Abstract
Biochemical reactions in living systems occur in complex, heterogeneous media with total concentrations of macromolecules in the range of 50  400 . Molecular species occupy a significant fraction of the immersing medium, up to 40% of volume. Such complex and volumeoccupied environments are generally termed 'crowded' and/or 'confined'. In crowded conditions nonspecific interactions between macromolecules may hinder diffusion  a major process determining metabolism, transport, and signaling. Also, the crowded media can alter, both qualitatively and quantitatively, the reactions in vivo in comparison with their in vitro counterparts. This review focuses on recent developments in particlebased Brownian dynamics algorithms, their applications to model diffusive transport in crowded systems, and their abilities to reproduce and predict the behavior of macromolecules under in vivo conditions.
Introduction
Intracellular organelles are packed with small solutes, macromolecules, membranes and skeletal proteins. Such crowding is characteristic not only of the cell's interior but also for extracellular tissues [1]. The kinetics and thermodynamics of macromolecular processes and biochemical reactions taking place in vivo are known to be affected by complex, volumeoccupied environments [25].
More and more experimental methods have become available to investigate macromolecules under in vivo conditions. These include nuclear magnetic resonance [68] and fluorescence spectroscopies [9]. Techniques such as SPT (singleparticle tracking) [10,11], FRAP (fluorescence recovery after photobleaching) [12,13], FCS (fluorescence correlation spectroscopy) [14] have been applied to measure diffusion constants of macromolecules in the cytoplasm and membranes. All experiments show that diffusion of proteins in vivo is significantly reduced compared to dilute conditions. In the cytoplasm of eukaryotic cells, diffusion of both large and small molecules is slowed down three to four times [1]. FRAP measurements of the diffusion coefficient of GFP in the Escherichia coli (E. coli) cytoplasm [15,16] yield about 10 times smaller values than at infinite dilution in water.
Theories of diffusion for colloidal soft matter are well established. Within their framework, the behavior of colloidal systems (both dilute and concentrated), consisting of mesoscopically large colloidal particles dispersed in lowmolecularweight solvent, can be simulated, reproduced, and more importantly predicted [1721]. In cell biology, however, an appropriate theory of diffusion phenomena is still lacking. Complications arise due to the cellular heterogeneity. Moreover, there are different cell types whose compositions vary over space and time on the scales associated with diffusive motion and the knowledge of the surroundings of diffusing species is essential for successful applications of theoretical approaches. Unfortunately, even for the beststudied systems, experimental data is too scarce to describe elementary processes involving macromolecules, their timing and spatial distribution.
Simulation methods used to study the in vivo diffusion of macromolecules should take into account their biological localization, sizes and shapes, and positions in three dimensional space. Intermolecular interactions (both specific and nonspecific, repulsive and attractive) should be also included [22]. Ideally, the modeling should be performed at atomic resolution. However, due to the μm size of an average cell and a large number of components occupying its volume, reductionist (coarsegrained) approaches are unavoidable. The simulation methods [22] that can be applied to directly account for the crowded conditions, sizes and shapes of molecules and their interactions, as well as other features of the cellular environment, include different variants of particlebased Brownian dynamics (BD) [2326]. In this review, we describe recent applications of BD to study diffusion in the crowded, invivolike systems. Examples of different modeling approaches, either coarsegrained or more elaborate, are presented and discussed.
Diffusion in dilute and crowded solutions
In dilute solutions, diffusional properties of molecules treated as rigid bodies are determined by their sizes, shapes, temperature, and solvent viscosity. The complete information required to describe translational and rotational motions of Brownian particles in dilute solutions is contained in their diffusion tensors. Singleparticle diffusion tensor, D, is represented with a symmetric, 6×6 matrix containing 3×3 blocks related to the translational (tt) and rotational diffusivities (rr) and their couplings (rt, tr) [26]:
The diffusion tensors of molecules can be obtained from rigidbody hydrodynamic calculations [27]. Translational diffusion coefficient is an average over the diagonal terms of D^{tt}. The rotational dynamics of a free diffusing molecule can be predicted from the eigenvalues of D^{rr }[28]. For a spherical molecule, diffusion is isotropic, the diffusion tensor is diagonal, and the molecule can be described with a single translational and a single rotational diffusion coefficient. Translational diffusion coefficient (D^{trans}) determines the time dependence of the diffusion of the particle's center of mass. The mean square displacement (MSD) of a freely diffusing (in three dimensions) particle's center of mass changes linearly with time t according to the following equation [29]:
with the D^{trans }as a proportionality constant.
In biological setup, solutes interact with their environments and other molecules which affects their Brownian motion, thus singleparticle diffusion tensors are no longer the unique determinants of diffusion. Nevertheless, the relation between the MSD and time t (not necessarily linear) can still be used to measure the apparent translational diffusion of molecules [4,11].
Particlebased Brownian dynamics algorithms
Brownian dynamics is a stochastic simulation approach with continuum space and time. In BD the molecules exhibit noise as they are propagated according to the Langevin equation [30]. The equation contains random and frictional forces that are intended to represent the interaction between the diffusing molecules and solvent (treated implicitly). Highfriction limit is assumed; momentum relaxation is much faster than position relaxation, thus the equations of motions do not contain particles' velocities. Below, we briefly describe two common BD algorithms.
Brownian dynamics of rigid, arbitrarily shaped objects
The propagation scheme for a number of rigid bodies, described either at atomic level or with coarsegrained models, can be written as [23,26]:
