On the Role of Vesicle Transport in Neurite Growth: Modelling and Experiments
Ina Humpert, Danila Di Meo, Andreas W. Püschel, Jan-Frederik Pietschmann
OOn the Role of Vesicle Transport in NeuriteGrowth: Modelling and Experiments
Ina Humpert ∗ , Danila Di Meo † ,Andreas W. P¨uschel † , Jan-Frederik Pietschmann ‡ August 7, 2019
Abstract
The processes that determine the establishment of the complex morphology ofneurons during development are still poorly understood. We present experimentsthat use live imaging to examine the role of vesicle transport and propose alattice-based model that shows symmetry breaking features similar to a neuronduring its polarization. In a otherwise symmetric situation our model predictsthat a difference in neurite length increases the growth potential of the longerneurite indicating that vesicle transport can be regarded as a major factor inneurite growth.
Keywords:
Neurite Growth, Vesicle Transport, Symmetry Breaking, Lattice-based Kinetic Models, Biologic Modelling, Cross Diffusion
1. Introduction
Neurons are highly polarized cells with functionally distinct axonal and dendritic com-partments. These are established during their development when neurons polarize aftertheir generation from neural progenitor cells and are maintained throughout the life of theneuron [25]. Unpolarized newborn neurons from the mammalian cerebral cortex initiallyform several undifferentiated processes of similar length (called neurites) that are highlydynamic ([9], [18]). During neuronal polarization, one of these neurites is selected tobecome the axon.The aim of this paper is to combine experimental results with modelling to betterunderstand the role of transport in this process. Indeed, while transport of vesicles indeveloping and mature neurons has been studied before [4, 5, 43], to the best of ourknowledge there are so far no models that examine its impact on neuronal polarization. ∗ Applied Mathematics M¨unster: Institute for Analysis and Computational Mathematics, Westf¨alischeWilhelms-Universit¨at (WWU) M¨unster, Germany ([email protected]). † Institute for Molecular Biology, Westf¨alische-Wilhelms-Universit¨at (WWU) M¨unster, Germany(d [email protected],[email protected]) ‡ Technische Universit¨at Chemnitz, Fakult¨at f¨ur Mathematik, Germany ([email protected]). a r X i v : . [ q - b i o . S C ] A ug or the experiments we use primary cultures of embryonic hippocampal neurons whichare widely used as a model system to study the mechanisms that mediate the transitionto a polarized morphology [32]. After attaching to the culture substrate, neurons extendmultiple undifferentiated neurites that all have the potential to become an axon. Beforeneuronal polarity is established, these neurites display randomly occurring periods ofextension and retraction ([9], [44]). Upon polarization, one of the neurites is specified asthe axon and elongates rapidly ([25], [32]). This neurite has to extend beyond a minimallength to become an axon ([11], [15], [46]).The extension and retraction of neurites depends on cytoskeletal dynamics and the exo-and endocytosis of vesicles ([27], [32], [37]). The growth of neurites and axons requiresan increase in the surface area of the plasma membrane by the insertion of vesicles in astructure at the tip of the developing neurite which is called the growth cone. Retraction,on the other hand, is accompanied by the removal of membrane through endocytosis ([28],[27], [37]). The material for membrane expansion is provided by specialized vesicles thatare characterized by the presence of specific vesicular membrane proteins ([17], [30], [40],[42]). The bidirectional transport of vesicles along neurites provides material producedin the cell body and recycles endocytosed membranes [22]. Molecular motors transportorganelles along microtubules in the anterograde direction towards the neurite tip (kinesins)and retrogradely to the soma (dynein). The nascent axon shows a higher number oforganelles compared to the other neurites due to a polarization of intracellular transport toprovide sufficient material for extension ([3], [32]). The net flow of vesicles into a neurite,thus, has to be regulated depending on changes in neurite length but it is not known howintracellular transport is adjusted to differences in the demand for vesicles in growing orshrinking neurites.Based on these findings, we aim to obtain a better understanding of the role of vesicletransport in the polarization process by means of modelling. We propose a lattice basedapproach for the transport of the vesicles between soma and growth cones. We modelantero- and retrograde vesicles as two different types of particles that are located on adiscrete lattice. To account for the limited space, we propose a maximal number of vesiclesthat can occupy one cell, see also Figure 4. Particles randomly jump to neighbouringcells with a rate that is proportional to a diffusion coefficient and (the relative change)of a given potential. Furthermore, only jumps into cells which are not yet fully occupiedare allowed. This is closely related to so-called (asymmetric) exclusion processes, see [10]and the references therein. Finally, at each end of the lattice, we introduce a pool thatrepresents either the vesicles present in the soma or, at the tip, those in the growth cones.Lattice based models (also called cellular automata) of this type are used in manyapplications, ranging from transport of protons, [21], to the modelling of human crowds,[2]. They also serve as a tool to understand the fundamental characteristics of systemswith many particles, [36]. Models with different species and size exclusion have also beenstudied, see [13, 29] and in particular [41] which deals with the case of limited resources.Here, we develop a model adopted to neuron polarization and present numerical simula-tions that analyze the relation between vesicle transport and neurite length. We are ableto show that an initial length advance of a single neurite leads to further asymmetries inthe vesicle concentration in the pools.We also present a system of nonlinear partial differential equations that arises as a(formal) limit from the cellular model and briefly discuss its properties.While our model still lacks major features of neurite growth, the presented results2a) (b)Figure 1: Vesicle transport in unpolarized neurons: a) Schematic representation ofneuronal polarization in primary cultures of neurons. b) Schematic representationof intracellular transport in neurons. Vesicles are transported in the anterogradedirection (green) from the cell body (soma) into the neurite towards the tip ofthe neurite along microtubules (yellow). Vesicles are inserted into the plasmamembrane by exocytosis in the growth cone at the tip of the neurite to pro-mote extension. Vesicles generated by endocytosis are recycled or transportedretrogradely to the soma (red).show high correspondence with real data: In live cell imaging experiments a neurite thathas exceed a critical length during the polarization process grows rapidly becoming thefuture axon. In our model we supply one neurite with an initial length advantage and thedynamics of the model result in a positive feedback that further increases its length advanceindicating that it becomes the future axon. We also observe oscillations in the vesicleconcentration in the pools that may be interpreted as cycles of extension and retraction.This paper is organized as follows: In section 2 we explain the biological background ofthe paper. In section 3 we introduce a discrete model for the numeric simulations and insection 4 we present the corresponding macroscopic cross diffusion model. In section 5 wepreformed the numerical simulations and interpreted the results. Finally, in section 6 wegive a conclusion.
