Experimental design depends on the purpose of an experiment. Generally, if the purpose is to find genes differentially expressed between two strains (eg mutant vs wild type) the design is more straightforward than if one wants to study expression profiles over time or condition. However if a mutation has pleiotropic effects - such as effecting growth rate, differentiation, etc - many genes that are differentially expressed will be due to "secondary" effects of the mutation. By comparing global expression data from numerous experiments can general/pleiotropic effects be effectively subtracted from a particular data set revealing effects specific to a particular stimulus. For example, data from numerous yeast experiments have defined a set of "shock" genes whose changes in expression are common to many stimuli. For such multi-experiment comparisons to be useful it is absolutely essential that the experiments are described adequately. We strongly recommend varying more than one parameter such as strain, condition or time relevant to the factors that are particular to your experiment.
cDNA vs cDNA hybridisations
Labelled cDNAs can be hybridised against another cDNA or against a common reference. Typical cDNA vs cDNA hybridisations are mutant vs wild type experiments. The ratios obtained from the array will give gene expression of one strain relative to the other. However, if more than two samples are to be compared a loop design is required in order to compare gene expression across all strains.
Loop design experiment. A simple loop where experimental samples are labelled once with Cy3 and once with Cy5 with the samples pairings forming a logical loop. Addition of new samples requires part of the loop to be replaced. The design feature are used during normalisation but poor arrays may "break" the loop. More complex loop structure can be more robust.
Another design strategy for this type of comparison is the balanced block design.
Using a common reference
A common reference can be used for an experiment with many samples. A specific RNA sample (eg T0 in a time course), a pool of all the experimental RNA samples or genomic DNA can be used. We recommend using genomic DNA from M145 for the common reference as this allows comparison between ALL experiments with this design. The ratios obtained from an array will be the gene expression relative to the common reference. For cDNA vs gDNA this is effectively expression per copy number of that gene. Plotting the ratios from two (or more) arrays gives the change in gene expression between the samples.
Experiment using a common reference. The common reference is always labelled with Cy5 and the experimental samples with Cy3. Additional samples can easily be added to the experiment.
It has been reported (for a different organism) that for some genes the Cy-dyes are differentially incorporated. In other words one Cy-dye gives more labels per average probe than the other Cy-dye resulting of a ratio significantly different from 1. When hybridising cDNA vs cDNA or genomic vs genomic repeat hybridisations with reciprocal labels is recommended. For cDNA vs genomic differential incorporation is unimportant because the same Cy-dye is used for each cDNA.
Replicate hybridisations are either "technical", where the same sample pair is repeated, or "biological", where the experiment is repeated and an equivalent sample used. We recommend doing at least four biological replicates as this allows statistical analysis on the results for each gene (such as t-tests). Since biological replication incorporates variation between arrays as well as between isolates we do not consider technical replication necessary.
Data from replicate hybridisations should be examined individually to reveal variations between replicates and highly variable data can be excluded. Generally further analysis uses the average of the replicates.
Bacterial mRNA is susceptible to rapid degradation by ribonucleases. Furthermore, the processing of the cells from collection to inactivation of ribonucleases and cell disruption may alter the original transcription pattern in the culture (e.g. by activating stress-response genes when cells are subjected to cooling shock and centrifugal forces during centrifugation at 4oC). In gene expression profiling experiments, particularly in a global scale, the full and precise representation of the transcription activity in the RNA preparation is a critical factor. It is then important to ensure that cells are harvested and killed as fast as possible in reagents that inactivate ribonucleases, such as RNAprotect Bacterial Reagent (Qiagen).
Unlike the use of mRNA in eukaryotic systems, microarray gene expression analyses in bacteria routinely include the use of total RNA in cDNA synthesis and labelling, although a method for enriching mRNA in bacteria has been reported (Su and Sordillo, 1998). For successful target synthesis a key issue is the purity and integrity of the isolated total RNA as well as the complete exclusion of any genomic DNA traces from the RNA preparation. Impurities are also thought to produce high background on glass microarrays. For this, adequate extraction and washing steps and treatment with DNase I are necessary. Prior to any use, RNA must be checked by both agarose gel electrophoresis and spectrophotometry.
The protocols cited below have given highly similar results in terms of RNA quality, yield and size distribution of synthesized Cy-cDNA, and signal intensities on microarrays. The methods were compared in parallel using the same culture to isolate RNA. However the RNeasy Protect Bacteria kit (Qiagen) is preferred because is faster, simpler and provides a standardized method for every Streptomyces lab.
Total RNA isolation with RNeasy Protect Bacteria Kit (Last updated 16/08/2003)
Total RNA isolation by Total Nucleic Acid precipitation and DNase digestion (Last updated 16/08/2003)