Purpose: The purpose of this study was to determine whether comprehensive transcriptional profiles (TPs) can be obtained from single-passage fine-needle aspirations (FNAs) of breast cancer and to explore whether profiles capture routine clinicopathological parameters.

Experimental Design: Expression profiles were available on 38 patients with stage I-III breast cancer who underwent FNA at the time of diagnosis. [33P]dCTP-labeled cDNA probes were generated and hybridized to cDNA membrane microarrays that contained 30,000 human sequence clones, including 10,890 expressed sequence tags.

Results: The median total RNA yield from the biopsies was 2 μg (range, 1–25 μg). The cellular composition of each biopsy was examined and, on average,, 79% of the cells were cancer cells. TP was successfully performed on all 38 of the biopsies. Unsupervised complete-linkage hierarchical clustering with all of the biopsies revealed an association between estrogen receptor (ER) status and gene expression profiles. There was a strong correlation between ER status determined by TP and measured by routine immunohistochemistry (P = 0.001). A similar strong correlation was seen with HER-2 status determined by fluorescent in situ hybridization (P = 0.0002). Using the first 18 cases as the discovery set, we established a cutoff of 2.0 and 18.0 for ER and HER-2 mRNA levels, respectively, to distinguish clinically-negative from -positive cases. We also identified 105 genes (excluding the ER gene) the expression of which correlated highly with clinical ER status. Twenty tumors were used for prospective validation. HER-2 status was correctly identified in all 20 of the cases, based on HER-2 mRNA content detected by TP. ER status was correctly identified in 19 of 20 cases. When the marker set of 105 genes was used to prospectively predict ER status, TP-based classification correctly identified 9 of 10 of the ER-positive and 7 of 10 of the ER-negative tumors. We also explored supervised cluster analysis using various functional categories of genes, and we observed a clear separation between ER-negative and ER-positive tumors when genes involved in signal transduction were used for clustering.

Conclusions: These results demonstrate that comprehensive TP can be performed on FNA biopsies. TPs reliably measure conventional single-gene prognostic markers such as ER and HER-2. A complex pattern of genes (not including ER) can also be used to predict clinical ER status. These results demonstrate that needle biopsy-based diagnostic microarray tests may be developed that could capture conventional prognostic information but may also contain additional clinical information that cannot currently be measured with other methods.

DNA microarrays represent an important new tool to analyze human tissues. The technology enables one to measure the expression of several thousands of mRNA species in a specimen; this process is often referred to as TP.3 There are several commonly used technology platforms that use glass slides, silicone chips, or plastic membranes to plant either short oligonucleotides or complete cDNA sequences onto the surface of these “chips” (1). Glass-based arrays require fluorescein-labeled probes and generally require larger amounts of total RNA to generate a successful signal. Membrane arrays mostly use 33P-radioisotope-labeled probes and, because of the increased sensitivity of radioactive signal detection, often require smaller amounts of RNA for hybridization. cDNA microarray technology has been applied to characterize cancer in general and breast cancer in particular (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13). The molecular profiles of hereditary breast cancer (BRCA1/BRCA2) are distinct from sporadic tumors based on the expression pattern of 6,500 genes (3). On the basis of the transcriptosome of 5500 genes, breast cancers may also be classified into luminal or basal epithelial cell type (4). ER-negative breast cancers seem to have distinct TPs compared with ER-positive tumors (5, 6, 7, 8). Gene expression profiles may also predict relapse in clinically node-negative breast cancers (8). All of these previously published studies used glass-slide-based methods without further cDNA amplification and used surgically dissected breast tumor specimens. There are some inherent limitations in using gross surgical tissue for TP. It is not practical for serial sampling and the TPs reflect the contribution of a large number of cellular components including normal breast epithelium, cancer cells, fibroblasts, adipocytes, infiltrating leukocytes and vascular components. For certain experimental questions, this tissue heterogeneity may be desirable, whereas for others, this may be a confounding factor. Also, a theoretical concern remains that the profiles are distorted by transcriptional response to surgical stress, tissue handling, and general anesthetics or other drugs administered before and during surgery. The expression of genes involved in signal transduction and response to stress or hypoxia may be particularly sensitive to rapid changes in the tissue microenvironment that occur during surgery.

For these reasons, we believe that obtaining reliable TPs from single-passage FNA of human tumors would be an important technological advance. FNAs are minimally invasive and, therefore, more acceptable for serial tumor sampling. Also, the cells removed with this method frequently represent relatively pure tumor cells. The contribution of infiltrating leukocytes and other stromal elements can be quantified in the same specimen that is profiled. Also, cells can be processed within minutes after removal from the tumor; therefore, gene expression profiles of FNA specimens likely resemble the in vivo profiles very closely.

In this article, we report our results on assessing the expression of 30,000 human sequence clones in 38 single-passage FNA biopsies obtained from breast cancer. Our goal was to test the feasibility of this sampling approach and to assess whether routine clinicopathological parameters such as ER and HER status are reliably captured by the mRNA profiles. The patient population does not yet have sufficiently long follow-up to correlate expression profiles with clinical outcome, which is the ultimate goal of this project.

Patients.

As part of an ongoing clinical trial at the MDACC, FNAs are being collected for TP from patients with newly diagnosed breast cancer before any systemic therapy. The study was approved by the Institutional Review Board of MDACC and all of the patients signed an informed consent. All of the patients were female, had clinically palpable lump in the breast, and were currently receiving therapy for their breast cancer at MDACC. In brief, three to four FNA passages were performed, and cells were collected from each pass and were placed into separate vials containing RNAlater solution (Ambion, Austin TX) or were snap-frozen and stored at −80°C. Ten to 14 cytological smears were also prepared from one of the passages. The cellularity and cellular composition of each aspirate that was processed for TP was assessed in matching smears by a cytopathologist (W. F. S.). If needed, separate tissue samples, usually from core needle biopsies, were obtained for diagnostic purposes. The status of the ER and HER-2 receptor was assessed on the diagnostic core biopsies by routine clinical pathology independent of the microarray experiments. The ER was assessed by IHC using 1D5 antibody from Zymed (San Francisco, CA). Tumors were considered hormone receptor-positive if ≥10% of cells showed nuclear staining for ER. HER-2 receptor status was determined by IHC using an anti-HER2 monoclonal antibody, AB8 (Neomarker, Fremont, CA). Patients were also tested for gene amplification by FISH using the PathVision kit (Vysis, Downers Grove, IL). Tumors with a gene copy ratio of >2.0 of HER2:chromosome 17 centrosome (CEP17), or with IHC staining intensity of 3+ were considered to have HER-2 overexpression.

RNA Isolation and cDNA Microarray Hybridization.