2. Experimental Results and Consequences for the Modelling
This section contains the results of live cell imaging of primary neurons that were preparedfrom rat embryos and a brief discussion of the consequences for the mathematical modelling.The final model with all details is then presented in section 3.
Unpolarized neurons extend several undifferentiated neurites called minor neurites. Uponpolarization, one of the minor neurites is specified as the axon and growths rapidly, seeFigure 1 a). As explained in the introduction, this increase in the length of a neuritedepends on the sum of the surfaces of all vesicles that fuse with the plasma membrane ofthe growth cone. Their intracellular transport from the soma to the tip of the neurite isdriven by molecular motors that transport the vesicles along microtubules, see Figure 1 b).3 .2. Exterimental Methods and Results µ -Dish, Ibidi) coated with poly-L-ornithine (SigmaAldrich)and cultured at 37 ◦ C and 5% CO for one day in BrightCell TM NEUMO Photostablemedium (Merck Millipore) with supplements. Neurons were transfected with an expressionvector for Vamp2-GFP by calcium phosphate co-precipitation as described previously [35].The pDCX-Vamp2 vector was constructed by replacing GFP in pDcx-iGFP (providedby U. M¨uller, Scripps Research Institute, La Jolla, CA, USA; [14]) by a new multiplecloning site (5’- GAATTC ACTAG TTCTA GACCC GGGGG TACCA GATCT GGGCCCCTCG AGCAA TTGGC GGCCG CGGGA TCC-3’) and Vamp2-GFP as an XbaI andBglII fragment from pEGFP-VAMP2 (addgene, ◦ C and 5% CO using a Zeiss LSM 800 laser scanning confocal microscope (Carl ZeissMicroImaging, Jena, Germany) and the Zeiss ZEN Software (Carl Zeiss MicroImaging).Images were taken at a frame rate of one scan per second for 2 minutes, followed by apause of 20 minutes for 2 to 12 hours (see Figure 2). The number, velocity and directionof vesicle movement were quantified using the ImageJ macro toolsets KymographClear and
KymoAnalyzer ([26], [23]).2.2.3. Statistical analysisStatistical analyses were done using the GraphPad Prism 6.0 software. Statistical signifi-cance was calculated for at least three independent experiments using the Wilcoxon SignRank test. Significance was defined as follows: If p > .
05 we regard the results as notsignificant and if (cid:63) p < .
05 as significant, see Figure 3 (a).2.2.4. ResultsCultures of hippocampal neurons were transfected with an expression vector for Vamp2-GFP as a marker for vesicles to analyze transport in the neurites of multipolar neurons [17].Antero- and retrograde movement of Vamp2-GFP positive vesicles was analyzed by live cellimaging 24 hours after transfection before axons are specified. To determine if the transportrates of vesicles change when neurites extend or retract we determined the number ofvesicles in the neurite that are immobile or move in the antero- or retrograde direction(Figure 3). Vesicle dynamics is markedly higher in neurites that undergo extension orretraction compared to those that do not show changes in length. A significant differencein vesicle transport was observed during the extension of neurites. The number of vesiclesmoving anterogradely (0,029 ± µ m ) was 37 % higher compared to thosebeing transported retrogradely (0,021 ± µ m ) in growing neurites. Therewas no significant difference between antero- and retrograde transport in retracting neuritesprobably because not all of the vesicles generated by endocytosis during retraction arepositive for Vamp2. No significant differences were found in the velocity of moving vesicles.4 unpolarized polarized anterograde retrograde D ns v e s i c l e s / s / u m Ant Ret * E µ m / s Ant Ret
F C
1 2 B anterograde retrograde exocytosis endocytosis growth cone cell body cell body + m e ( s ) distance (μm) growth cone cell body + m e ( s ) distance (μm) growth cone (a)(b)Figure 2: Quantification of vesicle transport in unpolarized neurons: a) Neuronsfrom the hippocampus of E18 rat embryos were transfected with an expressionvector for Vamp2-GFP and the transport of Vamp2-GFP-positive vesicles ana-lyzed by live cell imaging. The distribution of Vamp2-GFP-positive vesicles isshown at the indicated time points (hours) in unpolarized neurons. Neurite 1first undergoes retraction before it extends again while neurite 2 extends duringthe whole imaging time. A higher magnification of neurite 2 is shown in thelower panel. b) Representative colour-coded kymograph displaying the trajectoryof moving vesicles in the anterograde (green) and retrograde (red) direction.
