbreast cancer cell lines.The changes in gene expression due to heterotypic interactionwere subtle compared to the large intrinsic variation inexpression patterns among the involved cell types, as Figure1a illustrates for the cell pair MDA-MB231 and CCL-171. Toidentify the gene expression changes resulting from cell-cellinteraction, we needed to control for the simple additiveeffects that reflect the proportional contribution of the twocell types to the total population of each gene's transcript inthe co-culture. Eliminating these proportionally weightedadditive contributions would allow us to isolate supra-addi-tive interaction effects. The fact that transcript levels of mostgenes did not change in response to co-culture allowed a lin-ear regression model based on the transcript profiles of eachmonoculture to be fitted to the co-culture data for normaliza-tion. An example of such an analysis is shown in Figure 2a.For each gene, the ratio of the measured transcript level andthe level estimated by the linear model provides a measure ofthe heterotypic interaction effect. This is illustrated in Figure2b, which shows the distribution of the gene expressionchanges of the CCL-171/MDA-MB231 co-culture. The genesidentified by SAM as differentially expressed in co-cultureGenome Biology 2007, 8:R191[17]. Our results suggest that the interferon response pathwaymediates this induction. Although several of the genesinduced in this co-culture model have not previously beencompared to monoculture are highlighted to illustrate theperformance of this approach. Interaction effects, repre-sented as gene-expression changes, are converted toR191.4 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. http://genomebiology.com/2007/8/9/R191quantitative values that can be analyzed for similarities anddisparities over multiple different pair-wise interactionsbetween cells with the same tools we use to analyze conven-tional gene expression data.There was obvious heterogeneity in the responses of differentpairs of cells to co-cultivation. The patterns of gene expres-sion changes due to co-cultivation were mainly determined bythe type of the epithelial cell involved whereas the origin ofthe fibroblasts had a minor influence. Importantly, CCL-171,a lung fibroblast, and HTB125, a fibroblast derived from thebreast of a cancer patient, induced distinct but very similarinterferon responses in co-cultivation with different epithe-lial cells (Figure 2c). To highlight consistent features of theresponses of distinct normal or malignant epithelial cells,representing the distinct types of breast cancer, to co-cultiva-tion with fibroblasts, we collapsed our data into eight groups,one group for each epithelial cell co-cultured with three dif-ferent types of fibroblasts. There were 3,000 genes thatshowed a significant reproducible change (FDR < 1%) in tran-script levels in response to co-culture in at least one of thegroups. Clustering the averaged values of co-culture-inducedchanges for each group revealed specific and shared effects(Figure 2d). For several cell combinations, co-cultures led toan induction of smooth muscle actin (ACTA2), myosin regu-latory light chain interacting protein (MYLIP), myosin, lightpolypeptide kinase (MYL), myosin regulatory light chain 2,smooth muscle isoform (MYL9), calponin 2 (CNN2) andfibronectin (FN1). Induction of these genes has previouslybeen described to be associated with the acquisition of amyofibroblast phenotype [27]. The ability of the tumor cellsto induce this 'myofibroblast' expression program variedamong the breast cancer cell lines; the strongest effect wasseen with MCF7 cells. In a previous study, conditionedmedium of MCF7 cells was shown to induce a myofibroblastphenotype [28]. Targets of the TGF-β pathway, such as thegene encoding latent transforming growth factor beta bind-ing protein LTBP2 and transforming growth factor inducedgene TGFBI, were induced in parallel with the 'myofibroblastresponse'. In fact, TGF-β has previously been shown to inducea 'myofibroblast' phenotype [29], suggesting that theresponse observed in these co-cultures might be mediated bythe TGF-β pathway.The most consistent coordinated response, however, was aninduction of interferon-associated genes by cultivation offibroblasts with four of the seven breast cancer cell lines. Thisresponse was seen in the co-cultures involving the estrogen-receptor negative breast cancer cell lines MDA-MB231, MDA-MB436, Hs578T and BT549, but neither in HMECs nor in theestrogen-receptor positive breast cancer cells MCF7, T47Dand SKBR-3. Although the gene expression profiles of theseepithelial cells grown as monocultures reflected their molec-Effect of heterotypic interaction between breast cancer cell line MDA-MB231 and CCL-171 fibroblastsFigure 1Effect of heterotypic interaction between breast cancer cell line MDA-MB231 and CCL-171 fibroblasts. (a) Biologically independent replicates of the monocultured fibroblast CCL-171, the breast cancer cell line MDA-MB231 and the mixed co-culture of CCL-171 and MDA-MB231 were grown for 48 h at low serum conditions and characterized by DNA microarray hybridization. Hierarchical clustering of a total of 4,333 elements that display a greater than 3-fold variance in expression in more than 3 different experimental samples. Data from individual elements or genes are represented as single rows, and different experiments are shown as columns. Red and green denote expression levels of the samples. The intensity of the color reflects the magnitude of the deviation from baseline. Unsupervised hierarchical clustering of the experiments grouped the biological replicates together. Gene expression varied considerably between fibroblast and MDA-MB231 cultures, as expected for cells of mesenchymal or epithelial origin, respectively. The co-culture profile showed mainly intermediate expression levels. However, the vertical black bar marks a cluster of genes induced in all co-cultures compared to both monocultures, indicating that they are induced by heterotypic interaction. (b) Zooming in on the genes up-regulated in co-culture compared to monocultures reveals that they are associated with the response to interferon.PLSCR1IFIH1IFIT1MX1MX2EPSTI1STAT1IFITM1CCL7CXCL10ISGF3GINADLG1P2ISG20IRF7TRIM25IFRG28IFIH1PRKROAS1IFIT2IFI35AIM2IFI27OAS2CCL20CCL171MDA-MB231CCL171/MDA-MB231 >8<8(a) (b)Genome Biology 2007, 8:R191ular differences, including some consistent differencesbetween the estrogen-receptor negative and estrogen-recep-tor positive breast cancer cell lines, there were no consistent191.5http://genomebiology.com/2007/8/9/R191 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. Rdifferences between these groups in baseline expression ofthe interferon-induced genes in the monocultures. The cell-type specificity is a strong hint that the interferon-responseactivation is a specific effect of heterotypic interaction. Sincewe compared the gene-expression responses in the co-cul-tures with the responses in the corresponding monocultureskept under the same conditions, we can exclude responses toserum stimulation or withdrawal as sources of the interferonresponse observed in these experiments. The response doesnot represent an effect of crowding, which is a known inducerof an interferon response [30], since the cell density in ourexperiments was maintained below the threshold at whichthe interferon response genes were turned on (data notshown). Furthermore, we were unable to identify any infec-tive agent in any of the cultures despite extensive testing formycoplasma, reverse transcriptase activity and viral tran-scripts, using microarrays that provide a broad survey ofhuman viruses [31] (data not shown). The consistent cell-typespecific, coordinated response suggests that it depends on aspecific physiological feature shared among the estrogen-receptor negative human breast cancers, which is retained inlong-term culture, enabling them to activate this specificresponse upon contact with stromal cells.Localizing expression of interferon-response genes to breast cancer cell linesWe investigated in which cell the interferon-response geneswere induced in response to heterotypic interaction by differ-entially labeling the epithelial cells and the fibroblasts withdistinct fluorescent dyes prior to co-culture, then sortingthem after co-culture using FACS. Comparing gene expres-sion patterns