Feny? D. (Ed.) Computational Biology

Подождите немного. Документ загружается.

131

Computational Tools in Protein Crystallography

methods: molecular replacement (MR) and multi-wavelength

anomalous dispersion (MAD). The programs involved in structure

determination were ﬁrst developed for Unix platforms. These

programs have been redeveloped to run on Linux, and most of

them can now run on Windows and Mac OSX platforms. There

are different program packages available for different stages of

structure determination: data collection and processing, structure

solution, crystallographic reﬁnement, structure validation, and

documentation.

The diffraction data collected from exposure of crystals to X-rays

provide the amplitudes of structure factors. The phases cannot be

recorded by diffraction experiment, and hence, it is not possible

to generate an electron density map, which requires both ampli-

tudes and phases. Any strategy to determine the structure of a

molecule has to ﬁnd a solution for this “phase problem.” Once

protein crystals and X-ray diffraction data are obtained, several

methods can be used to address the phase problem (4).

The direct methods are mathematical tools which use the

probability theory and the assumption of approximate equal,

resolved atoms to estimate phases from the measured intensities

(5). Direct methods have been used to solve structures of small

molecules (typically a few hundred atoms). It is also possible to

solve structures of small proteins (250–1,000 non-H-atoms)

using these methods; however, it requires the resolution of X-ray

diffraction data to be better than 1.2 Å.

The experimental phasing of macromolecular structures is

achieved by the multiple (or single) isomorphous replacement

(MIR or SIR) method, which involves introducing heavy atoms

(like platinum, samarium, mercury) into the macromolecular

crystal. Heavy-atom derivatives are prepared by soaking the native

crystals in a buffer containing heavy-atom compound or by

cocrystallization. The requirement for this technique to work is

that the crystals of the native protein and that of the heavy-atom

derivative should be isomorphous, i.e., they should have the same

space group and cell dimensions. The heavy atom substructure is

then determined by difference Patterson, which is used to calcu-

late the phases. At least, two isomorphous derivatives are neces-

sary to unambiguously determine the phases. When one derivative

dataset is combined with the native protein dataset, the phasing

process is referred to as SIR. MIR is sometimes combined with

anomalous scattering of some of the heavy atoms, and the tech-

nique is then referred to as multiple isomorphous replacement

with anomalous scattering (MIRAS) or single isomorphous

replacement with anomalous scattering (SIRAS).

Another experimental phasing method is MAD (6, 7), which

uses anomalous scattering information from the datasets col-

lected at different wavelengths near the absorption edges of

1.1. Phasing Methods

132 Jain and Lamour

heavy metals. The protein structure to be determined should

contain anomalous scatterers like selenomethionine or metallic

centers. Different datasets on the same crystal are collected at

wavelengths where both the anomalous and the dispersive signals

are maximized. Also, this method requires tunable wavelength

available at synchrotron source. In some cases, a single dataset

collected at the single wavelength is sufﬁcient for structure solu-

tion by single wavelength anomalous dispersion (SAD) (8, 9).

SAD is attempted in cases where there is signiﬁcant radiation

damage preventing collection of several datasets on the same

crystal. Some elements such as Fe give signiﬁcant anomalous sig-

nal around CuKa wavelength; hence, a SAD dataset for a protein

containing Fe can also be collected on a rotating anode X-ray

generator. Most of the programs that work for MAD can also be

used with a SAD dataset.

Radiation damage induced by high-energy X-rays is usually

perceived as detrimental to the quality of diffraction data and

structure determination. Techniques are being developed to min-

imize this effect during data collection (10). However, a tech-

nique called radiation-induced phasing (RIP) is emerging which

takes advantage of site-speciﬁc radiation damage and the chemical

changes they induce in the protein structure (11) for phasing pur-

poses (12).

Molecular replacement (MR) uses the structural information

from a similar known structure. The calculated phases from the

available structure can be used as initial estimates for the target

dataset. This process is facilitated by thousands of crystal structures

deposited in the Protein Data Bank (www.rcsb.org), a number that

has seen a signiﬁcant rise in recent times owing to structural genom-

ics consortiums depositing coordinates of many proteins.

