Discoveries Through theComputational Microscope

Investigation of drug (Tamiflu) resistance of the “swine” flu virus demanded fast response! Klaus Schulten

Department of Physics and Theoretical and Computational Biophysics Group

University of Illinois at Urbana-Champaign

Accuracy • Speed-up • Unprecedented Scale

100 - 1,000,000 processors

The Computational

Microscope

Viewing the Morphogenesis of a Cellular Membrane from Flat to Tubular in 200 µs

DOE (INCITE)

Viewing the Morphogenesis of a Cellular Membrane from Flat to Tubular in 200 µs

Cell, 132:807 (2008)

Cryo-EM image

A. Arkhipov, Y. Yin, and K. Schulten. Four-scale description of membrane sculpting by BAR domains. Biophysical J., 95: 2806-2821 2008.

Ying Yin, Anton Arkhipov, and Klaus Schulten. Simulations of membrane tubulation by lattices of amphiphysin N-BAR domains. Structure 17, 882-892, 2009.

CPC-D-10-00292

Viewing How Proteins are Made from Genetic Blueprint

• Ribosome — Decodes genetic information from mRNA• Important target of many

antibiotics• Static structures of crystal forms

led to 2009 Nobel Prize• But one needs structures of

ribosomes in action!

protein-conducting channel

membrane

mRNA

new protein

ribosome

ribosome

protein-conducting channel

• Ribosome — Decodes genetic information from mRNA• Important target of many

antibiotics• Static structures of crystal forms

led to 2009 Nobel Prize• But one needs structures of

ribosomes in action!

Viewing How Proteins are Made from Genetic Blueprint

Low-resolution data High-resolution structure

Close-up of nascent protein

Viewing How Proteins Are Made from Genetic Blueprint

0

0.4

0.8

1.2

1.6

8 16 32 64

s/step

CPU cores

NCSA Lincoln Cluster performance(8 Intel cores and 2 NVIDIA Tesla GPUs per node, 1 million atoms)

2 GPUs

4 GPUs8 GPUs

16 GPUs

~2.8

GPUs reduced time for simulation from two months to two weeks!

Molecular dynamics simulation of protein insertion process

Faster

GPU Solution 3: Molecular Dynamics Simulations

Viewing Nanopore Sensors

1

10

100

1,000

10,000

100,000

1.5 2 2.5 3 3.5 4

88 b

p co

pies

voltage (V)

methylated DNA un-methylated DNA

Genetics: Genes control our bodies and experiences!Epigenetics: Our bodies and experiences control the

genes!Epigenetics made possible through DNA methylation

small changebig effect

methylation

Related pathologies: obesity, depression, cancer

Detect methylation with nanopores

hard to move

easy to move

un-methylated DNA

methylatedDNA

Viewing Nanopore SensorsCreate a Better Nanopore with Polymeric Materials

nanopre-cipitationof ions

New materials, new problems: Nanoprecipitation Radial distribution functions

identify nanoprecipitation

Precipitate

Liquid

0

2

4

6

8

2 4 6 8 10

g(r)

r / Angstrom

0.1

1

10

100

10,000 100,000 1,000,000 5,000,000

billi

ons

of a

tom

pai

rs/s

ec

Atoms

Intel X5550, 4-cores @ 2.66GHz

6x NVIDIA Tesla C1060 (GT200)

4x NVIDIA Tesla C2050 (Fermi)

Faster

GPU Solution 4: Computing Radial Distribution Functions

• 4.7 million atoms• 4-core Intel X5550 CPU: 15 hours• 4 NVIDIA C2050 GPUs: 10 minutes• Fermi GPUs ~3x faster than GT200

GPUs: larger on-chip shared memory

Precipitate

Liquid

0

2

4

6

8

2 4 6 8 10

g(r)

r / Angstrom

Collaborator: Bernard C. Lim (Mayo Clinic College of Medicine)

20ns SMD Simulation of fibrinogen, 1.06 million atoms, 1.2 ns/day with pencil decomposition, 15 days on PSC XT3 Cray (1024 processors)

A Blood ClotRed blood cells within a network of fibrin fibers, composed of polyme-rized fibrinogen mole-cules.

