| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Regular Articles |
Cancer Genetics Branch, National Human Genome Research Institute, NIH, Bethesda, Maryland 20892 [H. E. C., R. L. W., J. M. C., Y. C., P. S. M.]; Complex Systems Division, Department of Theoretical Physics, Lund University, SE-223 62 Lund, Sweden [M. R.]; and Advanced Technology Center, National Cancer Institute, Gaithersburg, Maryland 20877 [S. B.]
 |
ABSTRACT |
|---|
 Top ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONREFERENCES |
|---|
|
INTRODUCTION |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONREFERENCES |
|---|
Our understanding of signal transduction components and their interactions regulating growth and arrest of breast cancer cells is still limited. Microarray technologies provide a powerful method to explore the complexities of transcriptional profiles defined by selected pharmacological mitogens and inhibitors pivotal for breast cancer cell growth. Recently, our laboratory and others (11, 12, 13, 14, 15, 16, 17) have used expression arrays to subclassify breast tumors into categories possessing distinct biological and clinical properties. Among the distinctions made to date, the strongest separation is observed between ER+ and ER- tumors. This highlights the question of how genes contributing to the tumor subclassification are associated with a particular hormone or growth factor signal. Because this question is difficult to address using patient samples, we chose an in vitro model system using cultured cell lines treated with agents known to induce breast cancer cell mitosis or growth arrest. Human breast tumor cell lines have been used extensively as models of neoplastic disease, and accordingly, their expression profiles provide a frame of reference for assessing the biological significance of expression patterns in a specific tumor (11 , 18) . There have been several studies of gene expression analysis of breast cancer cells treated with a limited number of growth agonists and antagonists, producing catalogues of responsive genes (19, 20, 21, 22, 23, 24, 25) . These studies, although valuable, do not address the highly networked nature of signaling components with respect to regulation of gene expression. Individual growth regulators may impact (in a temporally distinct manner) on multiple axes of cell signaling, and conversely, multiple agents impinge on the same signaling cascades. For example, steroid hormones signal to the MAPK and protein kinase A/PKC axes (26, 27, 28, 29) , which strongly suggests that the gene expression responses to growth factors, steroids, and activators of second messengers may overlap.
In this article, we present a detailed analysis of the observed patterns of gene expression in MCF7, T-47D, and MDA-MB-436 breast cancer cells treated for 2, 8, and 24 h with E2, estrogen antagonists 4-OHT, ICI, a progestin (R5020), an antiprogestin RU486, atRA, EGF, a Mek1/2-specific inhibitor (U0126), and a phorbol ester (TPA). We hypothesized that by examining both the unique and interrelated time-sensitive patterns of gene expression and interpreting these in the context of GO and phenotypic response to each agent, we may identify pathways significantly impacted by breast cell growth regulators. We predicted that this study may also reveal individual or related groups of genes representing new candidates to investigate for their involvement in disease progression. The patterns of gene expression found in vitro were compared with in vivo gene expression profiles previously identified in breast cancer specimens. This comparative analysis reveals a significant association between specific gene clusters and discriminators of hormone receptor status and disease progression in tumor samples.
|
MATERIALS AND METHODS |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONREFERENCES |
|---|
with 10% charcoal/dextran-treated FBS (HyClone), 1 mM L-glutamine and nonessential amino acids]. Reagents are: 10 nM E2 (Sigma); 100 nM 4-OHT (Calbiochem); 100 nM pure antiestrogen ICI (Tocris); 1 µM R5020 (Perkin-Elmer); 1 µM RU486 (Sigma); 1 µM atRA (Sigma); 25 µg/ml EGF (Invitrogen); 10 µM U0126 (Cell Signaling Technology); and 100 nM TPA (Sigma). Cells were treated in parallel with either reagent or vehicle control. At 2, 8, and 24 h, two dishes of cells (one reagent-treated and one control) were harvested for RNA extraction. The control for each pair was the matched reference sample for hybridization. Cells were washed twice with PBS and lysed using TRIzol reagent (Invitrogen). Total RNA was extracted from TRIzol as recommended by the manufacturer, then further purified by addition of an equal volume of 70% ethanol, followed by RNeasy column extraction (Qiagen).
