DNA microarrays permit the analysis of the relative expression level of thousands of genes in a single experiment. Arrays can be membrane-based or slide-based. Nylon membranes are spotted with cDNA clones and probed with radiolabeled sample. Slide-based arrays are composed of glass microscope slides specially treated with an adherent such as polylysine or aminosilane. Glass arrays can be spotted with over 40,000 cDNA clones or presynthesized oligonucleotides using fine print tips or an ink jet printer, or prepared with oligonucleotide probes synthesized in situ using lithographic or ink jet technology. Slide-based arrays, which are generally probed with fluorescent dye-labeled sample, are smaller and easier to handle than membrane-based arrays for high throughput, although membrane-based arrays require less input RNA.
Tumor or breast tissue RNA is isolated from a snap-frozen specimen. In contrast to DNA, which may be extracted from tissue left at room temperature or from archival formalin-fixed tissue, RNA is less stable. Human tissue contains ribonucleases that contribute to RNA degradation, so the time between tissue devascularization and freezing at -80°C may affect both the quality of RNA and the genes that are expressed [23
]. Tissue specimen of less than 0.5 cm thickness, such as core needle biopsies, may be preserved at room temperature in solutions that permeate the tissue and stabilize its RNA (e.g. RNAlater
™, Ambion Inc., Austin, TX, USA, or RNAlater
™ TissueProtect Tubes, Qiagen Inc., Ventura, CA, USA). Recently, RNA isolated from paraffin-embedded tissue has been tested and compared to fresh specimen, generally on a gene-by-gene basis using real-time quantitative PCR assays. Studies on the suitability of paraffin-embedded RNA for array-based examinations are ongoing [24
]. Formalin preservation of tissue causes RNA and protein cross-linking that interfere with molecular analyses. In addition, RNA hydrolysis and fragmentation occur at the high temperatures required for paraffin embedding. Non-aldehyde-based tissue fixatives, such as ethanol and methanol, and low-melt polyester wax embedding compounds seem to hold promise, although long-term nucleic acid or protein stability are still in question and the performance of immunohistochemical staining antibodies would require reassessment. Recently developed commercial kits that facilitate the isolation of RNA from formalin-fixed paraffin-embedded tissues are undergoing testing.
For microarray experiments, either total RNA or mRNA is isolated from an experimental sample. The RNA is reverse transcribed to cDNA, directly or indirectly labeled with a fluorescent dye, and hybridized to the microarray. If RNA quantity is insufficient as a result of small tissue sample size, in vitro
transcription-based linear amplification [25
] may be performed. This can generate enough amplified antisense RNA, also known as complementary RNA, for array hybridization. When using cDNA microarrays, a differentially labeled reference sample is used with the experimental sample so that ratio measurements cancel out differences in hybridization kinetics and quantity of cDNA spotted on a given array. Total RNA, obtained from cell lines that reproducibly express a majority of human genes, may be used as a standard reference sample that allows comparisons among multiple experimental samples, even though they may be performed on different days and with different array print batches. By convention, the experimental (tumor) sample is labeled with a red fluorophore (Cy 5, which fluoresces at 635 nm) and the reference sample is labeled with a green fluorophore (Cy 3, which fluoresces at 532 nm). Based on the specificity and affinity of complementary base pairing, gene expression for each cDNA clone on the array is captured as signal intensities when the labeling dyes are fluoresced at the two appropriate wavelengths in an optical scanner. The measured signal intensities are normalized and a log ratio of the normalized signal intensities for the experimental sample compared to reference for each spot on the array is computed. This ratio essentially reflects the relative abundance of a particular gene in the experimental sample compared to the reference sample. The simultaneous measurement of relative gene expression of thousands of genes provides a genome-wide 'portrait' of gene expression for a tumor or other tissue. The data set is analyzed using bioinformatics tools [27
] to identify groups of genes that may define subtypes within an experimental set according to differences in their expression profiles. Correlations of the subtypes with histologic or clinical parameters are performed with the objective of identifying groups of genes that may define characteristic features of a tumor.
