research papers
MARTINI bead form factors for the analysis of timeresolved Xray scattering of proteins
^{a}Department of Chemistry and Molecular Biology, University of Gothenburg, Box 462, SE40530 Gothenburg, Sweden
^{*}Correspondence email: westenho@chem.gu.se
Timeresolved small and wideangle Xray scattering (SAXS and WAXS) methods probe the structural dynamics of proteins in solution. Although technologically advanced, these methods are in many cases limited by data interpretation. The calculation of Xray scattering profiles is computationally demanding and poses a bottleneck for all SAXS/WAXSassisted structural q < 0.75 Å^{−1}. The findings are promising for future attempts at Xray scattering data analysis, and may help to bridge the gap between timeresolved experiments and their interpretation.
and, in particular, for the analysis of timeresolved data. A way of speeding up these calculations is to represent biomolecules as collections of coarsegrained scatterers. Here, such coarsegraining schemes are presented and discussed and their accuracies examined. It is demonstrated that scattering factors coincident with the popular MARTINI coarsegraining scheme produce reliable difference scattering in the range 0 <Keywords: Xray solution scattering; proteins; coarsegraining; MARTINI; structural dynamics; smallangle Xray scattering (SAXS); wideangle Xray scattering (WAXS); protein structure refinement.
1. Introduction
Xray solution scattering is a popular technique for gathering structural information on biomolecules in solution (Petoukhov & Svergun, 2007; Koch et al., 2003; Svergun & Koch, 2003; Makowski, 2010; Ihee et al., 2010; Westenhoff et al., 2010; Andersson et al., 2009; Cho et al., 2010; Malmerberg et al., 2011; Kim, Muniyappan et al., 2012; Kim, Lee et al., 2012; Ibrahimkutty et al., 2011; Spilotros et al., 2012; Takala et al., 2014). The angular intensity distribution of scattered Xrays is recorded and advanced computational algorithms are available to determine threedimensional structures from the scattering patterns (Konarev et al., 2006; Petoukhov et al., 2012; Liu et al., 2012). Xray scattering at small angles (SAXS) provides information on molecular envelopes. At wider angles (WAXS), higherresolution information is encoded, but low scattering strength and a lack of uniqueness when assigning structural features to the data hinders its practical application.
Timeresolved Xray solution scattering is an emerging technique for observing structural changes of proteins (Ihee et al., 2010; Westenhoff et al., 2010; Andersson et al., 2009; Makowski, 2010; Cho et al., 2010; Malmerberg et al., 2011; Kim, Muniyappanet al., 2012; Kim, Lee et al., 2012; Ibrahimkutty et al., 2011; Spilotros et al., 2012; Takala et al., 2014). Xray scattering is recorded as a function of reaction time and referenced to the scattering patterns of the reactants. The difference technique makes it possible to access higher spatial resolution by detecting WAXS, since all background signals are very precisely canceled. At modern synchrotron facilities the time resolution is limited to approximately 100 ps, but freeelectron laser sources increase the resolution to <100 fs. This opens up the way for studies of elementary structural changes in proteins on the time scale of atomic motions.
Today the bottleneck in protein solution Xray scattering lies in interpreting the experimental data. One is forced to model it in an iterative fashion, and to calculate scattering patterns of a large number of trial structures. Since the total scattering is a result of pairwise interference between all the atoms in a protein, each such calculation is time consuming and
quickly becomes too computationally demanding.For a realistic representation of scattering from a molecule in solution, the contributions to the form factor from the electron density of the molecule, the electron density of the displaced solvent and any excess electron density of the solvation shell have to be evaluated. The first term is often computed from the atomic coordinates of the molecule. To represent the solvent displaced by the solute, it is common practice to use modified atomic scattering factors (Fraser et al., 1978). Whereas the use of this approximation is justified at small angles, systematic deviations are introduced for higher angles (Bardhan et al., 2009). The excess electron density in the solvation shell is often modeled as a homogeneous border layer, as implemented in the popular program CRYSOL (Svergun et al., 1995). However this strategy is problematic as two parameters describing the solvation shell are introduced and adjusted ad hoc. Recent developments include explicit solvent treatment (Grishaev et al., 2010; Park et al., 2009) or a more realistic representation of the solute–solvent boundary (Bardhan et al., 2009; Virtanen et al., 2011). By using these more sophisticated methods, the reliability q range can be extended to higher values. However these methods are computationally demanding and it is hard to prove their accuracy experimentally. We show explicitly below that the need for accurate computation of the solvent layer is relaxed when analyzing difference Xray scattering.