where i runs over molecules, Δt is the time step, and is the vector describing the position of the center of diffusion () and orientation () of the ith molecule, . is a generalized force acting on a given particle having two components: the force acting on the diffusion center () and the torque . (Δt) is a random displacement vector resulting from the Brownian noise, with zero mean and the variancecovariance given with:
Here, D is the precomputed and constant in the simulation 6×6 diffusion tensor of the ith molecule at its diffusion center. Random displacements of particles can be computed via Cholesky decomposition of D [23]. Intermolecular hydrodynamic interactions between diffusing bodies are neglected.
A simplified version of the above algorithm [25], which assumes isotropic properties of molecules and replaces their diffusion tensors with translational and rotational diffusion coefficients, has been widely used. Apart from treating the diffusing molecules as spheres, this simplification also ignores the coupling between translational and rotational diffusion. It is however questionable, whether using preaveraged diffusion tensors is justified for nonglobular molecules (e. g., see the discussion on the influence of anisotropy on transport and association rates of model molecules [31] or on the diffusion of an ellipsoid near planar surfaces [27]).
Brownian dynamics with hydrodynamic interactions
When hydrodynamic interactions (HI) are considered in BD simulations, the diffusion tensor of the entire system (6N×6N, with N being the number of spherical molecules or the number of spherical pseudoatoms in case of more elaborate coarsegrained molecular models [32]) is used to propagate particles. This tensor depends on the system's configuration and varies with time. The propagation scheme can be written as [23,24]:
When the RotnePragerYamakwa [33,34] form of the diffusion tensor is used (see below) the gradient term in the above equation (∇D) vanishes. The 6Ndimensional random displacement vector has zero mean and fulfills < (Δt) (Δt) > = 2DΔt. Again, random displacements of particles can be computed via the Cholesky decomposition of D [23]. However, the decomposition of D must be performed at each BD step due to its dependence on the positions of particles.
Intermolecular interactions
The influence of environments on the diffusion of macromolecules manifests through intermolecular forces acting on individual particles. The types and functional forms of interactions included in BD simulations vary depending on the level of detail used to describe the studied systems. Different types of intermolecular interactions that can be modeled in BD simulations are briefly described below.
Electrostatics
When diffusing molecules are described in a reduced way with spherical subunits (e. g., as in case of model spherical proteins [3537] or beadmodels of flexible polymers [38]), the DLVO (Derjaguin  Landau  Verwey  Overbeek) approach (or the DebyeHückel approximation), commonly applied in the studies of colloid suspensions, can be used [39]. Spherical molecules (of radii R_{i}, R_{j}) with central net charges (Q_{i}, Q_{j}) interact via pairwise additive potentials:
Where ϵ_{o }is the vacuum permittivity, ϵ is the dielectric constant of the immersing medium, κ is the inverse of the Debye screening length, and r_{ij }is the separation between molecules. However, the above description is not valid for highly charged systems such as nucleic acids where the linearized form of the PoissonBoltzmann equation [40] is not applicable and there is strong electrostatic coupling between the molecular and ionic species. In such cases, the net charges can be replaced with renormalized charges [41].
A more sophisticated approach to treat electrostatic interactions in BD simulations of rigid molecules with atomic detail [42,43] is based on the model of effective charges developed by Gabdoulline and Wade [44]. In this approach, each molecule is described with a limited set of optimized, discrete, effective charges immersed in an uniform dielectric. These effective charges are derived by fitting the electrostatic potential resulting from the DebyeHückel approximation to the external molecular potential obtained from the numerical solution of the PoissonBoltzmann equation. Electrostatic energies (and thus forces) are evaluated using the values of the electrostatic potential (derived from the solution of the PoissonBoltzmann equation on a threedimensional grid) generated by one molecule at positions of effective charges of the second molecule. The electrostatic potential grids and effective charges are precomputed so when multiple molecules are simulated, the electrostatic potential grids need to be translated and reorientated at each simulation step along with translating and rotating molecules. This approach ignores the dielectric polarization effects that occur when two molecules come close to each other. These effects may be particularly important in the crowded systems, where intermolecular distances are comparable with the sizes of molecules. Polarization can be accounted for by using an analytical formula proposed by Gabdoulline and Wade [44]. The performance of the effective charges approach is worsened for highly charged molecules in cases where the nonlinear form of the PoissonBoltzmann equation should be used to evaluate electrostatic potentials. The DebyeHückel approximation is not valid in these cases, especially in the vicinity of the electrostatic field sources [45].
LennardJones potentials
Nonspecific interactions between molecules can be modeled with the LennardJones type functions that are commonly used in molecular dynamics simulations to compute the van der Waals interactions between atoms; at large intermolecular separations molecules attract each other, while at small separations the interactions are repulsive and molecules are impenetrable. The LennardJones functional form and parameters for BD simulations with coarsegrained spherical models [35,37] can be, for instance, taken from the ones derived for large colloidal spheres [46,47].
In BD simulations employing allatom models, the forces acting between the surface atoms of molecules can be evaluated using standard (6/12) LennardJones potentials:
However, the well depth ϵ_{LJ }needs to be treated as a free parameter and adjusted to reproduce experimental data [42].
Hydrophobic effects
Nonpolar (hydrophobic) interactions can be included in atomistic BD simulations by assuming their proportionality to the amount of the solvent accessible surface area (SASA) that gets buried when molecules come close to each other. However, computing of SASA can be slow and to make SASAbased approaches efficient, approximations are needed. Examples of how to incorporate the hydrophobic effects based on SASA in BD simulations can be found in the work of Elcock and McCammon [48] or Gabdoulline and Wade [49].
