Diffraction Methods in the Study of Protein

Dynamics and Enzyme Mechanism

Pro Gradu

Esko Oksanen
Laboratory of Organic Chemistry
University of Helsinki
 Contents

1. Introduction 3

2. Brief Summary of Crystallographic Structure

Determination 5

2.1. Notes on Electron Density Maps and their Interpretation 7

3. Time averaged dynamics 8

3.1. Crystallographic Temperature Factors 8

3.2. The TLS Model 10

3.3. Applications of TLS Analysis 11

4. Time-Resolved Crystallography 14

4.1. General Requirements for Time-Resolved Crystallography 14

4.2. Trapping Catalysis Intermediates by Cryo-Cooling 17

4.3. The Laue Method 19

4.4. Processing of Laue data 21

4.5. Analysis of Laue Data 23

4.6. Some Applications of Time-resolved Laue Crystallography 24

5. Solution Scattering 28

6. Comparisons Between Dynamics Information from NMR and

Scattering Methods 30

References 31

2
 1. Introduction

Enzymes are biological catalysts that are capable of increasing reaction rates by many
orders of magnitude. The ways in which enzyme achieve this is a key question in biology and
chemistry and is widely debated. Not surprisingly much effort has been put to understanding
enzyme action. Experimental methods such as fast kinetics, FTIR spectroscopy or crystallography
are supported by computational approaches.
Diffraction of X-rays, electrons or neutrons by single crystals has provided most of the
available information on the structure of biological macromolecules such as proteins and DNA. Due
to developments in protein expression, purification and crystallisation, the availability of high
intensity synchrotron radiation and also improved software and sufficient computing power, new
structures are solved at an ever increasing rate by X-ray crystallography1.
Any structural investigations of enzymes are usually preceded by extensive
biochemical studies. These include the characterisation of the reactions kinetics with either steady-
state or stopped or quenched flow methods. Information on the characteristics of the enzyme and its
inhibitors or activators is often helpful for successful crystallographic experiments.
The traditional approach to study reaction mechanism with diffraction methods has
been to crystallise the enzyme of interest in complexes with competitive inhibitors that are substrate
or transition state analogues. Site-directed mutagenesis has also been used to investigate the role of
the active site residues. These studies have contributed to the understanding of enzyme action and
form a solid basis for mechanistic postulations.
One limitation of traditional X-ray crystallography is that the information obtained
from a typical crystal structure is mainly of a static nature. This is precisely the reason why stable
analogues of reaction intermediates are typically used in crystallographic studies to obtain snapshots
along the presumed reaction path. The time scale of a normal diffraction experiment with
monochromatic radiation and the oscillation method ranges from minutes at third generation
synchrotron sources such as the ESRF (European Synchrotron Radiation Facility, Grenoble, France)
to days with conventional X-ray generators. The electron density maps from such experiments
represent an average over the molecules in the crystal and over the time scale of the experiment.
When an atomic model is refined against the observed structure factor amplitudes, information
about the dynamic behaviour is incorporated into the crystallographic temperature factors. The
analysis of these temperature factors yields information about the mobility of different regions of

3
the molecule. Atomic resolution diffraction data allows the refinement of anisotropic temperature
factors, from which even more information can be extracted2. Such high resolution data also allow
the refinement of several alternative conformations that are discernible in the crystal. The
alternative conformations may also shed light on mechanisms.
In contrast to these methods in which the dynamics is observed indirectly, time
resolved diffraction methods can also be used3. One possibility is the white beam Laue technique. It
makes use of the polychromatic or ‘white’ X-radiation available from a synchrotron source. Using a
white beam allows a much larger region of reciprocal space to be covered by one image, allowing a
structure to be determined from only one image. Such high time resolution makes it possible to e.g.
monitor the structural changes during an enzymatic reaction. Another possibility to conduct time-
resolved crystallographic experiments is to monitor the progress of a reaction within a crystal with
such as UV-VIS or FTIR spectroscopy and cool the crystals at different identifiable states for data
collection by the normal monochromatic oscillation method4.
Various techniques apart from crystallography are available for the investigation of
protein dynamics in general and enzymatic catalysis in particular. They are not covered here in
depth, merely mentioned. The information obtained by these methods is often essential for the
validation of crystallographic data, for its interpretation and planning further experiments.
Time-resolved solution scattering methods can also be used to study biomolecular
dynamics, although not at atomic resolution. Small angle scattering experiments with X-rays can
provide time resolved information if performed at high flux sources, such as third generation
synchrotrons. The incoherent scattering of neutrons can also be used to identify thermal disorder in
proteins, thus complementing the time-averaged data available from crystal structures.
Indirect structural information on the active site during catalysis may also be obtained
by spectroscopic methods. The most traditional is electronic (UV-VIS) spectroscopy, which
unfortunately yields little specific chemical information. Specific data on individual vibrational
modes is available through FTIR spectroscopy. The use of difference spectra enables the extraction
of the relevant signals, but their identification and correlation to structural changes remains
somewhat problematic. Also many other spectroscopic techniques with reasonable signal-to-noise
ratios can be used for time resolved experiments, such as EXAFS, EPR or resonance-Raman
spectroscopies.
Nuclear magnetic resonance (NMR) spectroscopy is a very useful technique for
studying the dynamic behaviour of biomolecules, including domain movements and conformational
freedom of individual side chains. Even though NMR is widely used for the study of biomolecular
dynamics, it will not be discussed except for purposes of comparison to scattering techniques.

4
 The variety of computational methods used to study the dynamics of biological
macromolecules is very large. Classical molecular dynamics is a widely used method to simulate
protein dynamics, but it fails to describe processes involving bond formation or cleavage, such as
enzyme action. Quantum chemical calculations can be very helpful for a more thorough
understanding of enzyme catalysis, since the available experimental methods can rarely provide
specific information on electronic structure. The recent developments in both software and
hardware have made quantum chemical calculations feasible on systems large enough to be
biochemically relevant. Even the incorporation of nuclear quantum effects like tunnelling is
becoming possible. Accurate structural information, however, remains a prerequisite for the use of
such methods.

2. Brief Summary of Crystallographic Structure

Determination

The principles of crystallography are well covered in various textbooks1,5,6,7,8. This

summary is intended only to refresh the memory of the reader. The reader is strongly urged to
consult the literature for a more detailed exposition.
The scattering of X-rays from an infinite, periodic object, such as a crystal (the
number of molecules in a normal crystal is so large that it may be considered infinite for most
practical purposes), is non-zero only if the scattered waves interfere constructively. The periodicity
of a crystal allows the definition of a lattice of identical points, connected by lattice vectors. These
vectors are linear combinations of basis vectors, called a, b and c, that define the unit cell. A
central concept in crystallography is the reciprocal lattice, in which the basis vectors a*, b* and c*
are related to the real lattice basis vectors by a• a* = 1, b• b* = 1 and c• c* = 1.
If the incoming wave is represented by a wave vector k and the scattered wave by k’,
the scattered intensity will be non-zero only if k’-k = h, where h is a reciprocal lattice vector ( h a*,
k b* , l c* ), such that h, k and l have integral values. This is known as the Laue condition (not to be
confused with Laue crystallography). By considering the crystal as a set of planes, one arrives at an
equivalent formulation known as the Bragg law; nλ = 2d sinθ, in which λ is the wavelength, θ the
angle between the reflecting plane and the incoming beam and d the interplanar spacing.

5
 Figure 1 The Bragg law

A common way of describing a diffraction experiment is the Ewald construction. The

point Y corresponds to the origin of the reciprocal lattice, point X to an arbitrary reciprocal lattice
1
point h,k,l and A to the X-ray source. The radius of the sphere equals . A reciprocal lattice point
λ
satisfies the Laue conditions, i.e. is in diffracting position only at the surface of the Ewald sphere.

