A Bioinformatics Approach to Improving the Prediction of Programmed Ribosomal Frameshifting

1
A Bioinformatics Approach to Improving the
Prediction of Programmed Ribosomal Frameshifting
Yen Lin Huang Department of Computer Science,
National Tsing-Hua University
08/27/2007 InCOB 2007
2
Outline

• Introduction
• Our algorithm
• Experimental results and discussion
• Conclusion

3
Introduction

• Programmed ribosomal frameshifting ( PRF) is a
  recoding by which translating ribosome switches
  from initial (zero) reading frame to -1 or 1
  reading frame at a specific position and then
  continues its translation.

-1 PRF of SARS-CoV.
4
Introduction

• Consequently, the recoding of PRF leads to an
  expression of an alternative protein, which is
  different from that produced by standard
  translation.

5
Example -1 PRF of SARS-CoV

• Genomic organization of the SARS-CoV.
• If there is no frameshifting, polyprotein (pp) 1a
  is translated from ORF1a if there is a
  frameshifting, pp 1a/1b is translated from ORF1a
  and ORF1b.

6
Significance of PRF

• Many viruses, as well as bacteria, have been
  found to utilize the PRF mechanism for increasing
  the diversity of gene expression.
• This event can also be found in a few eukaryotes.
• It has been reported that for viruses, even small
  changes in their frameshifting efficiencies can
  inhibit viral propagation.
• This implies that frameshifting sites in viruses
  may present a potential target for antiviral
  therapeutics.

7
Signals of -1 PRF

• Two mRNA signals are critical for -1 PRF.
• Slippery sequence
• It is the place where -1 PRF event occurs.
• 3-stimulatory RNA structure
• It forces ribosome to pause over the slippery
  site such that the ribosome have a chance to
  switch from zero reading frame to -1 reading
  frame.

Stimulatory RNA structure
Slippery sequence
8
Model of -1 PRF
9
Slippery sequence

• Usually, the slippery sequence is a
  hepta-nucleotide (7-mer) of the general form X
  XXY YYZ.
• The spaces in X XXY YYZ separate codons in the
  zero frame.
• X and Z are any nucleotide and Y is mostly A or U.

10
Stimulatory RNA structures

• In most cases, the RNA structure is an H-type
  pseudoknot (or bulged helix), but in some cases,
  it is a simple stem-loop.

11
Other factors of -1 PRF

• The spacer between the slippery sequence and the
  RNA structure is also important for -1 PRF.
• The length of spacer alters the location of the
  paused ribosome and hence influences its shifting
  probability.
• For some bacteria (such as E. coli), an internal
  SD-like (Shine-Dalgarno-like) sequence often can
  be found upstream of the -1 PRF site.

12
Signals of 1 PRF

• It has been observed that the events of 1 PRF
  occur less than those of -1 PRF.
• Therefore, there is no general model that can be
  widely accepted to describe 1 PRF.
• The most known cellular genes with 1 PRF are
  prfB and oaz genes.
• prfB encode polypetide chain release factor 2
  (RF2) in E. Coli.
• oaz (ornithine decarboxylase antizyme ) encode
  antizyme 1 in mammals.

13
Signals of 1 PRF

• There is no general form of the slippery sequence
  for 1 PRF.
• EX the slippery sequences in prfB genes are CUU
  URA C and those in oaz genes are UUU UGA or YCC
  UGA, where R is A or G, and Y is C or U.
• Not all 1 PRF sites have a downstream RNA
  structure to function as the stimulator.
• EX the 1 PRF site in the bacterial prfB genes.

14
Proteins produced by PRF

• Protein products are longer than those by
  standard translation.
• This kind of PRF events are observed frequently.
• It occurs near the end of the zero reading frame.
  
• The ribosome switches to translate the new
  reading frame by extending beyond the terminator
  of the zero reading frame.

15
Proteins produced by PRF (cont.)

• Protein products are shorter than those by
  standard translation.
• Such a PRF is less observed currently.
• It takes place within the zero reading frame.
• The ribosome then slips backwards and terminates
  quickly, because it reaches a stop codon in the
  new reading frame near the slippery site.

