An Investigation into Selection Constraints in RNA Genes

1
An Investigation into Selection Constraints in
RNA Genes Naila Mimouni, Rune Lyngsoe and Jotun
Hein Department of Statistics, Oxford University
Aim A robust approach to investigate selection
patterns acting on ncRNAs. Motivation
Selection in RNA No RNA equivalent of
Ka/Ks. Comparing paired vs. unpaired bases
inconclusive. No trend valid for most
ncRNA families has been identified. RNA Gene
Prediction Improve identification of
ncRNAs reduce false positives.
Identification of RNA function, active sites, and
pseudogenes.
Results 1- Selection Patterns on the stem
classes We validate the hypothesis on
different ncRNA families. Pattern observed in
snoRNA, snoRNAtRNA. Not observed in miRNA
class III, because the conservation of the mature
miRNA drives class III conservation
upwards. 2- Significance
of the classification Does our classification
model perform better than no classification? Chi-s
quare test of nested models Real Data the
differences in inferred rates observed in our
ncRNA families are statistically significant with
a p-value of 0.01. Simulated Data we simulated
100 datasets according to the no-classification
model and with random classification. Fig. 2
Plot of differences in log- likelihoods under the
traditional no- classification model and our
classification. For comparison, the
corresponding percentiles of the chi- square
distribution with 58 degrees of freedom and the
observed difference for the ncRNA1 snoRNA1
dataset are also plotted. In 90 of the cases,
the difference of log likelihoods is less than
85.95, which is that at a p-value of 0.01. All
of the simulated data log likelihood differences
are smaller than that of the data given our
structural classification at 194.64. This
indicates that our classification is a better
fit than no classification. 4- Verification of
evolutionary origin of observed selection
pattern When applying our analysis to shuffled
alignments, the observed selection pattern
disappeared. For e.g. shuffled snoRNA1 alignments
we observed mutation rates of 0.59, 0.86 and 0.59
for class I, II and III respectively. The
observed evolutionary pattern remained when
consensus structure was determined using the
structure-first strategy of RNACast RNAforester
rather than the alignment-first strategy of
Pfold, see data sets marked with in the above
table.
Class I Class II Class III
snoRNA1 snoRNA1tRNA1 miRNA1 ncRNA1 ncRNA1snoRNA1 snoRNA2 miRNA2 ncRNA2 1.31 0.73 0.78 0.32 0.73 0.77 0.49 0.57 1.17 0.68 0.71 0.25 0.58 0.50 0.42 0.47 1.24 0.72 0.60 0.22 0.68 0.67 0.31 0.33
SnoRNA1 miRNA1 ncRNA1 ncRNA1snoRNA1 1.07 0.91 0.67 0.91 0.89 0.72 0.44 0.74 0.98 0.37 0.64 0.87
Method Idea Exploit conservation of RNA
secondary structure for homologous
sequences. By analogy with the contrasting
properties of different codon positions in
protein coding genes, we use the nearest
neighbour energy model for RNA structure to
infer the effects of disrupting base pairs at
different positions in a stem. We divide
stem positions into structural classes. We then
validate the class hypothesis for different ncRNA
families and datasets, and establish that
stem positions are under different selective
constraints. Approach A) Folding homologous
RNA sequences 1- Pfold mutational model,
SCFGs. 2- RNACast RNAforester Abstract
shapes, structure alignment. B) Classification
of base pairs Fig. 1 A stem of seven
contiguous base pairs annotated with the classes
as defined by thermodynamic considerations. It is
assumed that base pairs i,j, and i 6,j- 6
are adjoining loops. Based on the Turner
energy parameters, we can quantify the energy
cost of disrupting a bp depending on its
proximity to a loop. We also consider the
structural effects. Under the assumption that
only stacking base pairs are stable, we observe
that disrupting a class II base pair will also
disrupt adjoining terminal base pairs leading to
a larger structural effect than disrupting
class I and class III base pairs. C)
Evolutionary Analysis using Phase Use Phase
software to analyse the homologous sequences,
with conserved structure, annotated with
classes. Compare the average number of
substitutions in each class. Data Two
datasets Rfam RNA sequence alignments of human,
mouse, rat and chicken. 1st dataset
Alignments 112 snoRNA, 95 miRNA, 140 NcRNAs.
2nd dataset Alignments 54 snoRNA, 79 miRNA
alignments, 508 ncRNAs.
Terminal (Class I) Penultimate (Class II) Other (Class III)
Stability Structure Small Small Large Large Large Small
Overall Small Large Medium
Conclusion Most ncRNAs follow a selection
pattern whereby the penultimate base pair is
more conserved that the ultimate base pair. For
ncRNA families, class labelling is a better model
of homologous ncRNA evolution than treating all
stem base pairs the same.
1 http//www.bioinfo.rpi.edu/zukerm/cgi-bin/efil
es-3.0.cgi 2 Mathews, D. H. Sabina, J.
Zuker, M. Turner, D. H. 1999. J. Mol. Biol.,
288, 911-40. 3http//rana.lbl.gov/drosophila/wiki
/index.php/Noncoding_RNA 4 Knudsen, B. Hein,
J. 2003. Nucleic Acids Research, 31, 3423-8. 5
Reeder, J. Giegerich, R. 2005. Bioinformatics,
21, 3516-23. 6 Höchsmann, M. Voss, B. and
Giegerich,R. 2004. IEEE/ACM Transactions on
Computational Biology and Bioinformatics (TCBB),
1, 5362. 7 http//www.cs.man.ac.uk/gowridhv/
I thank the 6th Framework program of the European
Union for awarding me a travel fellowship.