Challenges For Structure Determination At Low Resolution Axel Brunger Stanford University - PDB40 Symposium, Cold Spring Harbor Laboratory, 2011
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Challenges For Structure Determination
At Low Resolution
Axel Brunger
Stanford University
PDB40 Symposium, Cold Spring Harbor Laboratory, 2011
Letter
PUBLIC ACCESS TO X-RAY D I F F R A C T I O N D A T A
ToI f the
requested
Editors:b y the a u t h o r s , the editors will ask the D a t a B a n k n o t to distribute the i n f o r m a t i o n
untilThe a specified date.have
undersigned F o r ac olong
o r d i nstanding
a t e lists concern
this d a t e with
m a y the
n o t problem
be m o r e of
t h apublic
n one access
year b eto
y othe
n d the ac-
results
coe fp tsingle
a n c e dcrystal
a t e o f the m a n diffraction
X-ray u s c r i p t . F o rstudies
the fullonstructure
biological a m pmacromolecules.
l i t u d e a n d p h a s e The
d a t aactual
the time data interval
from
before
such researchd i s t r i b uare
t i o nthe
m ameasured
y n o t exceed X-rayf o uintensities,
r years. Theand release d a t e specified
the primary results bare y the
thealists
u t h oof
r will
atomic a p p eco-
ar
inordinates
a f o o t n o derived
t e to thefrom p a p ethose
r a l o ndata.
g withWhile the s tmany
a t e m epapers
n t t h a t onthevarious
i n f o r m aproteins
t i o n has and
beennucleic
s u b m i acids
t t e d tohave the
Pbeen
r o t e i published
n D a t a Bank. In the absence
describing o f a specified
and interpreting releaseofdate,
the results such ita will be a s sanalysis,
structure u m e d that thethe i n f o rdata
actual ma-
and isresults
tion a v a i l aare
b l e often
i m m enot
d i a tmade
e l y oneasily
a p p e aaccessible,
r a n c e o f theif atp uall.
b l i cWe
a t i oare
n . ' requesting that the journals which
play a major role in the publication of such structural studies adopt and enforce rules for docu-
mentation
C o m p a r a bsimilar to those
l e letters have which been sentapplytoina all n u other
m b e r areas
o f j o uof r n ascientific
l s c o n c e research.
r n e d withStudies macrom where
o l e c u the
lar
structural Winformation
structure. e are a w a r e has t h a tnot been emade
several d i t o r iavailable
a l b o a r d s must be considered
are considering thisincomplete,
m a t t e r , a n dast hwould
a t a m aany jor
other on
effort piece of research
policy d e v e l o pwhere
m e n t istheu ndata
d e r wwere
a y by nottheprovided
I n t e r n a tin
i o npublished
a l U n i o n oro fdeposited
C r y s t a l l o form.
graphy and by
...
In theo fclosely
certain
al X-rayondata
formly)
its a d hrelated
thisand
e r i n g nfield
issue.
a t i o nof
a l small
bodies. molecule
W e hope crystallography
t h a t all will act there
the derived structural model are reported directly in the original article in hard
is no such
positively (and,problem. The actu-
if possible, uni-
copy
A list form. Even
o f the for a osmall
signers f thisprotein the amount
letter follows. The of data issignatures
original so large that are hard
on filecopy in thepublication
office o f has the
occurred only in one or two instances in the early years, but computer readable storage and distri-
undersigned.
bution as currently carried out so effectively by the Protein Data Bank at Brookhaven Frederic M.and its in-
Richards
ternational associated groups is quite practical, and is certainly the preferred form of University
Yale access to-
day. We note also that a policy requiring coordinate Department and of structure
Molecular factor deposition
Biophysics andwould ensure
Biochemistry
the preservation of important and expensively determined data which otherwise are likelyA venue
260 Whitney to be
lost with the ever-changing personnel and computer installations in numerous New Haven, laboratories
CT 06511
around the world. U.S.A.