Total RNA was extracted from single-pass FNA specimens with an RNeasy kit (Qiagen, Valencia CA). The amount and quality of RNA were assessed with Agilent 2100 Bioanalyzer RNA 6000 LabChip kit (Agilent Technologies, Paolo Alto, CA). First-strand cDNA synthesis was performed with Superscript II (Invitrogen, Carlsbad, CA) in the presence of [33P]dCTP (100 mCi/ml; Amersham, Little Chalfont, United Kingdom) from 1–2 μg of total RNA. The generated cDNA probes without further amplification were hybridized to high-density cDNA microarray membranes proprietary to Millennium Pharmaceuticals Inc. The membranes contain 30,000 human sequence clones, including 10,890 ESTs, obtained from Unigen and verified by direct sequencing.

Data Analysis.

Expression results were normalized to the median expression value of each membrane and were filtered for sequences with Alu-repeats (14). Because of the possibility of cross-hybridization between various Alu-containing sequences, sequences with Alu repeats were removed from the final analysis, which left 27,159 sequences for further analysis. To screen out genes with unreliable low signal and genes that would not contribute to molecular differences between individual samples, several approaches were used. In one approach, genes the normalized expression value of which fell within the first decile were removed, which left 19,387 genes for further analysis. To identify genes the expression of which was similar in all of the specimens, we calculated robust SD of normalized expression values for each gene across all of the specimens. Genes with a SD <0.17 were removed, which left 13,924 genes for further analysis. SD 0.17 is the first decile for the robust SD values calculated as SD = 2*IQR/3, where IQR is the interquartile range. To restrict analysis to highly expressed genes only, genes with normalized expression value of >4.0 in at least 5 of the 18 discovery cases were selected, which resulted in 5,110 genes. The value of 4.0 was chosen because it is above the typical mean value of gene expression in individual membranes. The median values were set to 1 by the normalization process, the post-normalization mean expression values ranged from 2.5 to 4.3 with a median of 3.2 for individual experiments. The most stringent analysis combined all of the filtering criteria including only genes with no Alu-repeats, large expression differences across specimens, and normalized expression value of ≥4.0, which resulted in 2,695 genes. Multidimensional scaling of logarithmic data and cluster analyses were used to explore similarities between expression profiles (15). Supervised and unsupervised single and complete linkage clustering were performed by using the 1-minus-the-rank correlation coefficient as the distance metric. We also performed hierarchical cluster analysis using genes assigned to distinct functional classes as defined by the National Cancer Institute, CGAP.4 Correlation between cluster outcome and clinical parameters were assessed by Fisher’s exact and exact Wilcoxon rank-sum tests. To identify genes the expression of which is associated with clinical ER status, we calculated two-sample T scores for each gene with and without log transformation. For all statistical analyses, Unix S-plus Version 3.4 software was used (MathSoft Inc., Seattle, WA).

RNA Yield from FNA Biopsies.

RNA extraction was attempted on 65 prospectively collected single-pass FNA biopsies. Total RNA was successfully extracted from 46 (71%) of the 65 specimens. In the remaining 19 cases, the quality of the RNA was not adequate for additional experiments assessed by the “Lab-on-chip” method of Agilent Technologies. The specimens were prioritized for TP based on the type of preoperative chemotherapy the patients received. Thirty-eight patients received or are currently receiving uniform paclitaxel/doxorubicin-based chemotherapy, and these were selected for TP. The clinical characteristics of these 38 patients are presented on Table 1. Median total RNA yield of the 38 cases was 2 μg/FNA passage (range, 1–25 μg; mean, 3.8 μg). Cytopathological assessment of the smears prepared from the aspirates showed that, on average, 80% of the cells were neoplastic. The remaining cells mostly represented infiltrating lymphocytes. Table 2 shows the RNA yield and cellular composition of each individual sample that was used for the profiling experiments.

ER and HER-2 Receptor Status Determined by Routine Clinical Methods Corresponds to Receptor Status Assessed by cDNA Microarrays.

The first 18 samples, including 9 ER-negative and 9 ER-positive cases, were used to assess the correspondence between clinical hormone receptor status and receptor status assessed by TP. The next 20 cases were used as an independent validation set to test the prospective predictive power of the retrospectively determined cutoff values. PR was not represented on our array; therefore, correlation for this hormone receptor could not be determined. There was a remarkable correlation between ER status measured by routine IHC and by TP; rank correlation coefficient was 0.91 and P = 0.0002 (Fig. 1). A cutoff value of >2.0 for normalized ER gene expression was chosen to assign cancers into the ER-positive group. This cutoff point was determined by inspecting the relationship between clinical ER status and transcription results (Fig. 1). The median ER expression in the ER-negative group (n = 9) was 0.7 (range, 0.4–1.1; mean, 0.7) compared with 7.7 (range, 1.4–29.8; mean, 11.3) in the ER-positive group (n = 9). The subsequent 20 samples (11 ER-positive, 9 ER-negative) were used to validate prospectively this empirically identified threshold. The profiling results correctly identified all 9 of the clinically ER-negative cases (range, 0.23–2.0) and 10 of the 11 ER-positive cases (range, 2.5–10.5). Among all of the cases, all 18 of the ER-negative tumors were correctly classified based on their ER mRNA content, and 18 of the 20 ER-positive tumors were correctly identified based on ER mRNA levels >2.0.

A similar analysis was performed to assess the correlation between HER-2 expression measured by TP and the expression measured by routine clinical IHC or FISH. In the training set, we established a threshold of 18.0 for the HER-2 mRNA level, which identified all four patients correctly with 3+ HER-2 overexpression or gene amplification; rank correlation coefficient was 0.9 and P = 0.0002 (Fig. 2). The median HER-2 gene expression in the HER-2-overexpressing group was 65.7 (range, 19.6–202.4; mean, 88.4) compared with 6.9 (range, 2.2–14.9; mean 7.9) in the HER-2-normal cases. The retrospectively fitted cutoff value 18 was prospectively tested in the next 20 cases, and it has correctly identified all 15 of the HER-2-normal (range, 1.9–17.7) and all 5 of the HER-2-overexpressing cases (range, 40.0–82.2). Among all of the 38 cases, the cutoff yielded a 100% accuracy compared with FISH.

ER-positive and ER-negative Cancers Have Distinct Gene Expression Profiles.

We also explored our data to search for a gene marker set, excluding the ER gene itself, that may reliably identify clinical ER status. Unsupervised complete linkage hierarchical clustering and multidimensional scaling were used to identify gene expression differences among the first 18 tumors. The first split of the complete linkage cluster diagrams generated with 13,962 genes, 5,110 genes, and 2,695 genes, respectively, all separated patients into two broad groups. When clinicopathological parameters, including tumor size, nodal status, nuclear grade, ER and HER-2 status, and age of each case, were added to the cluster tree diagrams, a correlation between ER status and the two major clusters emerged. The other clinical parameters (i.e., age, size, nodal status, HER2 status, or nuclear grade) did not correlate with the two large clusters. The best separation between ER-negative and ER-positive breast cancer was achieved when the most stringently filtered 2,695 genes were used. The first split of the cluster tree separated patients into two groups; one included 7 ER-negative and 3 ER-positive cases and the second, 6 ER-positive and 2 ER-negative (Fisher exact test, P = 0.054; Fig. 3). Multidimensional scaling produced similar results; the first principal coordinate of plots generated from the same 2,695 genes also separated patients into two groups, one with seven ER-negative and another with nine ER-positive and two ER-negative (data not shown).