Early studies of expression profiling of breast cancer were performed on cell cultures and invasive breast cancers [29
]. Tumor specimens contained mixed cell populations: epithelial cells, stromal fibroblasts, vascular and lymphatic endothelial cells, adipocytes, and tumor-infiltrating lymphocytes and macrophages. The important signaling between epithelial and adjacent non-epithelial cells (tumor microenvironment) was captured in the molecular profile of the whole tissue, and gene expression of non-epithelial populations could be distinguished. There are now multiple studies evaluating expression profiles of invasive breast cancer using different array technologies and on different patient populations [31
], including patients carrying BRCA
susceptibility genes [35
] and young breast cancer patients [37
Using their transcriptional profiles, invasive breast cancers may be divided by molecular subtype into groups with different responses to systemic therapy and different survival patterns [39
]. Tumor gene expression patterns from patients with locally advanced breast cancer, who were similarly treated with doxorubicin followed by tamoxifen, were distributed among five molecular subtypes. Two subtypes, denoted luminal A and B, were characterized by high relative expression of the estrogen receptor (ER) gene and other ER-associated genes, and showed cytokeratin expression patterns suggestive of luminal epithelial cell origin. The luminal subtypes comprised patients who had long-term survival, in spite of their advanced disease (luminal A), and patients with poor survival (luminal B), reflecting either differing tumor biology or differing responses to systemic therapy, including possible tamoxifen insensitivity. The other subtypes showed relatively little expression of ER-associated genes (most were ER-negative tumors) and were divided into three subtypes: an ERBB2
overexpressing group, a basal epithelial-like group (named for their high relative expression of basal cytokeratins), and a group that expressed normal-like genes, including genes known to be expressed in adipose and stromal tissue. The basal-like group (ER-negative and without ERBB2
overexpression) contained high-grade tumors that were associated with high proliferation rates and 82% harbored mutations in the TP53
gene. The expression patterns of luminal, basal, and ERBB2
-overexpressing tumors described in this study appear to correlate with the different tumor subtypes described by others using CGH or immunohistochemistry [40
Olopade and Grushko [42
] suggest that tumors with BRCA1
mutations may be consistent with a basal-like pattern of gene expression because six out of seven tumors from patients with BRCA1
mutations stained positive for basal keratins and none showed ERBB2
overexpression. They confirmed this in a larger study of BRCA1
-associated tumors that showed no or low ERBB2
amplification by fluorescence in situ
hybridization assays [43
]. This is in contrast to tumors from patients with BRCA2
mutations that, in a limited number, appeared to have a luminal, ER-positive pattern. The findings of estrogen and progesterone receptor negativity, lack of ERBB2
overexpression, and overall higher grade in tumors from patients with BRCA1
mutations, compatible with a basal-like molecular phenotype, was confirmed by Lakhani and colleagues [44
] in a larger series of 217 patients with BRCA1
mutations, comparing them to 103 patients with sporadic breast cancer. They also found that breast cancers caused by BRCA2
mutations had immunohistochemical profiles similar to sporadic breast cancers, although they were more likely to be ERBB2
Based on the CGH work described above, it is anticipated that noninvasive precursor lesions may be characterized by similar molecular phenotypes as invasive breast cancer. Expression profiling of pre-invasive lesions, however, is technically more complex. First, it is difficult to freeze this tissue prior to diagnosis. Atypical hyperplasias or DCIS frequently present as non-palpable mammographic abnormalities (e.g. microcalcifications). Patient care necessitates that the entire surgical biopsy specimen be analyzed, without saving tissue for molecular analyses, for the following reasons: ADH and DCIS may be adjoining; DCIS requires thorough histologic examination in order not to miss areas of microinvasion; and margin status is vital for treatment decisions if DCIS or microinvasive carcinoma is identified. Therefore, the immediate freezing of surgical biopsies of mammographic abnormalities is generally not performed. However, with proper informed consent, additional core needle biopsies may be obtained at the time of mammographic stereotactic or ultrasound-directed core needle biopsy and frozen or stored in a commercial reagent that preserves both tissue architecture and RNA integrity. Using RNAlater
™ (Ambion Inc.), Ellis and colleagues [45
] were able prospectively to obtain sufficient high-quality RNA for transcriptional profiling from preoperative or postoperative core needle breast biopsies.