Figure 2 The Ewald construction

The experimental data in crystallography consists of intensities of reflections indexed

according to the values of h, k and l. The scattering of X-rays is due to electrons, and the electron
density of the crystal can be reconstructed by Fourier synthesis. Unfortunately the phase of the
Fourier components, called structure factors F(h,k,l), is not measured in the intensities, which are
proportional to | F(h,k,l)|2. Since the phases of reflections dominate the outcome of the Fourier
synthesis, they must be obtained in some indirect manner. The most common methods for tackling

6
the phase problem include multiple isomorphous replacement, single or multiple wavelength
anomalous diffraction and molecular replacement. These have been well documented in the
literature1,7,8,9,10.
An atomic model is constructed based on the electron density maps and the atomic
coordinates and temperature factors (B-factors) are refined to fit the experimental data as well as
possible. The temperature factors model thermal motion and static disorder in the crystal.

2.1. Notes on Electron Density Maps and their Interpretation

The structural information from a crystal structure comes from the interpretation of
the electron density maps calculated by Fourier synthesis. The appearance of these maps, which
determines their interpretability, is dominated by the phases. The phases cannot be directly
measured and the error in the phases is difficult to estimate. Another source of error is the fact that
the electron density represents an average over the time for data collection and over all the
molecules in the crystal that are considered identical (i.e. related by crystallographic symmetry
operations). The phases are normally calculated from an atomic model, where static and dynamic
disorder is modelled with B-factors. In truly atomic resolution structures, alternative models can be
used and their occupancies refined, but this is usually limited in scope to individual side chains or
loops.
A problem associated with phases calculated from a model is model bias. Since the
phases dominate the appearance of the electron density map, incorrect features present in the model
may appear in the map. For this and other reasons simple Fourier maps are rarely used in
crystallography. Various kinds of difference maps, often denoted Fo-Fc-maps, are used to judge the
consistency of the model and the observed amplitudes. Fo stands for observed structure factor
amplitudes and Fc for amplitudes calculated from the model. A Fourier synthesis with these
coefficients produces a map in which positive density indicates features present in the data, but not
in the model, whereas negative density indicates features that are present in the model, but not in
the data. Since the phases (from the model) are not perfect, such difference maps tend to be rather
noisy and are often difficult to interpret. Difference maps can also be calculated from amplitudes
originating from two different crystals. Such difference maps have been used to locate inhibitors
bound to enzymes by measuring data with and without soaking the crystal in a solution of the
inhibitor and calculating a difference map with the phases from the free enzyme.

7
 The concept of crystallographic resolution in has a precise meaning; the smallest
observed Bragg plane separation dmin. A common misconception is that the quality and reliability of
a structure would be determined solely by the resolution. Since the quality and interpretability of the
maps is dominated by the phases, it is the quality of the phases that is often crucial. It is also
possible that some regions of the molecule are disordered and hence poorly or not at all modelled.
This can occur even in structures considered to have high resolution.

3. Time averaged dynamics

Since the time scale of a typical monochromatic diffraction experiment is typically

hours or days and the time scale of protein dynamics is femtoseconds to milliseconds, the crystal
structure represents an average over time and the molecules in the crystal. It turns out that this
information can be extracted and used for a better understanding of protein dynamics and function.

3.1. Crystallographic temperature factors

Atoms in a crystal are not perfectly ordered, but exhibit both static and dynamic
disorder. The static disorder results from the breakdown of crystal symmetry, i.e. not all atoms are
in equivalent positions. Dynamic disorder is due to the vibrations of atoms (and larger aggregates of
atoms, such as entire molecules) around their equilibrium position. The two different sources of
disorder cannot normally be separated, as this would require temperature dependent measurements9.
The way in which this disorder is modelled in the crystal structure is by B-factors, also known as
Debye-Waller-factors.
Assuming the scattering of X-rays is centred on atoms, the structure factor of a

reflection with indices h,k,l is F ( h ) = ∑ f j e

2π i ( h•x j )
, where fj represents the atomic scattering factor,
j

h is the reciprocal lattice vector h,k,l and xj is the coordinate vector of atom j. The B-factor is an
2
exponential attenuating factor associated with each atomic scattering factor f B = f 0 e − B (sinθ / λ ) , where

θ is the scattering angle and λ the wavelength. The B-factor is related to the mean square

displacement (MSD) <u2> of the atom by B = 8π 2 u 2 . In typical protein structure refinements, the

atomic displacements are assumed to be isotropic, i.e. the probability distribution is spherical. In

8
reality the movements of atoms rarely are isotropic; anisotropic disorder can be modelled by
replacing the B-factor, which is a scalar (or more formally a tensor of zeroth rank), with a 3×3
matrix (or tensor of second rank). A 3×3 matrix has nine elements but, since it is a tensor, only six
elements are independent10. The anisotropic displacement matrix Uj of atom j is symmetric and its
diagonal elements represent the variances of each coordinate value and the off-diagonal elements
 u12 u1u2 u1u3 
 
the respective covariances. U j =  u1u2 u 2
2 u2u3  The elements of the U-matrix are referred to
 u1u3 u2u3 u32 

as anisotropic displacement parameters (ADPs). In small molecule crystallography anisotropic Us
are routinely used, since the data extends to atomic resolution and the five additional parameters to
be refined per atom are not a problem. In proteins, however, the resolution is usually ‘near atomic’
and the number of parameters needed for anisotropic Us easily exceeds the number of observed
reflections. Therefore only structures with resolution higher than 1.2 Å are usually refined with
anisotropic Us11.
The anisotropic displacement parameters are usually represented graphically by
ellipsoids defined by the eigenvectors (or principal axes) of the U-matrix. Since the U-matrix is a
tensor, it has to be positive definite and hence its eigenvectors do define an ellipsoid instead of other
conics.

Figure 3 A loop region from a serine protease, illustrating anisotropic displacement ellipsoids

In macromolecular crystallographic structure refinements the values of parameters are

usually restrained based on prior chemical knowledge. The restrains used for ADPs for instance in
the program SHELXL12, which is often used in atomic resolution refinements, ensure that the

9
dynamic behaviour of bonded atoms is similar. For example the shape and direction of the thermal
ellipsoids are restrained to be similar (the DELU and SIMU restraints in Figure 4). For waters and
hydrogens (which can be modelled in high resolution structures) it is usual to impose isotropic
restrains to avoid fitting physically unreasonable models.

Figure 4 The restraints used for anisotropic displacement parameters in the program SHELXL2,33

3.2. The TLS model

The static and dynamic disorder present in crystals is rarely manifested in random,
non-correlated movements of individual atoms. Instead, movements of entire molecules or
relatively rigid domains within molecules are more likely. It is possible to describe any movement
of a rigid body with three matrices13,14; T (for translation), L (for libration) and S (for screw). T and
L matrices are symmetric, whereas S is not. The application of these matrices allows the calculation
of anisotropic displacement factors. The difference between the values calculated for the TLS
model and individual atoms is minimised by varying the elements of the T, L and S matrices, thus
resulting in a best fit of the rigid body displacement parameters. Incidentally, the same formulation
can also be used in refinements which do not allow individual anisotropic displacement parameters
to be fitted. This so called TLS-refinement15 fits rigid body displacement parameters to individual
molecules or domains (as defined by the user), thus drastically reducing the number of parameters
to be fitted. Whichever way the TLS parameters are obtained, one should always try to assess
whether the rigid body assumption actually is valid and physically reasonable. One obvious way is
to visualise the principal axes of the matrices. The rigid domains can also be identified, either prior
to TLS analysis, or afterwards for its validation, by analysing interatomic differences in
displacements. If two atoms A and B belong to a rigid body, their displacements along the

10
interatomic vector should be the same. The difference ∆AB of the displacements should therefore be
zero within experimental error. The n×n matrix, where n is the number of atoms formed by these ∆-
values is known as the ∆-matrix16. For even a small protein the calculation of the full ∆-matrix is
neither feasible nor very helpful.

