- Open Access
Analytical results on the Beauchemin model of lymphocyte migration
© Textor et al.; licensee BioMed Central Ltd. 2013
- Published: 17 April 2013
The Beauchemin model is a simple particle-based description of stochastic lymphocyte migration in tissue, which has been successfully applied to studying immunological questions. In addition to being easy to implement, the model is also to a large extent mathematically tractable. This article provides a comprehensive overview of both existing and new analytical results on the Beauchemin model within a common mathematical framework. Specifically, we derive the motility coefficient, the mean square displacement, and the confinement ratio, and discuss four different methods for simulating biased migration of pre-defined speed. The results provide new insight into published studies and a reference point for future research based on this simple and popular lymphocyte migration model.
- Random Walk
- Cell Track
- Convection Coefficient
- Lymphocyte Migration
- Functional Central Limit Theorem
A unique property of the immune system is that it mainly consists of constantly moving cells. Technological progress has facilitated the study of lymphocyte migration at increasingly higher resolutions, culminating in the application of two-photon microscopy to tracking single lymphocytes in the living animal [1–3]. Accompanying this progress, mathematical and computational models of lymphocyte migration have been developed and applied to both qualitative and quantitative questions. These models can be roughly categorized as follows according to their level of detail: Whole-population models, usually formulated using ordinary or partial differential equations, do not distinguish between individual cells, but treat cell subpopulations as continuous quantities. Particle-based models do consider each cell individually, but treat cells as freely moving particles without mass, volume, and shape. Finally, whole-cell models such as the Cellular Potts Model  explicitly represent the cell and its interaction with the environment. In the present paper, we analyze a particle-based model that was introduced by Beauchemin, Dixit, and Perelson . Throughout the paper, we refer to this model as the "Beauchemin Model".
Particle-based approaches such as the Beauchemin model are used for studying questions where a whole-population approach does not provide sufficient information, but a whole-cell model is not necessary, not feasible for computational reasons or would require too many assumptions on unknown parameters. For instance, Grigorova et al.  and ourselves  recently used the Beauchemin model to study the transit of T cells through a lymph node in the absence of antigen, and we used it to determine the amount of directional bias that could be detected in random T cell migration using contemporary two-photon imaging techniques .
The Beauchemin model is defined by three parameters - a speed vfree, a time interval tfree, and another time interval tpause. The model describes a particle moving freely in a three-dimensional space according to the following rules: The particle starts at a fixed position and turns in a random direction (more precisely, a direction is sampled from a uniform distribution on the unit sphere). The turn takes a fixed time tpause, during which the cell does not move. This reflects the time it takes to "displace the internal structure which brings about motion" . Then, the cell moves in the chosen direction with velocity vfree for a fixed time tfree, after which it stops and restarts the process.
The main distinguishing characteristic of the model is the introduction of the pause phase. In this sense, it can be seen as a generalization of the "ideal chain" model of polymer physics , in which a chain of rigid rods of fixed length ℓ is considered that are freely jointed to each other. This is mathematically equivalent to the Beauchemin model if we set vfree · t free = ℓ and tpause = 0. In particular, analytical results e.g. on the mean square displacement of the ideal chain  are directly transferable to this case. As we shall see later on in this paper, for some quantitative properties even the more general case with tpause ≥ 0 can be reduced to the ideal chain model.
This paper is structured as follows: In the upcoming section, we begin by formulating a mathematical framework for our analysis, which explicates sufficiently but not overly general notions of random walks and Brownian motion. We deliberately include a fairly large amount of detail, including proofs, in order to present our results in a self-contained fashion. Thereafter, we present analytical results about random motion in the Beauchmin model, and discuss four different ways for simulating biased migration with pre-defined speed. We wrap up with concluding remarks.
We point out that around half of the results discussed in this paper originate from our recent study that used the Beauchemin model . We chose to integrate these results into the present paper in rewritten and extended form because they were formerly only published in condensed form as supporting online material. In this manner, we hope to provide a coherent account of all available analytical results, which we anticipate to serve as a useful reference for future users of the model. We mark results that have appeared earlier in our previous study or elsewhere (e.g. Proposion 10) by giving the appropriate reference in proposition or theorem statements, while propositions or theorems stated without such a reference (e.g. Proposion 12) have, to our knowledge, not been published before.