Based on the previous experimental findings, we aim to formulate a mathematical modelfor the transport of vesicles based on the following assumptions: First, we consider neuritesas one dimensional lattices connected, on one end, to the soma and to a pool representingthe growth cone at the other end, see Figure 4. Vesicles that are currently transportedanterograde and those that are moving retrograde are modelled as two separate speciesmoving on these lattices. At the growth cone anterograde vesicles can fuse with themembrane while also vesicles can be separated from it. During this process, anterogradevesicles can be converted to retrograde ones and vice versa. The same can happen whenvesicles enter or leave the soma. The growth cones and the soma will be modelled separatelyas pools that can store a given number of vesicles. As vesicles have a positive volume,there is a maximal density within the neurites that depends on the size of the vesicles.This results in a lattice model that will be described in full detail in Section 3.Finally, let us briefly comment of the physical dimensions involved. In practice vesicles5a) (b)Figure 3:
Quantification of vesicle transport in unpolarized neurons: a) The num-ber of Vamp2-GFP positive vesicles moving in the anterograde or retrogradedirection (vesicles/s/ µ m ) was quantified in neurites that undergo extensionor retraction and in stationary neurites that do not show a change in length.The number of moving vesicles is higher in neurites showing changes in lengthcompared to stationary neurites. During extension, the number of vesicles movinganterogradely is higher (37 %) compared to those being transported retrogradely(Wilcoxon Sign Rank test; n = 4 independent experiments; values are means ± s.e.m, (cid:63) p < . µ m/s) of Vamp2-GFP positive vesicles movingin the anterograde and retrograde direction was quantified. No significant differ-ences were observed (Wilcoxon Sign Rank test; n = 4 independent experiments;values are means ± s.e.m)with different diameters varying from 80 to 150 nm have been described ([39], [40], [27])but for simplicity we assume that all vesicles have the same size (130 nm). Thus, as thelength of a neurite (which we consider as a one dimensional object) is approximately 1000nm, there is a natural maximal density of around 1000 nm / 130 nm ≈ x Figure 4: Sketch of the lattice based size exclusion model with the number of grid points n = 6. For illustration purpose the maximal number of vesicles is 4 whereas it ishigher in reality. 6igure 5: Sketch of a neuron and identification with a starshaped-domain:
Onthe left, a sketch of a neuron can be seen, where 1) corresponds to the cellnucleus, 2) to a dendrite and 3) to the axon. On the right, a union of six unitintervals portraits the shape of the neurite that is assumed in the modelling.After the branching of the neurites is neglected, the neuron is mapped to astarshaped-domain via a function f .
3. A discrete model
We will now present a mathematical model for the growth process described in the previoussection. In our approach, each neurite is modeled as a discrete lattice on which both antero-and retrograde vesicles, modelled as seperate particles, move. As the diameter of a neuriteis about 1000 nm and thus very small compared to its length that can be up to 1 m, wemodel neurites as one dimensional objects, i.e. a one dimensional lattice. On this lattice,the vesicles can jump to neighbouring cells with a probability that is determined by a givenpotential and a diffusion coefficient. Furthermore, we introduce a size exclusion effect byonly allowing jumps to cells which are not fully occupied (see the discussion in Section 2.3).These lattices are coupled to the soma at one end and to a vesicle pool corresponding tothe growth cone at the other end. See Figure 4 for a summarized version of the model.We will now describe the dynamics on the lattice as well as the coupling to the soma andpools in detail, in the simple case of a single neurite connected to a soma.We first present the detailed dynamics of a single neurite:1.
Lattice dynamics : Each lattice consists of i = 1 , . . . , N cells of width h . The midpointof cell i is denoted by x i and each cell can be occupied by a certain number of vesicles,depending on their size. Denoting by a i = a i ( t ) and r i = r i ( t ) the number of antero- andretrograde vesicles at time t in cell i , we have the following dynamics for the interior cells i = 2 , . . . , N − Ch ∂ t a i = − a i (1 − ρ i − ) e − ( V a,i − V a,i − ) + a i − (1 − ρ i ) e − ( V a,i − − V a,i ) − a i (1 − ρ i +1 ) e − ( V a,i − V a,i +1 ) + a i +1 (1 − ρ i ) e − ( V a,i +1 − V a,i ) , Ch ∂ t r i = − r i (1 − ρ i − ) e − ( V r,i − V r,i − ) + r i − (1 − ρ i ) e − ( V r,i − − V r,i ) − r i (1 − ρ i +1 ) e − ( V r,i − V r,i +1 ) + r i +1 (1 − ρ i ) e − ( V r,i +1 − V r,i ) , (1)where ρ i = a i + r i v max denotes the (relative) sum of antero and retrograde vesicles with v max denoting the maximalnumber of vesicles for a cell of width h . Furthermore, V a,i := V a ( x i ) and V r,i := V r ( x i ) aregiven potentials with V a , V r : R → R evaluated at the midpoint of cell i and C is one over7he diffusion constant, see Section 3.1 for details. Roughly speaking, on the right handsides of the above equations all terms with positive sign correspond to particles that jumpinto cell i from the neighbouring cells while negative terms remove those that jump out.2. Coupling to soma and pools : We assume that all lattices are connected to the somaat their first lattice site i = 1. There, we have the following effects: • Retrograde vesicles leave the neurite and enter the soma with a rate β r (Λ som ) r , ifthere is enough space, where Λ som denotes the number of vesicles currently in thesoma and β is a velocity that depends on this quantity. • Anterograde vesicles can leave the soma and enter the lattice, if there is enough space,i.e. if ρ <