Next, we used supervised analysis to search for a set of marker genes that differentiates ER-positive versus ER-negative tumors more accurately than unsupervised methods. Using the most stringently filtered 2,695 genes from the first 18 cases, we calculated two-sample T scores for each gene for which negative scores indicated a higher mean expression in ER-negative tumors and positive scores indicated higher expression in ER-positive tumors. A score of 2.120 was significant at the 5% level, and 2.921 at a 1% level (two-sided testing with 16 degrees of freedom). At the 5% significance level, 106 genes were positively associated and 249 negatively associated with ER status. At the 1% level, 42 genes were positively associated (including the ER gene) and 64 negatively associated with ER (Table 3). If none of the genes were truly differentially expressed between the two groups, then we would expect to see by chance alone, 135 genes significant at the 5% level and ∼27 at the 1% level. We were particularly interested in genes that form a molecular fingerprint of ER-negative cancers. Fig. 4 illustrates the differential expression of five genes the expression of which correlated highly with ER-negative status, including the “homologue of mouse quaking QKI,” “HSPC182 protein,” “interleukin 1 receptor-associated kinase,” “interleukin-10 receptor,” “platelet phosphofructokinase,” “ataxia-telangiectasia group D-associated protein,” and “LIM domain only 4.” Several of the genes that were differentially expressed in our experiments were also identified by Gruvberger et al.(5) using a different TP technique in surgical tissue. Eleven of the top 50 ER-associated genes published by that group also appear on our list of genes at the 10% significance level. Some of the genes with increased expression in ER-positive cancers identified by both sets of experiments included “trefoil factor 3,” “serine/cysteine proteinase inhibitor clad I,” “insulin-like growth factor-binding protein 2,” “cytochrome C oxidase subunit VI c,” and “cysteine-rich protein 1.” Genes associated with ER negativity included “serum constituent protein (H74163),” “solute carrier family 7 member 5,” “N-myc downstream regulated (AA489266),” platelet phosphofructokinase, ataxia-telangiectasia group D-associated protein, and LIM domain only 4.

To prospectively test the predictive power of the top 105 ER status-associated genes, excluding the ER gene itself, we repeated the cluster analysis on 20 independent cases (10 ER-positive and 10 ER-negative). Our marker genes have correctly classified 9 of the 10 ER-positive and 7 of the 10 ER-negative cases (Fig. 5). One of the ER-negative tumors that clustered with the ER-positive cases were clinically borderline, showing 5% ER expression on IHC; another had ER mRNA expression of 2.5. These results demonstrate that, by analyzing the expression profile of a select group of genes, one can identify most ER-negative cancers as distinct from ER-positive tumors without considering the ER gene itself.

ER-positive and ER-negative Breast Cancers Express a Distinct Set of Signal Transduction Molecules.

All previous analysis was performed without prior knowledge of the function of genes that were fed into the clustering algorithm. This provides an unbiased method to explore the transcriptosome and can identify novel genes previously not associated with a certain clinical parameter or outcome. Restricting hierarchical cluster analysis to functional groups of genes may provide more easily interpretable insights into biological differences among ER-positive and ER-negative tumors. Many of the known genes have been grouped into functional categories by the CGAP.5 We performed cluster analysis using genes classified by the CGAP as involved in cell signaling, signal transduction, development, pharmacology, immunology, cell cycle, transcription, angiogenesis, tumor suppression, gene regulation, and metastasis. Of these functional gene families, only genes involved in signal transduction (n = 249) yielded a cluster dendrogram that significantly correlated (P = 0.006, Fischer’s exact test) with ER status (Fig. 6). One major cluster contained 12 ER-negative tumors but only 2 ER-positive cancers, compared with the other main cluster, which had 7 ER-negative and 17 ER-positive cancers. This suggests that distinct signaling mechanisms operate in at least a subset of ER-negative tumors compared with ER-positive cancers.

Our results demonstrate that comprehensive gene expression profiles can be obtained from FNA biopsies of breast cancer. Sufficient quantity and quality RNA for microarray analysis was extracted from 71% of single passage FNAs with a median yield of 2 μg of total RNA. Since completion of these experiments, we have discovered that a substantial amount of intact RNA can be recovered from the supernatant of cells collected into RNAlater (Ambion) solution (16). Extraction of RNA from both the cell pellet and supernatant will further increase the total RNA yield from FNA specimens. Our study does not address directly what the minimal number of cells is, or the minimal amount of RNA that is needed, to perform a successful cDNA microarray hybridization experiment. Two studies (17, 18) addressed this question previously and reported that the median number of cells in a FNA specimen obtained from breast cancer is ∼200,000 cells (range, 30,000–2,500,000); and the minimum amount of RNA needed for reproducible hybridization is 0.5 μg.

In our data, ER and HER-2 mRNA expression measured by TP corresponded closely to clinical receptor status as assessed by routine clinical methods. The correlation between ER status assessed by IHC and by TP was very strong. A retrospectively identified threshold of 2.0 for ER mRNA expression correctly identified clinical ER status in all but 2 of the 38 cases, including 20 independent cases that were not included in the first case set used to determine the cutoff value. A similar strong correlation emerged between clinical HER-2 overexpression and expression of HER-2 mRNA assessed by TP. A threshold of 18.0 for HER-2 gene expression has correctly identified the clinical HER-2 status of all of the 38 patients including the 20 prospectively validated cases. For ER and HER-2, a close correlation between protein and mRNA levels measured by Northern blot or PCR has already been reported (19, 20, 21). Our results demonstrate that the mRNA expression profiles generated by cDNA microarrays also capture accurately the conventional single-gene diagnostic markers such as ER or HER-2 expression.

We also examined whether clinical ER status can accurately be predicted based on a TP using a set of marker genes, not including ER itself. Using the first 18 patients as a discovery set, we identified 105 genes that were highly associated with ER status. We prospectively tested the predictive power of this marker set on 20 new cases. Gene expression profiles correctly identified the clinical ER status of 16 (80%) of the 20 cases. To use cDNA microarrays to determine ER status, when there is a cheap and time-honored alternative, IHC, may not seem to be an important advance. However, TPs may be able to distinguish clinical subsets of patients within a hormone receptor group that is not currently possible with IHC. ER-positive (or ER-negative) breast cancers have a heterogeneous clinical course, and many ER-positive tumors do not respond to any form of hormonal therapy. Indeed, a recent report suggests that ER-positive cancers may form two distinct groups based on their TP with different clinical outcome (22). The patients in this present study do not yet have sufficiently prolonged follow-up to correlate their clinical outcome with TPs. However, their clinical outcome data are collected and will be analyzed in the future.