In this section, we outline the basic mathematical framework that we use to describe stochastic cell motion and, later on, the Beauchemin model. A similar, but slightly less general framework (based on random walk on a lattice) is used in Berg's textbook Random Walks in Biology , while a much more general account of random walks than given here is usually presented in probability textbooks (e.g., ). At the end of this section, we will outline the connection of this random walk to the corresponding macroscopic partial differential equation model, which is the convection-diffusion-equation.
First of all, we introduce some notational conventions. We denote the expectation of a random variable ξ by E[ξ], its variance by Var(ξ) = E [(ξ − E[ξ])2] and its probability density function by fξ. Elementary random variables are written in lowercase Greek letters (usually ξ i ) while derived random variables that are functions of elementary random variables are written in capital Latin letters. All random variables are assumed to be continuous and real-valued. Mean and variance of random variables will also be written as μ and σ2, if the referenced variable is clear from context.
Observed tracks and associated measures
The Beauchemin model is usually validated against, or used to generate, cell tracking data as it would be recorded in a two-photon microscopy experiment. As we will see later, there can be subtle differences between the real underlying state and the observed state of a cell. We thus make this difference explicit in our analysis by means of the following definitions for cell tracking terminology.
consisting of positions and increasing time points .
The time indices are written in brackets to indicate that the positions are discrete-time observations of an underlying continuous motion process. For population-based cell track analysis, it is common to align a set of tracks to a common starting point.
is called the zero-aligned version of T .
Throughout the paper, we will assume that all cell track sets are zero-aligned.
Different parameters can lead to similar motility coefficients.
Microscopic random walk with bias
In this section, we will define the notion of a one-dimensional random walk used in this paper and derive its relation to a one-dimensional convection-diffusion process. These concepts can be easily generalized to higher dimensions under the assumption that the dimensions are pairwise uncorrelated, while stochastic independence in the strict sense is not necessary. This assumption holds for the particular random walks we analyze in this paper.
is called a random walk with step bias μ and step variance σ2. If μ =0, (S t ) is called an unbiased random walk.
Intuitively, the sequence (S t ) is the random walk trajectory and the ξ i are the individual steps. The above definition is a slightly generalized version of the commonly used entirely discrete random walk with fixed steps to the left and the right: We assume that each step is sampled from a (possibly continuous) random variable with finite first and second moments. The following corollary provides an important characteristic of this random walk:
Proof. Since the ξ i are independent and thus uncorrelated, this is a direct consequence of the linearity of variance and expectation. □
For sufficiently long observations, we do not only know the variance and mean of a random walk, but can furthermore obtain a good approximation of the entire particle distribution from the central limit theorem.
converges to a normal distribution with mean 0 and variance σ2 .
Thus, after a sufficiently long time, the distribution of the random walk (S t ) at fixed times is approximately normal. Of course, the terms "sufficiently long" and "approximately normal" are rather vague. In the context of intravital two-photon imaging, which is limited to a finite, often axially thin imaging region, we can usually not safely assume that we are in the range of validity of the normal distribution approximation: We are observing only a few "steps" (prolonged periods of rather persistent motion followed by short pauses) of a random walk. Beltman et al.  and ourselves  have discussed the implications of this technical limitation for detecting directional bias and estimating the motility coefficient.
We now generalize the random walk to higher dimensions.
is called a d-dimensional random walk with step bias (μ(1), ..., μ(d)) and step variance σ2. If μ(1) = ... = μ(d) = 0, then (S t ) is called an unbiased d-dimensional random walk.
The following theorem gives the limit distribution of suitably scaled multi-dimensional random walks within our framework.
Hence, the result follows from the central limit theorem for one-dimensional random walks. □
The central limit argument shows that even complex cell migration processes might have a simple description when observed for a sufficiently long time: The diffusion coefficient is a function of all "microscopic" cell motility parameters such as persistence time, speed, turning angle distributions and so on. In the scaling limit, this single coefficient describes an unbiased random walk completely. Therefore, it is more common to describe random walk processes on such timescales in terms of a differential equation instead of a stochastic process. This can be done using a scale transition, as we will explain below.
The convection-diffusion equation
The random walk (S t ) introduced in Definition 4 is a plausible microscopic model for the position of particles. However, if the positions are observed on large time scales (e.g., only every 20th state of the random walk can be observed due to a limited temporal resolution), then a macroscopic continuous model can be used to describe the migration. This continuous model no longer depends on the detailed characteristics of the microscopic model such as the turning angle distribution or persistence length (i.e., the precise shape of the ξ i ).
