Biology Reference
In-Depth Information
to the average across all samples. In two-colour experiments, the hybridization sig-
nals of paired samples are compared thereby controlling for potential variation intro-
duced during sample labelling and array processing. Thus, the advantage of this
approach over one-colour experiments is its robustness in the face of day-to-day vari-
ations. It is, however, crucial that the reference sample is available in sufficient
amounts and consistent quality throughout the course of the entire study.
The availability of data on biological replicates, consisting of independent culti-
vations of the same genetic background in the same growth conditions, is important
for most applications in microbial transcriptomics, and crucial if the goal is to iden-
tify differential expression between different culture conditions and genetic back-
grounds. More replicates provide higher statistical power but directly increase the
cost of the experiment, thereby limiting the number of conditions/strains that can
be explored. More precisely, as discussed in
Bolstad
et al.
(2004)
, the statistical
power to detect differential expression depends on the level of variance between rep-
licates and the proportion and fold-change amplitude of the differentially expressed
genes. Therefore, the choice of the number of replicates may seem important but the
information needed for making a really wise choice is, in general, not available
beforehand. In practice, if the experimental set-up achieves low variance between
replicates, a design involving three biological replicates will already offer sufficient
statistical power to detect biologically relevant changes (i.e. either with important
fold changes or involving a large number of genes) and can therefore be considered
as a reasonable starting point. Of note, biological replicates account for both the bio-
logical and technical sources of variation. In contrast, technical replicates would be
helpful only to delineate the respective share of the biological and technical vari-
ances and are usually not included in the experimental design.
Another important issue that needs to be considered is between-sample normal-
ization. Global methods such as quantile normalization or median centring are
employed in the majority of transcriptome studies and will be elaborated in
Section 4.1
. However, one should be aware that such approaches rely on the assump-
tion that the distribution of expression levels is, at least to some extent, preserved
under the conditions to be compared. This will be true if the expression level of most
genes is unchanged and if the pattern of up- and down-regulated genes is roughly
symmetric. In the case of global changes in gene expression, these normalization
methods (based on endogenous transcript levels) will not accurately assess differen-
tial expression. Indeed, if global changes between samples are expected, an alterna-
tive is external normalization based on a strategy of “spike-in” controls added in
equal amounts to each total RNA sample before cDNA synthesis (
van de Peppel
et al.
, 2003
). This needs to be considered during the design or selection of the array
to ensure that it contains the appropriate control probes (
Thomassen
et al.
, 2009
).
For the Agilent platform, spike-in controls are available as mixtures of 10 different
in vitro
synthesized transcripts that hybridize to complementary control probes on
the arrays. By adding spike-in controls to each total RNA sample, the average dif-
ference of the control signal intensities between different arrays can be used to scale
the intensities of each individual array (
Chandriani and Ganem, 2007
). However,