Currently, MR and MAD represent the two most com -

mon methods to determine the 3D structure of biological

macromolecules.

X-ray data can be collected on home sources on a rotating anode

or using the X-ray beam generated by a synchrotron radiation.

The quality of the data collected on synchrotron beam lines is

much superior to that collected on a traditional home source due

to the high ﬂux, better polarization, and tunable X-ray optics of

the beam resulting in a substantial increase in the signal-over-

noise ratio. However, the ﬂux available from home-source gen-

erators has also improved signiﬁcantly over the years allowing

routine crystal structure determination from good diffracting

crystals. Several generations of synchrotrons have been built in

the past decades around the world, and information on synchro-

tron X-ray sources can be found in the following link: http://

biosync.rcsb.org. Currently, the entire data processing can usually

be done at the beam line during or immediately after data collection.

1.2. Data Collection

and Processing

Packages

133

Computational Tools in Protein Crystallography

There are a number of software packages available for X-ray data

processing. These programs usually attempt to index a few initial

frames and provide the correct Bravais lattice and cell constants of

the crystal. Once the data collection is complete, they can merge,

reduce, and scale the data to provide a set of unique indices, their

measured intensities, and corresponding standard deviations. The

HKL suite is the most popular of them, and it consists of three

modules: XdisplayF for visualization of the diffraction pattern,

Denzo for data reduction and integration, and Scalepack for

merging and scaling of the intensities obtained by Denzo or other

programs (13). In the current package HKL2000, all these three

modules have been integrated into a single window. The program

provides graphs of mosaicity, B factors, completeness, and other

statistics during data processing to assess the quality of the

dataset.

MOSFLM (14) (also available in GUI format called iMosﬂm)

is also a data processing program that is part of the CCP4 suite

(15). It is coupled with the scaling program SCALA and can pro-

cess data from a wide range of detectors and writes out the merged

and scaled data in the CCP4 binary format – with the extension

.mtz and can be directly read by other modules of the CCP4 suite

for further analysis and structure determination. Mosﬂm/iMos-

ﬂm is available free of charge whereas HKL2000 is a commercial

package.

The program XDS, X-ray detector software (16), is available

free of charge and contains the following components: XDS for

processing single datasets, XSCALE for scaling datasets,

XDSCONV that converts the output ﬁle into desirable formats,

and VIEW for visualizing. d*TREK (available with Rigaku

detectors) is another software suite for data collection and

processing (17). It contains three components: dtcollect to set

up parameters for data collection, dtdisplay to display X-ray

diffraction images, and dtprocess, an interface for integration

and processing of images.

The starting point and the total range of data collection

depend on the orientation of the crystal and its symmetry. Several

programs can be used to determine the best oscillation range to

get a complete dataset using information about the crystal param-

eters (Bravais lattice, cell constant) and the geometrical relation-

ships between the crystal and the detector. For this purpose, one

can use the program Strategy included in HKL and Mosﬂm after

collecting and indexing one or a couple of sequential images. The

program BEST takes into account the diffraction anisotropy and

evaluates the optimal oscillation width to avoid overlapping reﬂec-

tions after collecting one or two images with a 90° angle (18, 19).

Most synchrotron beam lines offer some useful tips and guidelines

for data collection (example: CHESS synchrotron website: http://

www.macchess.cornell.edu/MacCHESS/collect_strategy.html).

134 Jain and Lamour

The number of molecules in the asymmetric unit (AU) is usually

estimated through the calculation of the Matthews coefﬁcient,

which calculates the solvent content of protein crystals using the

molecular weight of protein as a parameter. It has been observed

that the value of the Matthews coefﬁcient usually lies within the

range 1.6–4.5, and this value can be used to calculate the num-

ber of molecules in the AU (20, 21). Several websites offer an

online calculation (see Table 1). The Matthews coefﬁcient can

also be calculated directly in the MR module of CCP4 (cell con-

tent analysis) or CNS (matthews_coef input ﬁle).