There are standard procedures now for the refinement of X-ray structures. Preparation of the
Cdata
o r r e and
s p o nresults
d i n g c ofor
s i g the
n a t oProtein
r y on b eDatah a l f oBank does not constitute
f the following individuals: ca. 1990
a significant burden for authors.
The remarks section of the Data Bank file provides plenty of space for any qualifying or expand-
U. Aebi, Basel, Switzerland W.E. Brown, Carnegie Mellon, USA
ing Agarwal,
R.C. remarksIBM, thatYorktown,
the authors USAmay wish to make.R.M. TheBurnett,
standard author
Columbia U.,comment
USA that a structure
The Low Resolution Problem
How can one obtain a “good” crystal
structure at low resolution?
There are generally more independent reflections than
torsion angles at 5 Å resolution
Problems:
• poor observable to parameter ratio
• electron density maps may be difficult to interpret
• potential model bias
Need for powerful reciprocal and real space methods
Outline • Overview of DEN-refinement • MR for a new structure at 3 Å (high B-factors) • Refinement at 7.4 Å resolution
Outline • Overview of DEN-refinement • MR for a new structure at 3 Å (high B-factors) • Refinement at 7.4 Å resolution
Target Function For Macromolecular Refinement
macromolecular structure
knowledge
Etotal = Egeometric + wx-ray Ex-ray + wmolecule Emolecule
improved
chemical → improved phases → improved
model knowledge electrondata
density maps
diffraction
in order to achieve a good fit to the data. If the cutoff
isDEN (Deformable
added Elastic Network)
from the reference model.Restraints
The reference mo
18 13
modeling and should have good local geometry
Etotal = Egeometric + wx-ray Ex-ray + wDEN EDEN
The elastic network energy term comprises a sum ov
4p
E DEN (n) = ! (d ij − d (n))
0
ij
N pairs i, j
where
Reference is theinter-atom
modeld➔ij selected distance between
distances atom
➔sparse pairs
distance i and
network d0ij(n)j i
• randomly selected atom pairs within a specified distance range,
the corresponding equilibrium distance at DEN upda
and separated by a specified interval in the primary sequence
• typically ~ 1 distance restraint per atom
corresponding distance of the reference model. The e
Schröder, Levitt, Brunger, Nature 2010
in order to achieve a good fit to the data
Implementation Of The DENfrom
is added Method
the reference model. The
Several (typically 10) macrocycles consisting of: 18 13
modeling and should have good lo
• torsion angle refinement (slow-cooling molecular
Chapter 3. Principles of Protein Structure.
dynamics)Use restricted to students enrolled in Chem130/MCB100A
• default option: restrained grouped B-factor refinement UC Berkeley, Fall 200
• default option: last two cycles
The peptide aregroupwithout shown DEN restraints
in Figure 3.4A is said to
Target function: The elastic network energy term compr
be in the trans conformation because the two C! atoms
are on opposite sides of the peptide bond (the C! atoms
4p
E
total =E
geometric C! aatoms
+w E
MLare on theDENsameDEN
+w
are said to be staggered). The alternative, in which the
E
side of theDEN
peptide bond
E ij (n) =0
ij ! (d − d (n))
FIGURE
i, j PRO?STRUCT??V
(Figure 3.4C), is referred to as the cis conformation,N pairs
During slow-cooling cycles,
and periodic
is rarely seenupdates
because itof
brings the R groups, some
the DEN equilibrium distances
of which are performed
are quite bulky, into close contact.
ij where d is the distance between atom
Y
F
3.2 the corresponding equilibrium
The backbone torsion angles # (phi) and $ distance
(psi) determine the conformation of the protein Y
γ: deformation factor (adjustable)
wDEN: weight for DEN restraints
chain
(adjustable) corresponding distance of the reference
Because of the planarity of the peptide groups, a F
κ: damping factor (usually setbackbone
protein to 0.1)only has freedom to rotate at the Y
drefij: distances from reference model
points where updated every six
the peptide groups meet (Figure 3.5). torsion angle molecu
These swivel
doij(n): current DEN restraint points are at each C! atom of the peptide
distances by by single bonds to the F
dij: current distances ofchain. The C! atom is connected
refinement model
carbonyl carbon (C) and the amide nitrogen (N) of the
Multiple trials for each (γ,same residue. Rotation about the N–C! and C!–C
wDEN) parameter pair and temperatureij d 0 (n + 1) = (1− κ )dij0 (n) + κ [γ dij
bonds define two torsion angles, denoted # and $,
Trial with the best Rfree isrespectively.