Inspecting the mechanical Strength of a blood clot

12

B. Lim, E. Lee, M. Sotomayor, and K. Schulten. Molecular basis of fibrin clot elasticity. Structure, 16:449-459, 2008.

Collaborator: Bernard C. Lim (Mayo Clinic College of Medicine)

20ns SMD Simulation of fibrinogen, 1.06 million atoms, 1.2 ns/day with pencil decomposition, 15 days on PSC XT3 Cray (1024 processors)

A Blood ClotRed blood cells within a network of fibrin fibers, composed of polyme-rized fibrinogen mole-cules.

12

B. Lim, E. Lee, M. Sotomayor, and K. Schulten. Molecular basis of fibrin clot elasticity. Structure, 16:449-459, 2008.

Inspecting the mechanical Strength of a blood clot

13

Petascale simulations will Permit Sampling For Example Carrying out a Second Simulation Required

by a Referee

model Szabo and Hummer, 2003.

Longest ever (1 us) SMD simulation closes gap between simulation and experiment!!!!!

Stretched is I91, one of the 300 domains of titin

Microsecond simulations of muscle elasticityReaching for Overlapping Time Scales

�H�

6SHFWURPHWHU

�D�

NLQDVH

�G��F�

�E�

NLQ�

99

Figure 1: Schematic representation of kinase sensor. (a) A single peptide is represented as a green line.After phosphorylation by protein kinase, a residue becomes negatively charged. The phosphorylated residueis pictured as red dot. A fluorescence marker is drawn as a blue star. (b) The sensor consists of peptidesthat are attached to a metallic gold surface. (c) When peptides are non-phosphorylated, spectroscopic signalis una↵ected by an applied electric field. (d) When peptides are phosphorylated, the negative charges bendthe peptides depending on the voltage polarity, enhancing the spectroscopic signal. (e) Experimental setupof the Raman spectroscopy. Kinase sensor is facing down. Peptides on SERS substrate are excited with alaser with wavelength of 785 nm, and the reflected Raman signal is collected with a spectrometer.

through an electrophoretic chamber, where they are detected by fluorescence spectroscopy.

We propose a modified sensor, where the peptides are covalently attached to a gold wafer

(Figure 1.b), immobilizing one end of the peptide to the metallic surface, while the other

peptide end is bound to a marker molecule. The functionality of the device relies on register-

ing di↵erent spectroscopy signatures for non-phosphorylated and phosphorylated peptides

under an external electric field. The intensity of the spectroscopic signal varies according to

the streching or coiling of the peptides. When a voltage biases is applied orthogonal to the

surface, non-phosphorylated peptides should emit a characteristic signal which is insensitive

to the changes in voltage (Figure 1.c). Conversely, phosphorylated peptides, due to their

extra negative charges, are sensitive to the voltage biases and change the conformation of

the peptides, either increasing or decreasing the spectroscopic signal (Figure 1.d).

Accordingly, we considered surface-enhanced Raman spectroscopy (SERS) as a sensitive

method to detect the chemical fingerprint of phosphorylation.The advantages of this method

are listed below: First, all optically excited molecules display a specific Raman shifts sig-

4

Design of Tyrosine Kinase Sensor

Rhodamine (R6g) Tyrosine residue interacts with kinases

by phosphorylation

�D�

����QP

�F� �G��E�)OXRUHVFHQFH�LPDJH�VKRZLQJ�SHSWLGH�SUREHV�ZHUH�VXFFHVVIXOO\�LPPRELOL]HG�RQ�WKH�QDQRFKLS�VXUIDFH

Figure 2: Proposed kinase sensor studied by experiments and simulations. (a) Optical image of sensor: A5⇥5 square array pattern are created by a SEPERISE method (see text). A thin layer of 5 nm titaniumfollowed by a layer of 80 nm gold is deposited on top of the nanocone structures to activate the sensorsurface. The nanostructures trap incident light within the surface, so they look darker than smooth surface.(b) MISSING FIGURE. (c) Nanoscopic structures of the activated areas of the kinase sensor. Peptide probesare immobilized on the gold covered surface of nanocones. (c) Atomic model of kinase sensor. Peptides arecolored in green, metallic surface in yellow. A single peptide is highlighted in licorice representation. Blueand red arrows point to the rhodamine cap and phosphorylated tyrosine residue, respectively.