While SASAbased models are commonly used to evaluate nonpolar interactions, it has been shown that including the solvent accessible volume (SAV) terms may be essential to accurately model the nonpolar solvation, especially at atomiclength scales [50]. Accurate treatment of nonpolar energies and forces is crucial for understanding any biomolecular process that involves changes in solvent accessibility. The described inability of SASAonly models [50] to reproduce and predict solvation energies and forces indicates that there is a need to include a more complete theory of nonpolar solvation in atomistic BD simulations.
It was also proposed that hydrophobic interactions can be accounted for by modifying the van der Waals interactions and assigning more favorable interaction energies to nonpolar surface atoms than to polar ones [42]. With such an approach hydrophobic interactions can be evaluated more rapidly, however its correctness is questioned (see the discussion in [42]).
Hydrodynamic interactions
In BD simulations water molecules are not treated explicitly but the influence of solvent on the diffusion of suspended molecules can be included by proper treatment of hydrodynamic interactions. However, even though their importance is widely recognized [32,37,51], accounting for HI is no simple matter due to their longrange and manybody character [52] resulting in high computational cost. Therefore, one is often tempted to neglect HI and, consequently, the correlations of motions of molecules.
In the ErmakMcCammon algortihm [23] (Equation 5), hydrodynamic interactions between spherical particles can be included by using the Oseen [53], RotnePrager [33] or RotnePragerYamakawa [34] forms of diffusion tensors (without HI the diffusion tensor is represented by a diagonal matrix). At each step of a BD simulation, the diffusion tensor of the whole system is factorized using the Cholesky decomposition [23] to obtain hydrodynamically correlated displacements of particles. The Cholesky decomposition scales as N^{3 }(with N being the number of particles), while evaluating direct intermolecular interactions, that are assumed pairwise, scales as N^{2}. It is possible to lower the cost of evaluating HI using the Chebyshev approximation proposed by Fixman [54], but the HIrelated computational overhead is still considerable (N^{2.5}). Recently, Geyer and Winter proposed a faster algorithm that scales as N^{2 }and is based on a truncated expansion of the hydrodynamic multiparticle correlations as twobody contributions [55].
An additional computational cost arises when HI are evaluated in finite simulation boxes containing molecules. This cost is due to the fact that for evaluating diffusion tensors Ewald summation has to be used [56,57]. While the direct intermolecular interactions in such periodic systems can be computed using less expensive cutoff based schemes, applying these schemes to evaluate HI leads to diffusion tensors that are not positive definite (a condition that has to be met to use diffusion tensors as covariance matrices [57].)
The treatment of hydrodynamics within the framework of the Oseen, RotnePrager or RotnePragerYamakwa tensors is appropriate for very dilute systems capturing only the farfield, twobody hydrodynamic effects. However, in crowded biological systems, the manybody HI and lubrication forces (resulting from the thin layer of solvent that separates the nearly touching particles) may significantly influence diffusion. One crude way to include the manybody HI in BD simulations is to use meanfield approaches [5860] that are based on scaling the diffusion coefficients of particles according to local volume fractions, as for example in the work of Sun and Weinstein [61].
The most advanced approach to treating HI, used recently in BD simulations of a biological system [37], was developed by Brady and Bossis [17]. Their Stokesian dynamics (SD) method includes both farfield and nearfield manybody contributions to Brownian forces acting on particles. Moreover, particular implementations of SD enable one to efficiently simulate longtime dynamics of large multicomponent systems. For example, the accelerated SD method, described in [62] scales as N^{1.25}log(N).
Unfortunately, modeling of HI in BD simulations is currently limited to systems composed of coarsegrained particles (either spherical molecules or composed of a small number of spherical units  pseudoatoms). While such models perform well for colloids, a question yet unanswered is whether they are sufficient for crowded, heterogeneous biological systems.
Applications to diffusion in crowded environments
Typically, the setup of BD simulations aimed at investigating diffusion in crowded solutions is the following. A number of molecules (usually 10^{2 } 10^{3}), described either at the atomistic level or with coarsegrained models, is contained inside a primary simulation box. Periodic boundaries are applied to reduce the edge effects and the influence of the finite size of the system. BD trajectories of all molecules are generated with the algorithm of choice. The BD simulation time scale should be long enough so the relative displacements in the system be at least comparable with the dimensions of molecules; the simulation time scales range from microseconds to milliseconds, while typical time steps are about a few picoseconds. When one considers the number of simulated molecules, the need to describe them in as detailed way as possible, and the complexity of interactions calculated in a single simulation step, it becomes clear that BD simulations of crowded systems require a lot of computational resources.
Below we present a few recent (except for one) examples of BD simulations of macromolecular diffusion in crowded media. We describe the studies that best illustrate the potential and shortcomings of the BD technique applied to in vivolike systems and the ones that introduce a significant element of novelty. Some of these computational studies relate to experiments  experimental data were either used for parameterization or for direct comparison with simulations.
Beginnings
One of the first studies that used the BD technique to investigate macromolecular diffusion in congested media was that of Dwyer and Bloomfield [63] in 1993. They performed BD simulations of probe and selfdiffusion of concentrated solutions containing short DNA polymers (modeled as wormlike chains, i.e. strings of spherical beads) and the bovine serum albumin (BSA) protein modeled as a sphere. Even though their DNA model lacked the detailed allatom description, it reproduced the overall translational and rotational behavior of linear DNA molecules [64]. The authors used an algorithm similar to that of ErmakMcCammon [23], however, they neglected HI, because the pairwise approaches are inadequate for concentrated solutions. They also used the DebyeHückel approach to model longrange electrostatics and the repulsive part of the LennardJones potential to ensure volume exclusion. This model, while very simple, turned out to be realistic enough to simulate diffusion in the DNA/BSA system and accurately reproduce experimental diffusion coefficients at 0.1 M ionic strength for the selfdiffusion of BSA and the probe diffusion of BSA in DNA (both over a wide range of BSA/DNA concentrations).