Given that solution scattering signals of proteins do not encode enough information to reveal atomic level details, a coarsegrained representation should often be suitable for interpreting SAXS and WAXS data. Such representations greatly decrease the computational cost of predicting Xray scattering curves (Yang et al., 2009; Stovgaard et al., 2010; Zheng & Tekpinar, 2011; Daily et al., 2012), rendering ambitious iterative schemes realistic for relatively large protein systems. In recent years, the MARTINI model, based on coarsegrained representations of biomolecules, has become popular for simulating the dynamics of various biological systems (Marrink et al., 2007; Monticelli et al., 2008; López et al., 2009; Yesylevskyy et al., 2010; de Jong et al., 2013; Marrink & Tieleman, 2013). It drastically reduces the computational cost of simulations, allowing simulation on longer time scales and of larger systems with modest computational resources. The force field is designed to reproduce thermodynamic data and has been successfully applied to several simulation problems, for example, protein–lipid interactions (van den Bogaart et al., 2011; Schäfer et al., 2011; Louhivuori et al., 2010).
In this study, we describe and compare methods for coarsegrained Xray scattering calculations, especially aiming at the analysis of timeresolved difference scattering. We show how difference Xray scattering profiles can be calculated efficiently from MARTINI coarsegrained representations of proteins and we assess the reliability limits for these calculations. We find that coarsegrained scattering calculations are reliable in a larger q range for difference scattering compared with absolute scattering. We conclude that this method opens up a way for structural routines of large proteins, especially in combination with timeresolved SAXS/WAXS experiments.
2. Theory and methods
2.1. Xray scattering from coarsegrained structures
The scattering amplitude from a collection of pointlike atomic scatterers (Warren, 1990) is described by
where q is the scattering vector and and f_{k} are the position and the scattering factor of atom k, respectively. For randomly oriented molecules in solution, the scattered intensity is obtained by multiplying this sum by its complex conjugate,
and then taking the spherical average with , where λ is the wavelength of the radiation, is the scattering angle and is the vector from scatterer l to scatterer k. Equation (2) becomes
The last result is known as the Debye equation and can be used to predict the vacuum scattering of a biomolecule.
A complication is that solution scattering patterns contain a large undesired solvent signal. This signal can be removed by subtracting a buffer background, but at the cost of including a negative term for the displaced solvent that must be accounted for in predicted data. It can be introduced in an approximate way, at the level of the atomic scattering factors, so that the unmodified Debye equation [equation (3)] can still be used. Such corrected atomic scattering factors f_{k}^{excl}(q) are derived from f_{k}(q) by subtracting a Gaussian sphere representing the scattering amplitude of the displaced solvent (Fraser et al., 1978):
Here is the tabulated (Fraser et al., 1978) volume for each atom and is the mean electron density of the bulk solvent. In the remainder of this paper, all scattering factors contain this displacedsolvent term unless stated otherwise.
For biomolecules, the computational cost for evaluating equation (3) can be quite high. This is especially important for iterative structural procedures where many test structures have to be evaluated. One strategy for decreasing computational cost is to use a coarsegrained representation of the structure, where each coarse bead represents a group of atoms. It is convenient if each bead is described by a scattering factor F(q), so that the scattering intensity is given by a coarsegrained Debye equation, where the indices (m, n) denote coarse beads and R_{mn} the distance between them:
Finding these F_{m}(q) is not trivial and we review some possibilities before describing the approach taken in this study.
2.1.1. The bead position approximation
The simplest way to express the overall scattering in terms of coarsegrained scattering factors is to consider each atom to be located at the center of its bead. All atom–atom distances are then approximated by the corresponding bead–bead distances. Where (k, l) are atomic indices and (m, n) denote coarse beads, the Debye equation can be written as follows:
We then have
In this approximation, the internal structure of the beads is completely ignored. We note that equation (7) holds exactly for q = 0.
2.1.2. The spherical `glob' approximation
Another option is to take, for each bead, the spherical average of the amplitude [equation (1)] before taking intensities. This is equivalent to smearing each atom out on a sphere of radius r_{k} around the center of the bead (Harker, 1953):
Here, the distance of each atom to the center of the beads is considered, but the angular arrangement of the atoms is ignored.
