Array-based approaches to bacterial transcriptome analysis - Methods in Microbiology - page 169

825 001 850 000

Genbank
annotation

yfmD yfmC
yfmB

yflS yflP nosA

yfl yflK yflH

mapB

yfkT yfkQ

yflT pel

citS citM nagP yflB

yflA

citT yflN yflE yfkR

Transcriptional
landscape (+)

Annotated genes
and new transcription
segments (+)

S257 S259 S263 S265 S269 S272

S258

Upshifts

( + )
( − )

Downshifts

and predicted TUs

Annotated genes
and new transcription
segments ( − )

S268 S271

Transcriptional
landscape ( − )

FIGURE 6.3
Bacillus subtilis transcriptional landscape determined by tiling arrays. A sample segment of
25 kbp of the chromosome is shown. The Genbank annotation is displayed above the
expression data collected in 269 hybridizations accounting for 104 different biological
conditions selected to maximize the diversity of lifestyles ( Nicolas et al., 2012 ). For both
strands of the chromosome, 50 transcriptional profiles (coloured curves) illustrative of the
diversity observed in the total data set are drawn along with the new transcription segments
delimited (coloured boxes). The positions of signal upshifts and downshifts, also represented,
were shown to correspond mostly to transcription start sites and terminators and the
associated transcriptional units (TUs) were predicted.

transcription signal (in log2-scale) associated with a 95% credibility intervals that inte-
grate out probe affinity effects as measured fromgenomic DNA hybridization ( Nicolas
et al. ,2009 ). The lower bound of credibility interval was then compared to a threshold
corresponding to 10 the chromosome median, used here as a proxy for the back-

ground. Of note, reverse transcription artefacts (limited by addition of actinomycin
D) cause well above background expression signals on the antisense strand that cannot
be identified by the aforementioned approaches. The authors found that visual inspec-
tion of transcriptional landscapes and quantitative measures of the reproducibility of
the signal could serve to filter out these artefacts.
Differential expression assessment plays a central role inmicroarray data analysis.
This is particularly true for experimental set-ups comparing a limited number of
biological conditions such that the research question can be easily addressed by estab-
lishing lists of differentially expressed genes. Hypothesis testing in linear models
represents the framework of choice for this purpose with well-established statistical
procedures (e.g. analysis of variance, ANOVA) that extend the popular Student
t -test for comparing the mean of two samples, see, for instance, Smyth (2005) .
After log-transformation or more general variance stabilizing transformation
( Huber et al. , 2002 ), expression levels are written as a sum of terms representing