Simulations of the diffusion of BSA in the DNA solution conducted at low ionic strength of 0.01 M gave worse agreement with experiments which the authors attributed to the lack of HI and limitations of the applied electrostatic model. The work of Dwyer and Bloomfield is significant because it shows that even very simple modeling approaches can accurately predict the behavior of rather complex systems.
We note that the models similar to the one of Dwyer and Bloomfield are still used, for example, in computational studies of facilitated diffusion of proteins on DNA. The description of facilitated diffusion phenomenon and models of nonspecific proteinDNA interactions can be found in references [65,66].
Proteinprotein association in crowded environments
Most biological reactions require some sort of diffusionmediated encounter. Here we describe BD applications that directly investigate the influence of crowded environments on diffusional encounter and association of molecules.
Sun and Weinstein [61] simulated systems of spherical particles, mimicking globular proteins. Particles were modeled as hard spheres; the authors used the elastic collision method to effectively deal with particle overlaps [67]. Electrostatic interactions were introduced using the DebyeHückel approach. A notable fact is that the authors attempted to model also manybody HI. They used the translational part of the ErmakMcCammon algorithm with the diagonal form of the diffusion tensor. However, they applied an average hydrodynamic correction and scaled the diagonal terms of the diffusion tensor (i.e., the single particle diffusion coefficients) with functions that depend on the instantaneous local volume fraction of each particle [5861]. The authors investigated how the size of background molecules, volume fraction, electrostatic interactions, and HI influence the diffusion of a tracer molecule. Their study also quantitatively described the dependence of diffusionlimited reaction rates on crowding in model systems. Using a very approximate approach, they showed a significant effect of HI on diffusion and association rates. While this review focuses on simulations of multimolecular systems, the importance of intermolecular HI in modeling the association kinetics in a twomolecular case was also described by FrembgenKesner and Elcock [32]. A number of simplifications was used in the described work of Sun and Weinstein [61]. Simulated systems were rather homogeneous in comparison with in vivo environments. Molecules were treated as spheres thus their association was nonspecific (the model of particles with uniformly reactive surfaces was used). Anisotropy of particles, which may play a role upon association, and specific intermolecular interactions were neglected. However, their work is important as it showed the necessity to include HI in BD simulations of crowded systems.
One of the conclusions of the work of Sun and Weinstein [61] was that crowding slows down the nonspecific association of molecules. The two more recent BD studies focused on the influence of crowders on specific interactions in proteinprotein association and considered the fact that partner proteins need to be properly oriented to associate. Wieczorek and Zielenkiewicz [68] used BD to simulate the diffusional encounter of HEL and HyHEL5 proteins in solutions of hardsphere crowders. Both proteins were modeled with atomistic details, however, only the excluded volume interactions were evaluated; simulation steps leading to overlaps were rejected and repeated with different random displacements, so no systematic forces were explicitly applied. The crowders and proteins were modeled as rigid bodies with isotropic diffusion tensors (both HEL and HyHEL5 were represented as spheres with equivalent hydrodynamic radii) and propagated with the algorithm described in [25]. Hydrodynamic interactions were neglected. The authors concluded that if the definition of an encounter complex accounts for the orientation of associating molecules, crowding can accelerate the association rate.
Similar observations were made by Kim and Yethiraj [69]. Using spherical protein models with nonuniformly reactive surfaces, they showed that association of model molecules can be either slowed down or accelerated, depending on the applied reaction criteria. These authors recently used BD to investigate the effects of crowding on the association at model membranes [70]. Both studies of Kim and Yethiraj [69,70] focus on how the excluded volume effects influence the reaction rates.
In general, the above simulation results are in agreement with purely theoretical predictions [3]. However, they lack a close connection to experiments and the ability to reproduce heterogeneous in vivo conditions.
Modeling bacterial cytoplasm
In the recent work of Elcock's [43] and Skolnick's [37] groups BD was applied to investigate the diffusion in cytosollike systems with the Escherichia coli cytoplasm as a model system. The cytoplasm model of McGuffee and Elcock [43] contains 50 different types of the most abundant proteins and nucleic acids present in the E. coli cytoplasm. Additionally, a few molecules of the Green Fluorescent Protein (GFP) were included, for the sake of comparison with experimental data. All molecules were represented with atomistic details. The algorithm of rigid bodies was applied in BD but all molecules were treated as isotropic, so the average rotational and translational diffusion coefficients were used instead of 6×6 diffusion tensors [25].
Intermolecular interactions included electrostatics, modeled using the effective charges approach of Gabadoulline and Wade [44], and combined van der Waals and hydrophobic interactions described with the LennardJones potential [42], with the well depth treated as an adjustable parameter. Hydrodynamic interactions were neglected. The authors investigated how different energy models affect the translational diffusion coefficient of GFP. When only steric interactions (modeled with the repulsive part of the 6/12 LennardJones potential) were considered, the BD simulated translational diffusion coefficient of GFP was 3 to 6 times higher than the coefficient estimated experimentally. Even including electrostatics did not significantly reduce the discrepancy between the experimental and computed values. Nevertheless, the authors obtained a close agreement with experiments after adding the shortrange LennardJones attraction and tuning its welldepth. The observation that purely steric interactions are not able to account for the reduction in proteins' mobilities in vivo concur with previous computational studies [71].