8

�D�

����QP

�F� �G��E�)OXRUHVFHQFH�LPDJH�VKRZLQJ�SHSWLGH�SUREHV�ZHUH�VXFFHVVIXOO\�LPPRELOL]HG�RQ�WKH�QDQRFKLS�VXUIDFH

Figure 2: Proposed kinase sensor studied by experiments and simulations. (a) Optical image of sensor: A5⇥5 square array pattern are created by a SEPERISE method (see text). A thin layer of 5 nm titaniumfollowed by a layer of 80 nm gold is deposited on top of the nanocone structures to activate the sensorsurface. The nanostructures trap incident light within the surface, so they look darker than smooth surface.(b) MISSING FIGURE. (c) Nanoscopic structures of the activated areas of the kinase sensor. Peptide probesare immobilized on the gold covered surface of nanocones. (c) Atomic model of kinase sensor. Peptides arecolored in green, metallic surface in yellow. A single peptide is highlighted in licorice representation. Blueand red arrows point to the rhodamine cap and phosphorylated tyrosine residue, respectively.

8

�D�

����QP

�F� �G��E�)OXRUHVFHQFH�LPDJH�VKRZLQJ�SHSWLGH�SUREHV�ZHUH�VXFFHVVIXOO\�LPPRELOL]HG�RQ�WKH�QDQRFKLS�VXUIDFH

Figure 2: Proposed kinase sensor studied by experiments and simulations. (a) Optical image of sensor: A5⇥5 square array pattern are created by a SEPERISE method (see text). A thin layer of 5 nm titaniumfollowed by a layer of 80 nm gold is deposited on top of the nanocone structures to activate the sensorsurface. The nanostructures trap incident light within the surface, so they look darker than smooth surface.(b) MISSING FIGURE. (c) Nanoscopic structures of the activated areas of the kinase sensor. Peptide probesare immobilized on the gold covered surface of nanocones. (c) Atomic model of kinase sensor. Peptides arecolored in green, metallic surface in yellow. A single peptide is highlighted in licorice representation. Blueand red arrows point to the rhodamine cap and phosphorylated tyrosine residue, respectively.

8

•Peptides are attached to a gold surface. The end of oligopeptide contains rhodamine.•Peptides are exposed to ATP and appropriate kinase.•Electric field is applied; change in conformation depends on phosphorylation state. •Distance from rhodamine determines magnitude of fluorescence signal.

Au surface, oligopeptide with 12 AA, bound via sulfide bond

initial sequence: {1 GLU} {2 GLY} {3 ILE} {4 TYR} {5 GLY} {6 VAL} {7 LEU} {8 PHE} {9 LYS} {10 LYS} {11 LYS} {12 CYS}

MD Simulations Revealed Problems in Design

•Improved sequence. Avoid residues that strongly bind to gold surface.

•Rhodamines molecules tend to aggregate. Aggregation affects peptide bending.

•MD systems include gold surface, water, ions and 5x5 peptide grid, ~ 100,000 atoms.•Different sequences tested under positive and negative volt biases.

initial sequence: {1 GLU} {2 GLY} {3 ILE} {4 TYR} {5 GLY} {6 VAL} {7 LEU} {8 PHE} {9 LYS} {10 LYS} {11 LYS} {12 CYS}

Initial sequence did not work due to surface adhesion and rhodamine aggregation

Current Design Improved

Rhodamine removed,Fluorescence signal discarded

MD Simulations show that

phosphorylated tyrosine bends the

peptide depending on voltage polarity

0 1 2 3 4 5 60

5

10

15

20

25

30

35

40

dist

Tyr-Gold /

A

time / ns

+60V-60V0V

700 1100 1500

0

200

400

600

800

Raman shift / cm-1

inte

nsity

/ A

U

+1.2V-1.2V

0VExperiments show that peptide bends

towards the surface at positive voltages,

increasing the intensity of Raman

shifts.Phosphorylated tyrosine

Peptide bending is now measured with Raman Spectroscopy. The detection relies on careful comparison of peak points