Figure 5 Displacements of two atoms belonging to a rigid body

Consequently the ∆-matrix is usually calculated for suitable subsets of atoms, such as Cα atoms.
Even including only Cα atoms of a protein would result in a huge matrix, so the ∆ values are binned
to represent larger groups of atoms. Once the validity of the TLS model has been assessed, the
biological relevance of the pseudo-rigid body motion can be interpreted with more confidence. It is
also important to take into account the effect of the restraints on ADPs in the structure refinement.

3.3. Applications of TLS Analysis

Although the underlying theory has been known for decades and is widely used in
small molecule crystallography, the applications to protein crystallography are mostly quite recent.
This may be due partly to the lack of atomic resolution data with anisotropic displacement factors
and partly to the lack of software to easily perform such analysis.
One early example is the study on ribonuclease A at 1.45 Å resolution17, in which the
movements of side chains as well as the entire molecule were analysed with the TLS model. While
the movement of the whole molecule was mostly translational and isotropic, the side chains on the
surface had librational movements. In the protein core the motion followed more that of the
environment and hence was mostly translational.
Arginine kinase catalyses a reversible phosphoryl transfer between adenosine
triphosphate (ATP) and arginine. A crystal structure of the enzyme was solved for a complex with
ADP (adenosine diphosphate) and nitrate which mimic the postulated trigonal transition state18. The
1.2 Å resolution of the structure allowed the refinement of ADPs and fitting of a TLS model of four
domains as rigid bodies. The domains were identified from the ∆-matrix. The movement of these

11
domains is largely librational, while the domains with catalytically important residues have least
mobility. The binding of the transition state analogue also restricts the domain movements
Calmodulin is a calcium-binding protein involved in regulation of other proteins. A
structure of calmodulin with Pb2+ instead of Ca2+ is significantly more disordered and diffracted
only to 1.75 Å resolution instead of 1.0 Å19. Despite this a TLS model revealed relatively large
anisotropic librational movements of the two TLS-groups in the N-terminal domain not observed in
the Ca2+ containing structure. These movements constitute a complex opening-closing domain
motion. This view is also supported by analysis of ∆-matrices.
Phospholipase A2 is an enzyme that hydrolyses the 2-acyl ester bond of 1,2-
diacylglycero-3-phospholipids. The bovine pancreatic enzyme has been solved at 0.97 Å resolution
but no TLS analysis was performed20. The bacterial enzyme from Streptomyces violaceoruber was
solved at 1.05 Å resolution and the TLS analysis of the bovine enzyme was performed and
compared to the bacterial enzyme21. This structure is remarkable because the data collection was
done in room temperature with a conventional rotating anode generator. Most crystals require cryo-
cooling and synchrotron radiation to diffract to atomic resolution. Therefore the displacement
factors in this structure are more due to thermal motion than positional disorder. The bovine protein
was, however, collected at a cryotemperature, and yet the disorder appeared at the same residues.
This suggests that the disorder is broadly similar in cryo and room temperatures, which is
encouraging as most crystals are cryocooled for data collection. An interesting comparison is also
made between the information obtained from an NMR and a crystal structure. For comparison
purposes an anisotropic displacement factor was calculated from the NMR ensemble with the

(  )
formula U NMR = r − r  r − r  , where ri is the i-coordinate of an atom in one conformer
ij i i  j j 

of the ensemble, < ri > is its average and i and j are x, y or z. Most of the residues deemed to be
mobile in NMR had high displacements in the crystallographic model, even though the magnitude
of the displacement was smaller in the crystallographic model. This would indicate that the
molecule has more flexibility in solution than in the crystal. However, the N-terminal region,
believed to be involved in substrate binding, that was mobile in NMR, did not show signs of being
disordered in the crystal. However, the structure is also different in the crystal and in solution, so it
is assumed to be an example of preferential crystallisation of one conformer.
Human fibroblast growth factor 1 was solved at 1.1 Å resolution and refined with
anisotropic temperature factors22. Rigid domains were identified from the ∆-matrix and TLS
analysis performed. The domain definitions were then optimised by repeating the TLS analysis with

12
various definitions and selecting the best fitting one. The β-strands in the C-terminus seem to move
as a rigid body, while the N-terminal β-strands move more or less independently.
The chaperonin GroEL is a large protein assembly that mediates protein folding. The
co-chaperonin GroES participates in the ‘catalytic’ cycle, which is driven by the hydrolysis of ATP.
In a study that combined ‘chemical trapping’ with TLS refinement23, GroEL was crystallised alone,
with ATPγS (unhydrolysable analogue of ATP), ADP and GroES, ADP-AlFx and GroES (a
transition state analogue of ATP hydrolysis). Even though the resolutions of the structures range
from 2.0 to 3.0 Å, the TLS parameters fitted in structure refinement are consistent between the
structures and improved the model. Furthermore, the directions of the rigid body movements are
consistent with the differences of domain orientations between the structures, and help to from a
picture of the dynamics of the complex.

Figure 6 Schematic representation of the domain movements in the GroEL-GroES system. The black line
represents an unfolded polypeptide, the GroEL domains are in red, green and blue and GroES in orange.

The bacterial light-harvesting complex II (LH2) is an assembly of proteins,

bacteriochlorophyll and carotenoids. It is an integral membrane complex that captures photons and
transfers the excitation energy on to a photosynthetic reaction centre. A crystal structure was solved
at 2.0 Å resolution and a TLS refinement significantly improved the quality of the model24. The
domain motions are consistent with spectroscopic results and represent an improvement over
stochastic models of thermal motion.

13
 These examples illustrate the potential of the rigid body TLS model in extracting
meaningful dynamic information from static crystal structures. While atomic resolution data is
clearly advantageous, it is not strictly necessary for the use of the TLS model, thus enabling studies
of large complexes or membrane proteins. In these cases the importance of the validation of the
model and checks of physical reasonability are particularly pronounced.

4. Time-Resolved Crystallography

Time-resolved crystallography attempts to obtain structural information in time scales

shorter than that of the dynamic process of interest. This dynamic process is often an enzymatic
reaction, but it may well be any other change of atomic coordinates. It should be noted that
transition states in enzymes can not be observed with crystallographic techniques. A transition state
corresponds to a saddle point in the multidimensional potential energy surface of a reaction. A
saddle point is a potential energy maximum along the reaction coordinate and a minimum along all
other coordinates. For a chemical reaction the reaction coordinate usually corresponds to some
vibrational mode of the system. Hence the lifetime of a transition state is in the order of one
vibrational period, typically in the picosecond scale. In fact only very specialised experimental
methods using femtosecond laser pulses are capable of directly observing transition states25.

