molecular dynamics simulations of large macromolecular ... · nih center for macromolecular...
TRANSCRIPT
-
T C B G
06/15/16
Molecular Dynamics Simulations of Large Macromolecular Complexes
PI Klaus Schulten
Theoretical and Computational Biophysics Group
NIH Center for Macromolecular Modeling and Bioinformatics
University of Illinois at Urbana-Champaign
Till Rudack
-
06/15/16 2
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
The Main Principle
“Everything that living things do can be understood in terms of the jigglings and wigglings of atoms.“
Richard Feynmen
Molecular dynamics (MD) simulations based on the structural data obtained in experiments are well suited to capture high-resolution dynamics of proteins and reveal their functional mechanisms.
-
06/15/16 3
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Computers Can Help to Bridge Disciplines
-
06/15/16 4
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Num
ber o
f Ato
ms!
1990!
106!
105!
107!
108!
1994! 1998! 2002! 2006! 2010! 2014!
Lysozyme!
(2 nm)3!
ATP Synthase!
STMV! Ribosome!
HIV Capsid!
Photosynthetic!Chromatophore!
(100 nm)3!
Year!
Aquaporin!
All-Atom Molecular Dynamics Today
-
06/15/16 5
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
HIV capsid (64M atoms)
Chromatophore (100M atoms)
Chemosensory Array (23M atoms)
Rous Sarcoma Virus (750K atoms 128 replicas)
Rabbit Hemorrhagic Disease (6M atoms)
A Sampling of TCBG‘s Petascale Projects
Perilla et al., “Molecular dynamics simulations of large macromolecular complexes.” Current Opinion in Structural Biology 2015 31:64-74.
-
06/15/16 6
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
1 Gigabit Network
Compressed Video
Handling Big Data
Solution: Interactive Remote Visualization and Analysis
Theoretical investigations of large macromolecular complexes is more than just running MD simulations. It also involves visualization and analysis of Terabytes of data.
-
06/15/16 7
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
• Enablevisualiza-onandanalysesnotpossiblewithconven-onalworksta-ons
• Accessdatalocatedanywhereintheworld– SameVMDsessionavailabletoanydevice
Remote Visualization Enables Petascale Anywhere
Bringing Blue Waters to your home
-
06/15/16 8
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Making MD Simulations Accessible to Everybody
Ribeiro, JV, Bernardi, RC, Rudack, T, et.al., “QwikMD—Integrative Molecular Dynamics
Toolkit for Novices and Experts.” Scientific Reports 6 (2016): 26536
Collaboration with: John Stone
Dr. Rafael Bernardi Dr. João Ribeiro Dr. Jim Phillips
Prof. Peter Freddolino
-
06/15/16 9
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
www.ks.uiuc.edu/Research/qwikmd
QwikMD: Easy preparation of an MD simulation Employing QwikMD, a user is able to prepare an MD simulation on a laptop or even a supercomputer in just a few minutes, through an intuitive “point and click” graphical interface connecting VMD and NAMD .