1.3. Estimation

of the Cell Contents

Table 1

List of crystallography programs and related websites cited in this chapter

Program name Link

3dSS http://cluster.physics.iisc.ernet.in/3dss/

ARP/wARP http://www.arp-warp.org

Auto-RIckshaw http://www.embl-hamburg.de/Auto-Rickshaw/

BALBES http://www.ysbl.york.ac.uk/~fei/balbes/

BEST http://www.embl-hamburg.de/BEST/

CASTp http://sts-fw.bioengr.uic.edu/castp/

CATH http://www.cathdb.info/

CCP4i http://www.ccp4.ac.uk

CNS http://cns.csb.yale.edu

Coot http://www.biop.ox.ac.uk/coot/

CRYSTAL http://crystal.uvm.edu/ (Updated link to crystallographic softwares)

d*TREK http://www.rigaku.com/software/dtrek_news.html

DALI http://ekhidna.biocenter.helsinki.ﬁ/dali_server/

Dino http://www.dino3d.orghttp://bioserv.rpbs.jussieu.fr/cgi-bin/PPG

EPMR http://www.epmr.info/

HIC-Up http://xray.bmc.uu.se/hicup/

HKL suite http://www.hkl-xray.com/

HKL2MAP http://schneider.group.ifom-ieo-campus.it/hkl2map/index.html

JCSG validation suite http://www.jcsg.org/scripts/prod/validation/sv_ﬁnal.cgi

LIGPLOT http://www.biochem.ucl.ac.uk/bsm/ligplot/ligplot.html

(continued)

135

Computational Tools in Protein Crystallography

Table 1

(continued)

Program name Link

Matthews constant http://www.ruppweb.org/Mattprob/http://pldserver1.biochem.