MTT Growth Assays.
Quantitation of cell growth rates was determined using the Vibrant MTT cell proliferation kit (Molecular Probes). Cells were seeded in 96-well plates in triplicate and treated with reagent or vehicle control. Growth medium and reagent were replaced at 48-h intervals. Solubilized formazan concentration was determined at absorbance 570 nm.
Immunoblotting.
Cells were seeded into 6-well plates and cultured for 48 h in CSS media as described above. Cells were serum starved (CSS media with 0.5% serum) for 24 h before stimulation. Whole cell lysate preparation and immunoblotting were performed as described previously (30)
. Antibodies were obtained from Cell Signaling Technology and ICN Biomedicals, Inc.
cDNA Microarrays.
Microarray slides containing 13,824 sequence-verified cDNA clones (10,536 unique genes) were obtained from the National Human Genome Research Institute/National Institute of Neurological Disorders and Stroke/National Institute of Mental Health microarray core facility. Gene names were listed according to build 154 of the UniGene human sequence collection.3
Sample labeling and microarray hybridization were conducted as previously described (31
, 32)
and using standard protocols.4
Fluorescence scanning and image analysis with DeArray software were performed as described previously (33
, 34)
. Microarray data presented herein is also available.4
Data from replicate experiments were averaged (see "Appendix"), and the averaged data were preprocessed in Filemaker Pro to identify genes with a mean minimum quality statistic (35)
of 0.6 (scale 0–1) and mean intensity in the Cy5 or Cy3 channel of >500 units (scale 0–65,535). Genes were also retained if the mean Cy5 or Cy3 intensity in any one cell line exceeded 1000 units. To identify responsive genes (deviating in expression ratio), we filtered for genes showing a ratio > 1.5 fold, with a quality > 0.75 and intensity measurement > 1100. A 1.5-fold cutoff for expression ratio was calculated to be statistically significant (P < 0.01); average ratio variation from duplicated (approximately eight repeats/array) housekeeping clones within an array was 1.22 with no single experiment exceeding 1.5; average ratio variation for the same array feature within paired duplicates for the set of 42 duplicate experiments was 1.34; ratio variation between three self-self experiments was 1.44, 1.41, and 1.42. We arbitrarily chose 19 ratio measurements between 1.5 and 2 for Quantitative Reverse Transcription-PCR validation 19 of 19 validated the direction and 18 of 19 the magnitude of response (data available upon request). For genes with a 1.5-fold expression change and an average Cy5 or Cy3 intensity of <2250 units in any cell line, we increased the fold expression ratio stringency in this range to 2. Lastly, we omitted genes that displayed a Cy5 or Cy3 dye bias across >28 of the 42 experimental conditions.
Hierarchical Clustering.
Data were analyzed using web-based resources available at the NIH.5
Uncentered Pearson’s correlation was used on log-transformed data, with induced genes indicated in red and repressed genes in green. Gray indicates data points with quality < 0.3 units and intensity (in both channels) < 300 units.
Statistical Comparison with ER Status and Metastasis Discriminator Genes from Tumor Expression Profiles.
An association between our gene expression clusters and the ER status and patient prognosis discriminator genes reported by van’t Veer et al. (15)
was established using a 
2 test. Because the number of discriminator genes present in some clusters was small, we estimated Ps using a Monte Carlo procedure: 100,000 replicates were generated by random sampling from the set of all contingency tables with marginal data identical to the original data, and the probability to get a 2 larger than the original data was calculated. Once an overall association between clusters and ER status or prognosis genes was established, each cluster was separately analyzed for association with ER status or prognosis genes. The odds ratio of the odds for a gene in a cluster to be a discriminator gene to the same odds for a gene in all other clusters (2 x 2 table) was calculated using the conditional maximum likelihood estimate. Ps for the odds ratios were calculated using Fisher’s exact test.
GO Analysis.