-
06/15/16 10
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
The Receycling System of the Cell
-
06/15/16 11
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
The ubiquitin proteasome proteolytic pathway
26S Proteasome
Substrate tagging by Ubq4
Ubq4-substrate recognition
Substrate degradation
Bortezomi
b
Kisselev Cancer Cell 2013
-
06/15/16 12
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Near-Atomic Model of the 26S Proteasome
Wolfgang Baumeister
Cryo-EM density Subunits from X-ray crystalography,
NMR, and homology modeling
PDB-ID 4CR2
EMDB-ID 2594
Resolution 7.7 Å
Unverdorben et al. PNAS 2014
Molecular Dynamics Flexible Fitting (MDFF): Trabuco et al. Structure 2008
-
06/15/16 13
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Functional Subunits of the 26S Proteasome
-
06/15/16 14
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Deubiquitylation subunit: Rpn11
Unverdorben et al. PNAS 2014 Chain V of PDB-ID 4CR2
Missing segments
- highly flexible
- ambigous density
Active site of Rpn11: substrate is cleaved from ubiquitin tag
Complete models are a basic prerequisite to perform MD simulations
-
06/15/16 15
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Rosetta/MDFF Interactive Model Building
incomplete structural model deposited in the PDB
complete structural model that fits cryo-EM data
de novostructure
prediction
energyranking
modelfiltering
interactiveMDFF of
cryo-EM data
incomplete structural model deposited in the PDB
complete structural model that fits cryo-EM data
de novostructure
prediction
energyranking
modelfiltering
interactiveMDFF of
cryo-EM data
Rosetta
incomplete structural model deposited in the PDB
complete structural model that fits cryo-EM data
de novostructure
prediction
energyranking
modelfiltering
interactiveMDFF of
cryo-EM data
Rosetta VMD/NAMDLeaver-Fay et al. Methods Enzymol. 2011 Porter et al. PLoS One 2015
Humphrey et al. J. Mol. Graph. 1996 Philips et al. J. Comput. Chem. 2005
www.ks.uiuc.edu/Research/MDFF
-
06/15/16 16
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Model Filtering by Secondary Structure Secondary structure histogram of
predicted ensembles of Rpn11‘s C-terminal tail
-
06/15/16 17
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Predicted Model
Represenative model of the predicted averaged secondary structure pattern
for Rpn11‘s C-terminal tail (purple)
Rosetta tends to build compact structures
Secondary structure pattern of amino acids 217-306 (purple)
-
06/15/16 18
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Visual Inspection of cryo-EM Density
-
06/15/16 19
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Predicted Model to Initiate MDFF
Represenative model of predicted ensemble for Rpn11‘s C-terminal tail
Secondary structure pattern of amino acids 217-306 (purple)
-
06/15/16 20
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Interactive Molecular Dynamics Flexible Fitting
MDFF Tutorial on You Tube and at http://www.ks.uiuc.edu/Research/mdff/
-
06/15/16 21
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Complete Model of Rpn11 Fitted to Density
-
06/15/16 22
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Incomplete vs. Complete Model Incomplete model Complete model
Rosetta/MDFF
Cross correlation 0.61 Cross correlation 0.63
-
06/15/16 23
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Low vs. High Resolution Density Model
Isolated lid cryo-EM model
Gabriel Lander / Andreas Martin
PDB-ID 3JCK EMDB-ID 6479
Resolution 3.5 Å
Dambacher et al. eLife 2016
Red: 3.5 Å cryo-EM model of Rpn11 within the isolated proteasomal lid
Purple: completed Rpn11 model within the 7.7 Å proteasomal cryo-EM density
26S proteasome cryo-EM density
Wolfgang Baumeister
EMDB-ID 2594
Resolution 7.7 Å
Unverdorben et al. PNAS 2014
-
06/15/16 24
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Low vs. High Resolution Density Model Structure predicted for low resolution matches structure of high resolution
Secondary structure pattern obtained by Rosetta/MDFF employing a 7.7 Å density
Secondary structure pattern of a structure modeled into a 3.5 Å density
-
06/15/16 25
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Functional Subunits of the 26S Proteasome
-
06/15/16 26
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Ubiquitin Recognition by Rpn10
K48-linked tetra-ubiquitin Rpn10 Recognition
Re-folding, electrostatic, and hydrophobic interactions lead to recognition
Zhang, Y, Vukovic, L, Rudack,T et al. “Recognition of poly-ubiquitins by the proteasome through protein re-folding guided by electrostatic and hydrophobic interactions.” Journal Physical Chemistry B, McCammon Festschrift 2016 in press
-
06/15/16 27
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Rpn10
PDB-ID 2X5N (S. Pombe)
PDB-ID 2KDE (human)
Ubiquitin interaction motif (UIM)
Ubiquitin recognition and deubiquitylation
Ubiquitin Transport
-
06/15/16 28
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Generalized Simulated Annealing – GSAFold
• Amino acid residues connecting Rpn10’s UIM with the proteasome are likely to be disordered and stochastic searching algorithms such as GSA can be used to explore their conformational space.
• GSAFold coupled to NAMD searches low-energy conformations to be used as starting points for molecular dynamics studies.