of cells in monoculture with those of the samecell type after co-cultivation showed that in the CCL-171fibroblasts, the interferon-response genes were induced onaverage by a factor of only 2.7 whereas in the MDA-MB231breast cancer cell line these genes were induced 11-fold (Fig-ure 3a). This result of a predominant induction in the tumorcell is in line with immunohistochemical evidence that in vivothe interferon- response genes STAT1, EPSTI1 [17] andEIF2AK2 [32] are expressed in the malignant epithelial cellsand to a much lesser extent in the stroma. To test whether asoluble factor is sufficient to induce the interferon responsegenes or whether direct cell-cell contact is needed for theirinduction, we let the cells interact in transwell co-cultures atlow serum conditions. In this setting, neither the MDA-MB231 breast cancer cell line nor the CCL- 171 fibroblastsshowed induction of interferon response genes, indicatingthat close cell-cell contact is necessary for interaction. If theinduction of interferon-response genes depended on short-range epithelial-mesenchymal interactions, we would expectto find the expression of interferon-response genes mainly atthe tumor-stromal interface. To test this hypothesis westained normal breast and breast cancer sections using anti-No staining was evident in normal breast samples. In tumortissue sections consisting of a homogenous tumor islandsurrounded by stroma, we typically observed a distinctivepattern of STAT1 expression concentrated at the periphery ofthe tumor islands, near the tumor-stroma boundary, support-ing the idea that the interferon-response genes are inducedpreferentially in the tumor cells in closest proximity to thestromal cells. The gradient in the response further suggestsinvolvement of a soluble factor acting over a short range.Induction of interferon in co-cultureTo investigate the possible roles of soluble factors or directcell-cell contact in triggering the observed interferonresponse, we tested the ability of conditioned medium fromselected cultures to induce the response in a monoculture ofMDA-MB231 cells. Conditioned medium from monoculturesof either CCL-171 or MDA-MB231 cells did not induce inter-feron-response genes. However, conditioned media from anMDA-MB231/CCL-171 co-culture did induce the interferonresponse genes in MDA-MB231 cells. Thus, interferon-response genes are induced by a soluble factor, the inductiondepending upon direct contact between the tumor cells andfibroblasts. In contrast to the MDA-MB231/CCL-171 co-cul-ture supernatant, the conditioned medium of the T47D/CCL-171 co-culture did not induce the interferon response geneswhen applied onto MDA-MB231 cells (Figure 4a). Con-versely, when T47D cells were exposed to MDA-MB231/CCL-171 co-culture supernatant, the interferon-response wasinduced (Figure 4b). However, the response of the T47D cellsto the co-culture supernatant was weaker than that of theMDA-MB231 cells. This implies that while the interferon-response genes can be induced in either tumor cell line, onlythe interaction of MDA-MB231 with fibroblasts released asoluble factor into the medium capable of inducing an inter-feron response. We speculated that the factor released by thefibroblasts might be a type I interferon. To confirm andlocalize the expression of type I interferon we used quantita-tive RT-PCR to analyze sorted cells after co-cultivation. Wefound over-expression of IFNβ in CCL-171 in response tointeraction with MDA-MB231 but not in response to T47D(Figure 4c). Expression analysis of IFNα gave us the sameresult (data not shown), indicating that the expression of typeI interferon genes by co-cultured fibroblasts might underliethe observed interferon response.Taken together, these results demonstrate that heterotypicinteraction between fibroblasts and a specific subset of breastcancer cells can induce the fibroblasts to express type I inter-ferons, resulting, in turn, in induction of interferon-responsegenes in the tumor cells and to a lesser extent in the fibrob-lasts (Figure 5). In our in vitro system, both estrogen-recep-tor positive and estrogen-receptor negative tumor cells areresponsive to type I interferons, but the ability to induceGenome Biology 2007, 8:R191bodies specific for STAT1, the key transcriptional activator ofthe interferon-response genes, and itself a protein over-expressed in response to interferon stimulation (Figure 3b).expression of interferons in co-cultivated fibroblasts was spe-cific to the estrogen-receptor negative breast cancer cells.R191.6 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. http://genomebiology.com/2007/8/9/R191Interferon-response"Myofibroblast"/ TGF-beta(a)(b)(d)Weighted average expressionbased on linear regression fitMeasured expression levelsFold change Genes>4<4Hs578TBT549MDA-MB436MDA-MB231HMECSKBR-3MCF7T47D-3-2-10123450123456(c)Fold change CCL-171/MDA-MB231HTB-125/MDA-MB231CCL-171/HMECHTB-125/HMECCCL-171/MCF7HTB-125/MCF7Genome Biology 2007, 8:R191Figure 2 (see legend on next page)191.7http://genomebiology.com/2007/8/9/R191 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. RGenomic analysis of epithelial-mesenchymal interaction effects in human cancersInteractions between cancer cells and non-malignant cells inthe surrounding microenvironment are important determi-nants of cancer development and progression [11,14,33,34].We reasoned that identifying and characterizing gene expres-sion programs characteristically induced by interactionbetween specific pairs of cells in culture might enable us torecognize and interpret specific features in the expressionprofiles of human cancer that reflect similar interactionsbetween tumor and stromal cells in vivo. The most consistentresponse to ex vivo co-cultivation of breast cancer and stro-mal cells was the induction of the interferon-response genes.We therefore looked for this response in the expression pat-terns in published data from 295 early stage (stage I and II)breast cancer samples from the Netherlands Cancer Institute(NKI) [35] (Figures 6a,b and 7). The interferon-responsegenes showed a strikingly coherent variation in expressionamong these cancers, enabling these cancers to be dividedinto two groups, one with relatively high expression and theother with relatively low expression of the interferon-response genes. Clustering the breast carcinomas based onlyon expression of the interferon response genes directed theminto two main clusters, one with high-level expression of mostof the interferon genes and the other with lower expression ofthese genes (Figure 6a). The same coordinated behavior andsegregation of tumors could be observed in a different set ofadvanced breast cancer samples [36,37], suggesting that var-iation in this interferon-response program is a general featurein breast cancer (Additional data file 3).As a first assessment of its potential biological relevance, wecompared distant metastasis-free survival and overall-spe-cific survival between the two groups distinguished by theinterferon-response genes (Figure 6b). We found that tumorswith high expression levels of interferon-response genes hada significantly shorter metastasis-free survival (p = 0.0014;58% at 10 years) and overall survival (p = 0.001; 59% at 10years) than tumors with low expression levels (metastasis freesurvival, 74% at 10 years; overall survival, 80% at 10 years).The same trend toward unfavorable outcome in patients withcancers showing high levels of interferon-response gene tran-scripts (p = 0.067) could be seen in an analysis of publisheddata from advanced-stage breast cancers [36,37]. As a metricthat can be compared to known prognostic parameters andapplied to other prospectively collected samples, we definedOverview of gene expression changes over multiple co-cultures of breast cancer cell lines and normal breast epithelial cells with fibroblastsFigure 2 (see pr vious page)Overview of gene expression changes over multiple co-cultures of breast cancer cell lines and normal breast epithelial cells with fibroblasts. (a) Correlation of the measured co-culture gene expression levels and their estimated expression levels based on the proportional contribution of each cell type determined by a linear regression fit of the monoculture to the co-culture data. (b) Fold change of each gene associated with co-culturing of MDA-MB231 and CCL-171. Genes of the interferon response gene