The GO (36)
provides annotation for 60% of the clones in our experiments. From the directed acyclic graph structure of the GO, each node of annotation is coupled to overlying and underlying nodes via "has a" and "part of" relations. This implies that a clone mapped to any given node in the GO is also mapped to that node’s parent nodes. As the GO provides a mapping from LocusLink identifiers6
to GO nodes, we mapped each clone to a UniGene cluster, then used the LocusLink identifier associated with each UniGene cluster to assign each clone to GO nodes. We removed clone redundancy and different clones representing the same UniGene cluster (leaving one representative copy in the dataset for ontology analysis). We used reshuffling7
to sort the list of expression patterns with respect to similarity. Briefly, this algorithm is based on a quality function, describing an interaction between the distance of any two expression patterns on the list of genes and the similarity of the patterns defined by Pearson’s correlation. The algorithm results in a sorted gene list with the most correlated expression patterns adjacent. We analyzed the sorted lists using the GO by investigating if genes belonging to a GO node were associated with specific parts of the list. Briefly, we calculated the number of genes belonging to each GO node in a sliding window moving along the sorted list. For each GO node, we set the size of the sliding window such that we would on average expect four genes belonging to the node in each window for a randomly sorted list. Because each GO node had a different number of genes in the list, this means that the window size was set separately for each GO node. For each window and GO node, we plotted the number of genes belonging to the GO node divided by the expected number. Aligning these gene density plots with the sorted expression data provides for a way to associate expression patterns with specific GO nodes.
|
RESULTS AND DISCUSSION |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONREFERENCES |
|---|
and were conducted in biological replicate. Cell line-specific variations in the proliferative response to each agent were measured to aid in subsequent biological interpretation of gene expression data (Fig. 1)
. The most potent growth inhibitor of MCF7 cells was ICI and atRA for T47D cells. Steroid hormones did not stimulate the growth of ER- MDA-MB-436 cells as expected. Eighty-four hybridizations were performed on microarray slides containing 13,824 cDNAs. Duplicate experiments were averaged using a quality weighted algorithm (see "Appendix"), resulting in 42 data points for each gene. We stringently filtered our data for quality, then identified genes that responded >1.5-fold in 2 or more of the 42 conditions (see "Materials and Methods"). A total of 1023 unique genes met these requirements. Numerical data for the 1023 responsive genes is provided in supplementary Table 1. Analysis of gene expression responsiveness across the entire dataset revealed a remarkable set of 47 genes that responded in six or more different cell line/treatment conditions (Fig. 2, A and B)
. A high proportion of these genes encode proteins known to be strongly associated with oncogenesis such as MYC, ERBB2, CCND1, CD44, and ETS family members. Five of these 47 genes are ESTs, which certainly warrant additional investigation. Global gene expression responses across this dataset are summarized in Fig. 2C
. Every cell line/treatment resulted in both activation and repression of gene expression, with the number of responsive genes increasing with time, except for EGF and TPA treatments, which exhibited an early peak in responsiveness. The most dramatic response was in MCF7 cells treated with TPA, and the smallest response was to RU486 in MCF7 cells.
|
|
|
, see also supplementary Fig. 1). atRA was included in this clustering to facilitate comparison of expression signatures between the agents that act on nuclear receptors. Each agent caused discrete but interrelated patterns of gene expression, including both induction and repression. Overall, we observed a reciprocal pattern of gene regulation in MCF7 cells treated with E2 and antagonists ICI and 4-OHT. Many of the E2-induced genes (Fig. 3
, i and iii) have previously been reported to be up-regulated by E2 (19, 20, 21
, 23, 24, 25)
. The antiestrogen effects of ICI are more profound than that of 4-OHT, consistent with previous studies (37
, 38)
. The pattern of progestin stimulation in the two hormone-responsive cell lines displayed significant differences, in particular, a large cluster of genes specifically up-regulated in T47D by R5020 (Fig. 3
, vii). This is not unexpected given the large amount of progesterone receptor expressed in T47D. Additionally, these cells express a high level of ER-ß compared with MCF7 (39)
, and given the slight estrogenic effect of R5020 observed at 10-6 M (40)
, the T47D/R5020 expression profile may, in part, be mediated by estrogen receptors. The strong T47D/R5020 response is consistent with the higher degree of responsiveness of T47D to RU486 compared with MCF7 (Fig. 2C)
. The expression profile of MCF7 treated with E2 or R5020 appeared remarkably similar, although the response to R5020 was less pronounced. This observation is consistent with previously reported evidence of a weak estrogenic effect of R5020 on MCF7 cells and the previously observed abrogation of progestin-induced growth of MCF7 in the presence of antiestrogens (40
, 41)
. Retinoic acid treatment resulted in a distinct pattern of gene expression, although this contains striking overlaps with ICI treatment (Fig. 3
, ii and iv), illustrating common features in the transcriptional effects of these negative growth regulators.