While adjusting the shortrange attractive interactions led to reproducing the in vivo GFP diffusion coefficient, it has been later pointed out [37] that the magnitude of attractive interactions can be tuned to set the diffusion coefficient to any chosen value. McGuffee and Elcock investigated also the character of translational diffusion[4] in cytoplasm with different interaction energy models. When only the steric interactions were considered, for the three chosen proteins they observed normal diffusion at short times (ps), subdiffusion at moderate time scales (ns) and normal diffusion at long times (μs). When electrostatic and shortrange attractions were included, subdiffusive behavior was noticed even for short observation times.
The approach to model diffusion in the E. coli cytoplasm taken by Ando and Skolnick [37] is different than the one described above. These authors investigated the effect of macromolecular shapes on the diffusion in crowded, heterogeneous environment, the role of HI, and the interplay between hydrodynamic and nonspecific attractive interactions. Their cytoplasm model consisted of 15 different types of molecules, each described with a set of spherical subunits positioned at the C_{α }protein atoms and the P, C4', N1, and N9 atoms of nucleic acids. The authors dealt with the anisotropy of macromolecular shapes using the rigid body BD algorithm [26] that describes molecules with precomputed, 6×6 diffusion tensors. Notably, this is the first BD study of a multimolecular system that takes into account the anisotropy of diffusion. To investigate the role of HI they also simulated the system of equivalent spheres corresponding to the cytoplasm model (all molecules in the system were represented as spherical particles and assigned hydrodynamic radii based on translational diffusion coefficients computed from their atomic structures [27]) with the farfield manybody HI interactions and lubrication forces treated with the method of Banchio and Brady [62]. Softsphere harmonic potential was used to model repulsive interactions between particles. Attractive interactions were described by the LennardJones potential function with a correction factor accounting for the roughness of molecular surfaces. Electrostatic interactions were modeled using the DLVO model.
Based on simulations of molecularshaped and equivalent sphere systems, the authors showed that the effects of macromolecular shapes on longtime translational diffusion are rather small in the studied concentration range of 250  350 . This is an important result that justifies the use of isotropic models to study translational diffusion in the crowded systems. The influence of macromolecular shapes on rotational dynamics, which is an interesting issue as well, was not investigated. Their BD simulations with only steric repulsion between particles also resulted in overestimated translational diffusion coefficients. However, when HI were switched on, the computed longtime translational diffusion coefficient of GFP was in good agreement with experimental value; this is another important result because this model did not include any tunable parameters. The authors concluded that the macromolecular motions in vivo are likely to be dominated by steric interactions and hydrodynamics. Additionally, they observed that when nonspecific attractive interactions between particles (either molecularshaped or spherical) are included, the reduction of translational diffusion coefficients depends much more on molecular sizes in comparison with the simulations where HI were enabled.
Both studies described above [37,43] are state of the art applications of BD in the field of macromolecular diffusion in biological systems, as they both take into account the heterogeneity of the environment and intermolecular interactions beyond simple excluded volume models. Both are also closely connected to experimental data.
Concluding remarks
This short review focuses on the current status of the particlebased Brownian dynamics technique and its applicability to model diffusive phenomena in biological environments. We described a few BD studies of threedimensional macromolecular diffusion in aqueous cellular compartments under molecular crowding conditions. However, there are various other BD applications, not presented here, that consider diffusive phenomena in cellular architectures, such as the cytoskeleton [72] or membranes [73].
We presented different modeling approaches, from simple excludedvolume models to most sophisticated atomic level descriptions of the in vivolike systems. The complexity of intra and extracellular environments, their dynamics, and the number of different interactions that influence the diffusive behavior of macromolecules, make computational studies of crowding extremely difficult. While it is desirable to construct models that treat biological systems at atomic resolutions and include sophisticated interactions [42,43], the recent model of the E. coli cytoplasm [37] shows that less elaborate, coarsegrained approaches should not be disregarded, at least as far as diffusive transport is concerned.
Obviously, current BD algorithms need development. These include accurate and computationally efficient models of direct, both specific and nonspecific, intermolecular interactions as well as HI models that up to date have been largely neglected in the simulations of biological systems.
BD simulations should be (at least ideally) able to reproduce and predict the in vivo dynamics. However, the computational cost of sophisticated BD simulations may be too high, even with the growth of computing power and available modern technologies. Thus, the longterm goal should be to systematically develop efficient algorithms interfacing different modeling approaches  from atomistic to coarsegrained models of macromolecules, mesoscopic models of biological environments, and appropriate boundary conditions for different cell compartments. New approaches and models should be verified experimentally to establish their quality and predictive power.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
MD studied literature and designed manuscript content. MD and JT wrote the paper.
Acknowledgements
The authors acknowledge support from University of Warsaw (BST 2010, G314), Polish Ministry of Science and Higher Education (N N301 245236, N N519 646640) and Foundation for Polish Science (Focus program and TEAM/20093/8 project cofinanced by European Regional Development Fund operated within Innovative Economy Operational Programme).
References