4.1. General Requirements for Time-Resolved Crystallography

If one wants to follow the course of an enzymatic reaction within a crystal with a
given time resolution, one of the first questions is that of synchronisation. A fairly recent review by
Stoddard26 addresses this question thoroughly. In order to observe a single, distinct species in the
crystal, that species has to have ~90 % occupancy. This imposes the requirement that the reaction
needs to be triggered by a process with a rate constant larger than that of the actual reaction.
A clear distinction should be made between reversible and irreversible photochemical
processes. Proteins such as bacteriorhodopsin (bR) or photoactive yellow protein (PYP) undergo
photocycles, which can be repeatedly observed in one crystal27. Such multiple turnover experiments
allow better signal-to-noise ratios as the data can be collected and averaged over multiple cycles.
Another well-studied example is that of carboxymyoglobin. The CO molecule bound to the heme
may be dislodged by laser photolysis, after which it will diffuse back to the binding site via various

14
intermediates in time frames ranging from pico to milliseconds28. On the other hand, most
enzymatic reactions are not reversible in a way that would allow repeated observations from one
crystal.
Single turnover experiments are less straight forward to perform, but are nonetheless
possible. The majority have been performed using photolabile substrate precursors. For very slow
reactions diffusion of the substrate into the crystal can be used. According to Stoddard, the
saturation of a protein crystal of typical size with substrate requires 15-100 seconds. This would
require that kcat of the enzyme would be < 10-3 s-1 in order to achieve synchronisation, which is very
slow for an enzyme in solution. However, the reaction rate in the crystal is usually slower than that
measured in solution. Crystallographic flow cells can be used to measure the actual steady-state
kinetics in a crystal29. If some particular intermediate along the reaction pathway is long-lived
enough, it can be visualised, as with cytochrome c peroxidase. The structure of the doubly oxidised
state, so called compound I, could be investigated by saturating the crystals with hydrogen peroxide
in a flow cell. The presence of this intermediate for some 30 minutes was verified by
microspectrophotometry30.
In the case of trypsin a pH jump has been used to study the acyl-enzyme complex
formed between Ser195 and p-guanidinobenzoate. The raising of pH in the crystal allows the
putative nucleophilic water molecule to move in position and this process was observed with Laue
crystallography31.
By far the most common method to synchronise enzymatic reactions in crystals is by
photolysis of a photolabile substrate precursor. Such precursors are often termed ‘caged’
compounds. Modern, commercially available lasers can routinely produce pulses with nanosecond
duration. With optical parameter oscillators wavelength tuneability and narrow band pass are
achieved. Even femtosecond laser pulses can be produced, so the excitation has very high time
resolution. Once an appropriate photolabile precursor compound is available, certain requirements
have to be met for successful time-resolved diffraction experiments. Firstly the rate of the
photolytic cleavage has to be as fast or faster than the enzymatic reaction to be studied. Secondly
the quantum yield of the process needs to be high enough, so that essentially all the molecules in the
crystal can be cleaved with a single laser pulse. Finally the wavelength of the excitation radiation
should be easy to produce and should not cause significant damage to the protein.
The most common way of obtaining photolabile precursor compounds is to synthesise
a (2-nitrophenyl)-2-ethyl (or o-nitrophenylethyl) derivative. 3,5-dinitrophenyl and similar
derivatives have also been used.

15
 hν 266 nm
+ H+
O R O R
+
NO2 N O
O

O R H+
ROH
N
O O +
O NO

aci-nitro intermediate
Figure 7 Photolysis mechanism of a o-nitrophenylethyl ester

The photolysis of o-nitrophenylethyl esters occurs generally in the millisecond regime

with high quantum yield32. The mechanism presented in (Fig. 7) is a minimal mechanism, and
recent studies by time-resolved IR spectroscopy and quantum chemical calculations suggest that
photolytic mechanism is in fact more complicated33. The photolysis rates of o-nitrophenylethyl
caged compounds have usually been measured from the decay of the UV absorption due to the aci-
nitro intermediate. This is based on the assumption that the decay rate is equal to the rate of release
of the caged compound. The more recent studies cast doubt on this assumption. In addition the pH
dependence of the photolysis rate is more complicated than previously thought.
A relatively simple and robust synthetic methodology is available for the synthesis of
o-nitrophenylethyl esters; examples include ‘caged’ phosphate34, ‘caged’ nucleotides, ‘caged’
pyrophosphate35, cholin, the hydrolysis product of the neurotransmitter acetylcholine36 and the Ca2+
channel blockers nifedipine and nisoldipine37.

H2N-NH2, H+ MnO2

O EtOH N NH2 +
N N
CHCl3
NO2 NO2 NO2

RO-
+
N N 1:1 H2O-CHCl3 O R
NO2 NO2

Figure 8 Synthesis of o-nitrophenylethyl esters

16
 It is also possible to genetically incorporate photolabile amino acids, such as o-
nitrobenzyl cysteine, into proteins using specifically engineered tRNA/aminoacyl tRNA-synthetase
pairs38. This would allow facile photochemical synchronisation of enzyme reactions dependent on
cysteine residues.
The major drawback of o-nitrophenylethyl cages is the short wavelength needed to
achieve high quantum yield. These UV wavelengths are strongly absorbed by proteins, causing
radiation damage and heating problems (especially relevant in cryocrystallographic applications).
The absorption also limits the maximum thickness of a sample in which the photolysis can be
performed. This problem is of course more pronounced at the high protein concentrations that occur
in crystals. The photolysis rates of o-nitrophenylethyl compounds are in the millisecond range,
which is a result of the various ‘dark’ steps occurring after the absorption of a photon. Compounds
photolysable with longer wavelengths and with faster photolysis rates have been developed, mainly
to release amino acid neurotransmitters like glycine or β-alanine. The caging groups include 2-oxo-
1,2-diphenylethyl39 (or desyl), 2-methoxy-5-nitrophenyl40 (MNP) and 2-(dimethylamino)-5-
nitrophenyl41 (DANP) moieties. The synthesis of such compounds is more complicated than for o-
nitrophenylethyl esters, and as the caging groups become bulkier and less water soluble, their
usefulness for triggering reactions in crystals decreases.

4.2. Trapping Catalysis Intermediates

A common approach to study the structure of intermediates along the catalytic

pathway of an enzyme is to prepare indefinitely stable chemical analogues of the intermediates.
These analogues can then be crystallised with the enzyme and data collected with the
monochromatic oscillation method. This approach is referred to as ‘chemical trapping’42. For
instance, in the case of dihydrofolate reductase six compounds were used to emulate the catalytic
intermediates and transition states43. An another example is inorganic pyrophosphatase, for which
the substrate complex is captured by inhibiting the hydrolysis with fluoride and two different
alternate conformations are observed for a high resolution product complex with phosphate. These
two conformations are thought to correspond to an immediate and a relaxed product complex44.
Additional mutant studies of the active site support the model of catalysis45. These studies illustrate
the range of information available from chemical trapping. The major problem associated with this
approach is the assumption that the stable analogues are indeed analogous to actual intermediates

17
and transition states. Hence the ultimate verification of such mechanistic postulations has to come
from structures of actual intermediates.
An alternative way of increasing the lifetime of reaction intermediates is by ‘physical
trapping’, which usually refers to reducing the temperature so much, that the intermediates are
stable in the time scales required for diffraction data collection. Two different approaches are
available, known as ‘freeze-trapping’ and ‘trap-freezing’. In freeze-trapping the reaction initiation is
performed photochemically at cryogenic temperatures. The reaction is greatly slowed down, and
specific intermediates can be accumulated by temporarily warming the crystals. Thymidylate
kinase, which catalyses the phosphorylation of TMP (thyminosine monophosphate) to TDP
(thyminosine diphosphate) was investigated with this method46. The appearance of a phosphate
group next to the TMP molecule could be observed. Trap-freezing refers to a technique in which the
reaction in the crystal proceeds at room temperature, whatever the method of synchronisation may
be, and the crystal is flash-cooled to a cryogenic temperature at a specific time point. The catalytic
mechanism of the hammerhead ribozyme (catalytic ribonucleic acid molecule) has been studied
with a trap-freeze approach47. The hammerhead ribozyme cleaves itself, requiring divalent metal
ions like Mg2+ for catalysis. The reaction does not occur at low pH, and it is fairly slow. Therefore
both pH and Mg2+ concentration can be used to trigger the reaction, and several intermediates could
be accumulated and captured by flash-cooling. It should be noted that the time resolution of this
method is ultimately limited by the rate of cooling of the crystal. For a large protein crystal it may
take almost 1 s to reach cryogenic temperatures48. In some earlier studies of intermediate trapping
the crystals were cooled to moderate temperatures to slow the reactions down. The mother liquor
was kept liquid by the use of cryoprotectants. Examples include a study on the serine protease
elastase, in which the formation of an acyl-enzyme intermediate could be observed49.
All the approaches outlined above can be combined with the use of specific mutants
designed to slow the reaction down at some steps and increase the lifetime of intermediates. Should
it not be possible to extend the lifetime of the intermediates enough to allow normal monochromatic
data collection, faster crystallographic methods, such as Laue diffraction have to be applied. Such
trapping approaches may also make it possible to perform time-resolved neutron crystallography
experiments50. Neutron crystallography allows the direct observation of protons, which are not
usually visible even in high resolution X-ray structures. As protons are crucial for many enzymatic
reactions, this would be a valuable addition to the repertoire of available methods. The experimental
difficulties involved are significant, but it is possible to solve neutron structures at cryogenic
temperatures51.