2.1.3. The selfconsistent set approximation
The most general approach described in this paper is to find, for a given set of proteins, a selfconsistent set of F(q) values that reproduces all pairwise bead–bead scattering intensity terms of the corresponding atomistic structures as well as possible. Considering two beads with scattering factors F_{A} and F_{B}, the total scattering intensity of the pair I_{AB} is given by equation (3). These quantities are related by applying equation (5) to the pair of beads:
If F_{B} is held constant in the comparison, F_{A} is obtained by solving this quadratic equation, choosing the correct root by comparing to equation (7), which actually holds for q = 0. A selfconsistent set of coarsegrained scattering factors can be found from the following scheme.
(a) Generate starting guesses for the bead form factors, for example, by using equations (7) or (8).
(b) Pick a random bead in the structure and call it A.
(c) Go through all other beads in the structure, letting each act as B, and calculate F_{A} for each case according to equation (9).
(d) Take the average of all these F_{A}, and assign it to bead A.
(e) Repeat (b)–(d) until the set form factors have converged.
2.1.4. The singlebead approximation
Although conceptually simple, the last approach is cumbersome, especially for large libraries of proteins. Yang et al. (2009) have presented a simpler approach, where form factors are chosen such that they reproduce the scattering intensities of isolated beads. In this approach, numerically correct coarsegrained form factors for entire amino acid residues can be obtained simply by taking the square root of the scattering intensity from a group of atoms:
We note that this equation can only produce positive form factors, which is not correct in general when the negative term for the displaced solvent is included. For q = 0, the value of the form factor of the bead must equal the sum of the atomic scattering factors:
With water as a solvent, f_{k}(q = 0) is negative for the hydrogen atom when the displacedsolvent contribution is included. Thus, negative values of F(q) can occur for beads containing hydrogen atoms [f(q = 0) = −0.72 electron units (e.u.)]. Therefore, form factors obtained using equation (10) that do not satisfy equation (11) must be corrected. In practice, this is the case for sidechain beads which contain only hydrogen and carbon atoms. When the beads consist of entire amino acids the f(q = 0) values are generally positive (Yang et al., 2009). To correct the scattering factors that do not satisfy equation (11), we use a common feature of all these scattering factors, which is the appearance of a minimum with two associated inflection points (Fig. 1). The data with q larger than the highq inflection point are used for a sixthorder polynomial fit, constrained at q = 0 to satisfy equation (11). This polynomial is then accepted as the actual form factor of the coarse bead (Fig. 1).
To illustrate the differences between the above four approaches, the coarsegrained Xray scattering of hen eggwhite lysozyme was calculated from a MARTINI (the MARTINI coarsegrained representation of proteins will be described later in §2.2) coarsegrained structure and compared with the allatom calculation (Fig. 2). To enable a direct comparison of the form factor calculation methods, the scattering was calculated without the displacedsolvent model. For q < 0.25 Å^{−1}, all approaches are in good agreement with the allatom calculation. This is reasonable because long interbead distances are probed in this q range. Regarding the highq region, the beadposition approximation shows significant deviations for q > 0.4 Å^{−1}. The spherical glob approximation reproduces the allatom calculation well for q < 1.2 Å^{−1} but deviates for larger q values. The selfconsistent set approximation and the singlebead approximation yield almost identical results and agree relatively well with the allatom calculations even for high q values.
This degree of agreement corresponds to the order with which the internal structure of the bead is accounted for: the beadposition approximation neglects the internal structure, the spherical glob approximation smears the atoms out on spheres around its center, and the singlebead approximation as well as the selfconsistent set approximation include the internal bead structure most accurately. The latter two approximations agree remarkably well with the allatom calculation. The singlebead approximation is computationally less expensive than the selfconsistent set approximation. We therefore chose to use the former for calculating coarsegrained scattering factors in the remainder of this study.
2.2. Application to the MARTINI model
We now turn our attention to applying the singlebead approximation to the MARTINI model (Marrink et al., 2007; de Jong et al., 2013; Marrink & Tieleman, 2013) as an efficient way to calculate Xray scattering from coarsegrained structures. In the MARTINI model, four nonhydrogen atoms and their associated hydrogen atoms are mapped, on average, onto one bead, with each amino acid residue composed of a backbone bead and up to four sidechain beads (Monticelli et al., 2008). The beads are grouped by their polarity and hydrogen bonding ability, yielding a total of 20 different bead types whose interactions are specified by the MARTINI force field.
