Oncogene (2006) 25, 2328–2338 & 2006 Nature Publishing Group All rights reserved 0950-9232/06 $30.00 www.nature.com/onc ORIGINAL ARTICLE Gene expression signatures and biomarkers of noninvasive and invasive breast cancer cells: comprehensive profiles by representational difference analysis, microarrays and proteomics GM Nagaraja1, M Othman2, BP Fox1, R Alsaber1, CM Pellegrino3, Y Zeng2, R Khanna2, P Tamburini3, A Swaroop2 and RP Kandpal1 1Department of Biological Sciences, Fordham University, Bronx, NY, USA; 2Department of Ophthalmology and Visual Sciences, University of Michigan, Ann Arbor, MI, USA and 3Bayer Corporation, West Haven, CT, USA We have characterized comprehensive transcript and Keywords: representational difference analysis; micro- proteomic profiles of cell lines corresponding to normal arrays; proteomics; breast carcinoma; biomarkers; breast (MCF10A), noninvasive breast cancer (MCF7) and copper homeostasis invasive breast cancer (MDA-MB-231). The transcript profiles were first analysed by a modified protocol for representational difference analysis (RDA) of cDNAs between MCF7 and MDA-MB-231 cells. The majority of genes identified by RDA showed nearly complete con- Introduction cordance withmicroarray results, and also led to the identification of some differentially expressed genes such The transformation of a normal cell into a cancer cell as lysyl oxidase, copper transporter ATP7A, EphB6, has been correlated to altered expression of a variety of RUNX2 and a variant of RUNX2. The altered transcripts genes (Perou et al., 2000; Becker et al., 2005). The identified by microarray analysis were involved in cell–cell expression of some of these genes is a direct result of or cell–matrix interaction, Rho signaling, calcium home- sequence mutation, whereas other changes occur due to ostasis and copper-binding/sensitive activities. A set of alterations in gene products that participate in specific nine genes that included GPCR11, cadherin 11, annexin pathways. The changes in gene expression have been A1, vimentin, lactate dehydrogenase B (upregulated in routinely characterized by classical subtraction hybridi- MDA-MB-231) and GREB1, S100A8, amyloid b pre- zation and differential display approaches (Cerosaletti cursor protein, claudin 3 and cadherin 1 (downregulated in et al., 1995; Alpan et al., 1996). With the availability MDA-MB-231) were sufficient to distinguishMDA-MB- of the human genome sequence and sequences for a 231 from MCF7 cells. The downregulation of a set of number of other model organisms, traditional methods transcripts for proteins involved in cell–cell interaction have largely been replaced by gene microarrays indicated these transcripts as potential markers for (Khan et al., 2001). These analyses have been used to invasiveness that can be detected by methylation-specific characterize the molecular basis of a variety of diseases PCR. The proteomic profiles indicated altered abundance including cancer. A comprehensive analysis of a large of fewer proteins as compared to transcript profiles. number of cancer cell lines allowed clustering of genes Antisense knockdown of selected transcripts led to into groups based on their expression patterns in inhibition of cell proliferation that was accompanied by phenotypically related cell lines (Khan et al., 2001; altered proteomic profiles. The proteomic profiles of Dan et al., 2002; Rosenwald et al., 2002; van’t Veer antisense transfectants suggest the involvement of pepti- et al., 2002). The results of profiling experiments dyl-prolyl isomerase, Raf kinase inhibitor and 80 kDa indicated expression of specific gene clusters in cell lines protein kinase C substrate in mediating the inhibition of that have the same origin or have arisen from the same cell proliferation. organ (Ross et al., 2000). A complementary approach Oncogene (2006) 25, 2328–2338. doi:10.1038/sj.onc.1209265; that has been used in limited ways is proteomics. published online 28 November 2005 Proteomics scores for changes in different proteins and peptides in cells with characteristic pheno- typic differences. However, a comparative analysis of transcripts and proteins to establish a relationship between transcript changes and protein levels has not Correspondence: Dr RP Kandpal, Department of Biological Sciences, yet become routine. Larkin Hall, Room 250, 441 E Fordham Road, Fordham University, Although expression profiling of tumor tissue and its Bronx, NY 10458, USA. E-mail: [email protected] comparison with normal tissue, in principle, is most Received 29 September 2005; revised 14 October 2005; accepted 14 appropriate to obtain the genetic signatures of a tumor October 2005; published online 28 November 2005 type, such comparisons have not been free of attendant Molecular signatures of breast