Ellis RJ: Macromolecular crowding: an important but neglected aspect of the intracellular environment.
Curr Opin Struct Biol 2001, 11:114119. PubMed Abstract  Publisher Full Text

Minton AP: The influence of macromolecular crowding and macromolecular confinement on biochemical reactions in physiological media.
J Biol Chem 2001, 276:1057710580. PubMed Abstract  Publisher Full Text

Zhou HX, Rivas G, Minton AP: Macromolecular crowding and confinement: biochemical, biophysical and potential physiological consequences.
Annu Rev Biophys 2008, 37:37597. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Dix JA, Verkman AS: Crowding effects on diffusion in solutions and cells.
Annu Rev Biophys 2008, 37:24763. PubMed Abstract  Publisher Full Text

Elcock AH: Models of macromolecular crowding effects and the need for quantitative comparisons with experiment.
Curr Opin Struct Biol 2010, 20:196206. PubMed Abstract  Publisher Full Text

Miklos AC, Li C, Sharaf NG, Pielak GJ: Volume exclusion and soft interaction effects on protein stability under crowded conditions.
Biochemistry 2010, 49:69846991. PubMed Abstract  Publisher Full Text

Sharaf NG, Barnes CO, Charlton LM, Young GB, Pielak GJ: A bioreactor for incell protein NMR.
J Magn Reson 2010, 202:140146. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Li C, Wang GF, Wang Y, CreagerAllen R, Lutz EA, Scronce H, Slade KM, Ruf RA, Mehl RA, Pielak GJ: Protein (19)F NMR in Escherichia coli.
J Am Chem Soc 2010, 132:321327. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Xie XS, Choi PJ, Li GW, Lee NK, Lia G: Single molecule approach to molecular biology in living bacterial cells.
Annu Rev Biophys 2008, 37:417444. PubMed Abstract  Publisher Full Text

Jin S, Verkman AS: Single particle tracking of complex diffusion in membranes: simulation and detection of barrier, raft, and interaction phenomena.
J Phys Chem B 2007, 111:36253632. PubMed Abstract  Publisher Full Text

Hall D, Hoshino M: Effects of macromolecular crowding on intracellular diffusion from a single particle perspective.
Biophys Rev 2010, 2:3953. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Verkman AS: Diffusion in cells measured by fluorescence recovery after photobleaching.
Method Enzymol 2003, 360:635648. Publisher Full Text

Pastor I, Vilaseca E, Madurga S, Garcés JL, Cascante M, Mas F: Diffusion of αchymotrypsin in solutioncrowded media. A fluorescence recovery after photobleaching study.
J Phys Chem B 2010, 114:40284034. PubMed Abstract  Publisher Full Text

Weiss M: Probing the interior of living cells with fluorescence correlation spectroscopy.
Ann N Y Acad Sci 2008, 1130:2127. PubMed Abstract  Publisher Full Text

Elowitz MB, Surette MG, Wolf PE, Stock JB, Leibler S: Protein mobility in the cytoplasm of Escherichia Coli.
J Bacteriol 1999, 181:197203. PubMed Abstract  PubMed Central Full Text

Konopka MC, Shkel IC, Cayley S, Record MT, Weisshaar JC: Crowding and confinement effects on protein diffusionin vivo.
J Bacteriol 2006, 188:61156123. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Brady JF, Bossis G: Stokesian dynamics.
Ann Rev Fluid Mech 1988, 20:111157. Publisher Full Text

Holmqvist P, Nägele G: Longtime dynamics of concentrated chargestabilized colloids.
Phys Rev Lett 2010, 104:058301058305. PubMed Abstract  Publisher Full Text

Gapinski J, Patkowski A, Nägele G: Generic behavior of the hydrodynamic function of charged colloidal suspensions.
J Chem Phys 2010, 132:054510054520. PubMed Abstract  Publisher Full Text