|
(see supplementary Fig. 2). Similar to the hormone data, we observed genes regulated in a reciprocal fashion between EGF and U0126 treatment, indicating these genes are likely to be regulated after activation of Mek1/2 kinase and downstream effectors. The magnitude of the response of MCF7 to TPA treatment was large and more pronounced than that of MDA-MB-436. This may reflect the differences in phorbol ester sensitivity and PKC isoform expression/activation between ER+ and ER- breast cancer cells (42, 43, 44)
. Many of the clusters in Fig. 4A
indicate cell type-specific gene regulation (Fig. 4A
, i and ii responsive to EGF and TPA in MCF7, (Fig. 4A
, iv, and v, responsive only in MDA-MB-436). We also observed a strong cluster of genes induced by TPA and EGF in both cell lines (Fig. 4A
, iii). We observed a striking similarity in the pattern of gene expression with EGF and TPA in MCF7 (and similarly EGF and TPA in MDA-MB-436). As the pattern of EGF regulation is likely to denote a MAPK activation signature, the pattern similarity with TPA is consistent with reports describing PKC signaling to the MAPK axis (45
, 46)
. Moreover, the profound induction of gene expression by TPA in MCF7 suggested dramatically enhanced activation of MAPK by TPA. We examined activation of MAPK in MCF7 after EGF or TPA by immunoblotting (Fig. 4B)
. EGF-mediated MAPK activation peaked at 15 min without an increase in p44/42 MAPK protein, in agreement with previous data (47)
. TPA treatment of MCF7 cells resulted in prolonged phosphorylation of MAPK without a parallel increase in p44/42 MAPK protein level. A similar but delayed pattern of TPA-mediated MAPK phosphorylation was observed in MDA-MB-436 cells. Data consistent with growth arrest of TPA-treated MCF7 cells were a 6-fold increase in p21WAF1/CIP1 expression by 2 h, a 4.5-fold decrease in ER expression by 24 h and significant changes in morphology (Fig. 4C)
. There are numerous studies discussing the role of MAPK overstimulation and PKC activation in breast cancer. The complex impact of growth factors on these signaling mechanisms (including phosphorylation of ER and associated coactivators, as well as the level of expression of ER itself; Refs. 28
, 48
, 49
), highlights the interaction between these signaling mechanisms.
|
, see supplementary Fig. 3). Immediately evident was a profound overlap in genes regulated by steroids and those regulated in the kinase signaling cascades. This is consistent with recent literature discussing the ability of E2 to stimulate cell growth via nonclassical and nonnuclear pathways, including direct interaction of membrane-associated ER and Src, E2 action via G protein-coupled receptors, ERBB2, as well as activation of adenylate-cyclase and protein kinase A/PKC (47
, 50, 51, 52, 53, 54)
. Fig. 5A
shows a number of very distinct clusters. Cluster vi harbors genes strongly induced by E2 and strongly repressed by the antiestrogens and TPA. Sixty percent of the genes in this cluster are associated with mitosis, DNA synthesis, RNA processing, cell cycle regulation, including a number of known S-phase targets of phospho-RB, including DHFR, TK1, CDC2, and CCNA. Moreover, regulation of p21WAF1/CIP1 expression (E2 repressed and strongly TPA induced) is consistent with regulation of these genes. In dramatic contrast to Fig. 5A
cluster vi are clusters ii and iii, which also contain genes highly induced by E2, but unlike cluster vi, these genes are induced by EGF and TPA and repressed by U0126 and thus represent an important set of genes involved in E2-mediated MAPK signaling, a pathway critical for cell migration and tumor progression. This separation of the bulk of E2-induced genes into distinct clusters illustrates the necessity of studying ER-mediated signaling in the context of other signaling pathways. A number of the cell type-specific clusters observed in Figs. 3
and 4A
are maintained in Fig. 5A
. For example, Fig. 5A
cluster iv, a group of 23 genes induced by EGF and TPA only in MDA-436 cells, contains a striking number of biologically related genes, including four modulators of tumor growth factor ß member signaling (FST, CKTSF1B1, SMURF2, and TSC22) and five immune, stress, and inflammatory response genes (CSF3, IL1B, IL6, CCL3, and CXCL1).