1. In this case, they enter with a given rate α a (Λ som )(1 − ρ N ), that alsodepends on the number of vesicles in the soma.At site N , the neurites are connected to their respective pools (growth cones) and we havethat: • Anterograde vesicles leave the lattice and enter the pool with rate β a (Λ N ) r N , whereagain the velocity β depends on the number of particles in the pool. • Retrograde particles move from the pool into the lattice with rate α r (Λ N ), againonly if space on the lattice is available. This yields the effective rate α r (Λ N )(1 − ρ ).Since we assume that both the pool and the soma have a maximal capacity that cannot beexceeded, we make the following choices for in- and out-flux rates α q (Λ j ) = α q Λ j Λ max j and β q (Λ j ) = β q (1 − Λ j Λ max j ) , k ∈ { a, r } , q ∈ { som , N } . This yields the following equations at the tips and the soma:2 Ch ∂ t a = − a (1 − ρ ) e − C ( V a, − V a, ) + a (1 − ρ ) e − C ( V a, − V a, ) + Chα a (1 − ρ ) , Ch ∂ t a N = − a N (1 − ρ N − ) e − C ( V a,N − V a,N − ) + a N − (1 − ρ N ) e − C ( V a,N − − V a,N ) − Chβ a a N , Ch ∂ t r = − r (1 − ρ ) e − C ( V r,x − V r, ) + r (1 − ρ ) e − C ( V r, − V r, ) − Chβ r r , Ch ∂ t r N = − r N (1 − ρ N − ) e − C ( V r,N − V r,N − ) + r N − (1 − ρ N ) e − C ( V r,N − − V r,N ) + Chα r (1 − ρ N ) . (2)3. Dynamics in soma and pools : Finally, we have to describe the change of number ofvesicles in the soma and the respective neurite pools. For now, we assume that the changeis only due to vesicles entering and existing, i.e. no growth or degradation terms areincluded. This yields the following ordinary differential equations ∂ t Λ N = − α r (cid:0) − ρ N (cid:1) Λ N Λ maxN + β a a N (cid:0) − Λ N Λ maxN (cid:1) , (3) ∂ t Λ som = β r r (cid:0) − Λ som Λ maxsom (cid:1) − α a (cid:0) − ρ (cid:1) Λ som Λ maxsom . (4)8. Extension to multiple neurites : In the case of M neurites, we denote by a i,l and r i,l theconcentration of retro- and anterograde vesicles in neurite l , l = 1 , . . . M at site i . Thepools are then called Λ N,l and we also change the names of all parameters accordingly, i.e.we have α r,l , β r,l , . . . . While the equations for the dynamics inside the neurites (1), at thetip (2) and for the respective growth cones (3) remain unchanged (despite the differentnotation for the constants), the equation for the soma becomes ∂ t Λ som = M (cid:88) l =1 (cid:20) β r,l r ,l (cid:0) − Λ som Λ maxsom (cid:1) − α a,l (cid:0) − ρ ,l (cid:1) Λ som Λ maxsom (cid:21) . (5) Remark 3.1 (On the Modelling) . a) Note that α has a different physical interpretationthan β . Whereas α is given in vessec and specifies an influx rate, β is given in µ msec andtherefore specifies an outflux velocity. This is essential for the boundary contributionsin (2) all having the same physical unit (using that − ρ i is already scaled).b) We are not dealing with the domain and the actual concentrations in the pool explicitlybut only model the total number of vesicles present. In particualar there is no diffusionor transport in the pools. Instead, we assume that the dynamics are fast compared tothose of the neurites. In that way we allow for vesicles that have left one neurite andentered a pool to immediately leave into another neurite. Remark 3.2 (Coupling) . Even though the equations in (1) describe the evolution ofconcentrations, the pools in (3) – (4) have the physical unit mass. Their coupling naturallyarises using the flux as a linking element. Indeed fluxes have the physical unit vessec as havethe terms on the right hand side of (2) that correspond to the boundary flow as well as thereaction terms for the time evolution of the pools in (3) – (4) . Remark 3.3 (Numerical Simulations) . One advantage of our model is that it immediatelyyields a discretisation for numerical simulations. Indeed, by construction it is alreadydiscrete in space and by applying an explicit Euler discretisation we arrive at a fully discretescheme. This scheme will be used to perform simulations in Section 5. There, we will alsopresent some of the scheme’s properties.
Next we transform all relevant variables into an appropriate scaled and dimensionless form,where we always indicate the corresponding dimensionless form with a bar and the typicalsize with a tilde. Thus e.g. r = r ˜ r is a dimensionless quantity. Note that we will then omitthe bar everywhere after this section for reasons of readability.Motivated by the discussion in Section 2 we make the following choices: The typicallength is ˜ L = 50 µ m, the typical time is ˜ t = 100 sec, the typical diffusion constant is˜ ε = 10 − µ m sec , the typical potential is ˜ V = 1 µ m sec . The typical influx is ˜ α = 1 vesiclessec andthe typical outflow velocity is ˜ β = 10 − µ msec , thus the different boundary conditions have thesame unit of measurement. As the typical diameter of one vesicle is 130 nm and the neuritediameter is 1 µ m, the maximal density is given by ρ max = , µ m ≈ vesicles µ m . The typicaldensity of anterograde and retrograde particles is ˜ a, ˜ r = 15 vesicles µ m , which corresponds to ahalf filled neurite. 9s 1 − ρ is already scaled, the equations (1) transform to, for i = 2 , . . . , N − λ ε ¯ Ch ∂ t a i = − a i (1 − ρ i − ) e − C ˜ C ˜ V ( V a,i − V a,i − ) + a i − (1 − ρ i ) e − C ˜ C ˜ V ( V a,i − − V a,i ) − a i (1 − ρ i +1 ) e − C ˜ C ˜ V ( V a,i − V a,i +1 ) + a i +1 (1 − ρ i ) e − C ˜ C ˜ V ( V a,i +1 − V a,i ) , λ ε ¯ Ch ∂ t r i = − r i (1 − ρ i − ) e − C ˜ C ˜ V ( V r,i − V r,i − ) + r i − (1 − ρ i ) e − C ˜ C ˜ V ( V r,i − − V r,i ) − r i (1 − ρ i +1 ) e − C ˜ C ˜ V ( V r,i − V r,i +1 ) + r i +1 (1 − ρ i ) e − C ˜ C ˜ V ( V r,i +1 − V r,i ) , (6)with ˜ C = ε and ¯ C = ε . Thus the product of all typical variables appearing in thesummands of the previous two equations are λ ε = ˜ t L ˜ ε = 100 sec2500 µ m − µ m sec = 4 · − and λ V = ˜ t L ˜ V = 0 .
