Maya Topf

Flexible Fitting of Atomic Models in cryoEM Density

(Flex-EM, RIBFIND & TEMPy)

S2C2 CryoEM CCP-EM Modeling Workshop11th Nov 2020

EMDB Statistics

https://www.ebi.ac.uk/pdbe/emdb/statistics_main.html/

Villa & Lasker, Curr Opin Struct Biol, 2014

Structural features vs. resolution

De novo

Flexible Fitting

Rigid-Body Fitting / Assembly Fitting

25 Å

Aims:

• To understand how to optimise your structural model inside the 3D-EM density map, with focus on Flex-EM and RIBFIND

• To understand approaches to model validation in the context of 3D-EM maps, with focus on TEMPy

Approach to modelling structures in 3D-EM maps

Multi-componentrigidfitting

Fitassessment

Map+Model

Flexiblefitting(refinement)

TEMPy

Flex-EM/RIBFINDγ-TEMPy

tempy.ismb.lon.ac.uktopf-group.ismb.lon.ac.uk/flex-em

• Uses simulated annealing rigid-body dynamics.

• During the refinement, the atoms (or rigid bodies) are displaced in the direction that maximizes their cross-correlation with the cryoEM density map (ECCC) and minimizes the violations of the stereochemical (ESC) and non-bonded contacts (ENB).

• Restraints are used to reduce over-fitting and maintain geometry.

• Rigid body restraints are defined based on inter-atomic or inter-residue contacts. Atoms in a rigid body move together during the course of refinement.

Topf et al. Structure 2008 http://topf-group.ismb.lon.ac.uk/flex-em/

Flex-EM: flexible fitting by real-space refinement

salilab.org/modeller

• Flex-EM can use rigid bodies calculated with RIBFIND.

• RIBFIND works by clustering secondary structure elements together (and intermittent regions) as rigid bodies.

Pandurangan et al. JSB 2012

Rigid-body restraints

SSE-based clustering

• Allows faster large body movements in the initial stages of refinement

• Useful when the resolution of density map is insufficient to fit smaller entities like individual residues or atoms.

0%

Cluster cutoff100%25% 35% 37% 50% 51%

Numberofclusters1 2 3 2 0

https://ribfind2.ismb.lon.ac.uk/

RIBFIND

PDB: 2driA 10Å resolution map

Pandurangan et al. JSB 2012

Initial Un-clustered Clustered

Cα RMSD from X-ray structure:

Hierarchical refinement

PDB: 1dpe 5Å resolution map Initial Final un-clustered Final clustered

Cα RMSD: 12.28Å 3.92Å3.69Å un-cluster0.81 0.870.89CCC:

Final two-stage refinement

RMSD:2.12ÅCCC:0.92

Pandurangan et al. JSB 2012; Clare et al., Cell 2012

Joseph et al. Methods 2016

Hierarchical fitting with Flex-EM

GroEL at 4.1 Å (EMD: 6422)

Flex-EM approach to high resolutions

• For large body motions, the maximum atom displacement along one axis at each MD step is limited by 0.39 Å.

• In cases where the model does not have multiple domains/sub-domains that can be identified as large RBs, one can start with the stage where SSEs are constrained as rigid. At this stage, maximum atom shifts were limited to ~0.2 Å at each MD step.

• At the final refinement stage, relative motions between all atoms were allowed without considering any rigid segments. The maximum atom displacement steps are reduced to 0.1 Å in this final stage for finer fitting refinement.

Implementation in CCP-EM

RIBFIND Flex-EM

Model assessment with TEMPy

http://tempy.ismb.lon.ac.uk/

Cragnolini et al. TEMPy2: A python library with improved 3D electron microscopy density fitting and validation workflows. Acta Crystallogr. D (in press).

http://tempy.ismb.lon.ac.uk/

Cragnolini et al. TEMPy2: A python library with improved 3D electron microscopy density fitting and validation workflows. Acta Crystallogr. D (in press).

TEMPy: Template and EM comparison using Python

Multiple TEMPy scores to compare model fits globally

Implementation in CCP-EM

Joseph et al. JSB 2017

• Cross-correlation coefficient (CCC)

• Mutual information-based score (MI)

• Model-map overlap

Local scoring

TEMPy + Chimera attribute files

Useful for calculating CCC locally on any defined segment

Probe density X

Target density Y

SCCC = SCCC

Segment-based CCC:

Pandurangan et al., 2014; Farabella et al. 2015; Atherton et al. eLIFE (2017)

Local scoring

Segment-based manders’ overlap coefficient:

TEMPy + Chimera

Useful to calculate local fit per residue

Joseph AP. et al. Methods 2016

SM

OC

The overlap coefficient is calculated over voxels covered by each residue (and the local neighbourhood), both individually (SMOCd) and along the chain (SMOCf)

Local scoring

Segment-based manders’ overlap coefficient:

Joseph AP. et al. Methods 2016

The overlap coefficient is calculated over voxels covered by each residue (and the local neighbourhood)

EMD-3488 (3.2Å) Deposited model PDB: 5NI1

329_1o: best by CCC (TEMPy) 460_1o: best by lDDT

Target contains errors

TS0984o

Example from CASP13

https://predictioncenter.org/casp13/Kryshtafovych, Malhotra et al. Proteins 2019

Joseph A.P. et al. Methods 2016

Unliganded GroEL at 4.2 Å resolution - (EMD-5001)ADP-bound GroEL (PDB: 4KI8)Refined modelDeposited model (Ludtke et al. Structure 2008, PDB: 3cau)

Hierarchical Refinement at intermediate to near-atomic resolutions

Implementation in CCP-EM

Joseph et al. JCIM 2020, Locke et al. PNAS 2017

Calculating difference maps

Agnel Joseph

Irene Falabella Daven Vasishtan

Arun P Pandurangan

Approach to modelling structures from 3D-EM maps

Harpal Sahota

Sony Malhotra

Tristan Cragnolini

Multi-componentrigidfitting

Fitassessment

Map+Model

Flexiblefitting(refinement)

TEMPy

Flex-EM/RIBFINDγ-TEMPy

Ben Blundell

Thomas Mulvaney

Aaron Sweeney

Thanks…

Topf group (current)Sony Malhotra Tristan Cragnolini Mauro Maiorca Aaron Sweeney Rebecca Brooker Eric Escriva Guendalina Marini  Sophie Knott Tomas Mulvaney Victoria Sanders Manaz Kaleel Karen Manalastas

CCP-EM (STFC)Martyn Winn Tom Burnley Colin Palmer Agnel-Praveen Joseph

EM groupNatasha Lukoyanova Helen Saibil Carolyn Moores Joe Atherton Alex Cook Elena Orlova

ISMB UCL/BirkbeckDave Houldershaw Kostas Thalassinos

EMDB (EBI)Ardan Patwardhan Gerard Kleywegt Ingvar Lagerstedt

eBICDan Clare

HPI (Hamburg)/OxfordKay Grünewald Daven Vasishtan