Microarrays are slides with spots of DNA deposited on them, and they can be used for a variety of purposes. There are two common types of microarrays, cDNA microarrays and oligo microarrays. The spots of cDNA microarrays hold sequences that are longer in length and are typically used to compare two samples. A control sample is labeled with one dye (e.g. Cy3) and a sample of interest is labeled with another dye (e.g. Cy5) and the ratio of the intensity of the sample emissions is used to determine the relative amount of the sample of interest (Cummings & Relman). Oligo microarrays use a set of spots that are around 20 bp in length to measure the expression for a single gene. Often each spot on an oligo microarray is paired with a "mismatch oligomer", a spot that contains a DNA sequence that differs by one base from the original sequence. In contrast to cDNA microarrays, oligo microarrays are used to measure one sample rather than compare two samples. The hybridization is measured by taking the difference in intensity between the perfect match and the mismatch (Wu). Although they are commonly used for gene expression profiling, microarrays can also be used for phylogenetic classification.
Target preparation and hybridization are only the first steps in a microarray experiment. After hybridization a digital image of the array must be made. Two common kinds of scanners are laser scanners and fluorescent microscope scanners. Fluorescent microscope scanners use a CCD camera to record the image of the microarray, and they tend to be cheaper than laser systems. However, laser systems are less influenced by background noise (Snijders, et. al). Once the image is captured the raw data must be transformed into an expression matrix, which requires identifying the spots, determining their boundaries, comparing their intensity to background noise, and normalizing the data (Brazma). Given the magnitude of the experiments and the quantitative nature of the data, analysis is not a simple task. Most often clustering methods are used, which requires the ability to measure the distance between either rows or columns in the expression matrix. Bioinformatics methods are very important because false predictions can happen at any point in the experiment, from microarray fabrication to data analysis.
There are several different strategies that have been used for bacterial identification with microarrays. One approach is to use a microarray to identify the presence of certain genes. The researchers examined genes that had previously been used in multiplex PCR assays, aligned available copies of the genes, and printed conserved regions of the gene on a microarray. In order to detect the genetic characteristics of interest the multiplex PCR assay was performed followed by hybridization to the array. The microarray technique has a much better specificity than multiplex PCR alone, presumably because nonspecific bands did not hybridize to the array (Chizhikov, et al.). While the technique was effective, it is not an approach that is generally applicable. An array similar to the one described above can not be fabricated unless there is already an existing PCR assay for the bacteria.
Troesch, et al. used a method for bacterial classification that interrogates the differences between certain genetic regions that contain important genetic information. They chose to spot the 16S rRNA because it contains species-specific polymorphisms and the RNA polymerase beta-subunit gene because missense mutations in the rpoB gene are associated with rifampin resistance in M. tuberculosis. The researchers were able to correctly discriminate between twenty-seven species and identify point mutations in the rpoB gene.
Another approach to bacterial typing with microarrays is to spot as many open reading frames (ORFs) as possible on the array. Behr, et al. spotted 99.4% of the M. tuberculosis ORFs on a microarray and used the experiment to identify genes from BCG, a related organism, that were not present in M. tuberculosis. Although the investigators used the array for genomic comparisons, with a few modifications the array could have been used for bacterial typing. Cho & Tiedje employed an interesting variation on the whole-genome approach to bacterial typing. Instead of spotting ORFs on a microarray, they randomly fragmented the genome by bead beating, extracted the DNA by gel purification, and spotted the DNA on a microarray. The target samples were prepared by random priming of the single Pseudomonas strain of interest and labeled with Cy3-dCTP. As a reference sample, a 1:1:1:1 mixture of the four Pseudomonas strains were prepared by random priming and labeled with Cy5-dCTP. The researchers used a UPGMA clustering algorithm with similarity coefficients as the distance metric to resolve the species that were tested. All four species were differentiated at a similarity coefficient values of 92-94%. The method outlined above is impressive because no sequence data was necessary in order to design a chip capable of differentiating the closely related bacterial species.