1. A biological crash course and introduction to prediction methods
RNA-RNA interaction
A biological crash course and introduction to prediction methods

2. Part I – Biological crash course
Bacteria
Plasmid copy control
Post-segregational killing systems
trans-encoded chromosomal RNAs
RNA interference (gene silencing)
Translation regulation
C. elegans developmental regulation
miRNA-miRNA interactions
Human telomerase

3. DNA vs. RNA Bases #Strands Structure DNA A,C,G,T 2 Double helix RNA
A,C,G,U
1 or 2
Stem-loop, pseudoknots, etc.

4. Gene expression
Central dogma of molecular biology

5. Translation mRNA -> protein via triplet code
What happens if mRNA is destroyed or otherwise can’t be translated?

6. Bacteria backgrounder
Single-celled organisms
Prokaryotes = no nucleus
Multi-cistronic transcripts -> multiple genes transcribed at one time, often with overlapping reading frames

7. Bacterial genetic information
Bacterial chromosome (1)
Genome of organism
Required for life
Plasmids (2)
Circular DNA molecules
Double-stranded
Independently self-replicating
Not required for life, often confer selective advantage such as antibiotic resistance

8. Plasmid replication (1),(2) – Genes encoded on plasmid
(3) – Origin of Replication (ORI)

9. Plasmid copy control Recall independent self-replication
Copy number fluctuations are unavoidable
Too many -> “runaway”, host dies
Too few -> increased risk of plasmid loss
Problem: How to control copy count?
Solution: negative feedback loop mediated by RNA-RNA interaction

10. R1 copy control Genes: oriR1 – origin of replication
repA – lots of this protein product is required for replication initiation
tap – translation of protein product is required for translation of repA protein
copA – product is antisense RNA
copB – product is a repressor protein (not covered here)
Draw it on the blackboard as we go

11. R1 copy control (2) copA – RNA with stem-loop structure
copT – target segment of repA/tap mRNA, also forms a stem-loop structure
Single loop-loop interaction

12. R1 copy control (3)

13. R1 copy control (4) copA RNA is unstable; it degrades
If not enough plasmids are producing copA antisense RNA (copy number is too low), more repA protein can be produced
Therefore the plasmid can replicate

14. Post-segregational killing systems
Plasmid self-preservation mechanism
Bacterial host losing plasmid results in host death
R1 plasmid hok/sok system is the prototype
All such systems work similarly

15. R1 hok/sok system hok/sok locus encodes:
hok protein – “host killing”
Overlapping reading frame – mok – “modulator of killing”
sok RNA – “suppressor of killer”
mok must be translated for hok to be expressed
mok cannot be translated if sok is present

16. R1 hok/sok system (2) hok mRNA is extremely compact
Many stem-loop structures
Flush 5’ – 3’ pairing
Highly stable -> long half-life
Translationally inert
mok segment is both:
Translationally active
Able to bind sok inhibitor RNA

17. R1 hok/sok system (3) sok RNA is highly unstable
Bacteria with R1 have lots of sok produced
sok binds mok, hok is not translated
Bacteria which lose R1 have:
Lots of stable hok mRNA
Quickly degrading sok RNA (low stability)
No new sok RNA being produced
hok is translated -> bacteria dies

18. Bacterial chromosomes
Plasmid antisense RNAs are generally cis-encoded
Implies complete Watson-Crick complementarity
Bacterial chromosomes contain trans-encoded antisense RNAs
Not necessarily complete complementarity
Often stress-related control systems
draw cis and trans examples on board here: worth at least a couple of minutes
Note that cis-acting = cis-encoded = cis-type, same for trans-

19. oxyS/fhlA in E. coli oxyS – RNA transcript induced by stress
fhlA – transcriptional activator site
oxyS/fhlA complex binds via two loop-loop interactions

20. RNA interference (RNAi)
a.k.a. post-transcriptional gene silencing
Double-stranded RNAs are introduced into the cell
Complementary to mRNA for a gene
Directly introduced in a wet lab, or
Produced by the cell itself

21. RNA interference (2)
dsRNAs are cleaved into nt segments (“small interfering RNAs”, or siRNAs) by an enzyme called Dicer

22. RNA interference (3)
siRNAs are incorporated into RNA-induced silencing complex (RISC)

23. RNA interference (4)
Guided by base complementarity of the siRNA, the RISC targets mRNA for degradation

24. RNA interference – why? Studying gene function Therapeutic suppression
Knock out or inhibit a gene’s normal function
Can the organism survive?
What phenotypic changes are observed?
Therapeutic suppression
E.g. cancer treatment