18
 4.3. The Laue Method

Laue crystallography refers to single-crystal diffraction experiments with

polychromatic (or ‘white’) X-rays. Although the very first X-ray diffraction images by Friedrich,
Knipping and von Laue were obtained by white X-radiation, most crystallography since then has
been performed with monochromatic X-rays. One of the major reasons was that the spectrum of X-
rays available from conventional generators has strong peaks due to the electronic transitions of the
anode material. The most commonly used has been the copper Kα edge at 1.54 Å. With
synchrotrons this problem is no longer relevant, since the emission from a synchrotron radiation
source is polychromatic.

Figure 9 Ewald construction for a Laue diffraction experiment

Figure 5 illustrates the principle of the Laue method in reciprocal space. An Ewald
sphere is associated with every wavelength present in the incoming radiation, Rmin corresponding to
the longest and Rmax to the shortest wavelength present. In reality these can be ‘soft’ rather than
‘hard’ limits. The reciprocal lattice point P comes to reflecting position at the wavelength
corresponding to a sphere with radius R. The sphere dmax is the resolution sphere, determined by
crystal quality, which also limits the number of reflections observed. In principle all the reciprocal
lattice points in the volume limited by Rmin, Rmax and dmax are simultaneously in reflecting position.

19
 Figure 10 A polychromatic Laue diffraction pattern

The first Laue images of protein crystals were obtained in 1984 at CHESS (Cornell
High Energy Synchrotron Source), Ithaca, USA52. A serious limitation to the feasibility of Laue
crystallography has been difficulty of indexing and integrating the reflections from Laue images.
The difficulties result from the large number of overlapping reflections in the images, which need to
be deconvoluted somehow. Even the mere indexing of Laue images requires knowledge of either
the unit cell dimensions or the crystal orientation. Fortunately the cell dimensions are always known
from previous single-wavelength experiments. The problem of harmonic overlaps, i.e. reflections
with indices (nh ,nk ,nl), plagues Laue crystallography. However, according to the analysis of Ren
et. al.53, the problem is only serious at low resolution. This results from the fact that even though
reflections may in theory be in reflecting position, they might not in practice be measurable. In
particular the high resolution reflections at short wavelengths, which are most likely to produce
harmonic overlaps, are not easily measurable, thereby reducing the number of overlaps. Fig. 11
illustrates the probability distribution of measurable reflections; another observation from this
figure is, that the Bragg angle acceptance θa is smaller than the Ewald construction alone would
suggest.

20
 Figure 11 The probability distribution of measurable reflections (for the ID09 beamline at the ESRF)

The lack of the low-resolution terms has an adverse effect to the connectivity of the
electron density maps, which hampers their interpretation. The advent of harmonic deconvolution
methods have alleviated this problem54.
Experimental facilities for time-resolved Laue crystallography exist at various
synchrotron sources, such as the APS (Advanced Photon Source) in Argonne, USA or the ESRF
(European Synchrotron Radiation Facility) in Grenoble, France55. The necessary software for data
processing is also available, the two most notable software packages being the Daresbury
Laboratory Laue Software Suite56 and LaueView22,26.

4.4. Processing of Laue data

The difference between a Laue diffraction experiment and a standard monochromatic

experiment is that polychromatic X-radiation is used. The coherent scattering of X-rays by a
perfect, infinite crystal is observed only if the Laue conditions are satisfied. The wavevector is

21
wavelength dependent, its length being uniform in a monochromatic experiment. Hence only a
limited number of diffraction spots is observed in any given image, as only a few scattering vectors
coincide with reciprocal lattice points. The success of the oscillation method requires sufficient
separation of the spots, which allows fairly straight forward indexing and integration of the
reflections. The disadvantage is that a fairly large number of 2D images are needed to cover the
asymmetric unit of reciprocal space resulting in relatively long data collection times.
If, however, a wide variety of scattering vector lengths is present, a much larger
number of reflections will satisfy the Laue conditions. This results in much more crowded images
than with monochromatic radiation.
The extraction of reflection intensities from the diffraction images consists of three
steps. The first step is indexing. After the positions of diffraction spots (usually spot centroids) are
determined, the smallest possible reciprocal lattice basis vectors and the crystal orientation are
fitted. The subsequent refinement of these diffraction parameters along with e.g. detector distance
yields the information required for the next step, integration. Knowledge of the crystal orientation,
unit cell dimensions and X-ray wavelength allows the prediction of reflection positions. The
integration of intensity around these predicted spot centroids gives the raw reflection intensities.
The background intensity is also estimated around the spots. In the next step, data reduction,
different correction factors, such as Lorenz, polarisation or absorption corrections are then applied
to the intensities. The intensities are then scaled to absolute scale.
The above applies to monochromatic data as well, but for Laue data additional
correction steps are required. Both the actual scattering and the response of detectors depend on
wavelength57. The scaling stage involves therefore wavelength dependent correction factors. This
wavelength normalisation is critical for the structure determination14. There are multiple methods
for obtaining the so called λ-curve, which describes the wavelength dependence of the source,
optics and detector.58 This can also be used to deconvolute the harmonic overlaps59. If a multiple
diffraction spot contains a contribution from the reflections F(h,k,l) and F(2h,2k,2l), they may give
rise to an observed intensity Iobs,a with wavelengths λ1a and λ2a, where λ2a = 2 λ1a. If the same
reflections give rise to an another multiple spot of observed intensity Iobs,b with wavelengths λ1b and
λ2b, then the observed intensities can be expressed as a pair of equations in terms of the ‘real’
I obs,a = g ( λ1a ) I1 + g ( λ2 a ) I 2
intensities I1 and I2: , where g(λi) are the wavelength normalisation
I obs,b = g ( λ1b ) I1 + g ( λ2 b ) I 2

factors. The deconvolution of n intensities is possible from n observations by solving a set of linear

22
equations like this. The accuracy of these deconvoluted intensities is therefore strongly dependent
on the quality of the wavelength normalisation.