used for By subsequent steps
convention, positive rotation in # and
where the parameter κ specifies the spe
How Does It Work? • DEN is a general refinement method that guides torsion angle molecular dynamics by restricting it to reasonable conformational changes (e.g., differences between homologous models or motions) • DEN introduces information that is specific for the particular starting structure (e.g., homology model) • Degree of deformation is determined by γ • γ=0: no deformations of reference distances allowed • γ=1: deformations track refined model (akin “jelly body”) • 0
Some Notes And Considerations • General modes • New refinement: initial model = reference model • Re-refinement: initial model ≠ reference model • Special options • DEN-restraints active throughout • wDEN=0 ↔ torsion angle simulated annealing (without DEN) • Implementation in CNS 1.3 and Phenix.refine (development) • Resource for grid search: SBGrid Science Portal (www.sbgrid.org)
Outline • Overview of DEN-refinement • MR for a new structure at 3 Å (high B-factors) • Refinement at 7.4 Å resolution
Structure Of Cgl1109 (JCSG HP3342)
(A Putative Succinyl-Diaminopimelate Desuccinylase
(dapE) From C. Glutamicum)
Collaboration with D. Das, A.Deacon, J.Grant,
T. Terwilliger, R. Read, P. Adams, M. Levitt, G. Schröder
• Anisotropic diffraction data to ~ 3 Å resolution
• MAD data available, but experimental map was not easily
interpretable
• Weak homology to known structures (1vgy with 25% identity)
• One of the cases used by DiMaio et al. (2011)
• MR solution for both 1vgy and Modeller model of Cgl1109
(Phaser: RFZ=3.2, TFZ=9.9, LLG=75, Rcryst=0.65)DEN Refinement
+ Automated Model Building With AutoBuild
• Reference model = initial model = MR solution
(search model: Modeller homology model)
• Standard DEN protocol: but restrained individual
B-factor refinement
• AutoBuild with “morphing” and “rebuild in place”Optimization Of DEN Parameters
optimum
Rfree
100.0
DEN wDEN
Rfree
10.0
Weight
0.475
0.470
w
0.465
1.0 0.460
0.455
0.450
0.445
0.0
0.0 0.2 0.4 0.6 0.8 1.0
Parameter γ
γLarger Radius Of Convergence For DEN Vs. Standard Refinement
Standard refinement (Rfree=0.52) vs. final (orange) Standard ref. + AutoBuild (Rfree=0.48) vs. final (orange)
DEN (Rfree=0.44) vs. final (orange) DEN + AutoBuild (Rfree=0.42) vs. final (orange)DEN-Refinement Produced Better 2mFo-DFc Maps
Than Standard Refinement
Standard ref. (blue) + AutoBuild (cyan) DEN (blue) + AutoBuild (cyan)
Orange: final modelDEN+Autobuild Map Showed How To Correct The Model
Magenta: model after first round of DEN+AutoBuild
Cyan: corresponding DEN+AutoBuild 2mFo-DFc map
Orange: final modelFinal Model
Rfree 0.258
Rcryst 0.234
Ramachandran favored 92.7%
Ramachandran outliers 0.8%
Molprobity score 2.41 (96th percentile)
Final 2mFo-DFcResults For The Refinement Of Cgl1109 • Synergism between DEN-refinement and AutoBuild • Improved model ➔ improved phases ➔ better starting point for AutoBuild • Semi-automated completion of the structure with AutoBuild
Outline • Overview of DEN-refinement • MR for a new structure at 3 Å (high B-factors) • Refinement at 7.4 Å resolution
Refinement Of Photosystem 1 (PSI) At 7.4 Å Resolution
Collaboration with Henry Chapman,Thomas White (CFELS, DESY),
Petra Fromme, Raimund Fromme (Univ. of Arizona)