Gögelein C, Nägele G, Tuinier R, Gibaud T, Stradner A, Schurtenberger P: A simple patchy colloid model for the phase behavior of lysozyme dispersions.
J Chem Phys 2008, 129:085102085114. PubMed Abstract  Publisher Full Text

Banchio AJ, Nägele G: Shorttime transport properties in dense suspensions: from neutral to chargestabilized colloidal spheres.
J Chem Phys 2008, 128:104903104923. PubMed Abstract  Publisher Full Text

Takahashi K, Arjunan SNV, Tomita M: Space in systems biology of signaling pathways  towards intracellular molecular crowding in silico.
FEBS Lett 2005, 579:17831788. PubMed Abstract  Publisher Full Text

Ermak DL, McCammon JA: Brownian dynamics with hydrodynamic interactions.
J Chem Phys 1978, 69:13521360. Publisher Full Text

Dickinson E, Allison SA, McCammon JA: Brownian dynamics with rotationtranslation coupling.
J Chem Soc Faraday Trans 2 1985, 81:591601. Publisher Full Text

Gabdoulline RR, Wade RC: Brownian dynamics simulation of proteinprotein diffusional encounter.
Methods 1998, 14:329341. PubMed Abstract  Publisher Full Text

Fernandes MX, de la Torre JG: Brownian dynamics simulations of rigid particles of arbitrary shape in external fields.
Biophys J 2002, 83:30393048. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Carrasco B, de la Torre JG: Hydrodynamic properties of rigid particles: comparison of different modeling and computational procedures.
Biophys J 1999, 76:30443057. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

de la Torre G, Harding SE, Carrasco B: Calculation of NMR relaxation, covolume and scatteringrelated properties of beadmodels using the SOLPRO computer program.
Eur Biophys J 1999, 28:119132. PubMed Abstract  Publisher Full Text

Einstein A: A new determinant of molecular dimension.
Ann Physik 1906, 19:289306. Publisher Full Text

Deutsch JM, Oppenheim I: Molecular theory of Brownian motion for several particles.
J Chem Phys 1971, 54:35473556. Publisher Full Text

Schluttig J, Korn CB, Schwarz US: Role of anisotropy for proteinprotein encounter.
Phys Rev E 2010, 81:030902030906. Publisher Full Text

FrembgenKesner T, Elcock AH: Absolute proteinprotein association rate constants from flexible, coarsegrained Brownian dynamics simulations: the role of intermolecular hydrodynamic interactions in barnasebarstar association.
Biophys J 2010, 99:L75L77. PubMed Abstract  Publisher Full Text

Rotne J, Prager S: Variational treatment of hydrodynamic interaction in polymers.
J Chem Phys 1969, 50:48314838. Publisher Full Text

Yamakawa H: Transport properties of polymer chains in dilute solution: hydrodynamic interaction.
J Chem Phys 1970, 53:436444. Publisher Full Text

Gorba C, Helms V: Diffusional dynamics of cytochrome C molecules in the presence of a charged surface.
Soft Mat 2003, 1:185202. Publisher Full Text

Ermakova E, Krushelnitsky AG, Fedotov VD: Brownian dynamics simulation of electrostatically interacting proteins.
Mol Phys 2009, 100:28492855. Publisher Full Text

Ando T, Skolnick J: Crowding and hydrodynamic interactions likely dominate in vivo macromolecular motion.
Proc Nat Acad Sci USA 2010, 107:1845718462. PubMed Abstract  Publisher Full Text

Dwyer JD, Bloomfield VA: Brownian dynamics simulation of probe diffusion in DNA: effects of probe size, charge and DNA concentration.
Bioph Chem 1995, 57:5564. Publisher Full Text

Bell GM, Levine S, N ML: Approximate methods of determining the doublelayer free energy of interaction between two charged colloidal spheres.
J of Colloid Int Sci 1970, 33:335359. Publisher Full Text

Gilson MK, Honig B: Energetics of chargecharge interactions in proteins.
Proteins: Struct Func Genet 1988, 3:3252. Publisher Full Text

Aubouy M, Trizac E, Bocquet L: Effective charge versus bare charge: an analytical estimate for colloids in the infinite dilution limit.
J Phys A: Math Gen 2003, 36:58355840. Publisher Full Text

McGuffee SR, Elcock AH: Atomically detailed simulations of concentrated protein solutions: The effects of salt, pH, point mutations, and protein concentration in simulations of 1000molecule systems.
J Am Chem Soc 2006, 128:1209812110. PubMed Abstract  Publisher Full Text

McGuffee SR, Elcock AH: Diffusion, crowding and protein stability in a dynamic molecular model of the bacterial cytoplasm.
PLoS Comput Biol 2010, 6:118. Publisher Full Text

Gabdoulline RR, Wade RC: Effective charges for macromolecules in solvent.
J Phys Chem 1996, 100:38683878. Publisher Full Text

Beard DA, Schlick T: Modeling saltmediated electrostatics of macromolecules: The discrete surface charge optimization algorithm and its application to the nucleosome.
Biopolymers 2001, 58:106115. PubMed Abstract  Publisher Full Text

Hamaker HC: The London van der Waals attraction between spherical particles.