4.5. Analysis of Laue Data

The systems studied with Laue crystallography are usually structurally very well
characterised by previous monochromatic experiments. Therefore the solution of the phase problem
is rather straight forward with molecular replacement, whereas the interpretation of the electron
density maps is less so. The conformational differences between the different intermediates may be
small and experimental noise in the map due to e.g. lack of low resolution terms, errors in
integrating the intensities or increased crystal disorder lower the signal-to-noise ratio of the map.
One strategy to avoid these effects is to perform crystallographic refinement against differences in
structure factor amplitudes instead of the amplitudes themselves60. This is advantageous because
much of the noise in the map results from phase error. If the difference map is calculated from two
independently refined sets of structure factors, the phase error will add up and reduce the signal-to-
noise ratio of the map. This may be avoided by using the same phase and an amplitude difference.
The map quality can be further improved with Bayesian weighting schemes for the amplitude
differences61. Another technique for the improvement of the signal-to-noise ratio is singular value
decomposition (SVD), which is another name for eigenvector analysis62. The time dependent data
in the data matrix A is expressed as a function of time-independent, orthonormal basis vectors in
matrix U, the time dependencies of the corresponding vectors in matrix VT and the singular values
in matrix S. A = USV T The singular values describe how much the corresponding vectors
contribute to the data, and can therefore be used to filter out the components that contribute only to
noise. A very useful property of the SVD method is that it identifies the largest independent
components that change during the time course of the experiment, thus yielding valuable
information about the mechanism.
Despite the sophisticated methods used to collect and analyse time resolved
crystallographic data, the interpretation of the electron density maps is far from trivial. Due to the
nature of the Fourier transformation the way in which experimental errors show in the maps is not
straight forward. Hence it is not always easy to distinguish real signals from noise, but despite these
difficulties time-resolved crystallography is capable of directly visualising reaction intermediates as
the reaction proceeds.

23
 4.6. Some Applications of Time-resolved Laue Crystallography

Much of the development of Laue crystallography has been done on systems which
enable multiple turnover. One of the best studied cases is that of the oxygen binding protein
myoglobin. It contains a heme prosthetic group which binds carbon monoxide with high affinity.
This CO molecule can be photolytically released and the structural relaxation along with the
migration of the CO observed with nanosecond Laue crystallography63,11. Time-resolved
crystallographic studies of myoglobin have also been used to validate molecular dynamics
calculations64. Figure 13 shows an example of difference electron density from a Laue experiment.
Photoactive yellow protein (PYP) is a photoreceptor of halophilic bacteria. The light-
induced isomerisation of the trans-4-hydroxycinnamyl chromophore causes conformational changes
that lead to signalling events. The hydrogen bonding network changes upon photoisomerisation,
which can be seen with Laue crystallography10. The SVD technique has been applied to PYP31,65.

OH

S
protein

Figure 12 4-hydroxycinnamyl chromophore

24
Figure 13 Difference Fourier maps of the heme region in myoglobin after various time delays. Positive
contours (density appearing due to photolysis) are shown in green and negative contours (density
disappearing due to photolysis) in red. A) 4 ns B) 1 µs C) 7.5 µs D) 50.5 µs E) 350 µs F) 1.9 ms

25
 One of the earliest Laue studies was on glycogen phosphorylase. The binding of the
substrate maltoheptose was visualised by calculating a difference map from the Laue data before
and after substrate presentation66. Later, a photochemically synchronised experiment was
performed67. The time scale was rather slow, as no phosphorylation of the substrate was observed
after 3 minutes, but a distinct phosphate group was seen in the difference map after one hour.
The Ha-ras p21 protein, which is a GTP (guanosine triphosphate) hydrolysing enzyme
involved in signal transduction, has also been studied by Laue crystallography68. The rate of GTP
hydrolysis is relatively slow, making time resolved studies less complicated. The reaction was
synchronised by the photolysis of ‘caged’ GTP. Even though the release of the substrate could be
confirmed and conformational changes were seen to take place upon hydrolysis, the difference
maps at the active site were inconclusive in the sense that the fate of the γ-phosphate could not be
tracked.
For γ-chymotrypsin, which is a serine protease, a photolabile trans-p-diethylamino-o-
hydroxy-α-methylcinnamate was bound to the active site serine69. The photolytic cleavage allows
the binding of an another inhibitor, 3-benzyl-6-chloro-2-pyrone. This process was observed by Laue
crystallography. Despite the additional disorder in the crystals caused by the photolysis,
conformational changes in the enzyme active site could be seen.

Figure 14 Reaction scheme for the photoinduced binding of 3-benzyl-6-chloro-2-pyrone

26
 Isocitrate dehydrogenase (IDH), which catalyses the oxidative decarboxylation of
isocitrate to yield α-ketoglutarate, was investigated by both steady-state and single turnover
experiments. The enzyme uses nicotineamide adenine dinucleotide phosphate (NADP+) as an
oxidant.

Figure 15 Catalytic scheme of isocitrate dehydrogenase

In the steady-state experiments, the lifetime of catalytic intermediates was increased by suitable
active site mutations in order to allow their observation by Laue crystallography70. The enzyme-
substrate (or Michaelis) complex was captured with the mutation Tyr160Phe and the oxalosuccinate
intermediate with Lys230Met.
Hydroxymethylbilane synthase (HMBS) is involved in the biosynthesis of porphyrins
(such as hemes) and catalyses the formation of hydroxymethylbilane from four molecules of
porhpobilinogen.

27
 Figure 16 a) Overall reaction catalysed by HMBS b) and c) further
metabolism of HMB d) catalytic cycle of HMBS

HMBS forms a sequence of enzyme-substrate complexes, labelled ES1 to ES4. The mutant
Lys59Gln shows an accumulation of the ES2 complex after some 2 min after reaction initiation.
The reaction was performed in a crystallographic flow cell under steady state conditions and
followed by Laue crystallography with millisecond exposures71.

5. Solution Scattering

The scattering of radiation (such as X-rays) by a non-periodic object is continuous in

contrast to the sharp diffraction spots from crystals. Since molecules in solution tumble rapidly, the
scattering by a molecule in solution is averaged over its rotational trajectory. Hence most atomic

28
resolution information is lost and only the general size and shape of the molecular object may be
inferred. Nevertheless even such low resolution information can be helpful in understanding
changes in quaternary structure or detecting large amplitude domain movements. Recent reviews on
solution scattering are available72,73.
The experimental data in solution scattering is the scattering curve as a function of
scattering angle. Often only the very low angle scattering is considered, even though the scattering
may extend to higher angles. The low angle data contains information about the size and shape of
the particle, whereas the high angle scattering results from the internal structure. In contrast to
diffraction from crystals, this diffraction data is rotationally averaged and extends to resolutions of
10-5 Å.
Various modelling methods are used to extract information from the solution
scattering curve. The most classical one is the Guinier plot which yields the radius of gyration of the
particle. More detailed structural information can be obtained by fitting molecular envelopes
expressed as an expansion in spherical harmonics to the experimental scattering curve74. Various
bead models can provide more detailed shape information. The volume defined by the maximum
particle diameter is considered as a lattice of points that may contain either protein or solvent. A
Monte Carlo simulation is then used to minimize the difference between the scattering calculated
from the bead model and the experimental scattering curve. Because the number of beads is much
larger than the number of parameters that may reasonably be fitted to the data, constraints must be
used in the simulation ensuring continuity and compactness of the model.
Many biologically interesting complexes involve a large number of protein or nucleic
acid components. Even if the crystal structures of the individual constituents is known, the entire
complex is usually difficult to crystallise and too large for NMR studies. If large amplitude
movements are involved in the dynamics, they may be restricted by crystal packing forces or
preferential crystallisation of only one conformer may occur. In such problems small angle
scattering data can be very useful. It is possible (although not at all trivial) to calculate a solution
scattering curve from an atomic model75. This allows the fitting of a rigid body motion model to the
experimental data. Such a model is basically equivalent to a TLS model discussed earlier, although
in the solution scattering case the rigid body postulate is more difficult to verify. However, since the
resolution of the data is much lower, it is likely that the postulate is valid for any motions that may
be resolved. An example case is aspartate transcarbamoylase, an enzyme that catalyses the first
committed step in the biosynthetic pathway of pyrimidines. The enzyme is a heterododecamer and
its allosteric regulation by various nucleotides is based on changes in quaternary structure. Solution
scattering measurements analysed with a rigid body model showed marked differences in the