• DEN-refinements using synchrotron (ALS) data of a PS1 crystal in harvesting
buffer used for nano-crystal generation (Chapman et al., Nature, 2011),
dmin ~ 5 Å
• Data truncated to 7.4 Å to make it comparable to FEL (LCLS) data of PS1
• Increasingly scrambled starting models (Cα rmsd to target 1JB0: 0, 2.3,...,4.4 Å)
• Omitted three Fe-S clusters (for validation)
• Reference model = starting model
• Standard DEN protocol, except DEN active throughout, sparse random selection
between all atoms, and only overall B-factor refinementIncreasingly Scrambled Starting Structures Green: PS1 (PDB ID1JB0) Magenta: maximum scrambled structure (Cα rmsd: 4.4 Å, right most helix displaced by 5.4 Å)
Molecular Replacement (7.4 Å Resolution)
Phaser TFZ Score Phaser LLG Score
30.0 400
22.5 300
LLG
TFZ
15.0 200
7.5 100
0 0
1 2 3 4 5 6 1 2 3 4 5 6
Starting Structure Starting Structure
Solutions were found for all starting structuresComparison Of Refinement Protocols
Rfree RMSD To Target
0.55 7
0.50 6
RMSD (Å)
5
0.45
Rfree
4
0.40
3
0.35
2
0.30 1
0.25 0
1 2 3 4 5 6 1 2 3 4 5 6
Starting structure
Average Significance () of FeS Peaks
12
overall rigid body refinement 9
standard refinement
6
torsion simulated annealing refinement
DEN-refinement 3
secondary structure restrained refinement
0
1 2 3 4 5 6
Starting structureResults For Maximally Scrambled Starting Structure
Mg ions are shown as spheres
DEN (orange) vs. starting (magenta) DEN (orange) vs. target 1JB0 (green)
DEN-refinement shifted helices by 5.5 ÅResults For The Maximally Scrambled Starting Structure
Mg ions are shown as spheres
DEN 2mFo-DFc map (blue) target 2mFo-DFc map (blue)
vs. target 1JB0 (green) vs. target 1JB0 (green)
1.5 σ contour levelDEN-Refinements With Two Helices Omitted
(Chain F, Residues 103:126 - Yellow)
DEN-refinement starting from 1JB0 DEN-refinement starting from max. scrambled
DEN mFo-DFc map (mesh) DEN mFo-DFc map (mesh)
vs. target 1JB0 (green and yellow) vs. target 1JB0 (green and yellow)
3 (orange), 2.5 (blue), 2 σ (light blue) contour levelsResults For The Maximally Scrambled Starting Structure:
Standard Refinement
Mg ions are shown as spheres
standard refinement (teal)
vs. target 1JB0 (green)Results For Maximally Scrambled Starting Structure:
Standard Refinement
Mg ions are shown as spheres
standard refinement 2mFo-DFc map (blue)
vs. target 1JB0 (green)
1.5 σ contour levelDEN-Refinement At 7.4 Å Resolution Is Beneficial
Compared to overall rigid body, standard, or simulated
annealing refinement, DEN-refinement
• moves closer to the true structure
• recovers information not included in the refined
model - more significant features in difference
mapsOverall Conclusions For DEN-Refinement • DEN-refinement is a general method that can produce better models than standard refinement or simulated annealing protocols • DEN-refined model phases are a better starting point for (automated) model building • Most benefit is for starting structures that are far from the true structure, low-resolution, or high B-factor cases
Acknowledgments
DEN refinement
Gunnar Schröder, Michael Levitt
Cgl1109/HP3342 structure determination
Debanu Das, Tom Terwilliger, Randy Read,
Paul Adams, et al.
DEN-refinement of photosystem 1
Henry Chapman, Thomas White,
Petra Fromme, Raimund Fromme, et al.
SBGrid Science Portal
Dan O’Donovan, Piotr SlizYou can also read