For Xray scattering calculations the geometrical similarity and the . The original mapping of atoms into beads according to the MARTINI model is retained for the Xray scattering calculations. This means that MARTINI coarsegrained structures can be used directly as an input for these calculations.
(number of electrons) of the beads is of main importance, not the polarity or hydrogen bonding ability. We therefore derive Xray form factors for each MARTINI bead as it appears in every amino acid residue type. This yields the 49 different scattering types listed in Table 1

The average form factors for the MARTINI beads following the singlebead approximation were acquired from a rationally selected library of protein structures that covers a wide range of protein folds and different secondary structure contents (Oberg et al., 2003). We excluded seven proteins with missing nonhydrogen atoms. This resulted in a library of 43 proteins shown in Table 2. Missing hydrogen atoms were added with the pdb2gmx tool which is part of the GROMACS suite (Hess et al., 2008). To investigate the effect of the bead size on accuracy, an additional coarsegrained mapping with one amino acid per bead was used (Yang et al., 2009; Zheng & Tekpinar, 2011). Both the amino acid and the MARTINI beads were positioned at the centers of mass of the respective atom groups.

To keep these calculations simple and universally applicable a few structural details were ignored. First of all, the N and Ctermini were not differentiated for the coarsegrained calculation. The respective amino acids were included in the calculation of the average bead scattering factors, and thus the larger number of electrons for the Cterminus is reflected by this averaging. Charged atoms were also ignored, both to limit the number of bead types and since information on the charge is not always available.
3. Results
3.1. The library average of bead form factors
A central aim of this work is to estimate the reliability of coarsegrained Xray scattering calculations with respect to the size of the beads. Two coarsegrained mapping schemes were used: the amino acid mapping as used by Yang et al. (2009) and our MARTINIbead approach. A first comparison can be made at the stage of form factor averaging. The smaller the variation between the individual form factors being averaged, the greater the reliability of the coarsegrained scattering calculation. This is especially important for small proteins or proteins of unusual structure or amino acid content.
The bead scattering factors for all methionines in the library are shown in Fig. 3. The data were computed with equation (10). Deviations for the amino acid bead and the backbone MARTINI bead at q = 0 are the result of the inclusion of Nterminal amino acids, which contain two additional hydrogen atoms. Since the hydrogen scattering with the displacedsolvent correction is negative for low q, the corresponding bead scattering factors for low q are below the majority of the curves.
The calculated form factors for the amino acid beads show a large variation (Fig. 3a), whereas the form factors of the finer MARTINI beads are more homogeneous (Figs. 3b and 3c). When the two MARTINI beads (backbone and side chain) are compared, the backbone bead scattering factors are more heterogeneous, whereas the sidechain scattering factors are remarkably well represented by the mean value.
To identify the structural origin of the variation in bead scattering factors, we clustered the protein library based on their prevailing secondary structure. According to the classification of Oberg et al. (2003), the scattering factors of the proteins that show a high percentage of αhelices and βsheets are shown in different colors (Fig. 3). The MARTINI backbone bead scattering factors clearly cluster into two groups according to secondary structures. In contrast, the amino acid beads do not show such a clear picture.
3.2. Calculation of difference scattering from coarsegrained protein structures
The analysis of timeresolved WAXS experiments often requires repeated evaluation of difference scattering from many different trial structures and thus depends on fast but reliable scattering calculations over a q range up to approximately 1 Å^{−1}. In order to estimate the accuracy of difference scattering calculations from coarsegrained structures, the predicted difference scattering between the crystal structures of human deoxy hemoglobin (PDB code 2hhb; Fermi et al., 1984) compared with human carbonmonoxy hemoglobin (PDB code 1bbb; Silva et al., 1992), as calculated with allatom and coarsegrained methods, are shown in Fig. 4. This model system has already been used for timedependent Xray scattering studies and highquality data are available (Cammarata et al., 2008). The aminoacidbased result deviates considerably from the allatom calculation for q > 0.4 Å^{−1}, but the MARTINI coarsegrained calculation is accurate for q < 0.75 Å^{−1}. This finding is reasonable considering that the level of structural detail is highest in the allatom representation, reduced in the MARTINI coarsegraining scheme and lowest in the amino acid approach. The computation times for the three curves in general scale approximately as 50 (all atom):1 (MARTINI):0.2 (amino acid approach).
