Bioinformatics and molecular modelling studies of membrane

proteins

Shiva Amiri

Professor Mark S.P. SansomJune 1, 2004

constitute approximately 25% of the genome

important drug targets- nerve and muscle excitation- hormonal secretion- sensory transduction- control of salt and water balance etc.

malfunctions result in various diseases

Membrane proteins

Nelson, M. Comparative Neurophysiology, 2000.

function is dependent upon the binding of a ligand.

examples of LGICs: nAChR, GABAA and GABAC receptors, 5HT3 receptor, Glycine receptor

sdf

Ligand gated ion channels (LGICs)

Sperelakis, N., Cell Physiology Source Book

problem: difficult to obtain high resolution crystallographic images of membrane proteins

Unwin et.al, Nature, 26 June 2003

some success using cryo-electron microscopy coupled with Fourier Transforms, i.e. Unwin’s 4Å image of the TM region.

but still no full structure of any LGIC

Structure prediction

Unwin et.al, Nature, 26 June 2003

to take available structural data and put the pieces together

main focus so far: using available information to predict the structure and motions of the α-7 nicotinic acetylcholine receptor (nAChR)

we have:4Å cryo-EM structure of AChR transmembrane domain2.7Å crystal structure of ligand binding domain homolog

task: to combine the two domains

the use of bioinformatics and simulation tools to study functionally relevant motions of LGICs

My project

α-7 nAChR

some properties –- cationic channel- homopentamer- four transmembrane regions

(M1-M4)

M2

M3

M4

M1

LB

TM

why nAChR? mutations in genes coding for nAChR can result in Parkinson’s disease, Alzheimer’s disease, myasthenia gravis, frontal lope epilepsy, etc. plays a role in nicotine addiction

The process …

homology modelling -Modeller, Procheck

2 PDBs {θmax, zmax}

ZAlign

termini distances bad contacts ( Unwin distances )

analysis – xfarbe plots

make model using chosen {θ, z}

procheck

GROMACS energy minimization

motion analysis:GNM

CONCOORDelectrostatics (Kaihsu Tai)pore dimensions - HOLE

homology models of other LGICs

transmembrane domain alignment

Homology modeling – transmembrane domain

the homology model of the TM region with the Torpedo marmorata structure

(PDB: 1OED - 4 Å) and the chick α-7 sequence using MODELLER

M1

M3

M2

M4

Homology modelling – transmembrane domain

ligand binding domain alignment

Homology modelling – ligand binding domain

the homology model of the LB domain with acetylcholine binding protein (AChBP) as the structure (PDB: 1I9B – 2.7 Å) and the chick α-7 sequence using MODELLER

Homology modelling – ligand binding domain

α

α

α

α

α

combining the transmembrane domain with the ligand binding domain

producing data upon rotations and translations to allow the user to choose an optimal model

The software

straighten and align each domain with respect to the z-axis

rotate and translate about z-axis- angle of rotation and steps of translations are user-defined

z

x

y

Unwin distance – distance between residues from the TM domain and the LB domain that are meant to come into close proximity

LYS 44

ASP 264

Scoring criteria

termini distance – distance between the N-terminus of the LB domain and the C-terminus of the TM domain

ARG 205

THR 206

Scoring criteria continued …

bad contacts – number of residues that are closer than a cut-off distance.

LB

TM

LB

TM

Scoring criteria continued …

termini distance

z translation (Å)

theta (radians)

theta (radians)

bad contacts

Unwin distance

z translation (Å)

theta (radians)

Plots of scoring criteria

termini + bad contacts

theta (radians)

theta (radians)

z translation (Å)

Linear combinations of scoring criteria termini + Unwin

theta (radians)

termini + bad contacts + Unwin z translation (Å)

x

chosen {θ, z}

model chosen based on scoring criteria data

once a good model was decided on, energy minimization using GROMACS was carried out to ensure the electrostatic legitimacy of the model- GROMACS joins the two domains at their termini- experimenting with how far can the domain be before GROMACS refuses to join them

procheck is run to check the validity of the structure

Choosing the best model

Putting ACRB together – test case

Plots for ACRB alignmentbad contacts

theta (radians)

theta (radians)

theta (radians)

z translation (Å) z translation (Å)

x

termini

termini + bad contacts

Gaussian network model (GNM)

