The targets for transcript profiling are complex RNA populations from cellular sources of interest, often cultured cells, surgically excised tissues or tumors, or isolated portions of such. Typical microarrays use many hundreds or thousands of gene-specific probes. cDNA probes have been used but the modern trend has been to use oligonucleotide probes. Two popular systems are Affymetrix GeneChip microarrays where oligonucleotides about 25 nucleotides long are synthesized in situ on an array, and Illumina microarrays where pre-synthesized oligonucleotides with a gene-specific component ~50 nucleotides long are attached to beads. More recently, high-density arrays of longer oligonucleotides have also become available.
Microarray-based expression analyses are often organized so as to compare two or more highly-related cellular or tissue sources that differ in an informative way. To carry out transcript profiling, the cellular RNA sample is normally reverse transcribed enmasse to form a representative complex cDNA population. Labeling of the cDNA can be achieved during synthesis (by including a fluorophore-conjugated nucleotide in the reaction mix). Alternatively, a two-step procedure is used. First, unlabeled cDNA is made. Then the cDNA is converted into labeled complementary RNA (cRNA) by incorporating biotinylated nucleotides (the biotin labels will later be detected with fluorophore conjugated streptavidin).
The labeled target cDNA or cRNA is applied to the array and allowed to hybridize. Each individual feature or spot on the array contains large numbers of copies of the same DNA sequence, and is therefore unlikely to be completely saturated in the hybridization reaction. Under these conditions, the intensity of the hybridizing signal at each feature on the array is proportional to the relative abundance of that particular cDNA or cRNA in the target population, which in turn reflects the abundance of the corresponding mRNA in the original source population. The relative abundance of thousands of different transcripts can therefore be monitored in one experiment. Multiple oligonucleotides are also used to help distinguish between closely related transcripts from individual genes. It is possible to monitor splice variants, for example, and to design oligonucleotides specific for every single known exon.
The huge amount of expression data generated by microarray-based hybridization analyses require careful statistical analyses. Stringent controls are also required to normalize expression data for cross-experiment variation. One way to avoid such problems is to hybridize cDNA populations labeled with different fluorophores to the same array simultaneously. Under nonsaturating conditions, the signal at each feature will represent the relative abundance of each transcript in the sample. If two samples are used, then the ratio of the signals from each fluorophore provides a direct comparison of expression levels between samples, fully normalized for variations in signal-to-noise ratio even within the array. The array is scanned at two emission wavelengths and a computer is used to combine the images and render them in false color. Usually, one fluorophore is represented as green and the other as red. Features representing differentially expressed genes show up as either green or red, while those representing equivalently expressed genes show up as yellow—see the example of using spotted cDNA arrays in Figure 1A.
Fig1. Comparative expression analysis with DNA microarrays. (A) Using spotted cDNA arrays. Here, comparative expression assays are usually carried out by differentially labeling two RNA or cDNA samples with different fluorophores; the labeled nucleic acids are hybridized to the arrayed cDNAs and then scanned to detect both fluorophores independently. Colored dots labeled X, Y, and Z at the bottom of the image correspond to three hypothetical genes present at increased levels in sample 1 (X, red), increased levels in sample 2 (Y, green), and similar levels in samples 1 and 2 (Z, yellow). (B) Using Affymetrix GeneChips. Here RNA is labeled in a two-step process to produce biotinylated cRNA. After hybridization and washing, biotin-cRNA bound to the array is stained by binding a streptavidin-conjugated fluorophore and the bound fluorophore is detected by laser scanning. Each gene is represented by 15–20 different oligonucleotide probe pairs (16 are shown here); one member of each pair is a perfectly matched oligonucleotide probe, the other is a control oligonucleotide with a deliberate mismatch. The example shows expression data for three hypothetical genes, representing genes that are preferentially expressed in sample 1 (X), preferentially expressed in sample 2 (Y), or show equivalent expression in samples 1 and 2 (Z). (Adapted from Harrington CA et al. [2000] Curr Opin Microbiol 3:285–291; PMID 10851158. With permission from Elsevier.)
Expression analyses with microarrays that have short oligonucleotide probes are similar in principle to those where cDNA probes or long (>50 nucleotide) oligonucleotides are used. However, when using oligonucleotides that are only 25 nucleotides long, the hybridization specificity is not so great. There is a higher tendency for probes to hybridize to other sequences in addition to their expected target sequences, and so additional controls are needed. Accordingly, in the case of Affymetrix GeneChip arrays, each gene is represented by 20 or so different oligonucleotide probes that are selected from different regions along the transcribed sequence. In addition to 20 perfect match (PM) oligonucleotides per gene, a corresponding series of 20 mis match (MM) oligonucleotides is designed to control for nonspecific hybridization by changing a single base in each of the PM sequences (Figure 1B). To deter mine the signal for a particular gene, the signals of all 20 PM oligonucleotides are added together and the signals from all 20 MM oligonucleotides are subtracted from the total.