Comparing the model calculations with the experimental difference scattering from Cammarata et al. (2008) shows that the agreement between the structural model and the experiment is excellent for q < 0.4 Å^{−1}, but that the model fails for higher q values. This is most likely because the does not represent the solution structure of hemoglobin very well (Cammarata et al., 2008). It is obvious that the MARTINI coarsegrained calculations could be used for any algorithm to improve the agreement, but that the amino acid approach would not contain enough structural detail to achieve this. Conversely, such schemes are very likely to benefit from the reduced computational cost of the MARTINI representation relative to the atomistic scattering model.
The calculations shown in Fig. 4 include the displacedsolvent term, but any excess electron density of the solvation layer was neglected. This is reasonable when considering difference scattering, as errors cancel to some degree when taking differences. The data in Fig. 5 demonstrate this. The difference scattering for three systems is shown: sperm whale myoglobin (deoxy and carbonmonoxy state), human hemoglobin (deoxy and carbonmonoxy state) and deinococcus radiodurans phytochrome (Pr and Pfr state). These systems were selected since they cover a wide range of magnitudes in conformational change as shown by the rootmeansquare deviations of the respective structure pairs (cf. Fig. 5). The two solution difference scattering curves for each test system are computed by (i) considering the atomic scattering and the displaced solvent [as formulated in equation (4)], and (ii) additionally including the scattering due to the solvation layer. The data were calculated using CRYSOL (Svergun et al., 1995), with its highest resolution (L = 50) and default parameters for the solvation shell and displaced solvent (Svergun et al., 1995). The program approximates the solvation layer effect by assuming a uniform electron distribution around the protein that differs from bulk water by +10%, which is known to be inaccurate in the highq region (Park et al., 2009). However, the simple solvation layer treatment in CRYSOL can be used as a prototype to estimate the effect of a solvation layer model on the calculation of difference scattering. It is evident that the calculation with displaced solvent is in good agreement with the one that additionally models the solvation layer scattering for all three test systems.
3.3. Reliability of absolute Xray scattering calculated from coarsegrained protein structures
We now turn our attention to assessing the accuracy of the calculations of absolute Xray scattering from coarsegrained structures. Fig. 6 shows the effect of coarsegraining on Xray scattering calculations for a hen eggwhite lysozyme [PDB code 6lyz (Diamond, 1974); Fig. 6(a)] and human carbonmonoxy hemoglobin [PDB code 2hhb, Fig. 6(a)]. These structures are not part of the protein structure library used to derive the coarsegrained form factors. For both structures, the agreement between the allatom calculation and the coarsegrained calculations are good for low q and start to become worse at higher q. As expected, coarsegraining according to the MARTINI scheme agrees with the allatom calculation to higher q than the amino acid approach. To quantify this agreement we use the average relative squared error in the range from 0 to q(N) with N being the number of data points in the respective range:
The dotted lines in Fig. 6 mark the maximum q value for the coarsegrained calculations (q_{threshold}), for which the error [equation (12)] is smaller than 0.2%. The error limit of 0.2% is arbitrary but was chosen with respect to the absolute scattering curves of hen eggwhite lysozyme (Fig. 6a). For hen eggwhite lysozyme and human carbonmonoxy hemoglobin, the q_{threshold} values are 0.31 and 0.25 Å^{−1}, respectively, for the amino acid approach and 0.48 and 0.47 Å^{−1}, respectively, for the MARTINI bead approach.
The q_{threshold} [equation (12)] values for the amino acid and the MARTINI approach for all proteins in the library are depicted as a histogram in Fig. 6(b). It is evident that the MARTINI coarsegrained calculations provide a wider q range (on average 0.53 Å^{−1}) compared with the amino acid bead approach (on average 0.27 Å^{−1}). This shows that the use of MARTINI beads significantly extends the range in which scattering can be reliably calculated compared with amino acid coarsegraining.
We note that the results presented in Fig. 6 do not include a model of the solvation layer around the protein. A number of sophisticated methods to incorporate this are already available, and this is critical for comparison with absolute experimental SAXS/WAXS data (Grishaev et al., 2010; Park et al., 2009; Bardhan et al., 2009). However, the underlying physics, which is that a coarser representation of structure leads to a loss in resolution, is well captured in the model that was used to compute the data shown in Fig. 6.