|
for possible correlations with breast tumor gene expression signatures that discriminate ER+ from ER- tumors and separately for genes associated with metastatic disease. We used the recent data of van’t Veer et al. (15)
because the large number of genes included on their microarrays was likely to provide a good overlap with our gene set. First, of their 2460 ER status discriminator genes, we associated 1720 to a UniGene cluster (build 154), and 1365 of these were present on our arrays. Comparing these 1365 genes to the 1023 responsive genes included in Fig. 5A
, we found 158 clones in common (54 were more highly expressed in ER+ tumors (ER+ discriminators) and 104 in ER- tumors (ER- discriminators), according to van’t Veer et al. (15)
. This suggests that the majority of tumor ER status genes are not associated with the regulatory events observed in this in vitro study. We mapped the positions of the 158 discriminators with respect to the 44 clusters in Fig. 5A
(determined from the hierarchical gene tree, refer to supplementary Fig. 3) and, established using a 2 test (see "Materials and Methods"), that these ER discriminators are not randomly distributed among the clusters (for ER+ discriminators, P < 0.0003, and for ER- discriminators, P < 0.0001; see supplementary Fig. 3 for gene comparison). Odds ratios calculated for each of the 44 clusters indicated that four clusters, iv–vii had statistically more ER discriminator genes than expected by random chance (P < 0.01), as summarized in Fig. 5C
. The average log expression ratios in each experiment for clusters iv–vii are shown in Fig. 5B
. Fig. 5A
, ii, showed a weaker correlation (P = 0.015) with ER- status genes. The highest correlation for this comparison was observed for Fig. 5A
, cluster vi (P = 9.0 x 10-8), which contains many steroid hormone-induced genes, and is the proliferation cluster discussed in the previous section. Remarkably, 30% of the genes in this cluster are ER discriminators, and all 26 of these genes are ER- discriminators. This dramatic correlation may, in part, reflect the higher flow cytometric S-phase fraction observed in ER- breast tumors (55)
, and the high proportion of S phase, mitosis, and cell cycle-related genes in this cluster. Conversely to cluster Fig. 5A
, vi, the estrogen-induced genes in cluster v contain four ER+ status genes, ASAHI, NRIP1, MYB, and XBP1, a larger number than expected by chance (P = 4.7 x 10-3). This cluster interestingly lacks T47D progestin-induced genes and lacks genes down-regulated by TPA (compare Fig. 5B
clusters v and vi). Importantly, Fig. 5A
, cluster iii, the second large cluster of strongly E2-induced genes, does not correlate at all with ER discriminator genes. Lastly, Fig. 5A
, cluster iv showed a significant correlation (P = 0.0012) with genes discriminatory for the more aggressive ER- tumor type. Remarkably, of the eight ER discriminators in this cluster (CCL3, CXCL1, IL1B, IL6, CALB2, CTSL, ALDH1A3, and VIM), four are known to be involved in acute inflammatory processes (chemokines CCL3 and CXCL1 and interleukins IL1B and IL6). VIM has previously been reported as a predictor of more aggressive/invasive breast cell behavior (56)
.
Similar to the comparison with the ER discriminators, we conducted a comparison of our 1023 drug-responsive genes with the 231 genes identified by van’t Veer et al. (15)
as significantly associated with disease outcome. One-hundred fifty-two of the 231 genes were present on our arrays (107 with higher expression in metastatic tumors and 45 in nonmetastatic tumors). Forty-one of these 152 discriminators were present in our set of 1023 drug-responsive genes (see supplementary Fig. 3 for gene comparison). Using the 2 test, we established that these genes were not randomly distributed among our clusters (metastatic, P < 0.0005; nonmetastatic, P < 0.05). Remarkably, 19 of these 41 prognosis predictors are present in cluster vi, and all 19 were metastatic indicators [having higher expression in poor prognosis samples as compared with good prognosis samples (P = 1.5 x 10-13; Ref. 15
)]. This is consistent with the known importance of S-phase fraction in determining patient prognosis (57)
. Of the 89 genes in Fig. 5
cluster vi, 37 are either ER- discriminators or indicators of poor prognosis (eight genes are discriminatory for both), which is consistent with the proliferation signature defining this cluster.