Most of the genes that account for the molecular differences between ER-negative and ER-positive breast cancer in our experiments, but also reported by others, are not known to be regulated by ER (5, 23). This suggests that the differences between ER-negative and ER-positive cancers may not simply be attributable to the absence or presence of ER function but rather reflect different molecular phenotypes. This is further supported by our observation that ER-negative and ER-positive tumors can be separated based on the gene expression signature of signal transduction molecules. The molecular differences between ER-negative and ER-positive tumors are particularly interesting in the light of clinical observations that indicate distinct clinical behavior of ER-positive and ER-negative tumors. ER-negative cancers tend to develop more commonly in premenopausal woman, may recur sooner, and may spread more commonly to lung, liver, and the central nervous system, compared with ER-positive tumors (24). Furthermore, when tamoxifen is used as chemopreventive agent, there is reduction only in the incidence of ER-positive cancers (25). This suggests that these two distinct forms of breast cancer may develop through different molecular mechanisms.

Genes associated with ER-negative status are of particular interest because they may reveal the biological causes of the distinct behavior of these tumors. They may also provide potential targets for drug development. We identified 64 genes from a larger pool of differentially expressed sequences that are highly expressed in ER-negative compared with ER-positive tumors. Many of these genes were differentially expressed with the same magnitude and consistency as seen with ER itself. Six of the top 50 genes strongly associated with “ER-negativity” identified in our analysis have also been reported by others to be ER-associated using a different profiling platform on gross surgically resected tissues (5). However, we could not confirm all of the previously reported observations, including preferential expression of cytokeratin 7, P-cadherin, and ladinin in ER-negative breast cancer. This may be attributable to technical differences such as using different nucleic acid sequences of the same gene on the arrays or to differences in the composition of the tissues used for the experiments. Nine of the top 20 genes highly expressed in ER-negative tumors seem to participate in the regulation of global gene expression and cellular signaling. The gene that is most significantly associated with ER-negative status in our study is the “human homologue of mouse quaking gene.” It is a KH domain RNA-binding protein that regulates intracellular RNA trafficking and RNA stability (26). “Nonhistone chromosomal protein high mobility group isoforms I/Y” and “nuclear autoantigenic sperm protein (NASP)” participate in histone remodeling and influence gene transcription (27, 28).“Spermidine synthase,” another gene closely associated with ER negativity is the source of spermidine, a polyamine that has a profound effect on cell proliferation and differentiation (29). Two nuclear proteins, exportin 1 (CRM1) and Ser/Arg-related nuclear matrix protein that regulate protein transport between the nucleus and cytoplasm were also highly expressed in ER-negative tumors (30, 31). Another set of genes codes for proteins that participate in signal transduction. These include interleukin 1 receptor-associated kinase 1 (IRAK-1) and interleukin 10 receptor, both strongly expressed by ER-negative tumors. It appears that the metabolic activity of ER-negative tumors may also be different, which is suggested by high expression of the glycolytic enzyme phosphofructokinase, phosphoglucomutase, and NADH-Coenzyme Q reductase.

In a separate study (32), we previously examined the cellular composition of FNA specimens and determined that this tissue-sampling method yields material that is highly enriched in tumor cells compared with core needle (CBX) or surgical biopsies. Microscopic cell counts from FNAs demonstrated means of 80% cancer cells, 15% lymphocytes, and 5% stromal cells; whereas CBXs contained 50% cancer cells, 20% lymphocytes, and 30% stromal cells. The implication of this observation is that TPs generated from FNA specimens are primarily dominated by mRNA signal originating from tumor cells whereas profiles from surgical specimens and CBX are composites of mRNA from stromal cells and tumor cells.

In summary, our results demonstrate that small amounts of breast cancer tissue obtained with FNA can yield sufficient RNA to perform comprehensive TPs. These profiles reliably capture conventional single-gene prognostic information such as ER and HER-2 status. The information on clinical ER status is independently imbedded in both the absolute expression level of ER mRNA and in a complex pattern of expression of ER-status associated genes. This latter is of importance because a larger sample size may provide an opportunity to classify patients within an ER group to clinically relevant subgroups. ER-associated genes can also further our understanding of the distinct biology of EP-positive breast cancer compared with ER-negative tumors. Furthermore, the supervised mathematical techniques applied in this research to identify ER-associated genes can also be used to define predictive gene sets for clinical outcomes that cannot be predicted with other methods. All of the patients included in this study currently receive preoperative chemotherapy with a paclitaxel/doxorubicin-based regimen, and their clinical outcome, including response to therapy, is available. Our next planned analysis of these data describes markers of chemotherapy response (33).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1

Supported by a research grant from Millennium Pharmaceuticals Inc., Cambridge, MA (to L. P.).

3

The abbreviations used are: TP, transcriptional profiling/profile; ER, estrogen receptor; PR, progesterone receptor; FNA, fine-needle aspiration; MDACC, University of Texas M. D. Anderson Cancer Center; FNA, fine-needle aspiration; IHC, immunohistochemistry; FISH, fluorescent in situ hybridization; EST, expressed sequence tag; CGAP, Cancer Genome Anatomy Project; CBX, core-needle biopsy.

4

Internet address: http://cgap.nci.nih.gov/Genes/CuratedGeneLists.

Fig. 1.

Concordance between ER mRNA expression measured by TP and by routine IHC. The bar graphs represent normalized gene expression levels of ER in the first 18 cases. The % values above each bar, the percentage of ER-positive cells in the tumor determined by the routine clinical laboratory.

Fig. 1.

Concordance between ER mRNA expression measured by TP and by routine IHC. The bar graphs represent normalized gene expression levels of ER in the first 18 cases. The % values above each bar, the percentage of ER-positive cells in the tumor determined by the routine clinical laboratory.

Close modal
Fig. 2.

Concordance between HER-2 receptor mRNA expression measured by TP and by routine IHC and FISH. The bar graphs represent normalized gene expression levels of HER-2 in the first 18 cases. The numbers above each bar, the clinical IHC score (1+ to 3+). Below each case, FISH results for HER-2; relative HER2 gene copy number is shown only if it is >2.0. If no gene amplification was detected, than a “−” sign is shown for clarity.

Fig. 2.

Concordance between HER-2 receptor mRNA expression measured by TP and by routine IHC and FISH. The bar graphs represent normalized gene expression levels of HER-2 in the first 18 cases. The numbers above each bar, the clinical IHC score (1+ to 3+). Below each case, FISH results for HER-2; relative HER2 gene copy number is shown only if it is >2.0. If no gene amplification was detected, than a “−” sign is shown for clarity.

Close modal
Fig. 3.

Unsupervised complete-linkage hierarchical clustering and its correlation with clinicopathological parameters for the first 18 patients. The distance metric used for the generation of this graph was the complement of the correlation coefficient (1 − r) computed on log-transformed gene expression values. A total of 2695 genes were entered into the analysis after gene filtering as described in the Text. Clinical parameters are also shown for each case. TxP, TP results; for ER, a cutoff value of 2.0 used to determine positivity; for HER-2, a cutoff value of 18.0 was used to determine positivity. Blacks Modified Nuclear grade, tumor size, nodal status (based on clinical TNM staging), and the age of the patient are also shown.