set (Additional data file 1) as determined by SAM are indicated in red. (c) Fold change in expression of the interferon response gene set (Additional data file 1) in co-culture of MCF-7, HMECs and MDA-MB-231 with either the CCL-171 lung fibroblast or the HTB-125 breast fibroblast, showing that CCL-171 and HTB-125 induce a distinct, but very similar response in co-culture with different epithelial cells. (d) Overview of collapsed data from repeat co-culture experiments of eight benign and malignant epithelial cells with three different fibroblasts. Hierarchical clustering of the interaction effects of 3,000 genes in co-cultures of 7 breast cancer cell lines and normal breast epithelial cells with fibroblasts. Red and green denote relative changes in expression associated heterotypic interaction. The magnitude of the relative change is given by color intensity.Interferon response gene induction in co-cultivated cellsFigure 3Interferon response gene induction in co-cultivated cells. (a) MDA-MB231 breast cancer cells and CCL-171 fibroblasts were labeled before co-culture with the fluorescent carbocyanine dye DiO and isolated after co-culture using FACS, which allowed a purification of 95%. Comparing gene expression patterns of the cells cultivated in monoculture to the same cell type after co-cultivation showed that the CCL-171 fibroblasts up-regulate the interferon response genes 2.8-fold on average, whereas the MDA-MB-231 breast cancer cell line up-regulates them about 11-fold. (b) Immunohistochemistry for STAT1. STAT1 expression in a normal breast (left panel) and in a breast cancer specimen (right panel). STAT1 is Normalized OAS2 mRNA expression1mm(b)(a)024681012141 2 3 4CCL171MDA-MB231CCL171with MDA-MB231MDA-MB231with CCL171Genome Biology 2007, 8:R191an 'interferon-response score' by averaging the gene expres-sion levels for the 42 genes of the interferon-response genelist. The interferon response did not significantly correlatepredominantly expressed in the malignant epithelial cells at the stromal interface in a centrifugal gradient.R191.8 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. http://genomebiology.com/2007/8/9/R191with clinical parameters such as age of the patient, tumor size,nodal stage or angio-invasion. It was, however, very signifi-cantly correlated with tumor grade and estrogen receptorstatus (p < 10-6; Additional data file 4), paralleling our in vitrofindings that cell lines representing estrogen-receptornegative tumors preferentially induce the interferon-response genes in co-culture.We also investigated the relationship between the interferon-response gene signature and three previously identified gene-expression signatures, which were useful prognosticators inthis dataset. The first signature is a set of 70 genes [38], whichwas identified in a supervised analysis of a subset of the NKIearly stage breast cancer dataset [35], to predict freedomfrom metastasis at 5 years. The second signature was identi-fied in vitro by exposing fibroblasts to serum to mimic awound response, and has been shown to predict risk of pro-gression [39]. The third signature, the response to hypoxia invitro [25], is also associated with a poor prognosis. The inter-feron signature was only very weakly correlated with eitherthe wound signature or the hypoxia signature, andmoderately correlated with the 70-gene prognostic profile,whereas the wound signature and the 70-gene score weremore strongly correlated to one another (Figure 7). Thus, theinterferon response appears to be a distinct feature of breastcancer biology, identifying a subgroup of cancers with ahigher propensity for progression.STAT1 protein expression in a second independent breast cancer datasetAs an independent test of the prognostic significance of inter-feron-response gene expression in primary early stage breastcancer we performed immunohistochemical staining forSTAT1 in a tissue collection derived from a case series ofwomen who underwent surgery for primary breast cancer atthe Vancouver General Hospital between 1974 and 1995 [40].Consistent with the variation we found in interferon-responsegene expression in breast tumors, we found a large variabilityin the expression of STAT1 protein, the principal transcrip-tional regulator of the interferon response genes, in thesetumors. Of the 353 primary tumors with interpretable results,102 displayed high (28.9%), 184 low (52.2%) and 67 absent(18.9%) STAT1 expression. Paralleling the results from theNKI dataset, patients from Vancouver with tumors displayinghigh STAT1 expression levels had a higher risk of death due tobreast cancer (33% dead from breast cancer at 10 years) thanpatients with tumors showing low or absent STAT1 expres-sion (25% dead at 10 years) (p = 0.056; Figure 8).DiscussionThe main objective of this study was to examine and charac-terize the effects of heterotypic cellular interaction, to gaininsight into the underlying biology of these effects in normalmammary tissue and breast cancer. To isolate specific, directinteractions from more complex interactions involving multi-ple cell types in a whole tissue or organism we used a simpleex vivo co-culture model. Since some important heterotypicinteractions can require direct cell-cell contact, we focused ona co-culture model where the two cell types were mixed. Achallenge in the analysis of a mixed co-culture model is theseparation of the interaction effects induced by signalexchange between the two cell types from the simple additivecombination of their intrinsic gene expression patterns in theoverall gene expression profile of the co-culture. Our strategyof normalizing for the simple additive effects based on a linearregression model proved to be advantageous, since it does notdepend on prior knowledge of the exact proportional contri-bution of the different cell types to the superposed geneexpression pattern. A similar approach has been described todefine the proportional contribution of different cell cyclestates in a mix of cells, although without taking into accountinteraction effects [41]. This strategy was effective in isolatingthe cell-cell interaction effects on gene expression.We examined the effects on global gene expression of themolecular crosstalk between stromal fibroblasts and each of adiverse set of breast cancer cell lines or normal breast epithe-lial cells as they interact in vitro. Not unexpectedly, thepicture of heterotypic interaction effects that emerged fromcombinatorial co-cultivation of multiple different cell typeswas complex, reflecting the different abilities of normal andmalignant cells to send and to respond to extrinsic signals.The overall pattern of gene expression changes were domi-nated by the type of epithelial cells. Against our expectations,which were based on the knowledge that fibroblasts from dif-ferent parts of the body show distinct gene expression pat-terns [42] leading to different p hysiological properties thatInduction of interferon response in two types of breast cancer cell linesFig re 4 (see following page)Induction of interferon response in two types of breast cancer cell lines. (a) MDA-MB231 cells were incubated in conditioned media from CCL-171 monoculture, MDA-MB231 monoculture, T47D monoculture, CCL-171/MDA-MB231 co-culture and CCL-171/T47D co-culture. OAS2 gene expression was determined by quantitative RT-PCR. The gene expression level of GAPDH was used for normalization between the samples. A strong induction of OAS2 by the supernatant from the CCL-171/MDA-MB231 co-culture can be seen in MDA-MB231. (b) Incubation of T47D cells with conditioned media from CCL-171 monoculture, MDA-MB231 monoculture, T47D monoculture, CCL-171/MDA-MB231 co-culture and CCL-171/T47D co-culture showed that only the CCL-171/MDA-MB231 co-culture supernatant induced OAS2 gene expression, although to a much lesser extent than in MDA-MB231 cells. (c) Gene expression levels of IFNβ were determined by quantitative RT-PCR. CCL-171 cells show much higher IFNβ expression levels when isolated by FACS after co-culture with MDA-MB231 than with T47D cells. Expression levels in tumor cells are shown as controls. The error bars show the standard Genome Biology 2007, 8:R191deviation from the normalized mean.191.9http://genomebiology.com/2007/8/9/R191 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. R0. 2 3 4 5Normalized OAS2 mRNAexpression level0. 