16
Outline

• Introduction
• Our algorithm
• Experimental results and discussion
• Conclusion

17
Previous results

• Based on the model described above, several
  computational approaches have been proposed for
  prediction of PRFs.
• Pattern recognition (Hammell et al., 1999 Moon
  et al., 2004)
• Statistical analysis (Shah et al., 2002)
• Machine learning (Bekaert et al., 2003)
• Hidden Markov models (Bekaert et al., 2005 2006)

Hammell, A. B. et al. (1999) Genome Res., 9,
417427 Moon, S. et al. (2004) Nucleic Acids
Res., 32, 48844892. Shah, A. A. et al. (2002)
Bioinformatics, 18, 10461053. Bekaert, M. et
al., (2003) Bioinformatics, 19, 327335. Bekaert,
M. et al., (2005) Mol. Cell, 17, 6168. Bekaert M
et al. (2006), Bioinformatics, 22, 2463-2465.
18
Our approach and web server

• In this study, we improved the pattern
  recognition method of detecting PRF sites in a
  genomic sequence with using structural and
  functional bioinformatics.
• In addition, we have implemented this algorithm
  as a web server, called PRooF (Programmed
  Ribosomal Frameshifting), that is open to the
  public for online analysis.
• http//bioalgorithm.life.nctu.edu.tw/PROOF

19
Flowchart of our algorithm
20
Step 1 Identification of ORFs

• All ORFs above a threshold (whose default is 100
  nt) are identified from an input sequence.

21
Step 2 Detection of slippery sites

• For the PRFs with longer products
• Find all pairs of partially overlapping ORFs.
• Use the pattern recognition to detect all
  possible slippery sites in the overlapping
  regions.
• The slippery sequences conform to the default
  patterns or user-defined patterns.

22
Step 2 Detection of slippery sites

• For the PRFs with shorter products
• We simply searches each identified ORF for its
  possible slippery sites that possess the required
  slippery sequences.

23
Detection of slippery sites (cont.)

• If the input is a bacterial sequence, we further
  looks for an internal SD-like sequence upstream
  of each slippery site.

24
Step 3 Verifying protein function

• For all candidate ORFs, their translated protein
  sequences are further verified by InterProScan to
  see if they have the potential protein
  motifs/domains already registered in the InterPro
  database.
• InterPro is an integrated database of protein
  families, domains and functional sites.
• InterProScan is a tool the InterPro that combines
  various protein signature recognition methods for
  the detection of motifs/domains.

25
Step 3 Verifying protein function

• For the cases of longer product
• Each of two overlapping ORFs is translated into a
  protein sequence, which is then examined by
  InterProScan.
• For the cases of shorter product
• The full-length ORF is cut into two fragments at
  the slippery site.
• These two fragments are then translated into
  protein sequences and are further examined by
  InterProScan.

26
Step 4 Predicting RNA structure

• We use a heuristic approach we developed before
  (Huang et al., 2005) to detect the H-type
  pseudoknot for the sequence fragment downstream
  of the slippery site of each PRF candidate.

C.-H. Huang et al. (2005), A heuristic approach
for detecting RNA H-type pseudoknots,
Bioinformatics, Vol. 21, pp. 3501-3508.
27
Step 4 Predicting RNA structure

• If no stable H-type pseudoknot is found, we
  continue to use RNAMotif to search for all
  possible bulged helixes and choose the most
  stable one.
• RNAMotif is an RNA structural motif search tool
  that can find the fragments with the possibility
  of forming a given structure.

28
Step 4 Predicting RNA structure

• If neither a stable H-type pseudoknot nor a
  bulged helix is found, RNAMotif is used to search
  for simple stem-loops.

29
Outline

• Introduction
• Our algorithm
• Experimental results and discussion
• Conclusion

30
Our web server PRooF

• Based on the algorithm we described above, we
  have implemented a web server called PRooF for
  online analysis.
• http//bioalgorithm.life.nctu.edu.tw/PROOF
• Our PRooF was tested with a number of sequences
  with one or two known PRF sites from different
  species.
• The experimental results were compared with those
  obtained by FSFinder2, which was developed by
  Moon et al. based on pattern recognition.
• Moon, S. et al. (2004) Nucleic Acids Res., 32,
  48844892.
• Byun, Y. et al. (2006) LNCS, 3991, 284-291.
• Song, J.J. et al. (2007) Comput. Biol. Chem., 31,
  298-302.