CONCOORD

Course grain methods of motion analysis

a course-grained model to approximate fluctuations of residues

Information on the flexibility and function of the protein

produces theoretical B-values

residues considered as ‘balls’ and the distance between neighbouring residues are ‘springs’

B-values generally in agreement with crystallographic data

Gaussian network model (GNM)

AChBP – theoretical vs. experimental B-values

0

20

40

60

80

100

120

140

160

1 67 133 199 265 331 397 463 529 595 661 727 793 859 925 991

number of residues

B-v

alu

es

experimentaltheoretical

Theoretical B-values of the model

0

20

40

60

80

100

120

140

1 101 201 301 401 501 601 701 801 901 1001 1101 1201 1301 1401 1501 1601

number of residues

B-v

alu

es

some results were as expected, with more freedom of motion for the outer helices of the TM region

identification of the ligand binding site and also of toxin binding sites

GNM results

ligand binding site

toxin binding

sites

nAChR model coloured by generated B-values

generates protein conformations around a given structure based on distance constraints

suggests plausible motions of the protein principal component analysis (PCA) is applied on the 500 resulting

structures from CONCOORD available at dynamite.biop.ox.ac.uk/dynamite (Paul Barrett)

- used to generate eigenvector (porcupine) plots and covariance line plots using CONCOORD’s output

CONCOORD

porcupine plots have an x number of spikes, each spike representing the element of the eigenvector associated with each c-alpha atom of the protein

although this is a homo-pentamer, there is asymmetry between the subunits (closed state)

Eigenvector plot - LB

the spikes show greater freedom of motion for the outer helices

the spikes are pointed either down or up, no uniform direction

Eigenvector plot - TM

when combined, the spikes have a more organized pattern, with LB region spikes all rotating to one side and the TM spikes rotating in the opposite direction, suggesting a twisting motion of the receptor

the middle of the structure is not as mobile

Eigenvector plot – nAChR model

first eigenvector shows twisting motion of receptor opening and closing of the pore as the subunits rotate

First eigenvector

GABA and glycine receptors (anion selective channel)- structure being used is the current model for the α-7 nAChR

Simulations on TM region of model and other LGICs – Oliver Beckstein- looking at the M2 helix and its relevant motions

Homology models of other LGICs

M2s of α-7 nAChR

models of other LGICs motion analysis of other LGICs looking at the hydrophobic girdle (M2) of LGICs to study patterns of

conservation and the behaviour of these residues during gating simulation studies of constructed models

modelling methods for LGICs

predicted structure of α-7 nAChR

used various methods (GNM, CONCOORD) to look at motions of the predicted structure of α-7 nAChR

models of anionic LGICs (GABA and glycine) using current α-7 nAChR structure

Summary

Future work

ACRB + TolC

Aligning other membrane proteins

Prof. Mark S.P. Sansom Oliver BecksteinDr. Phil Biggin Sundeep DeolDr. Kaihsu Tai Yalini PathyDr. Paul Barrett Jonathan Cuthbertson Dr. Alessandro Grotessi Pete BondDr. Andy Hung Katherine CoxDr. Daniele Bemporad Jennifer JohnstonDr. Jorge Pikunic Jeff CampbellDr. Shozeb Haider Loredana VaccaroDr. Zara Sands Robert D’Rozario Dr. Syma Khalid John HolyoakeDr. Bing Wu Tony IvetacGeorge Patargias Sylvanna Ho

Samantha Kaye

Thanks to:

covariance line plots indicate which parts of the protein are correlated or move together

Covariance line plot – nAChR model

Principal component analysisLoredana Vaccaro

Used to reduce the dimensionality of a data set

for a 3N dimensional data set

covariance matrix

diagonalisation

3N eigenvectors (orthogonal = independent of each other)

eigenvalues (contribution of each eigenvector to the whole motion)

keep the first eigenvectors

reduced data set

Cij = <(xi,t - <xi>t)(xj,t- <xj>t)>t

identify the major motions of the protein

Hydrophobic girdle

M2 alignment