2 3 4 5Normalized Interferon-βmRNA expression levelCCL171MDA-MB231T47DCCL171 isolated  afterMDA-MB231 exposureCCL171 isolated afterT47D exposure(c) 3 4 5Normalized OAS2 mRNAexpression level(b)(a)CCL171MDA-MB231T47DCCL171/MDA-MB231CCL171/T47DCCL171MDA-MB231T47DCCL171/MDA-MB231CCL171/T47DGenome Biology 2007, 8:R191Figure 4 (see legend on previous page)R191.10 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. http://genomebiology.com/2007/8/9/R191persist through many passages of in vitro cultivation, thesource of the fibroblast had only a minor influence on geneexpression responses to heterotypic interaction in our co-cul-ture system.We cannot exclude the possibility that fibroblasts isolatedfrom within a tumor might show additional specific interac-tion effects. Nevertheless, it would be surprising if carcinomaassociated fibroblasts failed to show the strong effects that weconsistently observed in co-cultures with fibroblasts ofdiverse origin. We recognize that these experiments might beinsufficient to detect subtle differences between co-culturesinvolving different types of fibroblasts. To rigorously evaluatethese differences a more extensive survey of co-culture condi-tions would be needed.In our co-culture system a subset of tumor-stroma combina-tions showed induction of a set of genes characteristic of a'myofibroblast' phenotype. In the same co-cultures, targetgenes of the TGF-β pathway were induced in parallel. Thiscoordinated induction is in line with reports describing TGF-β as the major trigger of a 'myofibroblast' phenotype [29]. Invivo, activation of a contractile 'myofibroblast' phenotype inthe tumor stroma occurs in a subgroup of patients, leading toshrinkage of the tumor environment causing skin dimplingand nipple retraction, both cardinal signs indicative of breastcancer. This example demonstrates how the analysis ofheterotypic interaction effects allows inferring signalingpathways involved in specific physiological and morphologi-cal changes of importance in breast cancer.The most prominent recurring theme arising from the heter-otypic interactions we examined was the induction of aninterferon-response program in cell lines derived from estro-gen-receptor negative breast cancers upon co-culture withfibroblasts. Interferon-response genes showed a strikinglymalignancies, including leukemias [43], ovarian cancer [44],gastric cancer [45], lung cancer [46] and breast cancer[30,43]. In breast cancer, in an attempt to validate the previ-ously described intrinsic gene signatures [36,37], Hu et al.[47] assigned a small group of tumors with very high geneexpression levels known to be induced by interferon as the'interferon subtype' with a poor clinical outcome. Despite itscommon occurrence, the origin and the consequences of thisphenomenon are unknown. Some reports have proposed thatthis program might reflect a viral infection or invasion ofinflammatory cells in response to the tumor [43]. Our datasuggest that the interferon response is not necessarilydependent on immune cells since our in vitro co-culture sys-tem comprises only fibroblasts and epithelial cells and noimmune cells. Despite considerable effort to identify infectiveagents, we could not find any evidence for an infection in ourcell culture causing the interferon response. Without exclud-ing these possibilities, we propose that in a subset of breastcancer, the interferon response arises as an effect of the inter-action of the malignant epithelial cells with the stroma.At a first glance, the proposed link between interferon signal-ing and tumor-stroma interaction is surprising. However,interferons are pleiotropic cytokines, and while best knownfor their function as a viral defense mechanism they are alsoinvolved in other biological processes [48], such as the induc-tion of cell cycle arrest, apoptosis, cell differentiation,immune stimulation and regulation of bone metabolism [49].The induction of interferons at the interface between tumorcells and the surrounding stroma may have profoundbiological significance. In response to viral infection, induc-tion of the interferon response genes, such as EIF2AK2, canlead to a global arrest of translation and subsequent apoptosis[50]. Interferon treatment has an anti-proliferative effect insome cultured cancer cells, and some human cancers shrinkin response to interferon [51], leading to the speculation thatan interferon response might be linked to a better prognosis[43]. In fact, our results show the opposite effect; patientswith breast cancers displaying high interferon-response geneexpression were 1.7 times more likely (95% confidence inter-val 1.1-2.6; p = 0.018) to develop metastasis and 1.8 timesmore likely to die of the disease (95% confidence interval 1.2-2.7; p = 0.006) than patients with tumors showing lowexpression levels of the interferon-response genes. Similarresults have been reported by others. For example, anincrease in EIF2AK2 expression and activity during tumorprogression had been described in melanoma and colorectalcancer [52]. In breast cancer cells EIF2AK2 was elevatedcompared to normal breast epithelial cells [53]. Also, IFI 27,known to be inducible by IFNα, is frequently over-expressedin breast cancer [54]. IFITM1 over-expression in gastric can-cer cells was reported to enhance migration and invasion invitro [55]. These findings along with the observation thatModel of interaction effectsFigure 5Model of interaction effects. Upon close cell-cell contact the tumor cells (red) interact with the fibroblasts (yellow) (1), which express type I interferon (IFNα and IFNβ) (2). They in turn induce the interferon response genes predominantly in the tumor cells (3).Interferon responsegenesIFNα1 +IFNβ1 2 3Genome Biology 2007, 8:R191coordinated variation in expression in an analysis of diversetumors and multiple datasets. Differential regulation of theinterferon response genes has been observed in many humaninterferon response gene expression in cancer is highly coor-dinated, suggests the possibility that the interferon responseprogram can promote cancer progression.191.11http://genomebiology.com/2007/8/9/R191 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. RThe role of STAT1, the main regulator of the interferonresponse genes, is controversial too. Our finding of a worseoutcome for patients with tumors with high levels of STAT1protein expression support our results of STAT1 mRNAexpression levels and is in accordance with the expressionlevels of the full set of interferon response genes. The mecha-of STAT1 was found to be associated with resistance to radio-therapy [56]. IFITM1 was reported in another model to beinvolved in IFNα induced radioresistance [57]. For patientstreated with radiotherapy, high expression of interferonresponse genes leading to radioresistance of the tumors couldcontribute to an unfavorable outcome compared to the moreInterferon response gene expression in early stage breast cancer (a) The expression values of genes in the 'interferon response gene set' were extracted from a published expr ssion study of 295 early stage breast cancers from the Netherlands Canc r Institut  [35]Figure 6Interferon response gene expression in early stage breast cancer (a) The expression values of genes in the 'interferon response gene set' were extracted from a published expression study of 295 early stage breast cancers from the Netherlands Cancer Institute [35]. Genes and samples are organized by hierarchical clustering. The tumors segregated into two groups defined by high (red) or low (blue) expression levels of 29 genes matching the 'interferon response gene set'. (b) Correlation of interferon response with distant metastasis free and overall survival. Kaplan-Meier curves for the clinical outcomes of indicated tumors exhibiting high (red curve) and low (blue curve) interferon responses are shown. 5 10 15 20 25Distant metastasis free survival time (y)Probability00. 