Transcript Protein structure and folding

Energy landscape for proteins



With statistical mechanics, one can try to
understand protein folding kinetics.
An extension from theories for glasses and
polymers.
Energy landscape for protein folding:




Must be rough, containing many local minimum.
Must exist guiding forces that stabilize the native
structure.
This is the “minimum frustration” principle.
Proposed and elaborated mainly by Peter G.
Wolynes.
A folding trajectory
Associative memory Hamiltonian: a
knowledge-based force field



Generalizing theory for spin glasses to treat
heteropolymers.
Wolynes and other people have developed a
method to extract “associative memory
Hamiltonian” from a database of known protein
structures.
Associative memory Hamiltonian automatically
have the funnel-like and rugged energy
landscape.
How well does Associative memory
Hamiltonian do in predicting
structures?
Fig. 4. Structural alignments of Qbest structures from simulated annealing of one training set
protein (434 repressor) and the three test set proteins to their x-ray structures
Hardin, Corey et al. (2000) Proc. Natl. Acad. Sci. USA 97, 14235-14240
Copyright ©2000 by the National Academy of Sciences
Fig. 1. Superpositions of Qbest structures onto the native state
Hardin, Corey et al. (2003) Proc. Natl. Acad. Sci. USA 100, 1679-1684
Copyright ©2003 by the National Academy of Sciences
II. Knowledge-based methods



More and more protein structures are solved
with either X-ray or NMR.
Observation: Many of them are similar either
globally or partially. (evolution?)
Can we use these structure as clues?



Derive empirical force field from the database.
Homology modeling; Phylogenomic inference.
Teach computers to deduce structural clues.
The use of homology (R. B. Altman
http://www.smi.stanford.edu/projects/helix/bmi214/5-15-01c2.pdf )

Homology Modeling (sequences with high
homology to sequences of known structure)


Given a sequence with homology > 25-30% with
known structure in PDB, use known structure as
starting point to create a model of the 3D structure of
the sequence.
Takes advantage of knowledge of a closely related
protein. Use sequence alignment techniques to
establish correspondences between known
“template” and unknown.
http://www.luc.edu/faculty/kolsen/lecture2/ppframe.htm
http://www.luc.edu/faculty/kolsen/lecture2/ppframe.htm
http://www.luc.edu/faculty/kolsen/lecture2/ppframe.htm
http://www.luc.edu/faculty/kolsen/lecture2/ppframe.htm
http://www.luc.edu/faculty/kolsen/lecture2/ppframe.htm
A more careful analysis can be
performed with phylogenomic
analysis
Phylogenomic analysis
http://phylogenomic.berkeley.edu
“Train” computers to deduce
rules for protein structures

For secondary structures: determining α
helixes,β sheets, or loops from sequence data

Neural networks: (>72% accuracy)
• B Rost: “PHD: predicting one-dimensional protein structure
by profile based neural networks.” Meth. in Enzymolgy, 266,
525-539, (1996)
• Available over the web.
http://cubic.bioc.columbia.edu/predictprotein

Support vector machines
• Developed recently in Taiwan.
• Slightly out-performed neural networks.
Neural networks

As an example, pattern recognition is very hard
to defined with simple “rules”..




The letter “A” can also be “A” “A” “A” or
.
Neural networks or SVM algorithms do not
assume any rules a priori. Instead they allow
the system to be “trained”, i.e. the undefined
parameters are to be determined by seeing
examples.
Needs a (large) training set, and a separate
testing set to see how well it does.
So part of the protein structure problems can be
solved by this “pattern recognition” process.
Secondary structure

Secondary structure prediction can be
done with more sophisticated algorithms.
Artificial intelligence such as neuronal
networks or support vector machines.
 Basically look at a local sequence and
recongnize its pattern.
 Usually such methods need a training set.
I.e. knowledge-based methods.

PredictProtein does more than
2º structure prediction
Figure 9-6 NMR structure of
protein GB1.