04 (7)after cancellation of a and r respectively on both sides. Note that the scaling parametersfor the boundary conditions can be calculated by multiplying the boundary conditionswith ˜ L ˜ t and additionally scaling terms corresponding to in- and outflux with ˜ γ . We obtain¯ Ch ∂ t a = − λ ε a (1 − ρ ) e − ¯ C ˜ C ˜ V ( V a, − V a, ) + λ ε a (1 − ρ ) e − ¯ C ˜ C ˜ V ( V a, − V a, ) + λ in ¯ Chα a (1 − ρ ) , ¯ Ch ∂ t a N = − λ ε a N (1 − ρ N − ) e − ¯ C ˜ C ˜ V ( V a,N − V a,N − ) + λ ε a N − (1 − ρ N ) e − ¯ C ˜ C ˜ V ( V a,N − − V a,N ) − λ out ¯ Chβ a a N , ¯ Ch ∂ t r = − λ ε r (1 − ρ ) e − ¯ C ˜ C ˜ V ( V r, − V r, ) + λ ε r (1 − ρ ) e − ¯ C ˜ C ˜ V ( V r, − V r, ) − λ out ¯ Chβ r r , ¯ Ch ∂ t r N = − λ ε r N (1 − ρ N − ) e − ¯ C ˜ C ˜ V ( V r,N − V r,N − ) + λ ε r N − (1 − ρ N ) e − ¯ C ˜ C ˜ V ( V r,N − − V r,N ) + λ in ¯ Chα r (1 − ρ N ) , (8)where we introduced the dimensionless scaling parameters λ in = ˜ t ˜ α L ˜ a = 100 sec · vessec µ m · ves µ m = 0 . , λ out = ˜ t ˜ β L = 100 sec · − µ msec µ m = 0 . . (9)Furthermore equation (12) that describes the pool concentration requires scaling. Applyingthe same time scale as above and the same scaling of vesicle concentrations yields1˜ t ∂ t ( ˜Λ N ¯Λ N ) = (cid:104) − ˜ α ¯ α r (cid:0) − ρ N (cid:1) Λ N Λ maxN + ˜ β a ¯ β a ˜ a N ¯ a N (cid:0) − Λ N Λ maxN (cid:1)(cid:105) . Multiplying by ˜ t and dividing by ˜Λ N gives ∂ t ¯Λ N = (cid:104) − ˜ α ˜ t ˜Λ N ¯ α rN (cid:0) − ρ N (cid:1) Λ N Λ maxN + ˜ t ˜ β aN ˜ a N ˜Λ N ¯ β aN ¯ a N (cid:0) − Λ N Λ maxN (cid:1)(cid:105) . Choosing ˜Λ N = 2 ˜ L ˜ a N = 50 µ m · vesicles µ m = 750 vesicles, we finally arrive at ∂ t ¯Λ N = (cid:104) − λ in ¯ α r (cid:0) − ρ N (cid:1) Λ N Λ maxN + λ out ¯ β aN ¯ a N (cid:0) − Λ N Λ maxN (cid:1)(cid:105) ∂ t ¯Λ som = λ out ¯ β r ¯ r (cid:0) − Λ som Λ maxsom (cid:1) − λ in ¯ α a (cid:0) − ρ (cid:1) Λ som Λ maxsom . Again, the generalization to more than one neurite is straight forward.
Remark 3.4 (Choice of Typical Parameters) . The identification ˜Λ N = 2 ˜ L ˜ a N in theparagraph above is natural as there is a proportion between the size of the pools and thesize of the neurites in reality, where the prefactor 2 corresponds to the fact that we dealwith two types of species. This proportion should be reflected in the typical value ˜Λ .
4. Macroscopic Cross Diffusion in a Model Neuron with Pools
It is well known that lattice models as the one described in the preceeding section have a(formal) correspondence to (systems of) partial differential equations [36, 7]. Let us brieflysummarize the procedure for a single neurite: First we chose h = 1 /N so that the latticehas exactly length one and fix the continuous domain Ω = [0 , x i ∈ Ω its midpoint and assume the existence of smooth functions r = r ( x, t )and a = a ( x, t ) such that r i ( t ) = r ( x i , t ) and a i ( t ) = a ( x i , t ). With this notation, we canformally apply Taylor’s formula to the right hand sides of equations (6), up to second order.For example, for the first equation in (6), this yields Ch ∂ t a ( x i , t ) = λ ε (cid:16) h ( a ( x i , t ) ∂ xx ρ ( x i , t ) + ∂ xx a ( x i , t )(1 − ρ ( x i , t )) (cid:17) + λ V (cid:16) Ch (cid:104) a ( x i , t ) ∂ x ρ ( x i , t ) ∂ x V a ( x i , t ) − ∂ x a ( x i , t )(1 − ρ ( x i , t )) ∂ x V a ( x i , t ) . − a ( x i , t )(1 − ρ ( x i , t )) ∂ xx V a ( x i , t ) (cid:105) + O ( h ) (cid:17) , where O ( h ) refers to remaining terms of order h . Then we divide both sides by h andtake the limit h → ∂ t a + ∂ x J a = 0 with J a := − ( λ ε [(1 − ρ )) ∂ x a + a∂ x ρ ] + λ V a (1 − ρ )) ∂ x V a ) ,∂ t r + ∂ x J r = 0 with J r := − ( λ ε [(1 − ρ )) ∂ x r + r∂ x ρ ] + λ V r (1 − ρ )) ∂ x V r ) , (10)on Ω × (0 , T ), having applied the same procedure to retrograde vesicles. Equations (2)results in the boundary conditions − J a · n = λ in α a Λ som Λ maxsom (1 − ρ ) at x = 0 ,J a · n = λ out β a (cid:0) − Λ N Λ maxN (cid:1) a at x = 1 ,J r · n = λ out β r (cid:0) − Λ som Λ maxsom (cid:1) r at x = 0 , − J r · n = λ in α r Λ N Λ maxN (1 − ρ ) at x = 1 , (11)where n and n denote the outward pointing unit vectors at x = 0 and x = 1, respectively.In the case of a single neurite with fixed in- and outflow boundary conditions (i.e. without11igure 6: Sketch of the model neuron:
The model neuron consists of two neurites andindicated boundary flow in the domain Ω = Ω ∪ Ω , where the two unit intervalsΩ and Ω correspond to two neurites. The squares correspond to pools wherevesicles can be stored, i.e. the pool in the middle corresponds to the soma andthe pools at the tips of the neurites correspond to the corresponding growthcones. For an easy visualization Ω is illustrated as a mirrored copy of Ω .considering the pools explicitly), this has been carried out in detail in [34]. Passing to thelimit in the ODEs for the pools yields ∂ t Λ N = − α r (cid:0) − ρ (1) (cid:1) Λ N Λ maxN + β a a (1) (cid:0) − Λ N Λ maxN ) (cid:1) ,∂ t Λ som = β r r (0) (cid:0) − Λ som Λ maxsom (cid:1) − α a (cid:0) − ρ (0) (cid:1) Λ som Λ maxsom , (12)i.e. the only difference in contrast to (3)–(4) is the fact that the concentrations a, r and ρ are now functions on a continuous domain Ω instead of a discrete grid. Therefore we wrote r (0) instead of r , etc.In the situation of two neurites, we will have equations (10) for each neurite withappropriate boundary condition and again the ODE for Λ som will contain as a right handthe sum of all in- and outfluxes. This situation is summarized in Figure 6. In particular,we see that formally the total mass of the system is preserved, as expected. Remark 4.1 (Analysis of the Model) . The focus of this paper is to gain an understandingof the distribution of vesicles during the growth of neurites based on the discrete modelintroduced in Section 3 and its numerical simulation. However, from a mathematical pointof view it is also very interesting to study the macroscopic counterpart of the model givenby the system of equations (11) – (12) . We therefore briefly point out the relevant questionsand difficulties in the mathematical analysis of this model.Clearly, most important is the question of existence and uniqueness of solutions. Froman application point of view, also the long time behaviour is relevant. As for existence, anumber of results on cross-diffusion equations of type (11) is available, [20, 6, 12], andalso the flux boundary conditions (11) have been analysed before, [8]. The main issue whenapplying these results to our model is the following. The present theory shows existence ofsolutions in the spaces r i , a i ∈ L ((0 , T ); L (Ω)) ∩ H ((0 , T ); ( H ) ∗ (Ω)) ,ρ i ∈ L ((0 , T ); H (Ω)) ∩ H ((0 , T ); ( H ) ∗ (Ω)) . hus, making use of the embedding of H into the space of continuous functions, it makessense to evaluate ρ at a point of the boundary (e.g. ρ (0) ). Unfortunately, this regularityis not available for the concentrations r and a so that we cannot evaluate them at theboundary as would be necessary for the boundary conditions (11) to be well-defined. Thusone would need an improved regularity theory (which seems out of reach at present) or oneneeds to modify the model in a way which is consistent with the biological modelling on thediscrete level (e.g. by allowing particles to switch places). As for the long time behaviour,the numerical simulations of Section 5 suggest that metastable states exist. Their analysisis another interesting problem and we postpone both issues to future work.