Fig. 3.

Unsupervised complete-linkage hierarchical clustering and its correlation with clinicopathological parameters for the first 18 patients. The distance metric used for the generation of this graph was the complement of the correlation coefficient (1 − r) computed on log-transformed gene expression values. A total of 2695 genes were entered into the analysis after gene filtering as described in the Text. Clinical parameters are also shown for each case. TxP, TP results; for ER, a cutoff value of 2.0 used to determine positivity; for HER-2, a cutoff value of 18.0 was used to determine positivity. Blacks Modified Nuclear grade, tumor size, nodal status (based on clinical TNM staging), and the age of the patient are also shown.

Close modal
Fig. 4.

Differential expression of six genes between ER-positive and ER-negative breast cancers. The expression of ER 1 (Estrogen receptor 1), interleukin-1 receptor associated kinase 1 (IRAK-1), human homologue of mouse quaking gene QKI, interleukin receptor 10α (IL-10R alpha), platelet phosphofructokinase, and LIM domain only 4 genes are shown in ER-negative (n = 19) and ER-positive cancers (n = 19). The box plots were generated using gene expression data from all 38 of the cases; and Ps are based on two-sample t statistics.

Fig. 4.

Differential expression of six genes between ER-positive and ER-negative breast cancers. The expression of ER 1 (Estrogen receptor 1), interleukin-1 receptor associated kinase 1 (IRAK-1), human homologue of mouse quaking gene QKI, interleukin receptor 10α (IL-10R alpha), platelet phosphofructokinase, and LIM domain only 4 genes are shown in ER-negative (n = 19) and ER-positive cancers (n = 19). The box plots were generated using gene expression data from all 38 of the cases; and Ps are based on two-sample t statistics.

Close modal
Fig. 5.

Complete-linkage hierarchical clustering and its correlation with ER status using 105 ER-associated genes. The distance metric used for the generation of this graph was the complement of the correlation coefficient (1 − r) computed on log-transformed gene expression values. Only 105 highly ER-status associated genes, not including ER itself, were entered into the analysis. These genes were identified by analysis of the first 18 cases. Results applied to all 38 cases are presented. All 18 of the cases included in the training set and 16 (80%) of the 20 new cases are correctly classified. N, ER-negative cases; P, ER-positive cases.

Fig. 5.

Complete-linkage hierarchical clustering and its correlation with ER status using 105 ER-associated genes. The distance metric used for the generation of this graph was the complement of the correlation coefficient (1 − r) computed on log-transformed gene expression values. Only 105 highly ER-status associated genes, not including ER itself, were entered into the analysis. These genes were identified by analysis of the first 18 cases. Results applied to all 38 cases are presented. All 18 of the cases included in the training set and 16 (80%) of the 20 new cases are correctly classified. N, ER-negative cases; P, ER-positive cases.

Close modal
Fig. 6.

Complete-linkage hierarchical clustering and its correlation with ER status using signal transduction-related genes. The dendrogram was generated by using a distance metric of 1 − the correlation coefficient calculated on log expression values. A total of 305 genes (see footnote 5) were included in the analysis after gene filtering to include only genes with expression values >10% quintile of each array and a CGAP functional annotation of “signal transduction.” P, ER-positive tumors; N, ER-negative tumors.

Fig. 6.

Complete-linkage hierarchical clustering and its correlation with ER status using signal transduction-related genes. The dendrogram was generated by using a distance metric of 1 − the correlation coefficient calculated on log expression values. A total of 305 genes (see footnote 5) were included in the analysis after gene filtering to include only genes with expression values >10% quintile of each array and a CGAP functional annotation of “signal transduction.” P, ER-positive tumors; N, ER-negative tumors.

Close modal
Table 1

Clinical characteristics of patients undergoing TP (n = 38)

The number of patients in each clinical category is presented in the table. HER-2 overexpression is defined as 3+ staining by IHC or gene amplification by FISH.

Age, yr median 51 (range 29–77) 
Race  
 Caucasian 25 
 African American 
 Hispanic 
 Asian 
Histology  
 Invasive ductal 34 
 Invasive mixed ductal and lobular 
 Invasive lobular 
Clinical stage before therapy  
 Tumor size  
  T1 
  T2 22 
  T3 
  T4 
 Node-positive 23 
 Node-negative 15 
Receptor status  
 ER+ 19 
 ER− 19 
 PR+ 19 
 PR− 19 
 ER+/PR+ 11 
 HER2 overexpressor 10 
 HER2 normal 28 
Age, yr median 51 (range 29–77) 
Race  
 Caucasian 25 
 African American 
 Hispanic 
 Asian 
Histology  
 Invasive ductal 34 
 Invasive mixed ductal and lobular 
 Invasive lobular 
Clinical stage before therapy  
 Tumor size  
  T1 
  T2 22 
  T3 
  T4 
 Node-positive 23 
 Node-negative 15 
Receptor status  
 ER+ 19 
 ER− 19 
 PR+ 19 
 PR− 19 
 ER+/PR+ 11 
 HER2 overexpressor 10 
 HER2 normal 28 
Table 2

Cellular composition and total RNA yield of single-pass FNA samples used for TP (n = 38)

The percentage of cancer cells and lymphocytes and the yield of total RNA (μg) from one passage of each FNA specimen is presented in the Table. When the percentage values do not add up to 100%, additional miscellaneous stromal cells account for the difference.

FNA% cancer cells% lymphocytesRNA yield
113 90 10 2.0 
128 95 2.0 
126 90 2.5 
106 90 3.5 
120 75 25 17.0 
123 not available not available 2.5 
115 85 12 2.0 
135 80 15 1.8 
139 65 18 2.8 
111 75 15 2.3 
110 85 13 2.5 
127 50 50 2.0 
102 25 70 3.9 
108 65 20 3.3 
133 95 1.9 
136 85 12 7.6 
117 85 11 6.7 
150 50 40 3.0 
116 75 20 1.7 
130 85 1.8 
153 100 15.0 
154 75 23 3.0 
155 100 1.5 
156 70 28 1.9 
157 95 1.5 
158 95 3.0 
159 70 26 1.2 
161 not available not available 1.0 
165 87 2.5 
170 72 1.0 
171 82 12 1.1 
174 75 15 3.0 
176 87 5.0 
177 79 15 2.0 
179 79 13 1.5 
180 88 3.2 
181 91 5.2 
182 88 1.0 
FNA% cancer cells% lymphocytesRNA yield
113 90 10 2.0 
128 95 2.0 
126 90 2.5 
106 90 3.5 
120 75 25 17.0 
123 not available not available 2.5 
115 85 12 2.0 
135 80 15 1.8 
139 65 18 2.8 
111 75 15 2.3 
110 85 13 2.5 
127 50 50 2.0 
102 25 70 3.9 
108 65 20 3.3 
133 95 1.9 
136 85 12 7.6 
117 85 11 6.7 
150 50 40 3.0 
116 75 20 1.7 
130 85 1.8 
153 100 15.0 
154 75 23 3.0 
155 100 1.5 
156 70 28 1.9 
157 95 1.5 
158 95 3.0 
159 70 26 1.2 
161 not available not available 1.0 
165 87 2.5 
170 72 1.0 
171 82 12 1.1 
174 75 15 3.0 
176 87 5.0 
177 79 15 2.0 
179 79 13 1.5 
180 88 3.2 
181 91 5.2 
182 88 1.0 
Table 3

A list of the top 50 genes associated with ER status

The table presents 50 of the most highly ER status-associated genes. The T score describes the strength of the association. A positive number reflects preferential expression in ER-positive tumors, whereas a negative number indicates expression in ER-negative tumors. As a reference point, the ER gene itself in this analysis had a T score of 3.1. GeneBank accession numbers are presented together with gene annotation including Unigene cluster identification (ID) number and finished sequence number if available.