4. Discussion
The increasingly popular method of timeresolved WAXS requires advanced computational structural q range of interest. For the case of human hemoglobin presented above, highresolution crystal models deviate from experimental solution data for q > 0.4 Å^{−1}. Thus a MARTINI representation and scattering model would be suitable for interpreting the available difference data up to q = 0.75 Å^{−1}.
schemes. In this paper, we have shown that coarsegraining the structures leads to a loss of accuracy at wide angles. Thus, when choosing the coarseness of the structural model, computational cost should be carefully balanced against the accuracy needed, a decision which critically depends on theTo the best of our knowledge, there are three et al. (2009) have successfully applied a biased simulation to timeresolved Xray scattering data, Andersson et al. (2009) and Ahn et al. (2009) moved rigid bodies, and Kim, Lee et al. (2012) used ab initio determination of the threedimensional structure. For these three approaches, a large number of Xray scattering calculations had to be performed and this was the limiting factor in the studies. A treatment of larger proteins becomes prohibitively expensive. The MARTINI coarsegrained calculations offer a good compromise between accuracy in the Xray scattering calculation and computational speed. The computation of Xray scattering from coarsegrained protein structures on the MARTINI level is 50 times faster than the corresponding allatom calculation (around 7–8 atoms go into the average MARTINI bead, 7^{2} = 49). This speedup could break new ground for applying difference scatteringassisted structural to larger proteins.^{1}
schemes for timeresolved Xray scattering experiments of proteins. AhnWhen many trial structures are to be evaluated, using computationally demanding stateoftheart methods to account for solvation effects is not feasible. We show here that less sophisticated solvation treatment can be used to reliably model difference WAXS. This is rationalized by the fact that the shape of the solvent shell does not change very much between different protein conformations, and that its contribution to the scattering simply cancels out in the difference scattering. Another advantage of using difference scattering as an experimental observable is that it is free of experimental artifacts stemming from incorrect subtraction of scattering from the buffer and the capillary. This means that the discrepancy between calculation and experiment significantly diminishes compared with standard SAXS/WAXS.
5. Conclusion
The speed of computations of Xray scattering from (bio)molecules can be controlled by coarsegraining the underlying structures. Our results provide the basis for matching the level of coarsegraining with the required resolution. When the beads contain entire amino acids and for the finer MARTINI scheme, we estimate reliability q ranges of 0–0.3 Å^{−1} and 0–0.5 Å^{−1}, respectively. The MARTINI coarsegrained model thus covers the q range available in standard SAXS experiments and is 50 times faster than the allatom calculation. When difference Xray scattering is analyzed, for example, in timeresolved SAXS/WAXS, the reliability q range is significantly extended to 0.75 Å^{−1}, which we showed for the model system human hemoglobin. We anticipate that the increased efficiency in computation of protein Xray scattering will enable more comprehensive structural analyses in the growing field of timeresolved difference Xray scattering of proteins.
Supporting information
Form factor library for MARTINI beads. DOI: https://doi.org//10.1107/S1600576714009959/aj5230sup1.pdf
Form factors for MARTINI beads. DOI: https://doi.org//10.1107/S1600576714009959/aj5230sup2.txt
Acknowledgements
The authors thank Professor Sichun Yang for fruitful discussions. Funding by the ERC grant `StructDyn' and the FFL4 program of the Foundation of Strategic Research, Sweden, is also acknowledged.