5. Numerical Simulations
In order to derive a fully discrete numerical scheme, see also Remark 3.3, we use anexplicit Euler discretisation for the time derivatives in (6) and (8). Subdividing the interval[0 , T ] into K intervals we denote by τ = T /K the step size and by a ki , r ki the respectiveconcentrations at time t k = kτ . Within the neurites this results in the scheme a k +1 i = a ki + τ HG ka ,r k +1 i = r ki + τ HG kr , i = 2 , ...N − , (13)where H = 2 λ ε εh and G kq = − q ki (1 − ρ ki − ) e − C ˜ C ˜ V ( V q,i − V q,i − ) + q ki − (1 − ρ ki ) e − C ˜ C ˜ V ( V q,i − − V q,i ) − q ki (1 − ρ ki +1 ) e − C ˜ C ˜ V ( V q,i − V q,i +1 ) + q ki +1 (1 − ρ ki ) e − C ˜ C ˜ V ( V q,i +1 − V q,i ) q ∈ { a, r } . The evolution at the boundary follows by discretising the time derivates in (8), e.g. foranterograde vesicles at the soma we obtain a k +11 = a k + τ (cid:16) HG ka, + λ in h α a (1 − ρ k ) (cid:17) , (14)with G ka, = − a k (1 − ρ k ) e − C ˜ C ˜ V ( V a, − V a, ) + a k (1 − ρ k ) e − C ˜ C ˜ V ( V a, − V a, ) . For the time discretisation of the ODEs (12) for the pools and the soma we also use anexplicit Euler discretisation with the same time step size. As Λ N , Λ som model a mass andto ensure that the total mass remains conserved we multiply the in- and outflux terms by h and finally obtain the evolution of the pool concentrations viaΛ k +1 q = Λ kq + τ (cid:104) λ in Influx Terms + λ out Outflux Terms (cid:105) , q ∈ {
N, som } . To further analyse the properties of this scheme, let us define the constants V − q,max := max ≤ i ≤ n − e − C ˜ C ˜ V ( V q,i − V q,i − ) , V + q,max := max ≤ i ≤ n − e − C ˜ C ˜ V ( V q,i − V q,i +1 ) as well as V max := max( V a,max , V r,max ) where V k,max := max( V + q,max , V − q,max ) . (15)13 emma 5.1 (Preservation of box constraints) . Assume that the initial concentrations a i , r i for i = 1 , . . . , N are non-negative and satisfy the density constraint a i + r i ≤ .Then if the (CFL-like) condition (1 − τ HV max − τ max(2 HV max , Chλ out max( β a , β r ) , Chλ in max( α a , α r ))) ≥ holds we also have ≤ a ki , r ki , a ki + r ki ≤ for k = 1 , . . . , M, i = 1 , . . . , N, with a ki , r ki computed from a k − i , r k − i via (13) – (14) .Proof. We argue by induction and assume that at time t k the constraints are satisfied.Indeed, according to (13), we have that for i = 2 , . . . , N − a k +1 i = (cid:16) − τ H (cid:104) (1 − ρ ki − ) e − C ˜ C ˜ V ( V a,i − V a,i − ) − (1 − ρ ki +1 ) e − C ˜ C ˜ V ( V a,i − V a,i +1 ) (cid:105)(cid:17) a Ni + a ki − (1 − ρ ki ) e − C ˜ C ˜ V ( V a,i − − V a,i ) + a ki +1 (1 − ρ ki ) e − C ˜ C ˜ V ( V a,i +1 − V a,i ) ≥ (1 − τ HV max ) a ki , where we used that by assumption (1 − ρ ki +1 ) and (1 − ρ ki − ) are bounded by one and thatthe last two terms are non-negative. Thus the condition(1 − τ HV max ) ≥ a k +1 i ≥ r k +1 i . To show that (1 − ρ k +1 i ) ≥ − ρ k +1 i ) ≥ (1 − τ HV max ) (1 − ρ k +1 i ) (17)holds. It remains to consider the boundary contributions. In order to preserve positivitywhen outflow conditions are present (i.e. for a n and r ) we obtain the condition(1 − τ HV max − τ Chλ out max( β a , β r )) ≥ , (18)while in order to preserve ρ ≤ − τ HV max − τ Chλ in max( α a , α r )) ≥ . (19)To have (16)–(19) satisfied simultaneously finally yields the assumption(1 − τ HV max − τ max(2 HV max , Chλ out max( β a , β r ) , Chλ in max( α a , α r ))) ≥ . Remark 5.2 (Natural choice of numerical scheme) . The classical upwind scheme is notapplicable in this context as it only considers the particle movement initiated by the driftterm. In this context the drift term of species A can push against the drift of species R.This aspect is not covered by the classical upwind scheme.