31
Testing data sets

• The tested sequences in our experiments were
  taken from the PseudoBase and RECODE.
• PseudoBase collects RNA pseudoknots, some of
  which are thought to function as the PRF
  stimulators .
• http//biology.leidenuniv.nl/batenburg/PKB.html
• RECODE contains the translational recoding events
  of PRFs in various biological species.
• http//recode.genetics.utah.edu/
• All the tests of PRooF and FSFinder2 were run
  with default parameters, unless otherwise
  specified.

32
Testing data sets of -1 PRF
33
Testing data sets of 1 PRF
34
Sensitivity and specificity

• Sensitivity (Sen) 100 ? TP /(TPFN)
• TP number of correctly predicted PRF sites
• FN number of known PRF sites that were not
  predicted
• Specificity (Spe) 100 ? TN /(TNFP)
• TN number of predicted non-PRF sites that
  possess a required slippery sequence but are not
  annotated as PRF sites in the database
• FP number of incorrectly predicted PRF sites

35
Average sensitivity and specificity
Prediction of PRF Average Sensitivity Average Specificity
PRooF
FSFinder2

• Indeed, our PRooF greatly improves detection
  sensitivity, when compared with FSFinder2.
• For the details, please refer to our paper.

36
Reduction of false positives

• To reduce false positives, FSFinder2 considered
  only two pairs of the partially overlapping ORFs
  whose zero reading frames are the largest two in
  length.
• Moon et al. (2004) reported that these two pairs
  had the highest probability to contain -1 and 1
  PRF sites.
• However, currently there seems to be no
  biological evidence to support their observation.

37
Reduction of false positives

• On the contrary, we utilized InterProScan to
  screen out the partially overlapping ORFs whose
  protein sequences contain no functional
  motifs/domains.
• As shown in our experiments, this approach of
  functional bioinformatics is very useful to
  reduce the number of false positives.

38
Predicted RNA structures

• Most of the RNA structures predicted by PRooF are
  H-type pseudoknots and bulged helixes, but many
  RNA structures identified by FSFinder2 are just
  simple stem-loops.
• Both H-type pseudoknots and bulged helixes are
  believed to be more constructive to promote the
  efficiency of -1 PRFs and some 1 PRFs.
• The reason is that they have a similar structure
  of bend conformation and are structurally more
  stable than simple stem-loops.

39
Predicted RNA structures

• The -1 PRF of HIV-1 was first thought to be a
  simple stem-loop, but it was then proved
  experimentally to be a bulged helix.
• Gaudin C. et al. (2005) J Mol Biol, 349,
  1024-1035
• The RNA structure predicted by PRooF for the -1
  PRF of HIV-1 is indeed a bulged helix, but the
  one predicted by FSFinder2 is just a simple
  stem-loop.
• It should be worthwhile to determine
  experimentally the RNA structures in other
  similar cases, where the structures predicted by
  PRooF are H-type pseudoknots or bulged-helixes,
  but just simple stem-loops by FSFinder2 or
  reported in the literature.

40
Outline

• Introduction
• Our algorithm
• Experimental results and Discussion
• Conclusion

41
Conclusion

• We improved the pattern recognition approach to
  automatically detecting PRF sites in a given
  genomic sequence with using both structural
  bioinformatics and functional bioinformatics.
• Based on this approach, we have developed a web
  server PRooF that is open to the public for
  online analysis.
• http//bioalgorithm.life.nctu.edu.tw/PROOF

42
Conclusion (cont.)

• In our experiments, the testing results showed
  that PRooF greatly improves sensitivity, when
  compared with FSFinder2.
• Most of the RNA structures predicted by PRooF are
  H-type pseudoknots and bulged helixes, whereas
  those predicted by FSFinder2 are simple
  stem-loops.
• PRooF was implemented in a flexible way that it
  allows the user to modify all the default
  parameters such that some exceptional PRF sites
  can still be detected.

43
Acknowledgement

• Prof. Chin Lung Lu
• Institute of Bioinformatics Department of
  Biological Science and Technology, National Chiao
  Tung University
• Mr. Chia-Jung Wu
• Institute of Bioinformatics, National Chiao Tung
  University
• Prof. Hien-Tai Chiu
• Department of Biological Science and Technology,
  National Chiao Tung University
• Prof. Chuan Yi Tang
• Department of Computer Science, National
  Tsing-Hua University

44

• Thank you for your attention