References
Ahn, S., Kim, K., Kim, Y., Kim, J. & Ihee, H. (2009). J. Phys. Chem. B, 113, 13131–13133. Web of Science CrossRef PubMed CAS
Andersson, M., Malmerberg, E., Westenhoff, S., Katona, G., Cammarata, M., Wöhri, A. B., Johansson, L. C., Ewald, F., Eklund, M., Wulff, M., Davidsson, J. & Neutze, R. (2009). Structure, 17, 1265–1275. Web of Science CrossRef PubMed CAS
Aranda, R., Levin, E. J., Schotte, F., Anfinrud, P. A. & Phillips, G. N. (2006). Acta Cryst. D62, 776–783. Web of Science CrossRef CAS IUCr Journals
Bardhan, J., Park, S. & Makowski, L. (2009). J. Appl. Cryst. 42, 932–943. Web of Science CrossRef CAS IUCr Journals
Bogaart, G. van den, Meyenberg, K., Risselada, H. J., Amin, H., Willig, K. I., Hubrich, B. E., Dier, M., Hell, S. W., Grubmüller, H., Diederichsen, U. & Jahn, R. (2011). Nature, 479, 552–555. Web of Science PubMed
Cammarata, M., Levantino, M., Schotte, F., Anfinrud, P., Ewald, F., Choi, J., Cupane, A., Wulff, M. & Ihee, H. (2008). Nat. Methods, 5, 881–886. Web of Science CrossRef PubMed CAS
Cho, H., Dashdorj, N., Schotte, F., Graber, T., Henning, R. & Anfinrud, P. (2010). Proc. Natl Acad. Sci. USA, 107, 7281–7286. Web of Science CrossRef CAS PubMed
Daily, M. D., Makowski, L., Phillips, G. N. Jr & Cui, Q. (2012). Chem. Phys. 396, 84–91. Web of Science CrossRef CAS PubMed
Diamond, R. (1974). J. Mol. Biol. 82, 371–391. CrossRef CAS PubMed Web of Science
Fermi, G., Perutz, M. F., Shaanan, B. & Fourme, R. (1984). J. Mol. Biol. 175, 159–174. CrossRef CAS PubMed Web of Science
Fraser, R. D. B., MacRae, T. P. & Suzuki, E. (1978). J. Appl. Cryst. 11, 693–694. CrossRef CAS IUCr Journals Web of Science
Grishaev, A., Guo, L., Irving, T. & Bax, A. (2010). J. Am. Chem. Soc. 132, 15484–15486. Web of Science CrossRef CAS PubMed
Harker, D. (1953). Acta Cryst. 6, 731–736. CrossRef CAS IUCr Journals Web of Science
Hess, B., Kutzner, C., van der Spoel, D. & Lindahl, E. (2008). J. Chem. Theory Comput. 4, 435–447. Web of Science CrossRef CAS PubMed
Ibrahimkutty, S., Kim, J., Cammarata, M., Ewald, F., Choi, J., Ihee, H. & Plech, A. (2011). ACS Nano, 5, 3788–3794. Web of Science CrossRef CAS PubMed
Ihee, H., Wulff, M., Kim, J. & Adachi, S. (2010). Int. Rev. Phys. Chem. 29, 453–520. Web of Science CrossRef CAS
Jong, D. H. de, Singh, G., Bennett, W. F. D., Arnarez, C., Wassenaar, T. A., Schäfer, L. V., Periole, X., Tieleman, D. P. & Marrink, S. J. (2013). J. Chem. Theory Comput. 9, 687–697.
Kim, K. H., Muniyappan, S., Oang, K. Y., Kim, J. G., Nozawa, S., Sato, T., Koshihara, S., Henning, R., Kosheleva, I., Ki, H., Kim, Y., Kim, T. W., Kim, J., Adachi, S. & Ihee, H. (2012). J. Am. Chem. Soc. 134, 7001–7008. Web of Science CrossRef CAS PubMed
Kim, T. W., Lee, J. H., Choi, J., Kim, K. H., van Wilderen, L., Guerin, L., Kim, Y., Jung, Y. O., Yang, C., Kim, J., Wulff, M., van Thor, J. & Ihee, H. (2012). J. Am. Chem. Soc. 134, 3145–3153. Web of Science CrossRef CAS PubMed
Koch, M. H., Vachette, P. & Svergun, D. I. (2003). Q. Rev. Biophys. 36, 147–227. Web of Science CrossRef PubMed CAS
Konarev, P. V., Petoukhov, M. V., Volkov, V. V. & Svergun, D. I. (2006). J. Appl. Cryst. 39, 277–286. Web of Science CrossRef CAS IUCr Journals
Liu, H., Hexemer, A. & Zwart, P. H. (2012). J. Appl. Cryst. 45, 587–593. Web of Science CrossRef CAS IUCr Journals
López, C., Rzepiela, A., de Vries, A., Dijkhuizen, L., Hünenberger, P. & Marrink, S. (2009). J. Chem. Theory Comput. 5, 3195–3210.