GO Mapping of Expression Profiles Identifies Functional Responses to Drug Treatments.
Mapping of genes to GO nodes is a powerful functional genomics tool suited to the analysis of microarray data because one can discover whether related groups of genes from expression clustering share significant functional annotation in the GO database. Our data were analyzed with respect to the ontology provided by the GO Consortium (36)
. We have developed a novel method to identify the most significant GO nodes associated with gene expression patterns in individual or combinations of drug treatment profiles. This method involves first re-sorting gene expression profiles of interest using a reshuffling algorithm (see "Materials and Methods"), which reorders expression patterns such that the most correlated patterns become adjacent. This method is especially suited to extract the structure of smoothly varying data patterns typical of time course analyses. As a result, we observe a time course-like dimensionality along the reordered genes that cannot easily be achieved using most existing clustering methods. GO terms harboring dense numbers of genes with shared annotations are then identified and positionally plotted relative to the peaks in gene expression changes in the reshuffling plot (see "Materials and Methods"). We present two examples using this method: first, comparing E2 and E2 antagonist data (Fig. 6A)
; and second, MCF7/E2 with MCF7/EGF data (Fig. 6B)
. Fig. 6A
shows genes regulated by E2 and antiestrogens in MCF7. Using reshuffling and GO analysis, we identified distinct biological responses in this data, including a strong signal related to regulation of DNA and RNA metabolism and cell cycle genes, as well as cell motility and adhesion genes. Similarly, in Fig. 6B
, comparison of the GO peaks characterizing the MCF7 E2 and EGF responses reveals differences between these two mitogens. For example, both ligands lead to striking GO signals for receptor signaling and for adhesion-related genes, but these genes were regulated in opposite directions by these two treatments. These examples demonstrate the power of the controlled vocabulary of the GO database to facilitate automated extraction of potentially useful biological information from expression data. These results can then be used to generate hypotheses to guide additional mechanistic studies of the distinct and overlapping pathways associated with various drug treatments.
|
Our data provides a number of additional opportunities for future studies. Many of the regulated genes, including some very responsive genes (Fig. 2B)
, are ESTs. Investigation of these genes is likely to identify biochemical pathways relevant to the growth and differentiation of breast cancer cells. GO mapping has identified regulation in a number of functional categories, including cell adhesion and transport molecules, which have not been previously recognized as candidate targets of signal transduction pathways in breast cancer. It will be of considerable interest to examine the group of regulated genes we have identified for the presence of regulatory elements, which may account for their behavior. Finally, it will be possible to extend the comparison of our data with other clinical subsets such as hormone refractory ER+ tumors.
Cross-Array Normalization and Ratio Averaging.
For a set of replicated experiments where RNA from the same test sample was replicated and hybridized, with the same reference RNA, to multiple arrays, we prefer to average expression ratios to reduce the noise level, as well as maintaining the characteristics of independence when compared with the other samples. Notice that the array-array variation may come from multiple sources such as print-batch variation and hybridization noises; we design the averaging process in two steps: cross-array normalization and weighted ratio averaging.
Cross-Array Normalization.
Cross-array normalization is based on the fact that reference channel and test sample channel from replicated array are from the same RNA source, respectively. Therefore, beside the requirement that two channels within each array shall be correlated (which is the theoretical base of normalization relative to the reference channel for all experiments), we additionally constrain that for any two replicate experiments, the intensities of test sample channels, as well as intensities from reference channels, shall be correlated, with some level of interarray noises. Assume there are n replicated arrays for one test RNA, and let ri,n and gi,n denote the logarithm-transformed intensities from ith gene of test sample and reference sample in nth replication, respectively. When ri,n and gi are correlated to some extent, the parameters (an and bn) in the following regression are meaningful.
 | (1) |
n, ß n} and
 | (2) |
 | (3A) |
 | (3B) |
 | (4) |
Weighted Ratio Averaging.