For the case to two neurites the algorithm was implemented in MATLAB using 400grid points in each domain and a time step size of τ = 10 − . See also the pseudo-code insubsection A.1 in the appendix. 14 .1. Choice of Parameters and their Interpretation If not stated otherwise, we use the symmetric initial data shown in Table 1 for the numericsimulations. The corresponding value to a parameter in physical units can be calculatedby multiplying the typical variable with the value used in the numerics (for reasonabilityof the data see [39], [40], [27]).The potentials V a ( x ) = 1 . x and V r ( x ) = − . x translate to that fact that particles oftype a move anterograde and particles of type r move retrograde with different velocities.In Figure 3 b) the velocity of vesicles that are marked by Vamp2 is shown. In practicedifferent species of vesicles with different velocities have been observed ([16], [33]). As therange of the velocities of the anterograde transport is 1 − . µ m / sec and the range of thevelocity of retrograde transport is 1 − µ m / sec and the typical velocity is 1 µ m / sec, wechose the mean of those ranges. The diffusion constant ε is not biological meaningful asvesicles do not diffuse but the formal inclusion of this effects justifies to neglect reversemovement of vesicles. The value of ε = 0 .
05 is purely estimated.For the maximal density in the pools we do not have a choice: A neurite of 50 µ m lengthhas volume V N = πr h = 39 , µ m . A vesicle with diameter d = 130 nm has volume V v = πr = 0 , µ m . Therefore, a neurite with a length of 50 µ m can contain amaximum of 34 000 vesicles. The soma is estimated to contain about 6000 vesicles and thepools in the growth cones at the tip of the neurites about 100 vesicles. Consequently, themass of the vesicles in the soma should be 0.175 times as big as the mass of vesicles in a50 µ m long neurite, i.e. Λ maxsom = 0 .
175 and Λ maxN,1 = Λ maxN,2 = 0 , µ m [0,1] [0,50 µ m]Ω µ m [0,3] [0,15 µ m] T
100 s 50 1 h 23 min a , a
15 ves /µ m 0.1 1.5 vesicles/ µ m r , r
15 ves /µ m 0.1 1.5 vesicles/ µ m ε − µ m / sec 0.05 5 · − µ m / sec α r, , α r, , α a, , α a, / sec 0.8 0.8 vesicles/s β r, , β a, , β r, , β a, − µ m / sec 15 1.5 µ m / s V a ( x ) 1 µ m / sec 1 . x . µ m/s V r ( x ) 1 µ m / sec − . x − . µ m/sΛ - 0.12 4114 vesiclesΛ , Λ - 0.0015 50 vesiclesΛ maxsom - 0.175 6000 vesiclesΛ maxN,1 , Λ maxN,2 - 0.0029 100 vesiclesTable 1: Initial data:
Symmetric initial data used for the numerical simulation and theircorresponding real data where the corresponding real data is the product of thetypical variable and the value in the numerics by construction.
In our numerical analyzation we performed two experiments with symmetric initial datafor both neurites except for their length, see Table 1. In the first experiment the domains15ad a similar length, i.e. Ω = [0 , , Ω = [0 , . = [0 , , Ω = [0 , . N,1 , thebar in the middle the value of Λ som and the right one the value of Λ
N,2 . The left diagramshows the current vesicle concentration of anterograde (green) and retrograde (red) movingvesicles in neurite 1 and the right the one the concentration in neurite 2.The aim of these experiments was the analysis of asymmetry-formation arising uponthe basis of symmetric initial data. In particular, we regard asymmetries in the vesicleconcentration in the growth cones as different growing potentials of the neurites. Therefore,if the concentration in the pools Λ
N,1 and Λ
N,2 are not equal, the neurite with the higherconcentration in the pool has a higher growing potential. This can be justified by the factthat we assume that vesicle merging with the membrane at the growth cone drive thegrowing process of the neurite.As visible in Figure 8 (b) the length difference of the neurites leads to a quick symmetrybreaking especially in the growth cones Λ
N,1 and Λ
N,2 . Our numerical experiments evensuggest that for a small initial length difference, there is nearly no asymmetry, see Figure 7.
Classically, in the context of dynamic systems a stable state without least energy is calleda metastable state. Therefore the system stays in that state if no external energy is addedwhereas a certain amount of energy can result in further time evolution and the systemcoming to its true stable state with least energy.In our model in the case with very different initial lengths we monitor a quite similarbut not equal effect where the particle concentration seems to have already converged toits equilibrium until a sudden rapid change in the vesicles concentration occurs at the tipof the longer neurite (Λ
N,2 ) after some time. With the initial data shown in Table 1 andΩ = [0 , , Ω = [0 , . t = 26 and t = 36, see Figure 9 (b). In the case where the initial length is nearlysimilar this feature does not occur, see Figure 9 (a).
6. Conclusion
Different experimental approaches have shown that a neurite has to extend beyond aminimal length to become an axon ([11], [15], [46]). During axon specification, intracellulartransport is polarized towards the nascent axon to allow its extension ([32], [3]), whereseveral molecular mechanisms have been proposed for the length-dependent specification ofaxons ([1], [31], [32]). Our live cell imaging experiments indicate that the flow of vesiclesinto a neurite increases when it extends indicating that vesicle transport rates depend onchanges in neurite length. The transport along microtubules connects pools of vesiclesin the cell body and at the tip of neurites. The number of vesicles in the pool at the tipof the neurites reflects their growth potential because it provides material for membraneexpansion ([28], [27]). Our simulations show that the number of vesicles in the pool ishigher in the longer neurite of the model neuron once the length advantage of the longerneurite has exceeded a certain length. In addition, the simulations show oscillations in the16a)(b)(c)(d)Figure 7:
Evolution over Time of the Vesicle Concentration in two Neurites withNearly Similar Length:
The vesicle concentration for Ω = [0 , , Ω = [0 , . t = 0, (b) t = 10, (c) t = 50,(d) T = 100. 17a)(b)(c)(d)Figure 8: Evolution over Time of the Particle Concentration in two Neuriteswith very Different Length:
The vesicle concentration for Ω = [0 , , Ω =[0 , .