We write the time index t in brackets in order to emphasize that the processes are given in continuous time. The process (Z n (t)) is obtained by observing n steps of the random walk (S t ) in every time interval (t − 1, t] with . It is scaled such that E [Z n (t)] = E [S t ] = tμ and Var(Z n (t)) = Var(S t ) = tσ2 for every . Between two successive steps, the process is constant.
Let (Z(t)), t ≥ 0, denote the limiting process obtained as n, the number of steps per time interval, tends to infinity. According to the Functional central limit theorem (sometimes also referred to as Donsker's Theorem, see ), (Z(t)) has the following two properties:
♦ Any finite-dimensional distribution of (Z(t)) is Gaussian.
♦ E[Z(t)] = tμ and Cov(Z(s), Z(t)) = min(s, t)σ2 for s, t ≥ 0.
As one can easily verify, this probability density function solves the following well-known differential equation for the initial condition ϕ(x, 0) = δ(x), where δ denotes the Dirac delta distribution.
Taken the initial condition given above, the solution of the convection-diffusion-equation exists and is unique .
As the sum of independent Brownian motions is again a Brownian motion, the convection-diffusion equation describes the dynamics of single diffusing particles as well as of ensembles of particles carrying out motions independently: Assuming the ensemble is large enough such that stochastic perturbations can be ignored, the convection-diffusion-equation describes the evolution of the particle concentration over time. This scaling argument justifies the use of the convection-diffusion equation for simulating large cell populations on timescales that are substantially longer than the rhythmicity of the cell migration process. For example, we used the convection-diffusion equation to simulate the transit of T cells through the lymph node, which takes approximately half a day .
Note that in the context of molecular motion, the quantity M is usually called diffusion coefficient. Throughout this paper we will use the term motility coefficient to indicate that we are talking about particles that are substantially larger than molecules; this is common in the immunological literature since Miller et al.'s early work .
Similarly as for the discrete model, the continuous model can be generalized to multiple dimensions. In this case, (Z n (t)) is obtained by observing n steps of a d-dimensional random walk (S t ) in every time interval of unit length. For n tending to infinity, the Functional central limit theorem yields convergence to a d-dimensional Brownian motion with independent components and convection coefficients C(1) = μ(1) , ..., C(d)= μ(d). The motility coefficient is equal to M = dσ2 /2, where d is the number of observed dimensions.
As a microscopic model for stochastic motility of particles, we have first introduced one-dimensional random walks. According to this model, particles move by taking independent and identically distributed steps at equidistant discrete time points. Therefore, just like in the Beauchemin model, our framework does not consider persistence to last longer than the duration of a single step; the model is fully characterized by the absolute time between successive steps and the distribution of steps. Important characteristics of the step distribution are the step bias μ and the step variance σ2. The random walk is called unbiased if μ = 0, and biased, otherwise. If we observe many particles carrying out simultaneous unbiased random walks, the particles will go nowhere on the average, but they will spread out increasingly, their distribution being approximately normal. The model can be easily generalized to d dimensions if we assume that the steps are dimension-wise uncorrelated (not necessarily statistically independent). Essentially, we can then treat the d dimensions like d separate one-dimensional random walks.
In this section, we analytically derive three quantitative characteristics of the particular random walk defined by the Beauchemin model, namely the motility coefficient, the mean square displacement, and the confinement ratio.
The fact that the model is three-dimensional might seem to complicate our analysis. However, fortunately, it is possible to decouple the dimensions of the model, as will be shown in the following. The upcoming result was stated by ourselves  and, more recently, by Donovan and Lythe .
Proof. The key ingredients to the proof are
(i) ξ i , υ i , ζ i are uniformly distributed on [−r, r].
(ii) Cov(ξ i , υ i ) = Cov(ξ i , ζ i ) = Cov(υ i , ζ i ) = 0.
In order to establish (i), note that P (ξ i < h), i.e., the probability that ξ i lies below the height h in the sphere, is proportional to the part of the sphere's surface area lying below the plane ξ i = h. The surface of a spherical cap is proportional to its height, and thus ξ i must be uniformly distributed on [−r, r] (see ). By isotropy, the same arguments hold for υ i and ζ i . The statement (ii) on pairwise uncorrelatedness is obtained by (ii) is obtained by observing that (ξ i , υ i ) is uniformly distributed on a circle with radius r − |z| given that ζ i = z, and hence Cov(ξ i , υ i ) = 0.