T scoreAccession numberAnnotation
8.2  R00275::T99925 CGI-49 protein (Hs.238126;NM_016002) 
6.5  AA403072::AA404352 solute carrier family 2 member 10 (Hs.305971, NM_030777) 
 −5.2 AA489386::AA489445 homolog of mouse quaking QKI (Hs.15020;) 
 −4.9 AA703536 HSPC182 protein (Hs.30026;NM_014188) 
4.9  AA156802::AA156926 laminin, β 2 (laminin S) (Hs.90291;NM_002292) 
4.5  AA976544 hypothetical protein MGC2771 (Hs.321130;NM_024101) 
 −4.4 AA683550 interleukin 1 receptor-associated kinase 1 (Hs.182018;NM_001569) 
4.3  AA873604 cysteine-rich protein 1 (intestinal) (Hs.17409;NM_001311) 
4.2  AA054287::AA054406 RNA-binding motif protein 3 (Hs.301404;NM_006743) 
4.2  AA669545 spermidine synthase (Hs.76244;NM_003132) 
 −4.1 AA504379::AA504120 ESTs (Hs.99743;) 
 −4.1 AA476305::AA433827 KIAA1089 protein (Hs.4990;) 
 −4 AA448261 high-mobility group protein isoforms I and Y (NM_002131) 
 −4 AA644128 nuclear autoantigenic sperm protein (Hs.243886;NM_002482) 
3.9  AA610066 homeo box B6 (Hs.98428;NM_018952) 
 −3.8 AA488373::AA488504 phosphoglucomutase 1 (Hs.1869;NM_002633) 
 −3.8 W87528::W87611 nuclear factor I/B (Hs.33287;NM_005596) 
 −3.8 AA436378::AA481311 potassium channel modulatory factor (Hs.5392;NM_020122) 
3.8  T80582::R38933 Hs.19673::Hs.274404 suppressor of S. cerevisiae (NM_007265) 
3.8  AA464152::AA464217 Hs.77266::Hs.71816! quiescin Q6 (Hs.77266;NM_002826) 
3.7  H78462::H78365 CGI-119 protein (Hs.283670;NM_016056) 
 −3.7 AA456161::AA443094 DNA segment on chromosome 19 (unique) (Hs.30928;NM_006114) 
3.7  AA521439 hypothetical protein FLJ20163 (Hs.92254;NM_017695) 
 −3.7 AA214053::AA214154 NADH dehydrogenase (ubiquinone) Fe-S protein 5 (NM_004552) 
 −3.7 AA417805::AA417806 hypothetical protein (Hs.54971;NM_016505) 
 −3.7 AA437226::AA442290 interleukin 10 receptor α (Hs.327;NM_001558) 
 −3.6 T59055::T59131 exportin 1 (CRM1, yeast, homolog) (Hs.79090;NM_003400) 
 −3.6 AA608558 phosphofructokinase, platelet (Hs.99910;NM_002627) 
 −3.6 AA626024 Homo sapiens mRNA for KIAA1750 protein, partial cds (Hs.173094;) 
 −3.6 AA976063 Ser/Arg-related nuclear matrix protein (Hs.18192, NM_005839) 
3.6  AA872153 EST (Hs.326288;) 
 −3.6 AA455935::AA456404 Homo sapiens mRNA; cDNA DKFZp564H1916 
 −3.5 R42685::R17337 inositol(myo)-1 (or 4)-monophosphatase 2//(Hs.5753;NM_014214) 
 −3.5 AA085917::AA085918 bromodomain-containing 4 (Hs.278675;NM_014299) 
3.5  AA599177 cystatin C (Hs.135084;NM_000099) 
 −3.5 W01400::N62924 Hs.75724::Hs.107127 coatomer protein complex, subunit β2 
 −3.5 AA464354::AA464246 major histocompatibility complex, class I, C (Hs.277477;NM_002117) 
3.5  AI732756::AA487468 ESTs, weakly similar to JE0350 anterior gradient-2 (Hs.100686) 
 −3.5 T56982::T56983 phosphodiesterase 7A (Hs.150395;) 
 −3.4 AA599073 PP3111 protein (Hs.168541;NM_022156) 
3.4  R98407 Homo sapiens cDNA FLJ12900 fis, clone NT2RP2004321 (Hs.21851;) 
3.4  W90224::W90128 X-box binding protein 1 (Hs.149923;NM_005080) 
 −3.4 AA482117 Ras homolog enriched in brain 2 (Hs.279903;NM_005614) 
 −3.4 AA293211::AA293653 Hs.278268::Hs.194673! homolog of mouse MAT-1 oncogene 
3.4  AA171913::AA171613 carbonic anhydrase XII (Hs.5338;NM_001218) 
3.4  H68663::H68664 Homo sapiens cDNA FLJ12900 fis, clone NT2RP2004321 (Hs.21851;) 
 −3.4 AA402875 U6 snRNA-associated Sm-like protein LSm7 (Hs.70830;NM_016199) 
3.4  AA025380::AA025379 GATA-binding protein 3 (Hs.169946;NM_002051) 
 −3.3 AA677517 polypyrimidine tract-binding protein 
 −3.3 R23083::R22977 Moesin (Hs.170328;NM_002444) 
T scoreAccession numberAnnotation
8.2  R00275::T99925 CGI-49 protein (Hs.238126;NM_016002) 
6.5  AA403072::AA404352 solute carrier family 2 member 10 (Hs.305971, NM_030777) 
 −5.2 AA489386::AA489445 homolog of mouse quaking QKI (Hs.15020;) 
 −4.9 AA703536 HSPC182 protein (Hs.30026;NM_014188) 
4.9  AA156802::AA156926 laminin, β 2 (laminin S) (Hs.90291;NM_002292) 
4.5  AA976544 hypothetical protein MGC2771 (Hs.321130;NM_024101) 
 −4.4 AA683550 interleukin 1 receptor-associated kinase 1 (Hs.182018;NM_001569) 
4.3  AA873604 cysteine-rich protein 1 (intestinal) (Hs.17409;NM_001311) 
4.2  AA054287::AA054406 RNA-binding motif protein 3 (Hs.301404;NM_006743) 
4.2  AA669545 spermidine synthase (Hs.76244;NM_003132) 
 −4.1 AA504379::AA504120 ESTs (Hs.99743;) 
 −4.1 AA476305::AA433827 KIAA1089 protein (Hs.4990;) 
 −4 AA448261 high-mobility group protein isoforms I and Y (NM_002131) 
 −4 AA644128 nuclear autoantigenic sperm protein (Hs.243886;NM_002482) 
3.9  AA610066 homeo box B6 (Hs.98428;NM_018952) 
 −3.8 AA488373::AA488504 phosphoglucomutase 1 (Hs.1869;NM_002633) 
 −3.8 W87528::W87611 nuclear factor I/B (Hs.33287;NM_005596) 
 −3.8 AA436378::AA481311 potassium channel modulatory factor (Hs.5392;NM_020122) 
3.8  T80582::R38933 Hs.19673::Hs.274404 suppressor of S. cerevisiae (NM_007265) 
3.8  AA464152::AA464217 Hs.77266::Hs.71816! quiescin Q6 (Hs.77266;NM_002826) 
3.7  H78462::H78365 CGI-119 protein (Hs.283670;NM_016056) 