queensu.ca/~rlc/pfd/links/calcs/vm_calc.shtml

http://csb.wfu.edu/tools/vmcalc/vm.html

Modeler http://salilab.org/modeller

Molprobity http://molprobity.biochem.duke.edu/

MolScript http://www.avatar.se/molscript/doc/index.html

MOSFLM and iMosﬂm http://www.mrc-lmb.cam.ac.uk/harry/imosﬂm/

MrBUMP http://www.ccp4.ac.uk/MrBUMP/

MSMS http://mgltools.scripps.edu/packages/MSMS/

O http://xray.bmc.uu.se/alwyn/A-Z_of_O/A-Z_frameset.html

PDB data deposit AUDIT http://deposit.rcsb.org/

PHENIX http://www.phenix-online.org

POVRAY http://www.povray.org/

PredictProtein http://www.predictprotein.org/

PRODRG http://davapc1.bioch.dundee.ac.uk/prodrg/

PyMOL http://www.pymol.org/http://www.pymolwiki.org/index.php/

Main_Page

RAPIDO http://webapps.embl-hamburg.de/rapido/

RASMOL http://www.bernstein-plus-sons.com/software/rasmol/

REFMAC http://www.ccp4.ac.uk/dist/html/refmac5.html

RIBBONS http://www.cbse.uab.edu/carson/papers/index.html#Ribbons

SHARP/autoSHARP http://www.globalphasing.com/sharp/

SHELX http://shelx.uni-ac.gwdg.de/SHELX/

SnB http://www.hwi.buffalo.edu/SnB/

SOLVE http://solve.lanl.gov/

Strategy http://www.crystal.chem.uu.nl/distr/strategy.html

Superpose http://wishart.biology.ualberta.ca/SuperPose/

WHATIF http://swift.cmbi.kun.nl/whatif/

XDS http://xds.mpimf-heidelberg.mpg.de/html_doc/downloading.html

136 Jain and Lamour

The correct choice of the search model is the main criterion to

ensure success in solving a structure using MR. The structural

homology of the search model with the target protein (i.e., the

root mean square deviation between the expected structure and

the known structure) and the primary sequence homology are two

of the main parameters considered for deﬁning the right search

model. Generally, the initial choice is based on sequence homol-

ogy, and the closer the target primary sequence is to the search

model (i.e., >25%), the higher the chance of ﬁnding a MR solu-

tion. When the protein sequence is divergent from the model and

it is known that the fold of the model and target protein is similar,

then the coordinate ﬁle can be transformed into a polyalanine chain

to reduce the errors introduced by side chains. Most MR modules

in crystallographic program packages offer this option. The poly-

alanine backbone can also be generated using programs such as

Moleman which can manipulate coordinate ﬁles (G.J. Kleywegt –

Uppsala Software Factory: http://xray.bmc.uu.se/usf/moleman_

man.html). To reduce the model bias, it is also advisable to set the

temperature factors (B factor) to the average value determined

from the Wilson plot (22) from the experimental data. Water mol-

ecules and other small ligands should be removed from the search

model.

If the structural homology of candidate search models is not

obvious, secondary structure prediction programs can be used to

predict the structure of the target protein to make a more informed

decision regarding the choice of the model. The PHD secondary

structure prediction program (23) uses evolutionary information

from multiple sequence alignments and is now available through

the PredictProtein website (24).

In addition, modeling the target structure using structures of

homologous proteins can be done to generate a search model.

One of the most popular program for this purpose is Modeler (25).

Large conformational changes in a structure might be an obstacle

for solving the structure by MR. Flexible and/or disordered loops

and small regions that could adopt a different conformation in the

target structure can be removed from the initial model before run-

ning a MR program. One must keep in mind that it will decrease

the completeness of the search model and might be detrimental to

MR, but every effort must be made to bring the structure of the

search model as close as possible to that of the target protein.

Most crystallographic packages offer one or several MR pro-

grams. Traditional programs are based on the Patterson func-

tion. An MR search is typically divided into the rotation and the

2. Molecular

Replacement

2.1. Choice

and Preparation

of the Search Model

2.2. Choice

of the Molecular

Replacement Program

137

Computational Tools in Protein Crystallography

translation functions, which are correlation functions between

observed and calculated model Pattersons. The self-rotation

function can be used to determine the orientation and nature of

noncrystallographic symmetry elements without the need of a

search model (26).

The CCP4 package includes a MR module with four different

programs: Phaser, BEAST, MOLREP, and AMoRe. BEAST (27)

and the improved version of Phaser (28) use a maximum-likelihood

objective function. AMoRe (29) carries out translation and rota-

tion search using the fast rotation function as well as a rigid body.

In case of a protein–protein or protein–DNA complex, traditional

MR programs such as AMoRe only allow searching for one mol-

ecule at a time. Once the ﬁrst one is found, the second molecule

of a complex can be searched. MOLREP (30) is an all-automated

program that takes into account the level of completeness and

identity of the search model. When looking for an oligomer, for

example, a dimer, either the monomer or the entire dimer can be

tried as a search model. Phaser allows the user to search for several

different molecules or protein domains in the AU at the same

time. The REPLACE suite of programs written by Liang Tong

also provides tools for MR: a rotation program GLRF and a trans-

lation program TF (31). Unlike the traditional MR programs,

EPMR is a program that does not separate the translation and

rotation steps but uses an evolutionary search algorithm simultane-

ously optimizing the orientation and position of a search model

(32). It is also referred to as a 6D MR. This program can be suc-

cessful for rather incomplete search models but requires more

computing time.

MR “pipelines” now provide an all-automated tool for faster

and user-friendly procedures integrating all steps, including the

choice of the search model. One such package, PHENIX, can

run Phaser using the AutoMR module and provides tools for

automated model building (RESOLVE). MrBUMP includes a

FASTA search using the target primary sequence to look for a

homologous structure (33). It is available through the CCP4i

interface and can run MR with Phaser or MOLREP. BALBES

also requires little to no user intervention during the entire MR

procedure (34) and can also be run through a CCP4i window.