Now, (i) implies E [ξ i ] = E [υ i ] = E [ζ i ] = 0, and hence (X t ), (Y t ) and (Z t ) are unbiased random walks. The statement on the asymptotic distribution of (X t , Y t , Z t ) follows by (ii) together with Theorem 8, and the fact that a random variable with uniform distribution on [−r, r] has variance r2/3. □
Hence, particles in the Beauchemin model behave exactly like the superposition of three uncorrelated random walks with uniformly distributed steps.
Mean square displacement
The corresponding diffusion coefficients for the 5 best-fitting parameter triplets determined by Beauchemin et al. are given in Table 1. It turns out that all the best-fitting triplets have very similar motility coefficients. On the other hand, the microscopic motility parameters vary quite a lot: For example, the pause time of the 4th best fitting triplet is 5 times longer than that of the 3rd best fitting triplet. The triplets were ranked by Beauchemin et al. according to the sum of squared residuals (SSR) to the experimental data. However, the SSR of the 5th best fitting triplet is only less than 4% higher than that of the best fitting triplet. Our upcoming analysis will provide some insight why this occurs.
The next two propositions will yield a precise characterization of the variance - and with it the mean square displacement - of the position of single particles and ensembles of observed particles in the Beauchemin model. It is mathematically more convenient to analyze the mean square displacement rather than the mean displacement because, as we have shown above, the dimensions can be decoupled. For mean displacement, this does not hold, and we would have to take into account the stochastical dependence between dimensions.
We start out with the variance of the position of one particle that starts its random walk at time t = 0. For simplicity, we will derive the following equations assuming that one dimension is observed. Due to the independence of the dimensions discussed above, multi-dimensional versions of these results are obtained simply by multiplying the derived quantity (i.e., mean square displacement or confinement ratio) with the number of observed dimensions.
with M as defined in equation (2).
Proof. The term 2Mt follows directly from Proposition 10 and the linearity of variance (see also Corollary 5). Thus we continue with t = 0. Then we have from Proposition 10 that the position at time tfree + tpause is uniformly distributed on the interval [−vfreetfree, vfreetfree], having variance 2M(tfree + tpause).
Combining the cases τ ≤ tpause and τ > tpause, we arrive at the proposition. □
Combining the variance and covariance terms as per expression 6, we obtain the claimed identity. □
where P is a parameter called the persistence length. Hence, in the Beauchemin Model, the persistence of migration carries on only for the predefined time tpause + tfree, after which it is immediately and completely lost. On the other hand, persistence in Fürth's equation gradually decays over time following an exponential term. This is why we previously used Fürth's equation to estimate the cell motility coefficient from short-term migration data . However, whether this really leads to more accurate results remains to be determined.
A second notable consequence of the above result is that although the Beauchemin model has three microscopic parameters, its step-wise variance has only two degrees of freedom. Consequently, while the value of tpause would become apparent when analyzing the track of a single simulated cell at a very high resolution, it does no longer matter when a large population of cells that migrate asynchronously is analyzed, as shown by Proposition 12: Given a fixed tfree, there are infinitely many combinations of tpause and vfree that yield an identical motility coefficient and thereby an identical expected square displacement of the population. For example, the empirical mean square displacement of the best-fitting triplet (0.5, 2.0, 18.8) is identical (up to some rounding) to that of the triplet (9.32, 2.0, 40). Hence, we cannot expect to get reliable estimates for all three parameters tfree,tpause and vfree from fitting the model to mean square displacement data. This is consistent with the observation by Beauchemin et al. that the parameter tpause can always be set to 0 without much influence on the quality of the best fit.
The confinement ratio is a measure of track straightness. In the context of lymphocyte migration analysis, it is also called "chemotactic index" or "meandering index" [2, 3]. The confinement ratio is usually defined as the quotient of a path's displacement from the origin versus the complete length of the path. If the cell changes directions very often, the confinement ratio will tend to zero over time, while for directed migration without a random component, the confinement ratio is equal to 1.
Instead of the confinement ratio, we consider here the squared confinement ratio, which can be easily determined for the Beauchemin model from the previous result.