GSAFold NAMD Plugin – Allows ab initio structure prediction
New implementation on Blue Waters also Allows the Conformational Search for Large Flexible Regions
Collaboration with: Marcelo Melo
Dr. Rafael Bernardi Prof. Pedro Pascutti
Melo, MCR, et al. "GSAFold: A new application of GSA to protein structure prediction." Proteins: Structure, Function, and Bioinformatics 80.9 (2012): 2305-2310.
Bernardi, RC., Melo MCR, and Schulten K. "Enhanced sampling techniques in molecular dynamics simulations of biological systems." Biochimica et Biophysica
Acta (BBA)-General Subjects 1850.5 (2015): 872-877.
-
06/15/16 29
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Conformation Space of Rpn10 Anchor
Fishing Rod
Fishing Hook
Globular Part
Center of Massof UIM
The conformational space of the Rpn10 linker is highly flexible.
-
06/15/16 30
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Ubiquitin Transport to Deubiquitinase Rpn11
Ubiquitin Transport
-
06/15/16 31
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Functional Subunits of the 26S Proteasome
-
06/15/16 32
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
The Motor of the Proteasome
-
06/15/16 33
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
3.9 Å Resolution Density of the Human 26S Proteasome
-
06/15/16 34
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
High-resolution Real Space Refinment with MDFF
Advantage: Positions of bulky side chains can be observed from density Challenge: no detailed side chain orientation X-ray structure refinement tools failed in the range of 4-5 Å resolution Solution: combining MDFF with monte carlo based backbone and side chain rotamer search algorithms in an iterative manner
Goh BC., Hadden JA, et al., “Computational methodologies for real-space structural refinement of large macromolecular complexes.” Annual Review Biophysics, 2016 in press.
-
06/15/16 35
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
The ATPase Motor of the 26S Proteasome
Schweitzer A, Aufderheide A, Rudack T, et al. “The structure of the 26S proteasome at a resolution of 3.9 Å.” PNAS 2016 in press.
-
06/15/16 36
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
The Motor Action of Protein Unfolding
-
06/15/16 37
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
NAMD QM/MM
Collaboration with:
Marcelo Melo Dr. Rafael Bernardi
Dr. Jim Phillips Prof. Gerd Rocha Prof. Frank Neese
The atomic structure enable detailed investigations of the unfolding process by QM/MM simulations combined with path sampling techniques.
NAMD QM/MM interface with MOPAC and ORCA will be
released in the second semester of 2016
Generic Interface will also
allow the use of any Quantum Chemistry Software
-
06/15/16 38
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Why Blue Waters
Running Simulations Investigating the functional processes of large protein machineries which occur on the millisecond timescale, is only possible on petascale computing resources, such as Blue waters. Visualization and Analyses Blue Waters is not only needed for running simulations of large macromolecular complexes but also for visualization and analysis of the produced data. Obtaining Atomic Structures De novo structure prediction and monte carlo based sampling algorithms are only efficient if thousands of models are predicted. Employing these algorithms for the numerous subunits of the proteasome is a well-suited task for the large-scale parallel architecture of Blue Waters.
-
06/15/16 39
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Work on Next Generation Computer Systems
Motor Action of Protein Unfolding We would like to investigate how the motor action of ATP hydrolysis is driving the subunit reorganization during the functional cycle of substrate processing, which requires extensive sampling of long trajectories. Drug Design In order to exploit the proteasome’s role as a ubiquitous drug target, high-throughput screening of thousands of drugs that possibly interact with the proteasome might be feasible by 2020 and would require pre-exascale computing for such a complex macromolecular machinery as the 26S proteasome.
-
06/15/16 40
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Acknowledgments Klaus Schulten
Wolfgang Baumeister
Ryan McGreevy John Stone
VMD NAMD MDFF
Jim Phillips
The Schulten Group
QwikMD GSA
Joao Ribeiro Marcelo Mello
-
06/15/16 41
T C B G
NIH Center for Macromolecular Modeling and Bioinformatics
Blue Waters Helped to Solve the Atomic Structure of the Human Receycling System