After cross-array normalization in which ratios from all genes are corrected according to the normalization parameters given in Eq. 2
, the weighted average expression ratio for ith gene, ti, and its corresponding quality measure qi, are
 | (5A) |
 | (5B) |
 | (6A) |
 | (9) |
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
1 To whom requests for reprints should be addressed, at National Human Genome Research Institute, NIH, 50 South Drive, Bethesda MD 20892-8000. Phone: (301) 594-5283; Fax: (301) 402-3281; E-mail: pmeltzer{at}nhgri.nih.gov 
2 The abbreviations used are: ER, estrogen receptor-; CSS, charcoal dextran-treated serum; E2, 17ß-estradiol; ICI, ICI 182,780; 4-OHT, 4-hydroxy tamoxifen; atRA, all-trans-retinoic acid; EGF, epidermal growth factor; TPA, 12-O-tetradecanoyl-phorbol-13-acetate; MAPK, p44/42 mitogen-activated protein kinase; Mek, MAP/ERK kinase; PKC, protein kinase C; EST, expressed sequence tag; MTT, 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide; GO, Gene Ontology.
3 Internet address: http://www.ncbi.nlm.nih.gov/UniGene/.
4 Internet address: http://research.nhgri.nih.gov/microarray/.
5 Internet address: http://arrayanalysis.nih.gov/.
6 Internet address: http://www.ncbi.nlm.nih.gov/LocusLink.
7 Internet address: http://www.thep.lu.se/pub/Preprints/00/lu_tp_00_18.pdf.
Received 6/19/03. Revised 8/15/03. Accepted 8/27/03.
|
REFERENCES |
|---|
|
TopABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONREFERENCES |
|---|
in breast cancer cells. Breast Cancer Res. Treat, 70: 89-101, 2001.[Medline]
in the differential sensitivity of MCF-7 and MDA-MB 231 human breast cancer cells to phorbol ester-induced growth arrest and p21(WAFI/CIP1) induction. Cancer Lett., 172: 43-53, 2001.[Medline]
. J. Biol. Chem., 277: 22558-22565, 2002. gene expression by epidermal growth factor. J. Endocrinol., 165: 371-378, 2000.[Abstract]
B kinase. Mol. Cell. Biol., 22: 3549-3561, 2002.This article has been cited by other articles:
|
|
 |
|
  M. Lupien, J. Eeckhoute, C. A. Meyer, S. A. Krum, D. R. Rhodes, X. S. Liu, and M. Brown Coactivator Function Defines the Active Estrogen Receptor Alpha Cistrome Mol. Cell. Biol., June 15, 2009; 29(12): 3413 - 3423. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 F. Wu, I. Ivanov, R. Xu, and S. Safe Role of SP transcription factors in hormone-dependent modulation of genes in MCF-7 breast cancer cells: microarray and RNA interference studies J. Mol. Endocrinol., January 1, 2009; 42(1): 19 - 33. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Y. Li, P. Hao, S. Zheng, K. Tu, H. Fan, R. Zhu, G. Ding, C. Dong, C. Wang, X. Li, et al. Gene expression module-based chemical function similarity search Nucleic Acids Res., November 1, 2008; 36(20): e137 - e137. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. L. Willis, N. L. Tran, J. M. Chatigny, N. Charlton, H. Vu, S. A.N. Brown, M. A. Black, W. S. McDonough, S. P. Fortin, J. R. Niska, et al. The Fibroblast Growth Factor-Inducible 14 Receptor Is Highly Expressed in HER2-Positive Breast Tumors and Regulates Breast Cancer Cell Invasive Capacity Mol. Cancer Res., May 1, 2008; 6(5): 725 - 734. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 E. P. Moiseeva, R. Heukers, and M. M. Manson EGFR and Src are involved in indole-3-carbinol-induced death and cell cycle arrest of human breast cancer cells Carcinogenesis, February 1, 2007; 28(2): 435 - 445. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. K Gruvberger-Saal, H. E Cunliffe, K. M Carr, and I. A Hedenfalk Microarrays in breast cancer research and clinical practice - the future lies ahead Endocr. Relat. Cancer, December 1, 2006; 13(4): 1017 - 1031. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J M W Gee, V E Shaw, S E Hiscox, R A McClelland, N K Rushmere, and R I Nicholson Deciphering antihormone-induced compensatory mechanisms in breast cancer and their therapeutic implications Endocr. Relat. Cancer, December 1, 2006; 13(Supplement_1): S77 - S88. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. J. Creighton, A. M. Hilger, S. Murthy, J. M. Rae, A. M. Chinnaiyan, and D. El-Ashry Activation of Mitogen-Activated Protein Kinase in Estrogen Receptor {alpha}-Positive Breast Cancer Cells In vitro Induces an In vivo Molecular Phenotype of Estrogen Receptor {alpha}-Negative Human Breast Tumors. Cancer Res., April 1, 2006; 66(7): 3903 - 3911. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S.R. Hansson, Y. Chen, J. Brodszki, M. Chen, E. Hernandez-Andrade, J.M. Inman, O.A. Kozhich, I. Larsson, K. Marsal, P. Medstrand, et al. Gene expression profiling of human placentas from preeclamptic and normotensive pregnancies Mol. Hum. Reprod., March 1, 2006; 12(3): 169 - 179. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Frasor, J. M. Danes, C. C. Funk, and B. S. Katzenellenbogen Estrogen down-regulation of the corepressor N-CoR: Mechanism and implications for estrogen derepression of N-CoR-regulated genes PNAS, September 13, 2005; 102(37): 13153 - 13157. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. Laganiere, G. Deblois, C. Lefebvre, A. R. Bataille, F. Robert, and V. Giguere From the Cover: Location analysis of estrogen receptor {alpha} target promoters reveals that FOXA1 defines a domain of the estrogen response PNAS, August 16, 2005; 102(33): 11651 - 11656. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 F. Al-Shahrour, R. Diaz-Uriarte, and J. Dopazo Discovering molecular functions significantly related to phenotypes by combining gene expression data and biological information Bioinformatics, July 1, 2005; 21(13): 2988 - 2993. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Laganiere, G. Deblois, and V. Giguere Functional Genomics Identifies a Mechanism for Estrogen Activation of the Retinoic Acid Receptor {alpha}1 Gene in Breast Cancer Cells Mol. Endocrinol., June 1, 2005; 19(6): 1584 - 1592. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. G. Son, S. Bilke, S. Davis, B. T. Greer, J. S. Wei, C. C. Whiteford, Q.-R. Chen, N. Cenacchi, and J. Khan Database of mRNA gene expression profiles of multiple human organs Genome Res., March 1, 2005; 15(3): 443 - 450. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. Becker, A. Sommer, J. R. Kratzschmar, H. Seidel, H.-D. Pohlenz, and I. Fichtner Distinct gene expression patterns in a tamoxifen-sensitive human mammary carcinoma xenograft and its tamoxifen-resistant subline MaCa 3366/TAM Mol. Cancer Ther., January 1, 2005; 4(1): 151 - 170. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Tang, S. L. Tan, S. K. Ramadoss, A. P. Kumar, M.-H. E. Tang, and V. B. Bajic Computational method for discovery of estrogen responsive genes Nucleic Acids Res., December 1, 2004; 32(21): 6212 - 6217. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. Frasor, F. Stossi, J. M. Danes, B. Komm, C. R. Lyttle, and B. S. Katzenellenbogen Selective Estrogen Receptor Modulators: Discrimination of Agonistic versus Antagonistic Activities by Gene Expression Profiling in Breast Cancer Cells Cancer Res., February 15, 2004; 64(4): 1522 - 1533. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Cancer Research | Clinical Cancer Research |
| Cancer Epidemiology Biomarkers & Prevention | Molecular Cancer Therapeutics |
| Molecular Cancer Research | Cancer Prevention Research |
| Cancer Prevention Journals Portal | Cancer Reviews Online |
| Annual Meeting Education Book | Meeting Abstracts Online |
/
n (95% confidence interval). c, a significant association between the 1023 responsive genes shown in a and two sets of genes reported by van’t Veer et al. 