 −3.7 AA456161::AA443094 DNA segment on chromosome 19 (unique) (Hs.30928;NM_006114) 
3.7  AA521439 hypothetical protein FLJ20163 (Hs.92254;NM_017695) 
 −3.7 AA214053::AA214154 NADH dehydrogenase (ubiquinone) Fe-S protein 5 (NM_004552) 
 −3.7 AA417805::AA417806 hypothetical protein (Hs.54971;NM_016505) 
 −3.7 AA437226::AA442290 interleukin 10 receptor α (Hs.327;NM_001558) 
 −3.6 T59055::T59131 exportin 1 (CRM1, yeast, homolog) (Hs.79090;NM_003400) 
 −3.6 AA608558 phosphofructokinase, platelet (Hs.99910;NM_002627) 
 −3.6 AA626024 Homo sapiens mRNA for KIAA1750 protein, partial cds (Hs.173094;) 
 −3.6 AA976063 Ser/Arg-related nuclear matrix protein (Hs.18192, NM_005839) 
3.6  AA872153 EST (Hs.326288;) 
 −3.6 AA455935::AA456404 Homo sapiens mRNA; cDNA DKFZp564H1916 
 −3.5 R42685::R17337 inositol(myo)-1 (or 4)-monophosphatase 2//(Hs.5753;NM_014214) 
 −3.5 AA085917::AA085918 bromodomain-containing 4 (Hs.278675;NM_014299) 
3.5  AA599177 cystatin C (Hs.135084;NM_000099) 
 −3.5 W01400::N62924 Hs.75724::Hs.107127 coatomer protein complex, subunit β2 
 −3.5 AA464354::AA464246 major histocompatibility complex, class I, C (Hs.277477;NM_002117) 
3.5  AI732756::AA487468 ESTs, weakly similar to JE0350 anterior gradient-2 (Hs.100686) 
 −3.5 T56982::T56983 phosphodiesterase 7A (Hs.150395;) 
 −3.4 AA599073 PP3111 protein (Hs.168541;NM_022156) 
3.4  R98407 Homo sapiens cDNA FLJ12900 fis, clone NT2RP2004321 (Hs.21851;) 
3.4  W90224::W90128 X-box binding protein 1 (Hs.149923;NM_005080) 
 −3.4 AA482117 Ras homolog enriched in brain 2 (Hs.279903;NM_005614) 
 −3.4 AA293211::AA293653 Hs.278268::Hs.194673! homolog of mouse MAT-1 oncogene 
3.4  AA171913::AA171613 carbonic anhydrase XII (Hs.5338;NM_001218) 
3.4  H68663::H68664 Homo sapiens cDNA FLJ12900 fis, clone NT2RP2004321 (Hs.21851;) 
 −3.4 AA402875 U6 snRNA-associated Sm-like protein LSm7 (Hs.70830;NM_016199) 
3.4  AA025380::AA025379 GATA-binding protein 3 (Hs.169946;NM_002051) 
 −3.3 AA677517 polypyrimidine tract-binding protein 
 −3.3 R23083::R22977 Moesin (Hs.170328;NM_002444) 
1
King H. C., Sinha A. A. Gene expression profile analysis by DNA microarrays.
JAMA
,
286
:
2280
-2288,  
2001
.
2
Khan J., Wei J. S., Ringner M., Saal L. H., Ladanyi M., Westermann F., Berthold F., Schwab M., Antonescu C. R., Peterson C. Classification and diagnostic prediction of cancers using gene expression profiling and artificial neural networks.
Nat. Med.
,
7
:
673
-660,  
2001
.
3
Hedenfalk I., Duggan D., Chen Y., Radmacher M., Bittner M., Simon R., Meltzer P., Gusterson B., Esteller M., Kallioniemi O. P. Gene-expression profiles in hereditary breast cancer.
N. Engl. J. Med.
,
344
:
539
-548,  
2001
.
4
Perou C. M., Jeffrey S. S., van de Rijn M., Rees C. A., Eisen M. B., Ross D. T., Pergamenschikov A., Williams C. F., Zhu S. X., Lee J. C., Lashkari D., Shalon D., Brown P. O., Botstein D. Distinctive gene expression patterns in human mammary epithelial cells and breast cancers.
Proc. Natl. Acad. Sci. USA
,
96
:
9212
-9217,  
1999
.
5
Gruvberger S., Ringner M., Chen Y., Panavally S., Saal L. H., Borg A., Ferno M., Peterson C., Meltzer P. S. Estrogen receptor status in breast cancer is associated with remarkably distinct gene expression patterns.
Cancer Res.
,
61
:
5979
-5984,  
2001
.
6
Sorlie T., Perou C. M., Tibshirani R., Aas T., Geisler S., Johnsen H., Hastie T., Eisen M. B., van de Rijn M., Jeffrey S. S., Thorsen T., Quist H., Matese J. C., Brown P. O., Botstein D., Lonning P. E., Borresen-Dale A. L. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications.
Proc. Natl. Acad. Sci. USA
,
98
:
10869
-10874,  
2001
.
7
West M., Blanchette C., Dressman H., Huang E., Ishida S., Spang R., Zuzan H., Olson A., Jr., Marks J. R., Nevins J. R. Predicting the clinical status of human breast cancer by using gene expression profiles.
Proc. Natl. Acad. Sci. USA
,
98
:
11462
-11467,  
2001
.
8
van’t Veer L. J., Dai H., Vijver M. J., He Y. D., Hart A. A. M., Mao M., Peterse H. L., Kooy K., Marton M., Witteveen A. T., Schreiber G. J., Kerkhoven R. M., Roberts C., Linsley P. S., Bernards R., Friend S. H. Gene expression profiling predicts clinical outcome of breast cancer.
Nature (Lond.)
,
415
:
530
-535,  
2002
.
9
Perou C. M., Sorlie T., Eisen M. B., van de Rijn M., Jeffrey S. S., Rees C. A., Pollack J. R., Ross D. T., Johnsen H. Molecular portraits of human breast tumors.
Nature (Lond.)
,
406
:
747
-752,  
2000
.
10
Alizadeh A. A., Eisen M. B., Davis R. E., Ma C., Lossos I. S., Rosenwald A., Boldrick J. C., Sabet H., Tran T., Yu X., Powell J. I. Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling.
Nature (Lond.)
,
403
:
503
-511,  
2000
.
11
Anbazhagan R., Tihan T., Bornman D. M., Johnston J. C., Saltz J. H., Weigering A., Piantadosi S., Gabrielson E. Classification of small cell lung cancer and pulmonary carcinoid by gene expression profiles.
Cancer Res.
,
59
:
5119
-5122,  
1999
.
12