29
quaternary structure in the crystal and in solution76. This shows that even though crystal packing is
unlikely to significantly change the structure of a folded protein, it may affect the interactions
between domains or subunits and constrain their motions.
Solution scattering measurements can also be performed in a time resolved manner.
As in diffraction experiments, the factor limiting the time resolution is often synchronisation. Many
of the same methods have been used, but diffusion is more practical in solution studies, since rapid
mixing devices may be used. High brilliance of the X-ray beam is required for millisecond time
resolution and therefore instrumentation for such experiments is available at synchrotron sources,
such as the ESRF (Grenoble, France) or ELETTRA77 (Trieste, Italy). Singular value decomposition
(SVD) methods similar to those described above can be used in the analysis of time dependent
solution scattering data. This technique allowed the characterisation of an intermediate state in the
refolding of the electron carrier protein cytochrome c78.
Data from solution neutron scattering experiments complements X-ray scattering data.
Scattering of neutrons is due to atomic nuclei instead of electrons, and consequently neutron
scattering lengths of elements can be radically different from X-ray scattering factors. The large
difference in the scattering lengths of 1H and 2H is particularly useful. 1H actually has a negative
scattering length, which means that the contrast difference between hydrogenated and deuterated
materials is pronounced. This allows contrast variation studies in which the scattering curve is
measured in solutions of different H2O/D2O ratios. The most remarkable application of solution
neutron scattering is the modelling of the ribosome based on solution scattering curves with X-rays
and neutrons and contrast variation by selective deuteration79. The results were later confirmed by
crystallography80.

6. Comparisons Between Dynamics Information from NMR and

Scattering Methods

NMR spectroscopy is based on the energy level difference that a magnetic field causes
between the spin states of nuclei. These spin states and coherences between them can be
manipulated with radiofrequency pulses. NMR spectroscopy is treated in various textbooks81. It is a
versatile method that yields information on dynamics in time scales from picoseconds to hours. The
investigation of dynamics in the faster regime is usually based on relaxation times. Rates of
exchange between conformations with different chemical shifts can be measured for instance by the
CPMG (Carr-Purcell-Meiboom-Gill) or R1ρ pulse sequences. These methods have been used to

30
study the active site dynamics of enzymes, such as cyclophilin A82 or the ribonuclease binase83. In
both cases the time scale of the active site motions during catalysis corresponded to the reaction
rates. Since the information obtained from NMR is time-averaged, it compares with the information
in crystallographic temperature factors, although NMR experiments are capable of charactering the
time scale of the motions. NMR does not provide information similar to time resolved
crystallography and scattering methods, where the progress of an enzymatic reaction can be directly
visualised.

References

1
Giacovazzo, C., Monaco, H.L., Artioli, G., Viterbo, D., Ferraris, G., Gilli, G., Zanotti, G., Catti,
M. Fundamentals of Crystallography, 2nd Edition, Oxford University Press (2002) Oxford, U.K.
2
Schneider, T.R. (1996) Proceedings of the CCP4 Study Weekend (eds. Dodson, E., Moore, M.,
Ralph, A., and Bailey, S.), SERC Daresbury Laboratory, Daresbury, U.K., 133-144
3
Moffat, K. Acta Cryst. (1998) A54, 833-841
4
Stoddard, B.L. Methods (2001) 24, 125-138
5
Drenth, J. Principles of Protein X-Ray Crystallography, 2nd Edition, Springer Verlag (1999) New
York, USA
6
Blow, D. Outline of Crystallography for Biologists, Oxford University Press (2002) Oxford, UK
7
McRee, D. Practical Protein Crystallography, 2nd Edition, Academic Press (1999) San Diego,
USA
8
Rhodes,G. Crystallography made crystal clear, 2nd Edition, Academic Press (2000) San Diego,
USA
9
Bürgi, H.B., Capelli, S.C., Acta Cryst. (2000) A56, 403-412
10
Sands, D.E., Vectors and Tensors in Crystallography, Addison-Wesley (1982) Reading, USA
11
Dauter, Z., Lamzin, V.S., Wilson, K.S., Curr. Opin. Struct. Biol. (1997) 7 681-688
12
Sheldrick, G.M., Schneider. T.R., Methods Enzymol. (1997) 276, 307-326
13
Cruickshank, D.W.J, Acta Cryst. (1956) 9, 754-756
14
Schomaker, V, Trueblood, K., Acta Cryst. (1968) B24,63-76
15
Winn, M.D., Isupov, M.N., Murshudov, G.N., Acta Cryst. (2001) D57, 122-133
16
Rosenfield, R.E., Trueblood, K.N., Dunitz, J.D., Acta Cryst. (1978) A34, 828-829
17
Howlin, B., Moss, D.S., Harris, G.W., Acta Cryst. (1989) A45, 851-861

31
18
Yousef, M.S., Fabiola, F., Gattis, J.L., Somasundaram, T., Chapman, M.S., Acta Cryst. (2002)
D58, 2009-2017
19
Wilson, M.A., Brunger, A.T., Acta Cryst. (2003) D59, 1782-1792
20
Steiner, R.A., Rozeboom, H.J., de Vries, A., Kalk, K.H., Murshudov, G.N., Wilson, K.S.,
Dijkstra, B.W., Acta Cryst. (2001) D57, 516-526
21
Matoba, Y., Sugiyama, M., Proteins: Struc. Funct. Genet. (2003) 51, 453-469
22
Bernett, M.J., Somasundaram, T., Blaber, M., Proteins: Struc. Funct. Bioinf. (2004) 57, 626-634
23
Chaudhry, C., Horwich, A.L., Brunger, A.T., Adams, P.D., J. Mol. Biol. (2004) 342, 299-245
24
Papiz, M.Z., Prince, S.M., Howard, T., Cogdell, R.J., Isaacs, N.W., J. Mol. Biol. (2003) 326,
1523-1538
25
Paik D.H., Yang D.S., Lee I.R., Zewail A.H., Angew. Chem. Int. Ed. Engl. (2004) 43(21), 2830-
2834
26
Stoddard, B.L. Pharmacol. Ther. (1996) 70(3), 215-256
27
Genick, U.K., Borgstahl, G.E., Ng, K., Ren, Z., Pradervand, C., Burke, P.M., Srajer, V., Teng,
T.Y., Schildkamp, W., McRee, D.E., Moffat, K., Getzoff, E.D., Science (1997) 275, 1471-1475
28
Bourgeois, D., Vallone, B., Schotte, F., Arcovito, A., Miele, A.E., Sciara, G., Wulff, M., Afinrud,
P., Brunori, M., PNAS (2003) 100(15), 8704-8709
29
Stoddard, B.L, Farber, G.K., Structure (1995) 3 991-996
30
Fülöp, V., Phizacerley, R.P, Soltis, S.M., Clifton, I.J., Wakatsuki, S., Erman, J., Hajdu, J.,
Edwards, S.L. Structure (1994) 2 201-208
31
Singer, P.T., Smalås, A., Carty, R.P., Mangel, W.F., Sweet, R.M., Science, (1993) 259, 669-673
32
Walker, J.W., Reid, G.P., McCray, J.A., Trentham, D.R., J. Am. Chem. Soc. (1988) 110, 7170-
7177
33
Il’ichev, Y.V., Schwörer M.A., Wirz, J., J. Am. Chem. Soc. (2004) 126, 4581-4595
34
Ferenczi, M.A., Holmsher, E., Trentham, D.R., J. Physiol. (1984) 352, 575-599
35
Oksanen, E., unpublished
36
Peng, L., Silman, I., Sussman, J., Goeldner, M., Biochemistry (1996) 35, 10854-10861
37
Morad, M., Goldman, Y.E., Trentham, D.R., Nature (1983) 304, 635-638
38
Wu, N., Deiters, A., Cropp, T.A., King, D., Schultz, P.G., J. Am. Chem. Soc. (2004) 126, 14306-
14307
39
Gee, K.R., Kueper, L.W., Barnes, J., Dudley, G., Givens, R.S., J. Org. Chem. (1996) 61, 1228-
1233
40
Ramesh, D., Wieboldt, R., Niu, L., Carpenter, B.K., Hess, G., PNAS (1993) 90, 11074-11078