25. micro RNA (miRNA) Gene expression regulation
Created by similar process to siRNA
Generally prevents binding of ribosome

26. Ex: C. elegans development
lin-4 and let-7 antisense RNAs
Regulate larval development in C. elegans
One of the two binding sites for lin-41 and let-7 interaction:

27. Human telomerase Telomerase = ribonucleoprotein complex
Ribo = ribosomal/RNA association
Nucleo = nuclear localization
Protein = contains a protein
Responsible for maintaining telomere length in eukaryotic chromosomes
Main components:
Telomerase reverse transcriptase
Human telomerase RNA (hTR)

28. Human telomerase (2) Reverse transcriptase
Transcribes RNA to DNA (rather than the usual DNA to RNA)
Telomeres – repeated regions at the end of eukaryotic chromosomes
hTR is the template for the repeated region

29. Human telomerase (3) hTR 11-nt templating region consists of:
Repeat template: CUAACCC
Alignment domain: UAAC
Positions telomerase on the DNA strand
Provides template for repeat region

30. Human telomerase (4)
A – secondary structure of hTR w/ proposed long-range interactions shown via connecting lines
B – new model for secondary structure of hTR
Catalytic domains: pseudoknot, CR4/CR5, template

31. Loop-loop interaction
Sometimes referred to as “kissing loops”
Recall that all of the RNA-RNA interaction discussed so far (excepting RNAi), involve loop-loop interaction
Predicting miRNA transcripts and targets involves loop structure prediction

32. References
Couzin, J. (2002) “Breakthrough of the year – Small RNAs make big splash.” Science 298(5602):
Lai, E.C., Wiel, C., and Rubin, G.M. (2004) “Complementary miRNA pairs suggest a regulatory role for miRNA:miRNA duplexes.” RNA 10(2):
Moss, E.G. (2001) “RNA interference – It’s a small RNA world.” Current Biology 11(19):R
Sharp, P.A. (2001) “RNA interference – 2001.” Genes and Development 15(5):
Shi, Y. (2003) “Mammalian RNAi for the masses.” TRENDS in Genetics 19(1):9-12.

33. References (2)
Ueda, C.T., and Roberts, R.W. (2004) “Analysis of a long-range interaction between conserved domains of human telomerase RNA.” RNA 10(1):
Wagner, E.G.H. and Flärdh, K. (2002) “Antisense RNAs everywhere?” TRENDS in Genetics 18(5):
Wagner, E.G.H., Altuvia, S., and Romby, P. (2002) “Antisense RNAs in bacteria and their genetic elements.” Advances in Genetics 45:

34. Part II – Prediction Identifying effective siRNAs Identifying targets
Neural network approach
Identifying targets
Mammalian miRNA target prediction

35. Prediction of siRNAs
Sequence properties that make a good antisense RNA an effective gene inhibitor are not well understood
Most computational models consider only:
RNA structure prediction
Motif searches

36. Neural net approach Training set: 490 known siRNA molecules
Input parameters:
Base composition
mRNA:siRNA binding energy properties
3’ and 5’ binding energy
Structure of siRNA (hairpin energy and quality)
Target function: efficacy

37. Neural net approach (2)

38. Neural net results 14 inputs, 11 hidden units, 1 output
Success rate of 92%
Average prediction of 12 effective siRNAs per 1000 base pairs
Stringent (high specificity)
Good for designing siRNAs for RNAi

39. Prediction of miRNA targets
Mammals/vertebrates
Lots of known miRNAs
Mostly unknown target genes
Initial method outline
Look at conserved miRNAs
Look for conserved target sites

40. micro RNAs in animals 0.5-1.0% of predicted genes encode miRNA
One of the more abundant regulatory classes
Tissue-specific or developmental stage-specific expression
High evolutionary conservation

41. micro RNAs in plants Finding targets in plants is relatively easy
Look for mRNA transcripts with near-perfect complementarity to known miRNAs
Signal-to-noise ratio exceeds 10:1 for Arabidopsis (model plant organism)
Naïve approach in C. elegans and D. melanogaster? No more hits than expected by random chance!