Su A. I., Welsh J. B., Sapinoso L. M., Kern S. G., Dimitrov P., Lapp H., Schultz P. G., Powell S. M., Moskaluk C. A., Frierson H. F., Jr. Molecular classification of human carcinomas by use of gene expression signatures.
Cancer Res.
,
61
:
7388
-7393,  
2001
.
13
Takahashi M., Rhodes D. R., Furge K. A., Kanayama H., Kagawa S., Haab B. B., The B. T. Gene expression profiling of clear cell renal cell carcinoma: gene identification and prognostic classification.
Proc. Natl. Acad. Sci. USA
,
98
:
9754
-9759,  
2001
.
14
Deininger P. L., Batzer M. A. Alu repeats and human disease.
Mol. Genet. Metab.
,
67
:
183
-193,  
1999
.
15
Eisen M. B., Spellman P. T., Brown P. O., Botstein D. Cluster analysis and display of genome-wide expression patterns.
Proc. Natl. Acad. Sci. USA
,
95
:
14863
-14868,  
1998
.
16
Dunmire V., Wu C., Symmans W. F., Zhang W. Increased yield of total RNA from fine-needle aspirates for use in expression microarray analysis.
Biotechniques
,
33
:
890
-896,  
2002
.
17
Ellis M., Davis N., Coop A., Liu M., Schumaker L., Lee R. Y., Srikanchana R., Russel C., Singh B., Miller W. R., et al Development and validation of a method for using breast core needle biopsies for gene expression microarray analyses.
Clin. Cancer Res.
,
8
:
1155
-1166,  
2002
.
18
Assersohn L., Gangi L., Zhao Y., Dowsett M., Simon R., Powles T. J., Liu E. T. The feasibility of using fine needle aspiration from primary breast cancers for cDNA microarray analyses.
Clin. Cancer Res.
,
8
:
794
-801,  
2002
.
19
Anderson J. M., Clarke E., Howell R., et al Effects of short-term antiestrogen treatment of primary breast cancer on estrogen receptor mRNA and protein expression and on estrogen-regulated genes.
Breast Cancer Res. Treat.
,
41
:
31
-41,  
1996
.
20
Sarkar F. H., Visscher D. W., Crissman J. D. Quantitative analysis of Her-2/neu (ERBB2) gene expression using reverse transcriptase polymerase chain reaction.
Diagn. Mol. Pathol.
,
2
:
210
-218,  
1993
.
21
Slamon D. J., Godolphin W., Jones L. A., Holt J. A., Wong S. G., Keith D. E., Levin W. J., Stuart S. G., Udove J., Ullrich A. Studies of the HER-2/neu proto-oncogene in human breast and ovarian cancer.
Science (Wash. DC)
,
244
:
707
-712,  
1989
.
22
Sorlie T., Perou C. M., Tibshirani R., Aas T., Geisler S., Johnsen H., Hastie T., Eisen M. B., van de Rijn M., Jeffrey S. S. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications.
Proc. Natl. Acad. Sci. USA
,
98
:
10869
-10874,  
2001
.
23
Charpentier A. H., Bednarek A. K., Daniel R. L., Hawkins K. A., Laflin K. J., Gaddis S., MacLeod M. C., Aldaz C. M. Effects of estrogen on global gene expression: identification of novel targets of estrogen action.
Cancer Res.
,
60
:
5977
-5983,  
2000
.
24
Clark M. Prognostic and predictive factors Harris J. R. Lippman M. Morrow M. E. Hellman M. S. eds. .
Diseases of the Breast
,
:
461
-485, Lippincott-Raven Philadelphia  
1996
.
25
Fisher B., Costantino J. P., Wickerham D. L., Redmond C. K., Kavanah M., Cronin W. M., Vogel V., Robidoux A., Dimitrov N., Atkins J. Tamoxifen for prevention of breast cancer: report of the National Surgical Adjuvant Breast and Bowel Project P-1 Study.
J. Natl. Cancer Inst. (Bethesda)
,
90
:
1371
-1388,  
1998
.
26
Li Z., Zhang Y., Li D., Feng Y. Destabilization and mislocalization of myelin basic protein mRNAs in quaking dysmyelination lacking the QKI RNA-binding proteins.
J. Neurosci.
,
20
:
4944
-4953,  
2000
.
27
Wisniewski J. R., Schwanbeck R. High mobility group I/Y: multifunctional chromosomal proteins causally involved in tumor progression and malignant transformation.
Intl. J. Mol. Med.
,
6
:
409
-419,  
2000
.
28
Richardson R. T., Batova I. N., Widgren E. E., Zheng L. X., Whitfield M., Marzluff W. F., O’Rand M. G. Characterization of the histone H1-binding protein, NASP, as a cell cycle-regulated somatic protein.
J. Biol. Chem.
,
275
:
30378
-30386,  
2000
.
29
Celano P., Berchtold C. M., Giardiello F. M., Casero R. A., Jr. Modulation of growth gene expression by selective alteration of polyamines in human colon carcinoma cells.
Biochem. Biophys. Res. Comm.
,
165
:
384
-390,  
1989
.
30
Gallouzi I. E., Steitz J. A. Delineation of mRNA export pathways by the use of cell-permeable peptides.
Science (Wash. DC)
,
294
:
1895
-1901,  
2001
.
31
Davido T., Getzenberg R. H. Nuclear matrix proteins as cancer markers.
J. Cell. Biochem. Suppl.
,
35
:
136
-141,  
2000
.
32
Symmans W. F., Pusztai L., Ayers M., Clark E., Stec J., Buchholz T. A., Sneige N., Hortobagyi G. N. RNA yield from needle biopsies for cDNA microarray analysis of breast cancer prior to neoadjuvant chemotherapy.
Breast Cancer Res. Treat.
,
69
:
301
2001
.
33
Ayers M., Symmans W. F., Stec J., Metivier J., Damokosh A., Clark E., Sneige N., Carter C., Whitman G., Hortobagyi G. N., Pusztai L. Gene expression profiling of fine needle aspirations of breast cancer identifies genes associated with complete pathologic response to neoadjuvant Taxol/FAC chemotherapy.
Breast Cancer Res. Treat.
,
76
:
S64
2002
.