Methods
RNA labeling
Previously, RNA half-lives were determined using a
technique that blocks transcription and monitors the decay of RNA over time
(Bernstein et al., 2002; Raghavan et al., 2002;
Yang et al., 2003; Narsai et al, 2007).
However, this method does
not lead to a high accuracy, since it has several drawbacks. While the blocking of transcription here
is assumed not to affect transcript half-lives, it has been shown that individual transcripts can be
stabilized as a consequence of a stress response such as transcriptional arrest
(Gorospe et al., 1998; Blattner et al., 2000). Also it has been shown
that medium-to-long-lived half-lives can not be determined precisely with this method
(Friedel et al, 2009).
In order to avoid these disadvantages and to improve the accuracy of half-life calculation a new method
has been developed, which is based on labeling of newly synthesized RNA with 4-thiouridine (4sU)
(Kenzelmann et al., 2007;
Dölken et al., 2008). The
4sU can be introduced into the transcription process over the nucleoside salvage pathway and thus gets
directly incorporated into newly transcribed RNA.
This labeling makes it possible to distinguish between pre-existing RNA (pre-existing RNA) and newly
transcribed RNA (labeled RNA) without disrupting the transcription process.
The separation of these two groups in a thiol-mediated way and subsequent quantification with microarrays
or mRNA-seq then serves as basis of half-life calculation, which can be performed based on the ratios of
pre-existing to total RNA, newly transcribed to total RNA or newly transcribed to pre-existing RNA with high accuracy. It has been shown
that transcript half-lives can help us to understand the regulation of biological processes on
a general level without being limited to specific conditions and, thus, maybe missing important
information (Yang et al, 2003;
Narsai et al, 2007; Friedel et al, 2009)
Filtering your data
HALO is a software framework that provides methods for the calculation of half-lives from such microarray
or mRNA-seq data, but also allows you to improve data quality through the introducing of filtering steps.
You can thus exclude unreliable probesets based on present/absent calls, low expression values or the
probeset quality score. This last method can be performed only after normalization, since it is based on
the distance of the RNA quantification values to the regression line. The higher the PQS, the lower the
distance to the linear regression line and thus the better the quality.
Assessing quality of data labeling
HALO also provides the possibility to assess the quality of data labeling and RNA quantification.
The insufficient labeling of short transcripts with 4sU, caused by a low content of uracil, could
induce reduced capture rates for these transcripts. As a consequence, a bias in calculating the
half-lives for these emerges. You can test for such a bias based on the number of uracils in the
RNA sequence compared to the ratio of transcript to total RNA. If a bias is present, a correlation
between these two values should be observed. If necessary such a bias can be corrected with the method of
Miller et al. (2009) implemented in HALO.
Normalization
Normalization of transcript quantification data is another possibility to control the quality of the
data. Based on the relationship between the three groups of RNA, pre-existing, newly transcribed and total RNA, a
simple linear regression can be performed for normalization
(Dölken et al., 2008). Since pre-existing and newly transcribed transcript
levels added up should equal total RNA levels, a negative linear correlation should be observed between
the ratios of newly synthesized/total and pre-existing/total RNA.
If only two of the three RNA fractions have been measured, data can be normalized based on the median
half-life.
Half-life calculation
RNA half-lives can be calculated with HALO using an exponential decay model either for individual
replicates or ratios averaged across all replicates. Three possible methods exist for the calculation
of half-lives, based on the newly transcribed/total RNA, pre-existing/total RNA and the newly transcribed/pre-existing RNA ratios,
respectively. Among these, the method based on newly transcribed (=labeled) to total RNA ratio has an overall high
accuracy, while the method using pre-existing/total RNA shows high precision for short transcripts, while
being unreliable for medium-to-long-lived transcripts (Friedel
and Döken, 2008). The ratio of newly transcribed to pre-existing RNA as a method
combines the good properties of the two other calculation methods.