5 10 15 20 25Overall survival time (y)Probability(a)(b)Genome Biology 2007, 8:R191nism for the negative association between interferonresponse gene induction and patient outcome is not yetunderstood. Several mechanisms are possible. Up-regulationradiosensitive tumors. Since all patients receiving breast con-serving therapy from the NKI dataset underwent adjuvantradiotherapy, this hypothesis cannot be further substantiatedR191.12 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. http://genomebiology.com/2007/8/9/R191from our data because of the lack of an appropriate controlgroup. Another possible mechanism, independent from aneffect on therapeutic efficacy, could be mediated by an effectThe finding that an interferon response can be induced inresponse to tumor-stroma interaction raises questions forfurther inquiry. First, our results do not allow us to distin-Correlation of the 70 genes signature [38], the wound signature [24], the hypoxia signature [25] and the interferon response score in the NKI datasetFigure 7Correlation of the 70 genes signature [38], the wound signature [24], the hypoxia signature [25] and the interferon response score in the NKI dataset. Pairwise scatterplot-matrix of four gene signatures. Pearson correlations are shown in the lower part of each plot.WoundHypoxiaInterferon70 geneGenome Biology 2007, 8:R191on invasiveness of the tumor. Up-regulation of STAT1 hasbeen reported in breast cancer micrometastasis in the bonemarrow [58], suggesting a more metastasis-prone phenotype.guish whether the interferon response has any role incontributing to tumor progression or is merely an incidentalfeature of certain cancers that tend to be more aggressive. If it191.13http://genomebiology.com/2007/8/9/R191 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. Rshould turn out that the interferon response has a contribut-ing role in progression and metastasis in some tumors, ther-apeutic application of interferon might be detrimental in suchcancers. Indeed, blocking the interferon response inductionas a therapeutic target using antibodies or small moleculeinhibitors might be beneficial in this situation. Second, apartfrom infectious agents, molecular signals that can induceinterferon secretion are not well defined and the signals thatinduce interferon secretion in the stromal fibroblasts in oursystem are still to be discovered. One molecular mechanismfor induction of interferon could be stimulation of a memberof the toll like receptor family by a tumor cell associatedligand. Endogenous ligands for toll like receptors have beenproposed, but further studies are needed to prove their exist-ence [59]. In our experimental system close cell-cell contactwas needed to induce the interferon-response, suggestingthat a short-range signaling mechanism, perhaps involving acell-surface ligand, might be involved.Molecular interactions between epithelial and mesenchymalcells represent only a small part of the molecular conversationamong all the interacting cells in the breast cancermicroenvironment. The approach used in this work, employ-ing an ex vivo model to develop gene expression signatures asan experimentally tractable window on the more complexinteractions in vivo can be deliberately extended to other cellstypes, such as endothelial, inflammatory and immune cells.This technique may allow us to explore complex interactionsamong the multiple molecules operating in these cells toorchestrate the process of cancer progression and metastasis.gene expression patterns in more complex heterogeneous invivo samples and identify effects of heterotypic cellularinteractions.Materials and methodsCell cultureHMECs (Cambrex Bio Science Walkersville, Inc.,Walkers-ville, MD, USA) were expanded in mammary epithelial basalmedium supplemented with bovine pituitary extract, humanepithelial growth factor, insulin and antibiotics (Clonetics,Cambrex Bio Science Walkersville, Inc.). MCF-7, T47D,MDA-MB231, MDA-MB436, SKBR-3, Hs578T, BT549, CCL-171, HTB-125 (ATTC) and HDF (Cambrex Bio Science Walk-ersville, Inc.) were propagated in DMEM supplemented with10% FBS (HyClone, Logan, UT, USA), glutamine, 100 U/mlpenicillin and 100 μg/ml streptomycin (GIBCO, GrandIsland, NY, USA). For co-culture experiments the cells werecultivated for 48 h at 50,000 cells/cm2 in endothelial basalmedia (Cambrex Bio Science Rockland, Inc., Rockland, ME,USA) with 0.2% FBS, which was a good universal medium forall cells involved. Separated co-cultures were kept in Tran-swell ® chambers with a 0.4 μm pore size (Costar, CorningInc., Corning, NY, USA). The cells for analysis were alwaysharvested from the bottom well and reciprocal interactionswere tested. Cells negatively tested for mycoplasma infectionusing MycoAlert™ (Cambrex Bio Science Rockland, Inc.) andVenorGem ® (Sigma, Saint Louis, MO) mycoplasma detectionkits used according to the manufacturers' instructions.FlowcytometryCells were fixed and stained using the Cytofix/Cytoperm™Kit (BD Biosciences, San Jose, CA, USA) according to themanufacturer's instructions using 20 μg/ml STAT1a mAB(Abcam, Cambridge, MA, USA) and a fluorescein-5-isothyo-cyanate labeled goat anti-mouse IgG (Sigma-Aldrich, StLouis, MO, USA) for detection. Goat serum 1:200 was usedfor blocking. Analytical flow cytometry was done on a modi-fied dual laser LSRScan (BD Immunocytometry Systems, SanDiego, CA, USA) in the Shared FACS Facility, Center forMolecular and Genetic Medicine at Stanford, using FlowJosoftware (TreeStar, Ashland, OR, USA) for data analysis.For FACS sorting, cells were stained with the lipophilic carbo-cyanine dye DiO (Vybrant ® DiO cell-labeling solution,Molecular Probes™ Invitrogen, Eugene, OR, USA) in serum-free DMEM medium for 20 minutes and washed three timesin calcium- and magnesium-free PBS according to the manu-facturer's instructions before co-culturing. After 48 h, thecells were detached by incubation in Trypsin/EDTA (GIBCO,Grand Island, NY, USA) for 3 minutes and washed in ice-coldPBS and then immediately put on ice. Cell sorting was doneon a MoFlow cell sorter (Becton Dickinson, Mountain View,Immunohistochemical staining of STAT1 in a cohort of primary breast cancers: Kaplan-Meier disease-specific survival curve for 353 primary tu o  as essed for STAT1Figure 8Immunohistochemical staining of STAT1 in a cohort of primary breast cancers: Kaplan-Meier disease-specific survival curve for 353 primary tumors assessed for STAT1. The red curve shows 102 patients bearing tumors with high STAT1 expression whereas the blue curve represents 251 patients with low or absent STAT1 expression. X = censored data. 5 10 15 20Survival (years)ProbabilityGenome Biology 2007, 8:R191Our experience suggests that in vitro modeling of specificprocesses and features of the tumor microenvironment canprovide a valuable interpretive framework for analyzing theCA, USA) in the Shared FACS Facility, Center for Molecularand Genetic Medicine at Stanford. The sorted cells wereharvested in TRIZOL ® LS Reagent (Invitrogen, Carlsbad, CA,R191.14 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. http://genomebiology.com/2007/8/9/R191USA). FlowJo software (TreeSTAR) was used for dataanalysis.RNA isolation and amplificationAfter discarding the culture medium and washing the celllayer once with PBS, total RNA was isolated by lysing the cellsin the culture dish with RLT buffer (Qiagen, Valencia, CA,USA) and extraction with the RNeasy ® Mini Kit (Qiagen).Total RNA (500 ng) was amplified using the Message Amp ™II aRNA Kit (Ambion, Austin, TX, USA). The amplificationproduct was checked for integrity by electrophoresis in a 1%agarose gel in MOPS buffer.cDNA microarrays and hybridizationWe used human cDNA microarrays containing 40,700 ele-ments that represent 24,472 unique genes based on UniqueClusters. Arrays were produced at the Stanford FunctionalGenomic Facility. Complete details regarding the clones onthe arrays may be found at Stanford: functional genomicsfacility [60]. cDNA produced from 6 μg amplified RNA werehybridized to the array in a two-color comparative format,with the experimental samples labeled with one fluorophore(Cy5) and a reference pool of mRNA (Universal human refer-ence, Stratagene, La Jolla, CA, USA) labeled with a secondfluorophore (Cy3). Fluorescent dyes were purchased fromAmersham Pharmacia Biotech (Piscataway, NJ, USA).Hybridizations were carried out using the standard protocoldescribed previously [61].Data analysis and clusteringArray images were scanned using an Axon Scanner 4000B(Axon Instruments, Union City, CA, USA), and image analysiswas performed using Genepix Pro version 5.0 (AxonInstruments). The raw data files were stored in the StanfordMicroarray Database [62]; the data used for the paper areavailable at the accompanying website [63]. Data wereexpressed as the log2 ratio of fluorescence intensities of thesample and the reference, for