In addition to the RNA isolation techniques discussed in Section 3.2, there are a number of sample selection issues key to the proper interpretation of microarray assays (27). Sample selection includes considerations of the population of patients selected for the study, the selection of tissue within the patient, and method of selecting cells for RNA collection. The proper selection of patients cannot be overemphasized because no analytical plan can overcome a major error in sample selection. For example, a microarray analysis of two groups of patients—one sick and one well, in which all sick patients are female and all well patients are male—will identify sex-specific genes as being associated with disease. This extreme example reminds us that the epidemiologic principles of confounding are still present even in genomics experiments. A more subtle but related observation is the following. Microarray assays are sensitive to experimental conditions, and arrays performed at different facilities or under different conditions will show systematic variations called batch effects. If all samples with a given phenotype are performed in one lab and all samples with another phenotype are performed at a second lab, it can be difficult to isolate batch effects from biologic effects. Batch effects can be overcome by ensuring that samples are analyzed in a consistent manner and careful study design.
The earliest microarray studies were performed on uniform populations of cells such as cell culture or the so-called liquid tumors of leukemia and lymphoma. Uniform samples such as these are not representative of the majority clinical relevant specimens. Tumors are composed of varying percentages of inflammatory, stromal, normal, and malignant cells. Each of these cell populations, including the cancer cells, might be comprised of a variety of subpopulations. At the gross level, the investigator needs to assess which portions of a sample are appropriate for analysis. Selection of an obviously tumorous section of a specimen might include necrotic areas of poor RNA quality. Sectioning at the edge of a tumor might include a large percentage of normal cells. Some tumors, such as Hodgkin's disease and prostate cancer, frequently have large amounts of nonmalignant cells throughout the tumor. There is no consensus on which approach is correct, but the investigator should at least consider how selection might influence the study outcome. Early concerns on the usefulness of microar-rays have focused considerable attention on proper sampling with these issues in mind.
4.1. SPECIAL TECHNIQUES One approach to microscopic sampling has been to estimate percent tumor by light microscopy and by sample RNA from tissue sections immediately adjacent. More explicit techniques to sample only cells of interest from a heterogeneous background have been developed, including cell culture, flow cytometry, needle dissection, and, most commonly, laser capture microdissection (LCM). In LCM, a tissue section, either frozen or paraffin embedded, is placed on a glass slide and covered with a thin film of ethylene vinyl acetate. Under direct microscope visualization, a low-power infrared laser is directed at areas of interest, heating the polymer above the cells. After laser capture in this manner, the cells adherent to the film are removed for analysis (40). The resolution of the methods allows for capture of single cells without degradation of the RNA.
Laser capture microdissection has been used by many investigators for genomic profiling of tumor cells (41-43). In other cases, however, investigators have argued that tumors are complex systems in which the microenvironment of interaction between cells is critical. Investigators have shown that tumors with a low percentage of malignant cells can have distinct profiles, such as the clear differentiation of colon metastases from lung cancer in samples of 30-50% malignant cells. The debate regarding proper sample preparation has not yet been resolved. Most likely, the two approaches each have different merits, depending on the question the investigator is trying to answer (44,45). If the concern is to clarify an aspect of tumor cell biology, perhaps selective sampling is appropriate. If the interest is gross tumor behavior, then whole tumor sampling is preferred.
Selective sampling methods such as LCM and some nonse-lective processes such as fine-needle aspiration (FNA) biopsies will often fail to collect sufficient RNA for microarray analysis. A single LCM session will typically select 103-104 cells, at most a few nanograms of RNA, far short of that required. Similarly, many clinically available samples including but not limited to FNA also harvest insufficient RNA. To overcome this shortcoming, a variety of RNA amplification techniques have been developed (46-49). Initial validation in a variety of clinical and experimental settings demonstrate that RNA-amplified specimens will likely be acceptable for microarray analysis (41-43).
4.2. SELECTION OF NORMAL Mentioned in Section 3.2 were technical concerns for selection of a reference sample for cDNA microarrays that relate to the problem of variable spot size. The goal of reference selection was to achieve a nonzero-level mRNA expression across all genes of interest so that stable ratios could be calculated. Many investigators have used "normal" RNA for that purpose, although, as we will discuss, the selection of control group involves different concerns than the selection of the reference RNA. The selection of a control group, be it normal tissue or otherwise, applies equally to olgonucleotide and cDNA arrays.
In considering what constitutes a normal control, the investigator can either collect histologically normal tissue from the same diseased individual or normal tissue from another source. Uninvolved adjacent tissue is often used for this purpose, although there is ample evidence that such tissue might be influenced by local molecular effects related to a disease process such as cancer. These histologically normal but molec-ularly abnormal controls might mask underlying biology of interest in the analysis. Alternatively, the use of tissue from another individual might reveal gene expression differences that do not relate to the disease of interest. Again, consider the comparison of sick females to healthy males in which sex-specific genes confound the relationship between disease and normal. The usual confounders of age, ethnicity, medications, and others still need consideration in the study design and interpretation. Likewise, the diseased group might be heterogeneous in factors unrelated to the biology of their illness that could influence gene expression and deserve similar attention.
Was this article helpful?
Are You Prepared For Your First Baby? Endlessly Searching For Advice and Tips On What To Expect? Then You've Landed At The Right Place With All The Answers! Are you expecting? Is the time getting closer to giving birth to your first baby? So many mothers to be are completely unprepared for motherhood and the arrival of a little one, but stress not, we have all the answers you need!