A web interface has been developed for CaspR also starting from

a primary sequence and generating a homology model for the

MR search (35).

One has to remember that more than using a particular pro-

gram, a key step for MR is the choice of the search model, along

with careful selection of the resolution range for the search as the

quality of the data will greatly inﬂuence the outcome. The correct

estimation of the number of molecules in the AU is helpful, since

that allows the MR programs to know how many molecules of

the search model to look for in the AU.

138 Jain and Lamour

The MR solutions output from the search can be evaluated

with the help of three parameters: the R-factor, the correlation

coefﬁcient (CC), and the resulting packing in the target unit cell.

The possible solutions in the form of transformation matrices or

transformed coordinates and their corresponding R-factor and

correlation coefﬁcients are listed in the MR output ﬁles. The pres-

ence of a solution with signiﬁcantly higher correlation coefﬁcient

than the rest is an indication that the MR search has succeeded.

A few issues of Acta Crystallographica have been dedicated to MR

and give a more detailed review about all aspects of this method

(36, 37). A special chapter in Methods in Molecular Biology Series

also describes different strategies and programs for MR (38).

Modeler or other current modeling programs can be used

after running MR to substitute the correct primary sequence in

the MR model. Automated building programs such as ARP/

wARP (see Subheading 5) can be used for model building. When

only a few residues are different between the search model and

the query, amino acid substitution can be done directly in most

model-building programs (see Subheading 5). After a MR solu-

tion is obtained, electron-density maps can be calculated, and the

structure has to go through several rounds of model building and

reﬁnement to reduce the bias introduced by the starting model.

The procedures for crystallographic structure reﬁnement are basi-

cally the same for different phasing methods and are detailed in

Subheading 5 of this chapter. We provide in Table 2 an example

of structure determined by MR with the list of programs used at

each step.

The MAD method exploits the observation that when the heavy

atoms absorb X-rays, there is a violation of Friedels law and the

Bijovet pairs (h,k,l) and (-h-k-l) will have different intensities.

The difference in intensities is referred to as anomalous difference

(41). For most atoms, anomalous dispersion is negligible, only

select elements (Se, Hg, Au, Pt, Zn, W, Os, Br, lanthanides, etc.)

have their X-ray absorption edges in the energy range that are

accessible by synchrotron radiation. In addition, the presence of

natural metal centers containing iron, copper, and zinc can be

exploited to conduct a MAD experiment. Nucleic acids with bro-

mouracil residues can be used for determining their structures

and that of their complexes with proteins. Bound phosphate ions

have been replaced by tungstate (42, 43), which can provide good

anomalous signal. More recently covalent modiﬁcation of nucleic

acids has been done by solid-phase synthesis using selenium for

crystallographic phasing (44). In proteins, sulfur atoms show

3. Multi-

wavelength

Anomalous

Dispersion

139

Computational Tools in Protein Crystallography

Table 2

Steps leading to structure determination of Tth Gyrase ATPase domain/novobiocin

complex by molecular replacement

Programs and

packages used

Steps for solving

the structure Result

HKL2000 Data processing

and scaling

Data collected on a synchrotron beamline (ID14, ESRF)

P21 space group with a = 44.9 Å, b = 125.5 Å, c = 79.8 Å,

and g = 96.4°

Data scaled between 15 and 2.3 Å, Rsym = 6.4%

Solvent content (Matthews coefﬁcient) 51.5%

indicating two molecules/AU

AMoRe CCP4

Moleman

Molecular

replacement

Translation and

rotation search

Scalepack output ﬁle transformed into CCP4 format

for molecular replacement (MR) using AMoRe

Search Model: one monomer from the structure

of Escherichia coli 43 kDa ATPase domain

(PDB ID 1EI1); PDB transformed into a

polyalanine chain with Moleman due to low primary

sequence homology with the target protein; water

molecules, and ligand removed for the search (search

using the whole dimer failed)