Henderson D, Duh DM, Chu X, Wasan D: An Expression for the Dispersion Force between Colloidal Particles.
J Colloid Int Sci 1997, 185:265268. Publisher Full Text

Elcock AH, McCammon JA: Calculation of weak proteinprotein interactions: the pH dependence of the second virial coefficient.
Biophys J 2001, 80:613625. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Gabdoulline RR, Wade RC: On the contributions of diffusion and thermal activation to electron transfer between phormidium laminosum plastocyanin and cytochrome f: Brownian dynamics simulations with explicit modeling of nonpolar desolvation interactions and electron transfer events.
J Am Chem Soc 2009, 131:92309238. PubMed Abstract  Publisher Full Text

Wagoner JA, Baker NA: Assessing implicit models for nonpolar mean solvation forces: The importance of dispersion and volume terms.
Proc Natl Acad Sci USA 2006, 103:83318336. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

FrembgenKesner T, Elcock AH: Striking effects of hydrodynamic interactions on the simulated diffusion and folding of proteins.
J Chem Theory Comput 2009, 5:242256. Publisher Full Text

Mazur P, van Saarloos W: Manysphere hydrodynamic interactions and mobilities in a suspension.
Physica A 1982, 115:2157. Publisher Full Text

Kirkwood JG, Riseman J: The intrinsic viscosities and diffusion constants of flexible macromolecules in solution.
J Chem Phys 1948, 16:565574. Publisher Full Text

Fixman M: Construction of Langevin forces in the simulation of hydrodynamic interaction.
Macromolecules 1986, 19:12041207. Publisher Full Text

Geyer T, Winter U: An O(N2) approximation for hydrodynamic interactions in Brownian dynamics simulations.
J Chem Phys 2009, 130:114905114913. PubMed Abstract  Publisher Full Text

Beenakker CWJ: Ewald sum of the RotnePrager tensor.
J Chem Phys 1986, 85:15811582. Publisher Full Text

Smith ER, Snook IK, van Megen W: Hydrodynamic interactions in Brownian dynamics simulations. I. Periodic boundary conditions for computer simulations.
Physica A 1987, 143:441467. Publisher Full Text

Heyes DM: Brownian dynamics simulations of self and collective diffusion of near hard sphere colloidal liquids: inclusion of manybody hydrodynamics.
Molec Phys 1996, 87:287298. Publisher Full Text

UrbinaVillalba G, GarcíaSucre M, ToroMendoza J: Average hydrodynamic correction for the Brownian dynamics calculation of flocculation rates in concentrated dispersions.
Phys Rev E 2003, 68:061408061417. Publisher Full Text

Tokuyama M, Oppenheim I: Dynamics of hardsphere suspensions.
Phys Rev E 1994, 50:R16R19. Publisher Full Text

Sun J, Weinstein H: Toward realistic modeling of dynamic processes in cell signaling: Quantification of macromolecular crowding effects.
J Chem Phys 2007, 127:155105155115. PubMed Abstract  Publisher Full Text

Banchio AJ, Brady JF: Accelerated Stokesian dynamics: Brownian motion.
J Chem Phys 2003, 118:1032310332. Publisher Full Text

Dwyer JD, Bloomfield VA: Brownian dynamics simulations of probe and selfdiffusion in concentrated protein and DNA solutions.
Biophys J 1993, 65:18101816. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Hagerman PJ, Zimm BH: Monte Carlo approach to the analysis of the rotational diffusion of wormlike chains.
Biopolymers 1981, 20:14811502. Publisher Full Text

Florescu AM, Joyeux M: Description of nonspecific DNAprotein interaction and facilitated diffusion with a dynamical model.
J Chem Phys 2009, 130:015103015112. PubMed Abstract  Publisher Full Text

Florescu AM, Joyeux M: Dynamical model of DNAprotein interaction: effect of protein charge distribution and mechanical properties.
J Chem Phys 2009, 131:105102105114. Publisher Full Text

Strating P: Brownian dynamics simulations of hardsphere suspension.
Phys Rev E 1999, 59:21572187. Publisher Full Text

Wieczorek G, Zielenkiewicz P: Influence of macromolecular crowding on proteinprotein association rates  a Brownian dynamics study.
Biophys J 2008, 95:50305036. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Kim SK, Yethiraj A: Effect of macromolecular crowding on reaction rates: a computational study.
Biophys J 2009, 96:13331340. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Kim JS, Yethiraj A: Crowding effects on association reactions at membranes.
Biophys J 2010, 98:951958. PubMed Abstract  Publisher Full Text

Ridgway D, Broderick G, LopezCampistrous A, Ru'aini M, Winter P, Hamilton M, Boulanger P, Kovalenko A, Ellison MJ: Coarsegrained molecular simulation of diffusion and reaction kinetics in a crowded virtual cytoplasm.
Biophys J 2008, 94:37483759. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Klann MT, Lapin A, Reuss M: Stochastic simulation of signal transduction: impact on the cellular architecture on diffusion.
Biophys J 2009, 96:51225129. PubMed Abstract  Publisher Full Text  PubMed Central Full Text

Walther D, Kuzmin P, Donath E: Brownian dynamics simulation of the lateral distribution of charged membrane components.
Eur Biophys J 1996, 24:125135. PubMed Abstract  Publisher Full Text