42. So what can we use?
Pairing to nucleotides 2-8 at the 5’ end of the miRNA
Target recognition
Target regions enriched for genes involved in transcriptional regulation

43. Goals for algorithm Predict 100s of miRNA targets
Estimate false-positive rates
Provide computational and experimental evidence of authenticity
Identify common functionality classes other than transcriptional regulator genes

44. TargetScan Algorithm developed by Lewis et al 2003 Input: Output:
miRNA that is known to be conserved across multiple organisms
Orthologous 3’ UTR sequences
Cut-off values for two parameters
Value for one free parameter
Output:
Ranked list of candidate target genes

45. TargetScan (1) Search UTRs in one organism
Bases 2-8 from miRNA = “miRNA seed”
Perfect Watson-Crick complementarity
No wobble pairs (G-U)
7nt matches = “seed matches”
Choice of first organism is arbitrary

46. TargetScan (2) Extend seed matches Allow G-U (wobble) pairs
Both directions
Stop at mismatches

47. TargetScan (3) Optimize basepairing Remaining 3’ region of miRNA
35 bases of UTR 5’ to each seed match
RNAfold program (Hofacker et al 1994)

48. TargetScan (4)
Folding free energy (G) assigned to each putative miRNA:target interaction
Ignores initiation free energy
RNAeval (Hofacker et al 1994)
Can someone tell me what initiation free energy is?

49. TargetScan (5) Z score for each UTR (no match -> Z=1.0)
Term Z-score is misleading – it’s not really a statistic
This is one of the cut-off parameters that can be set
n = number of seed matches in UTR (may be more than one)
Gk = free energy of miRNA:target site interaction of kth seed match
T = parameter influencing relative weighting of UTRs with few high affinity target sites against UTRs with lots of low affinity target sites (experimentally determined)

50. TargetScan (6) Order UTRs by Z score Assign rank to each UTR
Repeat this process for each of the other organisms with UTR datasets
Z sub C and R sub C are cut-off values – experimentally determined

51. TargetScan (7)
UTR i is a predicted target if for all organisms:

52. Datasets nrMamm (mammalian – 79 sequences)
Homologs in human, mouse, and pufferfish
Identical between human and mouse, not necessarily pufferfish (fugu)
nrVert (vertebrate – 55 sequences)
Identical between human, mouse, and fugu
Non-redundant: if multiple miRNAs had the same seed, one representative chosen
Major criticism of this method – how to choose that one representative??
Simplified signal-to-noise calculation

53. Sample program flow

54. Results for nrMamm
nrMamm searched against human, mouse, and rat orthologous 3’ UTRs
451 miRNA:target interactions predicted for 400 unique genes
Average 5.7 targets per miRNA
Signal:noise ratio of 3.2:1
Very high number of targets per miRNA

55. Results for nrVert Additional search against fugu UTRs
Signal:noise ratio improves to 4.6:1
Relaxed cut-off values
115 predicted miRNA:target interactions for 107 unique genes
2.1 putative targets per miRNA

56. Signal:noise ratio calculation
Signal = number of predicted targets from nrMamm dataset
Noise = number of predicted targets from randomly shuffled miRNAs
Shuffled control sequences screened to ensure preservation of relevant features – don’t underestimate the noise!

57. Screening control sequences
Features to consider:
Expected frequency of seed matches
Expected frequency of matching to 3’ end of miRNA (after seed extension)
Observed count of seed matches in UTR datasets
Predicted free energies for seed:match interactions

58. Signal:noise results Filled bars are for authentic miRNAs
Open bars show the mean and standard deviation for shuffled sequences
nrMamm set used for first two, nrVert used for set including fugu

59. Biological relevance
Hypothesis: 5’ conservation of miRNAs is important for mRNA target recognition
Highest signal:noise ratio observed when seed positioned close to 5’ end
Hypothesis: highly conserved miRNAs are more involved in regulation
High degree of conservation -> more predicted targets
Membership in large miRNA family -> more predicted targets

60. Experimental verification
15 predicted target sites chosen
All with known biological function
Representative of the entire list of candidates
11 target sites confirmed
Expression of upstream ORF influenced
27% false positives – close correspondance to predicted 30% false positives

61. References
Chalk, A.M. and Sonnhammer, E.L.L. (2002) “Computational antisense oligo prediction with a neural network model.” Bioinformatics 18(12):
Hofacker, I.L., Fontanta, W., Stadler, P.F., Bonhoeffer, S., Tacker, M., and Schuster, P. (1994) “Fast folding and comparison of RNA secondary structures.” Monatshefte fur Chemie 125:
Lewis, B.P., Shih, I., Jones-Rhoades, M.W., and Bartel, D.P. (2003) “Prediction of mammalian microRNA targets.” Cell 115(7):