each element on the array.The (Cy5/Cy3) ratio is defined in the Stanford MicroarrayDatabase as the normalized ratio of the background-cor-rected intensities. Spots with aberrant measurements due toobvious array artifacts or poor technical quality weremanually flagged and removed from further analysis. A filterwas applied to omit measurements where the fluorescentsignal from the DNA spot was less than 50% above the meas-ured background fluorescence surrounding the printed DNAspot in either the Cy3 or Cy5 channel. Genes that did not meetthese criteria for at least 80% of the measurements across theexperimental samples were excluded from further analysis.Valid data were filtered to exclude elements that did not haveat least a three-fold deviation from the mean in at least threesamples. Data were evaluated by unsupervised hierarchicalDetermination of the heterotypic interaction effect on gene expressionTo facilitate the identification of heterotypic interactioneffects on global gene expression in a mixed co-culture exper-iment, we normalized the gene expression data based on theproportional contribution of each cell type to transcript abun-dance. Given that the average gene does not change due toheterotypic interaction and that there are simple additiveeffects to account for, we used a linear regression fit for nor-malization. To determine the contribution of each cell type tothe combined gene expression pattern in the linear regressionmodel, the expression levels of the monocultures are the pre-dictors and the expression levels of the co-culture, theresponse.Specifically, a set of equations (1-n) is established (one pergene), as illustrated in the additional data file 5 in which theexpression level of gene n (en, co-culture) in the mixed co-cul-ture equals the fraction a of mRNA from cell type 1 times therelative expression level of gene n in type 1 mono-culturedcells plus the fraction (1-a) of mRNA from cell type 2 times therelative expression level of gene n in type 2 mono-culturedcells multiplied by the interaction coefficient In. We assumethat the average gene is not influenced by heterotypic interac-tion in the mixed co-culture represented as I = 1. Since thedataset over e1-n is skewed, we empirically identified a linearregression fit based on Gamma errors and identity link as agood model to calculate a. Then the equations 1-n can then besolved for I1-n, which results in a profile of interaction effectsfor the genes1-n. These interaction effects can be analyzed inmuch the same way as conventional gene expressionmeasurements.Real time quantitative PCRTotal RNA (500 ng) was mixed with dT16 primer in a volumeof 11 μl, incubated at 65°C for 10 minutes and immediatelyput on ice. Following addition of 100 units Superscript IIreverse transcriptase (GIBCO, Carlsbad, CA, USA), reversetranscription was performed for 2 h at 42°C in 1× RT reactionbuffer (GIBCO), 10 μM dithiothreitol, 500 μM dNTP (Amer-sham Biosciences, Pittsburgh, PA, USA) with 2.5 μM dT16primer in a volume of 20 μl. PCR reactions were performed ina final volume of 20 μl with cDNA prepared from 20 ng RNAand a final concentration of 1× SYBR ®Green PCR Master Mix(ABI, Foster City, CA, USA) and 200 nM of each primer(sequences: GAPDH, forward GAAGGTGAAGGTCGGAGTC,reverse GAAGATGGTGATGGGATTTC; OAS2, forwardGGAATACCTGAAGCCCTACGAA, reverse CCTGCAGACGT-CACAGATGGT; IFNα, forward CCTCGCCCTTTGCTT-TACTG, reverse GCCCAGAGAGCAGCTTGACT; IFNβ,forward ACCTCCGAAACTGAAGATCTCCTA, reverse TGCT-GGTTGAAGAATGCTTGA). The reaction was run in an ABI7700 Sequence Detection System with the following cyclingGenome Biology 2007, 8:R191clustering [64] and SAM [26] and displayed using Treeview[65].conditions: 50°C for 2 minutes, 94°C for 10 minutes, then 40cycles of 94°C for 15 s and 60°C for 60 s. For each gene a191.15http://genomebiology.com/2007/8/9/R191 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. Rstandard curve was prepared and triplicate measurementswere performed for each sample.ImmunohistochemistryLarge biopsy or tissue microarray sections were cut from par-affin blocks, deparaffinized in xylene, and hydrated in agraded series of alcohol. The slides were pretreated with cit-rate buffer and a microwave step. Immunostaining wasperformed using the DAKO Envision+ System, PeroxidaseDAB, (DAKO, Cambridgeshire, United Kingdom) for STAT1amonoclonal antibody (1:100 dilution; Abcam).We stained 1,024 tissue cores from 521 donor blocks. Immu-nohistochemistry images were acquired with the BLISSMicroscope System (Bacus Laboratories, Lombard, IL, USA).Staining results were assessed using a three-point scoringsystem, where 0 = invasive tumor cells present in the tissuecore and no staining seen, 1 = invasive tumor cells presentwith weak staining intensity and/or < 20% of the cellsstained, and 2 = invasive tumor cells present with strongstaining in > 20% of the cells. Tissue cores that failed toadhere to the glass slide, did not contain invasive carcinomaor were otherwise uninterpretable were excluded. Scoring ofthe arrays was analyzed using the Deconvoluter software aspreviously described [66], with each sample receiving thehigher of the scores for two replicate cores.Human breast cancer datasetThe dataset for breast cancer contained 295 tumors analyzedon a 25,000 spot oligonucleotide array as described [35]. Inbrief, patients were diagnosed and treated at the NKI for earlystage breast cancer (stage I and II) between 1984 and 1995.The clinical data were updated in January 2005. The medianfollow-up for patients still alive is now 12.3 years.The interferon response gene list consists of 42 genes repre-sented by 49 image clones on the cDNA Stanford array.Clones having the same Unigene locus were removed. Thegene sequences were mapped to spots on the NKI array usingUnigene build number 184 (released on 9 June 2005) to give29 unique spots. In order to overcome possible overweightingof clones from Unigene clusters that were matched to morethan one probe on the NKI array, expression values derivedfrom probes that were not matched to the same Unigenecluster were averaged. Expression measurements for eachgene were mean centered. The resulting dataset wassubjected to hierarchical clustering with average linkage clus-tering [64] and displayed with Treeview [65].Distant metastasis was analyzed as first event only (distantmetastasis-free probability). If a patient developed a localrecurrence, axillary recurrence, contralateral breast cancer ora second primary cancer (except for non-melanoma skin can-cers could be a source for distant metastases. An ipsilateralsupra-clavicular recurrence was soon followed by a distantmetastasis in all but one patient. An ipsilateral supra-clavicu-lar recurrence was thus considered the first clinical evidencefor metastatic disease for this analysis and patients were notcensored at the time of ipsilateral supra-clavicular recur-rence. Overall survival was analyzed based on death from anycause and patients were censored at last follow up. Kaplan-Meier survival curves were compared by the Cox/Mantel log/rank test using Winstat for Microsoft Excel (RFitch Software,Staufen, Germany). Multivariate analysis by the Cox propor-tional hazard method was performed using the software pack-age SPSS R 11.5 (SPSS, Inc., Chicago, IL, USA).A dataset of gene expression patterns from advanced breastcancers was described by Sorlie et al. [36,37]. Expression datafrom 19 image clones representing the interferon responsegene list were included in this dataset. Genes and sampleswere organized by hierarchical clustering. Relapse-free andoverall survival were calculated as described above.The independent breast cancer tissue microarray validationseries is as described [40]; immunohistochemical scores forSTAT1 were related to breast cancer-specific survival by Kap-lan-Meier analysis with log-rank test.AbbreviationsDMEM, Dulbecco's modified Eagle's medium; EIF2AK2,eukaryotic translation initiation factor 2-alpha kinase; FACS,fluorescent assisted cell sorting; FBS, fetal bovine serum;FDR, false discovery rate; HMEC, human mammary epithe-lial cell; IFIT, interferon-induced protein with tetratricopep-tide repeats; IFN, interferon; OAS, 2',5'-oligoadenylatesynthetase; NKI, Netherlands Cancer Institute; PBS, phos-phate-buffered saline; SAM, significance analysis of micro-array data; STAT, signal transducer and activator oftranscription; TGF, transforming