Result: Two peaks corresponding to the two molecules

in the asymmetric unit; correlation factor 21.1

DM CCP4 Density

modiﬁcation

DM density modiﬁcation with non-crystallographic

symmetry (NCS) averaging

(self-rotation function gave position of dimer twofold

axis)

CNS Reﬂection ﬁle

formatting for

reﬁnement

Mtz reﬂection ﬁle transformed into CNS format (CCP4)

7.5% reﬂections put aside for free R

calculation for reﬁnement in CNS

O Model building Major conformational change between the two domains

of each monomer had to be adjusted manually and

the active site ATP lid had to be rebuilt

CNS minimization

and simulated

annealing

Reﬁnement, R

and R

free

values

monitoring

Iterative steps of model building and CNS reﬁnement

Map calculation after each major building step and

reﬁnement (2F

obs

− F

calc

, F

obs

− F

calc

)

B factors included after most of the model was

corrected

Novobiocin ligand ﬁt last and included in

reﬁnement

Addition of water molecules and ﬁnal round

of minimization

CNS Ramachandran

Plot and PDB

submission

formatting

Model validation

PDB submission

Manual adjustment of residues in disallowed regions

for ﬁnal PDB submission followed by a few cycles

of minimization

Final R = 20%; R

free

= 26%

An example of molecular replacement: Tth Gyrase 43 kDa ATPase domain in complex with the antibiotic novobiocin

(PDB ID 1KIJ)

Source: Lamour et al. (39, 40)

140 Jain and Lamour

weak anomalous signal, and more and more structures are being

determined using this property (45). Some other choices of heavy

atoms for structure solution have been described in Boggon and

Shapiro (46).

However, the most preferred method for MAD is the use of

Selenomethionine-substituted protein. SeMet protein is pre-

pared by standard protocols involving suppression of methion-

ine biosynthesis (47) or through the use of an Escherichia coli

strain that is auxotrophic for methionine. Most often, the crys-

tals of the derivatized protein grow from conditions nearly simi-

lar to the native crystals. In case there are problems getting

crystals of the SeMet-substituted protein, microseeding using

seeds from native crystals can be attempted. Amino acid or mass

spectrometry analysis can be done to conﬁrm the presence as

well as to determine the percentage of incorporation of SeMet in

the protein.

MAD relies on good-quality X-ray diffraction data collected at

three different wavelengths preferably from a single crystal.

Initially, a ﬂuorescence scan of the SeMet crystal is done to verify

the presence and location of the absorption edge for the crystal.

The ﬂuorescence scan can be analyzed with a program called

Chooch (48) to give the recommended wavelengths and esti-

mates of f ′ and f ″. Usually the peak wavelength is the maximum

in the ﬂuorescent scan. High energy remote is usually 150–

250 eV higher than the peak. The inﬂection value is halfway

down the f′ absorption edge, usually about 3 eV less than the

peak. It is advisable to collect the peak data ﬁrst so that the data-

set can be used to attempt SAD in case excessive radiation dam-

age does not permit data collection at other wavelengths.

A good MAD dataset should be highly redundant and complete

to accurately determine the anomalous differences. It is advis-

able to collect a low-resolution (~3 Å), high-redundant dataset

initially to solve the structure as compared to higher resolution

dataset that is not redundant. Data are usually collected in

wedges of 10–30° at two values of phi, which are 180° away

from each other (inverse beam method). This minimizes the

effect of radiation damage on the intensities of (h,k,l) and (-h,-

k,-l) by collecting them close in time. Since this method takes

twice as long to collect the data, for higher symmetry space

groups, data collection by the inverse beam method can be avoided.

While the data are being collected and processed, scale factors

and B factors should be monitored in the log ﬁle. These values

should change very gradually during entire data collection

reﬂecting how much the crystal is affected by radiation. Data for

each of the three wavelengths should be integrated and scaled

separately. Any of the packages mentioned above can be used for

data reduction and scaling.

3.1. MAD Data

Collection