Note that , because the path length tvfree is measured in three dimensions while the expected observed square distance is projected to one dimension.
Thus, the normalized squared confinement ratio converges to tfree/3. When there is evidence that the observed migration process really behaves roughly like the Beauchemin model, this could be a good rule of thumb for quick estimations of the persistence time from experimental data.
For our recent study of the detection limits of two-photon imaging , we extended the Beauchemin model for simulating three different types of biased migration, for which we used the term taxis modes  to indicate that such biased migration in cells usually occurs in response to an external stimulus. These three taxis modes were orthotaxis, where the movement speed is adjusted, klinotaxis, where the duration of the "free runs" is changed, and topotaxis, where the turning angle distribution of the simulated cell is changed. Thus, these modifications can be viewed as semi-mechanistic ones that attempt to describe ways in which a cell could re-adjust its migration behaviour from pure random migration to biased random migration in response to an external stimulus. Before we review these modifications, we start by presenting a fourth method to simulate biased migration which is not mechanistically motivated, but is much easier to analyze.
In all cases, taxis is spatially and temporarily uniform and controlled by two parameters: The vector , assumed to have unitary length, sets the direction of the taxis, and the parameter p ∈ [0, 1] determines the magnitude of taxis, with p = 0 equivalent to an unbiased random walk and p = 1 to a maximally biased random walk; the precise meaning of the parameter p will be defined for each case below.
Simple phenomenological model
A simple ad-hoc method to extend the Beauchemin model for simulating biased migration is the following: During the pause phase, the cell is no longer kept stationary, but moves into the bias direction with speed p · vfree. This model is very simple to analyze, because the modification is deterministic and so changes only the mean, but not the variance of each step. Hence, the motility coefficient remains the same.
Proof. During each pause phase, the cell moves a distance p · vfree · tpause into the bias direction. ||C|| is then simply the product of this distance and the total fraction of time that the cell spends in the pause phase. □
There might be some use for the simple biased migration model in situations where it is necessary to study the effect of bias independently of the motility coefficient. However, it is difficult to imagine how such a situation could arise biologically. Below we therefore discuss model modifications that have a more realistic motivation. For each of these modifications, it is possible to analytically derive the convection coefficient, i.e., the expected speed of the bias.
Throughout the rest of this section, we let the vector denote the orientation vector of the Beauchemin model (i.e., a vector randomly sampled from the unit sphere), and the bias direction. Moreover, we let denote the vector of random variables representing a step of the modified model, while (ξ, υ, ζ) denotes a step in the original model.
In the orthotaxis model, the migration speed is no longer constant but depends on the chosen target direction. The more consistent this direction is with the target direction, the faster the simulated cell will move. Such bias could occur for instance in response to some external dragging force, and might be illustrated by imagining a random stroll through a city on a very windy day.
Here 〈·,·〉 denotes the scalar product and p the bias strength parameter. Note that the movement perpendicular to is not affected at all, while movement against the direction is slower and in direction it is faster. More precisely, the relation between p and the bias speed vtaxis is given by the following proposition.
Dividing this expected location by the step duration tfree + tpause yields the claimed expression for ||C||. □
In the case of topotaxis, the cell adjusts its turning behaviour but keeps the speed constant. This also affects both the variance and the mean of the cell step distribution. Topotaxis could be a likely bias mechanism for cells migrating on some sort of structural cell network in the tissue, like it was suggested to be the case for T cells in secondary lymphoid organs .
Hence, for p = 0 we obtain the distribution of the unbiased case, while for p = 0.5 we get the slightly skewed distribution and for p = 1 we obtain the maximally skewed distribution .
This modification of the Beauchemin model yields the same convection coefficient as in the previous case.
Proof. This is simply shown by calculating for the probability density function defined above. □
Lastly, we can also induce bias by modifying the duration tfree of the free run depending on the picked orientation . A well-known biological example of such bias is the "run-and-tumble" motion by which E. Coli searches for food: Because the bacterium is too small to sense concentration gradients in situ, it is capable of comparing concentration measurements along a persistent path until it figures out whether it is going in the right direction. When it senses that it is moving in the right direction (run), it will try to keep going, while otherwise it will stop its motion soon (tumble) and orient itself in a new random direction.
Again, this modification leads to exactly the same bias speed as found for the previous two models.