growth factor.Authors' contributionsMB and POB designed the research; MB and RP performedthe research; MB, TH and TN contributed new reagents/ana-lytical tools; MB, DSAN, TH and POB analyzed the data; andMB and POB wrote the paper. The authors declare no conflictof interestAdditional data filesThe following additional data are available with the onlineversion of this paper. Additional data file 1 is a table listing theinterferon response genes. Additional data file 2 shows theexpression of OAS2 measured by RT-PCR in the co-cultureGenome Biology 2007, 8:R191cer), she was censored at that time and subsequent distantmetastases were not analyzed. This is based on the theoreticalpossibility that the locally recurrent or second primary can-CCL171/MDA-MB-231 and the expression of STAT1 meas-ured by immunofluorescent staining and FACS analysis.Additional data file 3 shows the analysis of the interferonR191.16 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. http://genomebiology.com/2007/8/9/R191response signature in advanced human breast cancers. Addi-tional data file 4 shows box and scatter plots illustrating thecorrelation of the interferon score to clinical parameters withknown prognostic significance. Additional data file 5 is anillustration of the linear regression model used to normalizefor additive effects in the mixed co-culture gene expressiondata.Additional data file 1Interferon response genes.Click here for file 2Exp ession of OAS2 measured by RT-PCR in the co-culture CL171/MDA-MB- 31 and the expression of STAT1 measured by immunoflu esecent st ining and FACS analysis.3naly is f th  int rfero  respo se signature in advanced humanbreast ca ce s. 4orr lati  f the i f ro  score to clinic l parameters with k own progno tic si nif ca cB x a d scatt r p ots illu trating the correl tion of the inte feroscore t  c inical ara ters with kn w pr gno tic significance.5Lin r egr s i  mod l us d t  normalize for ad i iv  f cts in th  m xed -cultu e g n  x r s ion data.AcknowledgementsWe would like to thank Michael Fero, John Collar, Elena Seraia and the staffof the Stanford Functional Genomics Facility for supplying us with thehuman cDNA microarrays used for this study. We thank our colleagues ofthe Brown lab for discussion, Janos Demeter, Jeremy Gollub, Gavin Sher-lock, Catherine Bell and other staff of the Stanford Microarray Database.This work was supported by NIH grant CA111487 and by the HHMI andthe Dutch Cancer Society NKB 2002-2575 (DSAN). POB is an investigatorof the Howard Hughes Medical Institute. MB is a fellow of the Krebsliga bei-der Basel, the M. and W. Lichtenstein-Stiftung Basel, and the Huggenberger-Krebsstiftung, Zürich, Switzerland.References1. Cunha GR, Hom YK: Role of mesenchymal-epithelialinteractions in mammary gland development.  J MammaryGland Biol Neoplasia 1996, 1:21-35.2. Levine JF, Stockdale FE: Cell-cell interactions promote mam-mary epithelial cell differentiation.  J Cell Biol 1985,100:1415-1422.3. Weaver M, Batts L, Hogan BL: Tissue interactions pattern themesenchyme of the embryonic mouse lung.  Dev Biol 2003,258:169-184.4. Howlett AR, Bissell MJ: The influence of tissue microenviron-ment (stroma and extracellular matrix) on the developmentand function of mammary epithelium.  Epithelial Cell Biol 1993,2:79-89.5. Howard B, Ashworth A: Signalling pathways implicated in earlymammary gland morphogenesis and breast cancer.  PLoSGenet 2006, 2:e112.6. Bissell MJ, Radisky DC, Rizki A, Weaver VM, Petersen OW: Theorganizing principle: microenvironmental influences in thenormal and malignant breast.  Differentiation 2002, 70:537-546.7. Bissell MJ, Radisky D: Putting tumours in context.  Nat Rev Cancer2001, 1:46-54.8. Hanahan D, Weinberg RA: The hallmarks of cancer.  Cell 2000,100:57-70.9. Arias AM: Epithelial mesenchymal interactions in cancer anddevelopment.  Cell 2001, 105:425-431.10. Sneddon JB, Zhen HH, Montgomery K, van de Rijn M, Tward AD,West R, Gladstone H, Chang HY, Morganroth GS, Oro AE, BrownPO: Bone morphogenetic protein antagonist gremlin 1 iswidely expressed by cancer-associated stromal cells and canpromote tumor cell proliferation.  Proc Natl Acad Sci USA 2006,103:14842-14847.11. Olumi AF, Grossfeld GD, Hayward SW, Carroll PR, Tlsty TD, CunhaGR: Carcinoma-associated fibroblasts direct tumorprogression of initiated human prostatic epithelium.  CancerRes 1999, 59:5002-5011.12. Camps JL, Chang SM, Hsu TC, Freeman MR, Hong SJ, Zhau HE, vonEschenbach AC, Chung LW: Fibroblast-mediated accelerationof human epithelial tumor growth in vivo.  Proc Natl Acad Sci USA1990, 87:75-79.13. Kuperwasser C, Chavarria T, Wu M, Magrane G, Gray JW, Carey L,Richardson A, Weinberg RA: Reconstruction of functionally nor-mal and malignant human breast tissues in mice[seecomment].  Proc Natl Acad Sci USA 2004, 101:4966-4971.14. Bhowmick NA, Chytil A, Plieth D, Gorska AE, Dumont N, Shappell S,Washington MK, Neilson EG, Moses HL: TGF-beta signaling infibroblasts modulates the oncogenic potential of adjacentcinoma growth and invasion through upregulation of TGF-alpha-, MSP- and HGF-mediated signaling networks.  Onco-gene 2005, 24:5053-5068.16. Petersen OW, Nielsen HL, Gudjonsson T, Villadsen R, Rank F, Nie-buhr E, Bissell MJ, Ronnov-Jessen L: Epithelial to mesenchymaltransition in human breast cancer can provide a nonmalig-nant stroma.  Am J Pathol 2003, 162:391-402.17. Nielsen HL, Ronnov-Jessen L, Villadsen R, Petersen OW: Identifica-tion of EPSTI1, a novel gene induced by epithelial-stromalinteraction in human breast cancer.  Genomics 2002,79:703-710.18. Khodarev NN, Yu J, Labay E, Darga T, Brown CK, Mauceri HJ, YassariR, Gupta N, Weichselbaum RR: Tumour-endothelium interac-tions in co-culture: coordinated changes of gene expressionprofiles and phenotypic properties of endothelial cells.  J CellSci 2003, 116:1013-1022.19. Allinen M, Beroukhim R, Cai L, Brennan C, Lahti-Domenici J, HuangH, Porter D, Hu M, Chin L, Richardson A, et al.: Molecular charac-terization of the tumor microenvironment in breast cancer[see comment].  Cancer Cell 2004, 6:17-32.20. West RB, Nuyten DS, Subramanian S, Nielsen TO, Corless CL, RubinBP, Montgomery K, Zhu S, Patel R, Hernandez-Boussard T, et al.:Determination of stromal signatures in breast carcinoma.PLoS Biol 2005, 3:e187.21. Whitney AR, Diehn M, Popper SJ, Alizadeh AA, Boldrick JC, RelmanDA, Brown PO: Individuality and variation in gene expressionpatterns in human blood.  Proc Natl Acad Sci USA 2003,100:1896-1901.22. Lamb J, Ramaswamy S, Ford HL, Contreras B, Martinez RV, KittrellFS, Zahnow CA, Patterson N, Golub TR, Ewen ME: A mechanismof cyclin D1 action encoded in the patterns of gene expres-sion in human cancer.  Cell 2003, 114:323-334.23. Iyer VR, Eisen MB, Ross DT, Schuler G, Moore T, Lee JC, Trent JM,Staudt LM, Hudson J Jr, Boguski MS, et al.: The transcriptional pro-gram in the response of human fibroblasts to serum[seecomment].  Science 1999, 283:83-87.24. Chang HY, Sneddon JB, Alizadeh AA, Sood R, West RB, MontgomeryK, Chi JT, van de Rijn M, Botstein D, Brown PO: Gene expressionsignature of fibroblast serum response predicts human can-cer progression: similarities between tumors and wounds.PLoS Biol 2004, 2:E7.25. Chi JT, Wang Z, Nuyten DS, Rodriguez EH, Schaner ME, Salim A,Wang Y, Kristensen GB, Helland A, Borresen-Dale AL, et al.: Geneexpression programs in response to hypoxia: cell type specif-icity and prognostic significance in human cancers.  PLoS Med2006, 3:e47.26. Tusher VG, Tibshirani R, Chu G: Significance analysis of micro-arrays applied to the ionizing radiation response.  Proc NatlAcad Sci USA 2001, 98:5116-5121.27. Lazard D, Sastre X, Frid MG, Glukhova MA, Thiery JP, Koteliansky VE:Expression of smooth muscle-specific proteins in myoepithe-lium and stromal myofibroblasts of normal and malignanthuman breast tissue.  Proc Natl Acad Sci USA 1993, 90:999-1003.28. Valenti MT, Azzarello G, Balducci E, Sartore S, Sandri M, Manconi R,Sicari U, Bari M, Vinante O: Conditioned medium from MCF-7cell line induces myofibroblast differentiation, decreased cellproliferation, and increased apoptosis in cultured normalfibroblasts but not in fibroblasts from malignant breasttissue.  Histochem J 2001, 33:499-509.29. Desmouliere A, Guyot C, Gabbiani G: The stroma reactionmyofibroblast: a key player in the control of tumor cellbehavior.  