carcinoma cells GM Nagaraja et al 2329 complications. These complications arise due to hetero- involvement of the majority of these cDNAs is well geneity of tumor specimens wherein any cell type- characterized either in tumorigenesis or in metastasis. specific changes are likely to be masked by other cell The phenotypic characteristics of MCF-7 and MDA- types that constitute the tumor specimen. For this MB-231 ideally match with the biological significance of reason, well-characterized cell lines established these genes. The alterations in transcripts for Rho from tumor tissue may prove more informative and signaling proteins, Ca2 þ binding/requiring proteins, have been considered useful by cancer researchers. tight junctions/anchoring junctions/gap junctions, cop- Comparing gene profiles between cell lines has the per binding or sensitive proteins, and RUNX2 are potential to reveal genes that could be causative for particularly noteworthy. the phenotype and other genes that can serve as tumor The differential expression of a representative number biomarkers. of RDA clones was validated by semiquantitative PCR. Our investigations are aimed at designating a subset As shown in Figure 1, these transcripts were either of transcripts that could distinguish a normal breast cell specific to or upregulated in the cell line that was used as from a breast cancer cell and help to predict tumorigenic a tester. Such analyses demonstrated that more than or metastatic potential of a transformed cell. We 90% of the clones were differentially expressed. The describe here transcript and proteomic profiles of a abundance of transcripts and the results of RT–PCR normal breast cell line, a tumorigenic but noninvasive were also confirmed by Northern blotting (Figure 2). breast carcinoma cell line and an invasive breast The pattern of hybridization clearly indicates that all carcinoma cell line, and summarize them as a set of these transcripts showed differential expression in candidate biomarkers or targets for therapeutic inter- MCF7 and MDA-MB-231 cells that were used as vention. The comparison of transcript profiles with driver/tester combinations for the RDA. proteomic profiles demonstrated that altered proteins were not always represented in the microarray desig- nated profiles and vice versa. Furthermore, we have Gene microarrays targeted five transcripts that were upregulated in MCF7 After obtaining preliminary molecular signatures of cells for investigating their role in cell proliferation MCF7 and MDA-MB-231 cells by RDA, we used pathways. The proteomic profiles have revealed that Affymetrix gene micoarrays to expand the above inhibition of cell proliferation by antisense knockdown analysis to identify a comprehensive set of transcripts was mediated by a specific set of proteins. that is deregulated in invasive breast carcinoma cells. The comparisons of cell lines on the basis of transcripts that are either present or absent as shown in Figure 3 Results revealed that a set of 123 genes distinguishes MDA-MB- 231 cells from MCF7 and MCF10A. These genes can be Representational difference analysis classified by their involvement in functional classes such As described in the Materials and methods section, as transcription, signal transduction, cell adhesion, cell RDA was performed by using cDNAs from MCF7 and cycle, metabolism, transport, response genes and devel- MDA-MB-231 as tester/driver or driver/tester combina- opment (Figure 4). The majority of these genes tion. The difference product in the first case represents participated in the process of signal transduction the genes that are either upregulated in or specific to followed by transcription, cell adhesion and metabo- MCF7. On the other hand, the difference product of lism, respectively. A few transcripts in these classes were MDA-MB-231 (tester) and MCF7 (driver) hybridiza- tested by real-time RT–PCR to confirm their altered tion resulted in the isolation of cDNAs that are either abundance. The selected transcripts showed changes upregulated in or specific to MDA-MB-231. The initial ranging between two- and 10-fold, 11- and 20-fold and linkers used in this protocol had internal BglII sites. greater than 20-fold, and were in close agreement with One strand of the linker was used to amplify both the the results of microarray analysis. The qualitative tester and driver cDNAs after linkers had been ligated pattern of change observed in microarrays analysis to cDNAs. After removal of linkers from amplified was readily reproduced by real-time or semiquantitative cDNAs by digestion with BglII, a dephosphorylated RT–PCR for all transcripts tested. BglII adaptor was ligated to tester DNA. The BglII The number of altered transcripts was over 1000 adaptor had an internal EcoRI site. The difference