Louhivuori, M., Risselada, H. J., van der Giessen, E. & Marrink, S. J. (2010). Proc. Natl Acad. Sci. USA, 107, 19856–19860. Web of Science CrossRef CAS PubMed
Makowski, L. (2010). J. Struct. Funct. Genomics, 11, 9–19. CrossRef CAS PubMed
Malmerberg, E., Omran, Z., Hub, J. S., Li, X., Katona, G., Westenhoff, S., Johansson, L. C., Andersson, M., Cammarata, M., Wulff, M., van der Spoel, D., Davidsson, J., Specht, A. & Neutze, R. (2011). Biophys. J. 101, 1345–1353. Web of Science CrossRef CAS PubMed
Marrink, S. J., Risselada, H. J., Yefimov, S., Tieleman, D. P. & de Vries, A. H. (2007). J. Phys. Chem. B, 111, 7812–7824. Web of Science CrossRef PubMed CAS
Marrink, S. J. & Tieleman, D. P. (2013). Chem. Soc. Rev. 42, 6801–6822. Web of Science CrossRef CAS PubMed
Monticelli, L., Kandasamy, S., Periole, X., Larson, R., Tieleman, D. & Marrink, S. (2008). J. Chem. Theory Comput. 4, 819–834. Web of Science CrossRef CAS
Oberg, K., Ruysschaert, J. & Goormaghtigh, E. (2003). Protein Sci. 12, 2015–2031. Web of Science CrossRef PubMed CAS
Park, S., Bardhan, J., Roux, B. & Makowski, L. (2009). J. Chem. Phys. 130, 134114. Web of Science CrossRef PubMed
Petoukhov, M. V., Franke, D., Shkumatov, A. V., Tria, G., Kikhney, A. G., Gajda, M., Gorba, C., Mertens, H. D. T., Konarev, P. V. & Svergun, D. I. (2012). J. Appl. Cryst. 45, 342–350. Web of Science CrossRef CAS IUCr Journals
Petoukhov, M. V. & Svergun, D. I. (2007). Curr. Opin. Struct. Biol. 17, 562–571. Web of Science CrossRef PubMed CAS
Schäfer, L. V., de Jong, D. H., Holt, A., Rzepiela, A. J., de Vries, A. H., Poolman, B., Killian, J. A. & Marrink, S. J. (2011). Proc. Natl Acad. Sci. USA, 108, 1343–1348. Web of Science PubMed
Silva, M. M., Rogers, P. H. & Arnone, A. (1992). J. Biol. Chem. 267, 17248–17256. PubMed CAS Web of Science
Spilotros, A., Levantino, M., Schiro, G., Cammarata, M., Wulff, M. & Cupane, A. (2012). Soft Matter, 8, 6434–6437. Web of Science CrossRef CAS
Stovgaard, K., Andreetta, C., FerkinghoffBorg, J. & Hamelryck, T. (2010). BMC Bioinformatics, 11, 429.
Svergun, D., Barberato, C. & Koch, M. H. J. (1995). J. Appl. Cryst. 28, 768–773. CrossRef CAS Web of Science IUCr Journals
Svergun, D. I. & Koch, M. H. (2003). Rep. Progr. Phys. 66, 1735. Web of Science CrossRef
Takala, H., Björling, A., Berntsson, O., Lehtivuori, H., Niebling, S., Hoernke, M., Kosheleva, I., Henning, R., Menzel, A., Ihalainen, J. & Westenhoff, S. (2014). Nature, 509, 245–249. Web of Science CrossRef CAS PubMed
Virtanen, J. J., Makowski, L., Sosnick, T. R. & Freed, K. F. (2011). Biophys. J. 101, 2061–2069. Web of Science CrossRef CAS PubMed
Warren, B. E. (1990). Xray Diffraction. New York: Dover Publications Inc.
Westenhoff, S., Nazarenko, E., Malmerberg, E., Davidsson, J., Katona, G. & Neutze, R. (2010). Acta Cryst. A66, 207–219. Web of Science CrossRef CAS IUCr Journals
Yang, S., Park, S., Makowski, L. & Roux, B. (2009). Biophys. J. 96, 4449–4463. Web of Science CrossRef PubMed CAS
Yesylevskyy, S., Schäfer, L., Sengupta, D. & Marrink, S. (2010). PLoS Comput. Biol. 6, e1000810. Web of Science CrossRef PubMed
Zheng, W. & Tekpinar, M. (2011). Biophys. J. 101, 2981–2991. Web of Science CrossRef CAS PubMed
This is an openaccess article distributed under the terms of the Creative Commons Attribution (CCBY) Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original authors and source are cited.