Int J Dev Biol 2004, 48:509-517.30. Perou CM, Jeffrey SS, van de Rijn M, Rees CA, Eisen MB, Ross DT,Pergamenschikov A, Williams CF, Zhu SX, Lee JC, et al.: Distinctivegene expression patterns in human mammary epithelialcells and breast cancers.  Proc Natl Acad Sci USA 1999,96:9212-9217.31. Wang D, Coscoy L, Zylberberg M, Avila PC, Boushey HA, Ganem D,DeRisi JL: Microarray-based detection and genotyping of viralpathogens.  Proc Natl Acad Sci USA 2002, 99:15687-15692.32. Suzuki T, Urano T, Tsukui T, Horie-Inoue K, Moriya T, Ishida T, Mura-matsu M, Ouchi Y, Sasano H, Inoue S: Estrogen-responsive fingerprotein as a new potential biomarker for breast cancer.  ClinCancer Res 2005, 11:6148-6154.Genome Biology 2007, 8:R191epithelia.  Science 2004, 303:848-851.15. Cheng N, Bhowmick NA, Chytil A, Gorksa AE, Brown KA, MuraokaR, Arteaga CL, Neilson EG, Hayward SW, Moses HL: Loss of TGF-beta type II receptor in fibroblasts promotes mammary car-33. Parrinello S, Coppe JP, Krtolica A, Campisi J: Stromal-epithelialinteractions in aging and cancer: senescent fibroblasts alterepithelial cell differentiation.  J Cell Sci 2005, 118:485-496.34. Orimo A, Gupta PB, Sgroi DC, Arenzana-Seisdedos F, Delaunay T,191.17http://genomebiology.com/2007/8/9/R191 Genome Biology 2007,     Volume 8, Issue 9, Article R191       Buess et al. RNaeem R, Carey VJ, Richardson AL, Weinberg RA: Stromal fibrob-lasts present in invasive human breast carcinomas promotetumor growth and angiogenesis through elevated SDF-1/CXCL12 secretion.  Cell 2005, 121:335-348.35. van de Vijver MJ, He YD, van't Veer LJ, Dai H, Hart AA, Voskuil DW,Schreiber GJ, Peterse JL, Roberts C, Marton MJ, et al.: A gene-expression signature as a predictor of survival in breast can-cer [see comment].  New Engl J Med 2002, 347:1999-2009.36. Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, HastieT, Eisen MB, van de Rijn M, Jeffrey SS, et al.: Gene expression pat-terns of breast carcinomas distinguish tumor subclasses withclinical implications.  Proc Natl Acad Sci USA 2001, 98:10869-10874.37. Sorlie T, Tibshirani R, Parker J, Hastie T, Marron JS, Nobel A, Deng S,Johnsen H, Pesich R, Geisler S, et al.: Repeated observation ofbreast tumor subtypes in independent gene expression datasets.  Proc Natl Acad Sci USA 2003, 100:8418-8423.38. van 't Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, Mao M,Peterse HL, van der Kooy K, Marton MJ, Witteveen AT, et al.: Geneexpression profiling predicts clinical outcome of breastcancer.  Nature 2002, 415:530-536.39. Chang HY, Nuyten DS, Sneddon JB, Hastie T, Tibshirani R, Sorlie T,Dai H, He YD, van't Veer LJ, Bartelink H, et al.: Robustness, scala-bility, and integration of a wound-response gene expressionsignature in predicting breast cancer survival.  Proc Natl AcadSci USA 2005, 102:3738-3743.40. Makretsov NA, Huntsman DG, Nielsen TO, Yorida E, Peacock M,Cheang MC, Dunn SE, Hayes M, van de Rijn M, Bajdik C, Gilks CB:Hierarchical clustering analysis of tissue microarray immu-nostaining data identifies prognostically significant groups ofbreast carcinoma.  Clin Cancer Res 2004, 10:6143-6151.41. Lu P, Nakorchevskiy A, Marcotte EM: Expression deconvolution:a reinterpretation of DNA microarray data reveals dynamicchanges in cell populations.  Proc Natl Acad Sci USA 2003,100:10370-10375.42. Chang HY, Chi JT, Dudoit S, Bondre C, van de Rijn M, Botstein D,Brown PO: Diversity, topographic differentiation, and posi-tional memory in human fibroblasts.  Proc Natl Acad Sci USA2002, 99:12877-12882.43. Einav U, Tabach Y, Getz G, Yitzhaky A, Ozbek U, Amariglio N, IzraeliS, Rechavi G, Domany E: Gene expression analysis reveals astrong signature of an interferon-induced pathway in child-hood lymphoblastic leukemia as well as in breast and ovariancancer.  Oncogene 2005, 24:6367-6375.44. Schaner ME, Ross DT, Ciaravino G, Sorlie T, Troyanskaya O, DiehnM, Wang YC, Duran GE, Sikic TL, Caldeira S, et al.: Gene expres-sion patterns in ovarian carcinomas.  Mol Biol Cell 2003,14:4376-4386.45. Chen X, Leung SY, Yuen ST, Chu KM, Ji J, Li R, Chan AS, Law S, Troy-anskaya OG, Wong J, et al.: Variation in gene expression pat-terns in human gastric cancers.  Mol Biol Cell 2003, 14:3208-3215.46. Garber ME, Troyanskaya OG, Schluens K, Petersen S, Thaesler Z,Pacyna-Gengelbach M, van de Rijn M, Rosen GD, Perou CM, WhyteRI, et al.: Diversity of gene expression in adenocarcinoma ofthe lung [erratum appears in Proc Natl Acad Sci USA99:1098].  Proc Natl Acad Sci USA 2001, 98:13784-13789.47. Hu Z, Fan C, Oh DS, Marron JS, He X, Qaqish BF, Livasy C, Carey LA,Reynolds E, Dressler L, et al.: The molecular portraits of breasttumors are conserved across microarray platforms.  BMCGenomics 2006, 7:96.48. Stark GR, Kerr IM, Williams BR, Silverman RH, Schreiber RD: Howcells respond to interferons.  Annu Rev Biochem 1998, 67:227-264.49. Takayanagi H, Kim S, Matsuo K, Suzuki H, Suzuki T, Sato K, YokochiT, Oda H, Nakamura K, Ida N, et al.: RANKL maintains bonehomeostasis through c-Fos-dependent induction ofinterferon-beta.  Nature 2002, 416:744-749.50. Donze O, Jagus R, Koromilas AE, Hershey JW, Sonenberg N: Abro-gation of translation initiation factor eIF-2 phosphorylationcauses malignant transformation of NIH 3T3 cells.  EMBO J1995, 14:3828-3834.51. Jonasch E, Haluska FG: Interferon in oncological practice:review of interferon biology, clinical applications, andtoxicities.  Oncologist 2001, 6:34-55.52. Kim SH, Gunnery S, Choe JK, Mathews MB: Neoplastic progres-sion in melanoma and colon cancer is associated withkinase, PKR.  Oncogene 2000, 19:3086-3094.54. Rasmussen UB, Wolf C, Mattei MG, Chenard MP, Bellocq JP, Cham-bon P, Rio MC, Basset P: Identification of a new interferon-alpha-inducible gene (p27) on human chromosome 14q32and its expression in breast carcinoma.  Cancer Res 1993,53:4096-4101.55. Yang Y, Lee JH, Kim KY, Song HK, Kim JK, Yoon SR, Cho D, Song KS,Lee YH, Choi I: The interferon-inducible 9-27 gene modulatesthe susceptibility to natural killer cells and the invasivenessof gastric cancer cells.  Cancer Lett 2005, 221:191-200.56. Khodarev NN, Beckett M, Labay E, Darga T, Roizman B, Weichsel-baum RR: STAT1 is overexpressed in tumors selected forradioresistance and confers protection from radiation intransduced sensitive cells.  Proc Natl Acad Sci USA 2004,101:1714-1719.57. Kita K, Sugaya S, Zhai L, Wu YP, Wano C, Chigira S, Nomura J, Taka-hashi S, Ichinose M, Suzuki N: Involvement of LEU13 in inter-feron-induced refractoriness of human RSa cells to cellkilling by X rays.  Radiat Res 2003, 160:302-308.58. Woelfle U, Cloos J, Sauter G, Riethdorf L, Janicke F, van Diest P, Brak-enhoff R, Pantel K: Molecular signature associated with bonemarrow micrometastasis in human breast cancer.  Cancer Res2003, 63:5679-5684.59. Ohashi K, Burkart V, Flohe S, Kolb H: Cutting edge: heat shockprotein 60 is a putative endogenous ligand of the toll-likereceptor-4 complex.  J Immunol 2000, 164:558-561.60. Stanford: Functional Genomics Facility   [http://www.microarray.org/sfgf/jsp/servicesFrame.jsp#productionArrays]61. Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Pol-lack JR, Ross DT, Johnsen H, Akslen LA, et al.: Molecular portraitsof human breast tumours.  Nature 2000, 406:747-752.62. Ball CA, Awad IA, Demeter J, Gollub J, Hebert JM, Hernandez-Bous-sard T, Jin H, Matese JC, Nitzberg M, Wymore F, et al.: The StanfordMicroarray Database accommodates additional microarrayplatforms and data formats.  Nucleic Acids Res 2005,33:D580-582.63. SMD : View Publication : Characterization of heterotypicinteraction effects in vitro to deconvolute global gene expres-sion profiles in cancer.   [http://microarray-pubs.stanford.edu/tumor-stroma-interaction]64. Eisen MB, Spellman PT, Brown PO, Botstein D: Cluster analysisand display of genome-wide expression patterns.  Proc NatlAcad Sci USA 1998, 95:14863-14868.65. Eisen Lab: Software   [http://rana.lbl.gov/EisenSoftware.htm]66. Liu CL, Prapong W, Natkunam Y, Alizadeh A, Montgomery K, GilksCB, van de Rijn M: Software tools for high-throughput analysisand archiving of immunohistochemistry staining dataobtained with tissue microarrays.  Am J Pathol 2002,161:1557-1565.Genome Biology 2007, 8:R191increased expression and activity of the interferon-inducibleprotein kinase, PKR.  Oncogene 2002, 21:8741-8748.53. Kim SH, Forman AP, Mathews MB, Gunnery S: Human breast can-cer cells contain elevated levels and activity of the protein


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items