Page 280


Green: residues 23-33.
Cyan: residues 42-53.
Chm-alpha: a new sequence
replaces green part.
Chm-beta: the same new
sequence replaced cyan part.
Both are structurally similar to
native GB1.
The same sequence can be
either an alpha helix or a beta
sheet structure, depending on
their context.
Threading



Instead of searching over a global range of
structure, threading uses existing protein
conformations.
Still needs an energy function.
Improves the speed and accuracy if:


the constrained conformation set has the target
native structure (unknown), and
within this set the energy function is good enough to
distinguish the native structure from others.
Threading is often combined
with other methods
Classification of protein folds


Common structural motifs are often seen. They are
classified into hierarchical groups.
SCOP


CATH


http://scop.mrc-lmb.cam.ac.uk/scop/
http://www.biochem.ucl.ac.uk/bsm/cath/index.html
Can such knowledge helps us recognize “folds” in
proteins? (again I am quoting R. B. Altman)



Fold recognition (sequences with no sequence identity (<=
30%) to sequences of known structure.
Given the sequence, and a set of folds observed in PDB, see if
any of the sequences could adopt one the known folds.
Takes advantage of knowledge of existing structures, and
principles by which they are stabilized (favorable interactions).
Representatives of some popular folds
(J. Mol. Graphics Modell. 19, 157, (2001))
Fold Recognition (Altman)

New sequence:
MLDTNMKTQLKAYLEKLTKPVELIATLDDSAKSAEIKELL…

Library of known folds:
Fold Recognition (Altman)

Library of protein structures, suitably processed




Scoring function



All structures
Representative subset
Structures with loops removed
contact potential
environmental evaluation function
Method for generating initial alignments and/or
searching for better alignments.
Fold recognition server: 3D-pssm
http://www.sbg.bio.ic.ac.uk/~3dpssm/
Additional useful links

On-line course materials:




PredictProtein Server:


http://www.cryst.bbk.ac.uk/PPS2/course/index.html
http://www.expasy.org/swissmod/course/courseindex.htm
http://scpd.stanford.edu/SOL/courses/proEd/RACMB
/materials.htm
http://www.emblheidelberg.de/predictprotein/predictprotein.html
3D-pssm server:

http://www.sbg.bio.ic.ac.uk/~3dpssm/
Folding Accessory proteins
1.
2.
3.
Protein Disulfide
Isomerases
Peptidyl prolyl cistrans Isomerases
Chaperones
Electron micrographs - low
resolution molecular structures
Diseases associated with
protein structures
The story of -sheets twists
Why right-handed?
Voet Biochemistry 3e
Page 251
© 2004 John Wiley & Sons, Inc.
Figure 8-49
Retinol binding protein.
Why right-handed twist?
• A question that is easier to answer for helix. (L encounters steric hindrances).
Voet Biochemistry 3e Page 224
© 2004 John Wiley & Sons, Inc.
Figure 8-11
The right-handed  helix.
References
• Chou and Scheraga PNAS (1982) 79,
7047-7051.
• Wang et. al, J. Mol. Biol. (1996) 262, 283293.
Chou (1982)
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
Voet Biochemistry 3e Page 300
© 2004 John Wiley & Sons, Inc.
Table 9-1
Propensities and Classifications of Amino Acid
Residues for  Helical and  Sheet Conformations.
Conclusions from Chou 1982
• Intrastrand twisting (I.e. single-strand twisting) is
an important factor.
• For a single peptide strand, twisted strand is
more stable than a flattened strand.
 This is because a higher torsional energy exists in flat
structure.
 Also a more favorable non-bonding interaction in the
twisted structures.
• Right-handed twist is more stable than lefthanded twist, due to a sum of many nonbonding interactions.
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
14 years later…
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
What was done?
• Molecular dynamics simulation, instead of
previous conformational energy minimization.
• Studied poly-Ala, poly-Val and poly-Gly, each
with 2 or 3 strands, and each strand has 3 or 5
residues.
• Examined effects of electrostatic vs. van der
Waal interactions; effect of single vs. multiple
strands; effects of solvents.
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
Single
molecular
manipulation of
a protein
Protein folding and
unfolding experiments
Science 288, 143 (2000)
Bacteriorodopsin
Science 303,
1674 (2004)
Force Clamp?
• A. F. Oberhauser, P. K. Hansma,
M. Carrion-Vazquez, and J. M.
Fernandez, PNAS 468, 98
(2001)
Science 303,
1674 (2004)
120 pN
50 pN
120 pN
120 pN
35 pN
120 pN
120 pN
35 pN
120 pN
120 pN
23 pN
120 pN
In Conclusion
• The authors said: “Our results contradict the
generally held view that folding and unfolding
reactions correspond to transitions between
well-defined discrete states. In contrast, we
observed that ubiquitin folding occurs through a
series of continuous stages that cannot be easily
represented by state diagrams.”
• What do you think?