Towards RNA structure prediction: 3D motif prediction and

knowledge-based potential functions

Christian Laing

Tamar Schlick’s lab

Courant Institute of Mathematical SciencesDepartment of Chemistry

New York University

RNA folding is hierarchical

5‘gGACUCGGGGUGCCCUUCUGCGUGAAGGCUGAGAAAUACCCGUAUCACCUGAUCUGGAUAAUGCCAGCGUAGGGAAGUUc3'

Sequence Secondary Structure 3D StructureAnnotated diagram

• Tertiary motifs serve as modular building blocks in the RNA architecture. • To understand the role of RNA tertiary motifs in RNA folding will help to understand RNA 3D prediction.

TPP riboswitch (PDB: 2GDI)

Annotating 3D RNA

• Selected seven key RNA tertiary motifs: coaxial helix, A-minor motif, ribose zipper, tetraloop-tetraloop receptor, pseudoknot, kissing hairpin, and tRNA D-loop:T-loop.

• Searched RNA tertiary motifs via different computer programs

• Annotate tertiary interaction motifs. • Perform analysis over the diagrams

produced.

Examples

RNA junctions have a high probability (84%) to contain at least one coaxial helix.

Kissing hairpin 6 (1%)

Loop-loop receptor 16 (3%)

Coaxial helix 182 (30%)

A-minor motif 229 (38%)

Ribose zipper 121 (20%)

Pseudoknot 40 (7%)

tRNA D-loop;T-loop 7 (1%)

Distribution of tertiary motifs

• For 54 high-resolution RNA structures, 615 RNA tertiary interactions were found.

• Ribose zippers, coaxial helices and A-minor interactions are highly abundant (88%).

Correlated motifs• Several A-minor motifs

(64%) are involved with coaxial helices.

• Coaxial helices (70%) interact with A-minor.

• Most ribose zippers (70%) contain an A-minor.

• Every loop-loop receptor contains a ribose zipper, which in turn contains one or more A-minor motifs 87% of the time.

Group I intron (PDB: 1HR2)RNA V.5 1119 (2001)

Can we predict coaxial stacking?

HCV virus fragment (PDB: 1KH6)NAT.STRUCT.BIOL. V. 9 370 2002

Topology of 4-way junctions in folded RNAs

Analysis on 24 junctions:

GUIDELINES• A coaxial staking Hi-Hi+1 is enhanced when Ji,i+1 is small.• There is a strong preference for a H1-H4 stacking.• Coaxial helices in family H have similar lengths.• Pseudoknots usually stack their helices.

Family H Family K Family X Family t

Family <J12> <J23> <J34> <J41> Coaxial H.

H: 10

H: 1

2.0

0

0.5

0

2.4

0

0.0

0

H1-H4; H2-H4

H1-H2; H3-H4

K: 3

K: 4

K: 1

5.0

4.3

2

2.3

3.8

0

4.3

1.8

1

0.3

1.3

0

H1-H4

H3-H4

H2-H3

X: 5 4.4 2.2 5.6 3.8

t: 1 6 4 5 1 H2-H4

Structural context of the inserted A in A-minor

0

5

10

15

20

25

30

35

40

Helix (WC) Internal Terminal Junction Other SS

Structural context

Per

cen

tag

e

A-minor involved in long-range interactions

Structural context of the Watson-Crick pair in A-minor

0

10

20

30

40

50

1 2 3 4 5

Helical context

Per

cen

tag

e

Statistical Potentials for A-minor prediction• The adenosine (A) can be located in four

single stranded regions: internal (I) and

terminal (T) loops, junctions (J), and other (O). • The helix receptor (R) can be located in five

positions relative to the end site of a helix.• The potential function for an A-minor is defined by:

• is the observed number of interacting pairs• is the expected number of interacting pairs• The statistical free energy for each A-minor is:

( , ).i jA R

{ , , , },i I T J OA A A A A( , )( , ) ,

( , )Obs i j

i jExp i j

N A RP A R

N A R

( , )Exp i jN A R

1 2 3 4 5{ , , , , }jR R R R R R

.,Bk Boltzmann const T Temperature ( , ) ln( ( , ))i j B i jG A R k T P A R

( , )Obs i jN A R

( , ).i jA R

( , )i jA R

Improving coarse-grained modelRosetta:- 1-bead model (only the base).- Neglect sugar and

phosphate.- Only one RNA (23S

rRNA) was considered.

Possible improvements:- 3-bead model to consider

base, sugar, and phosphate.

- Consider our 54-non-redundant RNA dataset. PNAS V. 102 No. 19 6789-94 (2005)

PNAS V. 104 No. 37 14664-9 (2007)

From 2D to 3D: RNA assembleAssume we know the RNA secondary structure (and possible some constrains), how can we use this knowledge into our Mesoscopic model?

• Stems can be described by type-A helices.• Programs such as Assemble (Westhof) and RNA2D3D (Shapiro) can be used but require manual intervention.

Suggested future directions• Design statistical potentials for coaxial stacking.• The combination of A-minor/coaxial helix

prediction may lead to stronger arguments. • Design statistical potentials to predict non

canonical basepairs (Leontis/Westhof), and explore the possibility to use them with dynamic programming.

• Test the statistical potentials (threading, decoys).• There are 272 non-redundant 4-way junctions in

the RNA junction database that can be analyzed topologically and geometrically.

Acknowledgements

Yurong Xin and Tamar Schlick

Many thanks to:Hin Hark GanSean McGuffeeShereef Elmetwaly

Human Frontier Science Program

NSF/NIGMS initiative in Mathematical Biology (DMS-

0201160)

Topology of 3-way junctions in folded RNAs (Lescoute & Westhof RNA 2006 12: 83-93)

Analysis over 33 junctions.

Back up slide

Back up slide

Back up slide