Note that the klinotaxis model no longer fits within our mathematical framework because the duration of a step in the model is no longer constant. Thus, the central limit theorem we applied in that section no longer applies, and a different central theorem would be needed to show formally that the model converges to a Brownian motion.
In this article, we have compiled both old and new analytical results on the Beauchemin model of lymphocyte migration, which was originally a model of purely random motility, but is now able to accommodate partially biased migration as well. An important practical consequence of these results is that validation of the model against experimentally determined motility coefficient, mean square displacement data, and confinement ratio curves does no longer need to be performed by simulation, because these quantities can be calculated directly from the model parameters. This also allows for the use of standard numerical fitting procedures, such as implemented in popular computer algebra packages, to estimate the model parameters from such data by least-squares-fitting. An important observation was that when simulating a large population of unsynchronized cells, the model is equivalent to the ideal chain model of polymer physics, which is however only true with respect to the quantities named above and not for others like the mean displacement. Interestingly, this shows that we can sometimes infer information from mean displacement data that we cannot infer from mean square displacement data. However, an analytical derivation of the mean displacement of the model has not yet been achieved, and can be expected to be complicated judging from the experiences with similar models . For now, we leave this as an important open problem.
Subtle biased migration of lymphocytes can point to important biological functions or phenomena [22, 23], and the use of mathematical models has helped to understand the quantitative impact of such effects [23, 7]. The modifications discussed above facilitate simulations of such subtle biases with pre-defined speed, again removing the need for fitting the bias strength parameter to data by using simulations. However, all but one of the discussed modifications also affect the random migration component, which is likely also true in biological scenarios where randomly migrating cells respond to an external stimulus. Therefore, another important task that remains to be solved is to analytically derive the motility coefficient matrices for the topotaxis, klinotaxis, and orthotaxis models.
We thank Joost Beltman for helpful discussions. JT was supported by the Netherlands Organisation for Scientific Research (NWO) grant 912.10.066 (to RJdB). The research of Mathieu Sinn was conducted while he was a Postdoctoral Research Fellow at the David R. Cheriton School of Computer Science, University of Waterloo, Waterloo, Ontario, Canada N2L 3G1.
This article has been published as part of BMC Bioinformatics Volume 14 Supplement 6, 2013: Selected articles from the 10th International Conference on Artificial Immune Systems (ICARIS). The full contents of the supplement are available online at http://www.biomedcentral.com/bmcbioinformatics/supplements/14/S6.
- Miller MJ, Wei SH, Parker I, Cahalan MD: Two-photon imaging of lymphocyte motility and antigen response in intact lymph node. Science. 2002, 296: 1869-1873. 10.1126/science.1070051.View ArticlePubMedGoogle Scholar
- Cahalan MD, Parker I: Choreography of cell motility and interaction dynamics imaged by two-photon microscopy in lymphoid organs. Annual Reviews in Immunology. 2008, 26: 585-626. 10.1146/annurev.immunol.24.021605.090620.View ArticleGoogle Scholar
- Beltman JB, Marée AFM, de Boer RJ: Analysing immune cell migration. Nature Reviews Immunology. 2009, 9: 789-798. 10.1038/nri2638.View ArticlePubMedGoogle Scholar
- Beltman JB, Marée AFM, Lynch JN, Miller MJ, de Boer RJ: Lymph node topology dictates T cell migration behavior. Journal of Experimental Medicine. 2007, 204: 771-780. 10.1084/jem.20061278.PubMed CentralView ArticlePubMedGoogle Scholar
- Beauchemin C, Dixit NM, Perelson AS: Characterizing T cell movement within lymph nodes in the absence of antigen. Journal of Immunology. 2007, 178 (9): 5505-5512.View ArticleGoogle Scholar