Laser microdissection may be used to isolate pre-invasive lesions from adjacent 'normal' ductolobular tissue [46
]. A purified population of epithelial or stromal cells may be obtained, and in conjunction with RNA amplification techniques [47
], expression profiling of the cells can be performed. From a single modified radical mastectomy specimen, Sgroi et al
] microdissected normal epithelial cells, malignant invasive epithelial cells, and cells metastatic to an axillary lymph node and used the RNA from these specimens for studies on nylon membrane arrays containing approximately 8000 genes. Verifying gene expression with duplicate hybridizations, real-time quantitative PCR and immunohistochemistry, they confirmed the feasibility and validity of this technique. Luzzi and colleagues [49
] compared the expression profiles of non-malignant human breast epithelium and adjacent DCIS microdissected from three breast cancer patients and identified several differentially expressed genes that had been previously implicated in human breast cancer progression.
Adeyinka et al
] compared six cases of DCIS with necrosis (4 of high nuclear grade and 2 with intermediate nuclear grade) to four cases of DCIS without necrosis (all with low nuclear grade) using microdissection and 5544 spot membrane arrays. Similar to CGH studies, distinct expression changes associated with DCIS grade and morphology were found. Some of the genes that differed between the two groups included those involved in cell cycle regulation, signaling, apoptosis, and response to hypoxia. In particular, the upregulation of AAMP
, angio-associated, migratory cell protein gene, in high grade DCIS with necrosis was demonstrated using array technology, real-time PCR, and in situ
hybridization – a gene considered to function in migrating cells and which may be hypoxia-mediated in tumors. The four DCIS samples without necrosis demonstrated little gene expression variability, in contrast to the highly variable DCIS samples with necrosis, and consistent with the hypothesis that low-grade DCIS may represent a single molecular phenotype.
Ma et al
] compared microdissected epithelial cells captured from normal breast lobules, ADH, DCIS, and invasive ductal carcinoma. They examined 39 breast specimens, 36 containing cancer (5 of the 36 had DCIS only) and three from reduction mammoplasties. Comparing gene expression profiles of premalignant, pre-invasive, and invasive cells to normal cells isolated from the same specimen, but distant from the tumor, or from reduction mammoplasties, they observed no consistent major transcriptional differences between ADH, DCIS and invasive ductal carcinoma from the same specimen. There were, however, distinct tumor signature differences between low-grade and high-grade tumors. Grade II tumor expression profiles were mixed, showing either low-grade or high-grade signatures. This corroborates previous limited data showing similarity between DCIS and invasive breast cancer from Porter et al
] using serial analysis of gene expression, and immunohistochemical data from Warnberg et al
] suggesting that well differentiated DCIS progresses to well differentiated invasive cancer and that poorly differentiated DCIS progresses to poorly differentiated invasive cancer. Ma et al
. also showed that a small subset of genes whose expression increased between DCIS and invasive breast cancer, predominantly in high-grade lesions, were related to cellular proliferation/ cell cycle regulation. Significantly, compared to normal epithelium, ADH appeared to be a genetically advanced lesion with an expression profile that resembled DCIS and invasive breast cancer within the same specimen. This study by Ma, Erlander, and Sgroi is the first to use transcriptional profiling to demonstrate that ADH and DCIS are direct precursors to invasive ductal carcinoma, confirming the work by Boecker [54
] using double-immunofluorescence staining techniques, which suggested that ADH is a committed precursor lesion to different molecular phenotypes of invasive breast cancer.
Analyzing data obtained using 16,000 gene oligonucleotide arrays, Ramaswamy et al
] suggested a set of 17 genes whose common expression across multiple primary solid tumor types and their metastases identified tumors with metastatic potential. van 't Veer et al
] described a 70 gene prognosis profile in women less than 55 years of age that outperformed standard prognostic criteria in a follow-up validation study [38
]. One might hypothesize that if (i) breast epithelial cells are committed to a neoplastic subtype in the ADH stage, and (ii) gene expression profiles of pre-invasive lesions presage the molecular phenotype of invasive cancers, and (iii) different molecular phenotypes of invasive breast cancer vary in their clinical outcome, then examination of pre-invasive lesions for unfavorable expression signatures may distinguish breast tissue that may ultimately evolve into metastatic breast cancer. By eradicating more aggressive subtypes of pre-invasive lesions using surgery, radiation, or targeted chemoprevention, the development and clinical outcome of invasive breast cancer might be favorably influenced.