New sequence: rhodamin removed and measurement replaced by Raman spectroscopy

0 1 2 3 4 5 60

5

10

15

20

25

30

35

40

dist

Tyr-Gold /

A

time / ns

+60V-60V0V

700 1100 1500

0

200

400

600

800in

tens

ity /

AU

Raman shift / cm-1

+1.2V-1.2V

0V

0 1 2 3 4 5 60

5

10

15

20

25

30

35

40

dist

Tyr-Gold /

A

time / ns

+60V-60V0V

700 1100 1500

0

200

400

600

800

inte

nsity

/ A

U

Raman shift / cm-1

+1.2V-1.2V

0V

EGIYGVLFKKKC-AU EGI(pY)GVLFKKKC-AU

Unphosphorylated Phosphorylated

WLPH���QV� � � � � �

�

�

�

�

�

GLVW������������QP

<�JROG

�D� UKR�(*,<*9/)...&�JROG

WLPH���QV� � � � � �

�

�

�

�

� �E� UKR�(*,�S<�*9/)...&�JROG

�L�

GLVW������������QP

<�JROG

�LL�

�L�

�LL�

Figure 4: Peptide phosphorylation revealed by molecular dynamics simulations. Panels (a) and (b) showthe distance from the tyrosine residue to the gold surface for di↵erent voltage polarities. Error bars represent± standard deviation. (a) Non-phosphorylated peptide sensor under +60 V (blue), -60 V (red) and 0 voltagebiases. (b) Phosphorylated peptide sensor under +60 V (blue), -60 V (red) and 0 voltage biases. Panels(i) and (ii) show snapshots of two peptides after 5 ns for +60 V (i) and -60 V (ii) voltage biases. Thepeptides are represented as gray tubes. Rhodamine and tyrosine residues are shown in blue and red colors,respectively.

14

WLPH���QV� � � � � �

�

�

�

�

�

GLVW������������QP

<�JROG

�D� UKR�(*,<*9/)...&�JROG

WLPH���QV� � � � � �

�

�

�

�

� �E� UKR�(*,�S<�*9/)...&�JROG

�L�

GLVW������������QP

<�JROG

�LL�

�L�

�LL�

Figure 4: Peptide phosphorylation revealed by molecular dynamics simulations. Panels (a) and (b) showthe distance from the tyrosine residue to the gold surface for di↵erent voltage polarities. Error bars represent± standard deviation. (a) Non-phosphorylated peptide sensor under +60 V (blue), -60 V (red) and 0 voltagebiases. (b) Phosphorylated peptide sensor under +60 V (blue), -60 V (red) and 0 voltage biases. Panels(i) and (ii) show snapshots of two peptides after 5 ns for +60 V (i) and -60 V (ii) voltage biases. Thepeptides are represented as gray tubes. Rhodamine and tyrosine residues are shown in blue and red colors,respectively.

14

WLPH���QV� � � � �

WLPH���QV� � � � �

�

�

�

�

�

GLVW���������QP

<�JROG

�

�

�

�

�

GLVW���������QP

<�JROG

�D�

�E�

(*,<*9/$$$$&�JROG

(*,�S<�*9/$$$$&�JROG

Figure 5: Optimized sequence. A new sequence is proposed from molecular dynamics simulations. Thereis no rhodamine caps; avoiding aggregation. Unnecessary lysine residues changed by alanine residues. Thenon-phosphorylated peptide sensor is still insensitive to an electric field (a), while the phosphorylated oneis responsive to an external electric field (b); therefore it should produce distinctive raman signatures foropposite voltage polarities. Error bars represent ± standard deviation. Color blue, red an green represent+60 V, -60 V and 0 voltage biases.

16

WLPH���QV� � � � �

WLPH���QV� � � � �

�

�

�

�

�

GLVW���������QP

<�JROG

�

�

�

�

�

GLVW���������QP

<�JROG

�D�

�E�

(*,<*9/$$$$&�JROG

(*,�S<�*9/$$$$&�JROG

Figure 5: Optimized sequence. A new sequence is proposed from molecular dynamics simulations. Thereis no rhodamine caps; avoiding aggregation. Unnecessary lysine residues changed by alanine residues. Thenon-phosphorylated peptide sensor is still insensitive to an electric field (a), while the phosphorylated oneis responsive to an external electric field (b); therefore it should produce distinctive raman signatures foropposite voltage polarities. Error bars represent ± standard deviation. Color blue, red an green represent+60 V, -60 V and 0 voltage biases.

16

A new sequence is proposed from molecular dynamics simulations. There are no rhodamine caps, avoiding aggregation. Unnecessary lysine residues are changed to alanine residues. The resulting non-phosphorylated peptide sensor is still insensitive to an electric field, while the phosphorylated one is responsive to an external electric field; therefore the sensor should produce distinctive Raman signatures for opposite voltage polarities. Error bars represent standard deviation. Colors blue, red, and green represent +60 V, -60 V and 0 voltage biases.

Simulation-Optimized Sequence