32
41
Banerjee, A., Grewer, C., Ramakrishnan, L., Jäger, J., Gameiro, A., Breitinger, H-G.A., Gee,
K.R., Carpenter, B.K., Hess, G.P., J. Org. Chem. (2003) 68, 8361-8367
42
Moffat, K., Henderson, R., Curr. Opin. Struct. Biol. (1995) 5, 656-663
43
Sawaya, M.R., Kraut, J., Biochemistry (1997) 36, 586-603
44
Heikinheimo, P., Tuominen, V., Ahonen, A.K, Teplyakov, A., Cooperman, B.S., Baykov, A.A.,
Lahti, R., Goldman, A., PNAS (2001) 98, 3121-3126
45
Tuominen, V., Heikinheimo, P., Kajander, T., Torkkel, T., Hyytiä, T., Käpylä, J., Lahti, R.,
Cooperman, B.S., Goldman, A., J. Mol. Biol. (1998) 284, 1565-1580
46
Ursby, T., Weik, M., Fioravanti, E., Delarue, M., Goeldner, M., Bourgeois, D., Acta Cryst. (2002)
D58, 607-614
47
Scott, W.G., Murray, J.B., Arnold, J.R.P., Stoddard, B.L., Klug, A. Science (1996) 274, 2065-
2069
48
Teng, T., Moffat, K., J. Appl. Cryst. (1998) 31, 252-257
49
Ding, X., Rasmussen, B.F:., Petsko, G.A., Ringe, D., Biochemistry (1994) 33(9), 9285-9293
50
Blakeley, M.P., Cianci, M., Helliwell, J.R., Rizkallah, P.J., Chem Soc Rev. (2004) 33(8), 548-557
51
Blakeley, M.P., Kalb, A.J., Helliwell, J.R., Myles, D.A., PNAS (2004) 101(47), 16405-16410
52
Moffat, K., Szebenyi, D., Bilderback, D., Science (1984) 223, 1423-1425
53
Ren, Z., Bourgeois, D., Helliwell, J.R., Moffat, K., Srajer, V., Stoddard, B.L., J. Synchrotron
Rad. (1999) 6, 891-917
54
Ren, Z., Moffat, K., J. Appl.Cryst. (1995) 28, 482-493
55
Nieh, Y.P., Raftery, J., Weisgerber, S., Habash, J., Schotte, F., Ursby, T., Wulff, M., Hädener, A.,
Campbell, J.W., Hao, Q., Helliwell, J.R., J. Synchrotron Rad. (1999) 6, 995-1006
56
Campbell, J.W., J. Appl. Cryst. (1995) 28, 228-236
57
Sweet, R.M., Singer, P.T., Smalås, A, Acta Cryst. (1998) D49, 305-307
58
Ren, Z., Moffat, K., J. Appl.Cryst. (1995) 28, 461-481
59
Campbell, J.W., Hao, Q. Acta Cryst. (1993) A49, 889-893
60
Terwilliger, T.C., Berendzen, J., Acta Cryst. (1995) D51, 609-618
61
Ursby, T., Bourgeois, D., Acta Cryst. (1997) A53, 564-575
62
Rajagopal, S., Schmidt, M., Anderson, M., Ihee, H., Moffat, K., Acta Cryst. (2004) D60, 860-871
63
Srajer, V., Teng, T., Ursby, T., Praderwand, C., Ren, Z., Adachi, S., Schildkamp, W., Bourgeois,
D., Wulff, M., Moffat, K., Science (1996) 274 1726-1729
64
Hummer, G., Schotte, F., Anfinrud, P.A., PNAS (2004) 101(43) 15330-15334

33
65
Schmidt, M., Pahl, R., Srajer, V., Aderson, S., Ren, Z., Ihee, H., Rajagopal, S., Moffat, K., PNAS
(2004) 101(14), 4799-4804
66
Hajdu, J., Machin, P.A., Campbell, J.W., Greenhough, T.J., Clifton, I.J., Zurek, S., Gover, S.,
Johnson, L.N., Elder, M., Nature (1987) 329, 178-181
67
Duke, E.M.H., Wakatsuki, W., Hadfield, A., Johnson, L.N., Protein Sci. (1994) 3, 1178-1196
68
Schlichting, I., Almo, S.C., Rapp, G., Wilson, K., Petratos, K., Lentfer, A., Wittinghofer, A.,
Kabsch, W., Pai, E.F., Petsko, G.A., Goody, R.S., Nature (1990) 345, 309-315
69
Stoddard, B.L., Koenigs, P., Porter, N., Petratos, K., Petsko, G.A., Ringe, D., PNAS (1991) 88,
5503-5507
70
Bolduc, J.M., Dyer, D.H., Scott, W.G., Singer, P., Sweet, R.M., Koshland, D.E. Jr., Stoddard,
B.L., Science (1995) 268, 1312-1318
71
Helliwell, J.R., Nieh, Y-P., Raftery, J., Cassetta, A., Habash, J.,Carr, P.D., Ursby, T., Wulff, M.,
Thompson, A.W., Niemann, A.C., Hädener, A., J. Chem. Soc., Faraday Trans. (1998) 94(17) 2615-
2622
72
Koch, M.H., Vachette, P, Svergun, D.I., Q. Rev. Biophys. (2003) 36(2), 147-227
73
Svergun, D.I., Koch, M.H., Curr. Opin. Struct. Biol. (2002) 12, 654-660
74
Svergun, D.I., Volkov, V.V., Kozin, M.B., Stuhrmann, H.B., Barberato, C., Koch, M.H.J., J.
Appl. Cryst. (1997) 30, 798-802
75
Svergun, D.I., Barberato, C., Koch, M.H.J., J. Appl. Cryst. (1995) 28, 768-773
76
Fetler, L., Vachette, P., J. Mol. Biol. (2001) 309, 817-832
77
Amenitsch, H., Rappolt, M., Kriechbaum, M., Mio, H., Laggner, P., Bernstorff, S., J. Synchrotron
Rad. (1998) 5, 506-508
78
Segel, D.J., Elizier, D., Uversky, V., Fink, A.L., Hodgson, K.O., Doniach, S., Biochemistry
(1999) 38, 15352-15359
79
Svergum, D.I., Nierhaus, K.H., J. Biol. Chem. (2000) 275, 14432-14439
80
Schluenzen, F., Tocilj, A., Zarivach, R., Harms, J., Gluehmann, M., Janell, D., Bashan, A.,
Bartels, H., Agmon, I., Franceschi, F., Yonath, A., Cell (2000) 102, 615-623
81
Cavanagh, J. Protein NMR spectroscopy: priciples and practice, Academic Press (1996) San
Diego, USA
82
Eisenmesser, E.Z., Bosco, D.A., Akke, M., Kern, D., Science (2002) 295, 1520-1523
83
Wang, L., Pang, Y., Holder, T., Brender, J.R., Kurochkin, A.V., Zuiderweg, E.R.P., PNAS (2001)
98, 7684-7689

34