- Grigorova IL, Panteleev M, Cyster JG: Lymph node cortical sinus organization and relationship to lymphocyte egress dynamics and antigen exposure. Proceedings of the National Academy of Sciences of the United States of America. 2010, 107: 20447-20452. 10.1073/pnas.1009968107.PubMed CentralView ArticlePubMedGoogle Scholar
- Textor J, Peixoto A, Henrickson SE, Sinn M, von Andrian UH, Westermann J: Defining the Quantitative Limits of Intravital Two-Photon Lymphocyte Tracking. Proceedings of the National Academy of Sciences of the United States of America. 2011, 108 (30): 12401-12406. 10.1073/pnas.1102288108.PubMed CentralView ArticlePubMedGoogle Scholar
- Flory PJ: Statistical Mechanics of Chain Molecules. 1969, Interscience PublishersGoogle Scholar
- Berg HC: Random Walks in Biology. 1993, Princeton University PressGoogle Scholar
- Kallenberg O: Foundations of modern probability. 2002, SpringerView ArticleGoogle Scholar
- Murray JD: Mathematical Biology: I. An Introduction. 2002, Springer, 3Google Scholar
- Billingsley P: Convergence of Probability Measures. 1999, Wiley, John & Sons, 2View ArticleGoogle Scholar
- Evans LD: Partial Differential Equations. 1998, American Mathematical SocietyGoogle Scholar
- Donovan GM, Lythe G: T-cell movement on the reticular network. Journal of Theoretical Biology. 2012, 295: 59-67.View ArticlePubMedGoogle Scholar
- Marsaglia G: Choosing a point from the surface of a sphere. Annals of Mathematical Statistics. 1972, 43 (2): 645-646. 10.1214/aoms/1177692644.View ArticleGoogle Scholar
- Selmeczi D, Mosler S, Hagedorn PH, Larsen NB, Flyvbjerg H: Cell Motility as Persistent Random Motion: Theories from Experiments. Biophysical Journal. 2005, 89: 912-931. 10.1529/biophysj.105.061150.PubMed CentralView ArticlePubMedGoogle Scholar
- Grabher C, Cliffe A, Miura K, Hayflick J, Pepperkok R, Rørth P, Wittbrodt J: Birth and life of tissue macrophages and their migration in embryogenesis and inflammation in medaka. Journal of Leukocyte Biology. 2007, 81: 263-271.View ArticlePubMedGoogle Scholar
- Ionides EL, Fang KS, Isseroff RR, Oster G: Stochastic models for cell motion and taxis. Journal of Mathematical Biology. 2003, 48: 23-37.PubMedGoogle Scholar
- Bajénoff M, Egen JG, Koo LY, Laugier JP, Brau F, Glaichenhaus N, Germain RN: Stromal cell networks regulate lymphocyte entry, migration, and territoriality in lymph nodes. Immunity. 2006, 25: 989-1001. 10.1016/j.immuni.2006.10.011.PubMed CentralView ArticlePubMedGoogle Scholar
- de Gennes PG: Chemotaxis: the role of internal delays. European Biophysics Journal. 2004, 33: 691-693. 10.1007/s00249-004-0426-z.View ArticlePubMedGoogle Scholar
- Codling EA, Plank MJ, Benhamou S: Random walk models in biology. Journal of the Royal Society Interface. 2008, 5 (25): 813-834. 10.1098/rsif.2008.0014.PubMed CentralView ArticleGoogle Scholar
- Castellino F, Huang AY, Altan-Bonnet G, Stoll S, Scheinecker C, Germain RN: Chemokines enhance immunity by guiding naive CD8+ T cells to sites of CD4+ T cell-dendritic cell interaction. Nature. 2006, 440: 890-895. 10.1038/nature04651.View ArticlePubMedGoogle Scholar
- Beltman JB, Allen CD, Cyster JG, de Boer RJ: B cells within germinal centers migrate preferentially from dark to light zone. Proceedings of the National Academy of Sciences of the United States of America. 2011Google Scholar
- Mempel TR, Henrickson SE, von Andrian UH: T-cell priming by dendritic cells in lymph nodes occurs in three distinct phases. Nature. 2004, 427: 154-159. 10.1038/nature02238.View ArticlePubMedGoogle Scholar
- Miller MJ, Hejazi AS, Wei SH, Cahalan MD, Parker I: T cell repertoire scanning is promoted by dynamic dendritic cell behavior and random T cell motility in the lymph node. Proceedings of the National Academy of Sciences of the United States of America. 2004, 101: 998-1003. 10.1073/pnas.0306407101.PubMed CentralView ArticlePubMedGoogle Scholar
- Miller MJ, Wei SH, Cahalan MD, Parker I: Autonomous T cell trafficking examined in vivo with intravital two-photon microscopy. Proceedings of the National Academy of Sciences of the United States of America. 2003, 100 (5): 2604-2609. 10.1073/pnas.2628040100.PubMed CentralView ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.