3] and initial data (1). (a) The initial concentration at t = 0, (b) t = 10, (c) t = 50, (d) T = 100. 18a)(b)Figure 9: Evolution over time of concentration in the pools
The time evolution ofΛ som , Λ N,1 , Λ N,2 solving (12) and initial data (1). (a) For Ω = [0 , , Ω = [0 , . = [0 , , Ω = [0 , . Motivated by the fact that the length of neurites is changing during the polarizationprocess, the most urgent extension of our model is the feature of a growing and shrinkingdomain. Therefore we have to consider a free boundary value problem, i.e. the lengths ofthe domains Ω ( t ) and Ω ( t ) become time dependent.The difficulty arising in this method is the following: In reality, growing and shrinkingis a continuous process but numeric simulations are always discrete. Thus in numericsimulations the domain has to shrink by segments but currently the approximation of whathappens to the vesicles located on these intervals is not obvious as in reality situations likethese never occur.Furthermore, as neurites grow by exocytosis of vesicles in the membrane, a production19erm of vesicles in the soma is necessary since at present the total mass of vesicles isconstant which prevents the neurite from intensive growth which requires a huge amountof vesicles. Furthermore, as there is a maximal concentration in the neurites it can happenthat two waves of particles are pushing onto each other resulting in traffic jams that arebiological not meaningful. Consequently, we have to choose the values of the productionterm carefully in a way that jams around the soma are prevented.The second feature, that would be suggestive to include, is an age-based populationstructure, i.e. the probability of vesicles leaving the pool increases with the length of theduration of its stay in the pool. Currently vesicles that enter a pool can immediately leaveit in the next time, but this additional delay could result in concentration oscillations inthe growth cones that reflect the cycles of stochastically occurring periods of extensionand retraction of neurites mentioned in the introduction.Finally, as pointed out in remark 4.1, challenging analytical problems arise in the contextof this model. Acknowledgments:
The authors acknowledge support by EXC 1003 Cells in Motionand EXC 2044 Mathematics M¨unster, Clusters of Excellence, M¨unster, funded by theGerman science foundation DFG. pEGFP VAMP2 was a gift from Thierry Galli (Addgeneplasmid http://n2t.net/addgene:42308 ; RRID:Addgene 42308). We thankUlrich M¨uller (Scripps Research Institute, La Jolla, CA, USA) for pDcx-iGFP. Furthermorethe authors would like to thank Martin Burger (FAU) for several interesting discussions.
A. Appendix
A.1. Pseudocode and Computing Time
For a better understanding of the numerics, we give a small overview on the computingtime and how the implementation works.As we only analyzed a one dimensional numerical problem, the solving algorithm is forsure very fast. The elapsed time for the algorithm for model neuron with pools was about6 minutes for T = 100 on MATLAB R2017b. For a better understanding of the code apseudocode is given in Algorithm 1. A.2. Glossary anterograde
Direction of transport from the soma to the tip of a neurite. 2, 3, 4, 8, 9, 15,16 endocytosis
Internalization of an area of cell membrane as a vesicle. 2, 3, 4 exocytosis
Insertion of a vesicle into the plasma membrane. 3, 19
GFP
Green fluorescent protein. 3 growth cone
Dynamic cellular structure at the tip of neurites that contains cytoskeletalelements and vesicles. 2, 3, 4, 7, 8, 12, 15, 16, 19, 20 microtubuli
Cytoskeletal elements that serves as tracks for intracellular transport. 2, 3, 1620 lgorithm 1:
Solving the time evolution of the vesicle concentrations in the modelneuron
Input :
Typical values for all parameters; initial pool concentrations and theirmaximum capacity; influx values and outflux velocities; potentials V a , V r ;parameter ε ; initial concentration of anteo- and retrograde moving vesiclesin both neurites Init :
Grid on space and time discretisation; Initialize each neurite as a structurearray N i consisting of its initial vesicle concentrations; influx rates andoutflux velocities; initial values of neighbouring pools, empty array for pooldevelopment;Calculate Scaling Parameters;Plot initial concentration for every time step do Update concentrations in N i with the particle-hopping algorithm; for every th time step do Update each figures that shows the current vesicles density in a neurite or apool; end
Save current pool concentrations in an array;Update concentrations in the pools Λ som and Λ N i ; end Plot development in the pools; polarization
Establishment of an asymmetric organization of cells. 1, 2, 3, 2, 3, 19 progenitor cell
Stem cell that generates neurons by cell division. 1 retrograde
Direction of transport from the tip of a neurite to the soma. 2, 3, 4, 6, 7, 9,15, 16 soma
Cell body of neuron. 2, 3, 4, 7, 8, 12, 13, 15, 19
Vamp2
Vesicular membrane protein. 4, 15
Vamp2-GFP
Fusion protein of Vamp2 to GFP. 3, 4 vesicle
Organelle separating its contents from the cytoplasm by a membrane (lipid bilayer).1, 2, 3, 4, 5, 4, 5, 4, 6, 7, 8, 9, 10, 12, 15, 16, 19, 20
References [1] Christin A. Albus, Ida Rishal, and Mike Fainzilber. Cell length sensing for neuronal growthcontrol.
Trends in Cell Biology , 23(7):305–310, July 2013.
2] Saad Ali, Ko Nishino, Dinesh Manocha, and Mubarak Shah. Modeling, simulation and visualanalysis of crowds: a multidisciplinary perspective. In
Modeling, simulation and visual analysisof crowds , pages 1–19. Springer, 2013.[3] F. Bradke and C. G. Dotti. Neuronal polarity: vectorial cytoplasmic flow precedes axonformation.
Neuron , 19(6):1175–1186, December 1997.[4] Paul C. Bressloff. Stochastic model of protein receptor trafficking prior to synaptogenesis.
Phys. Rev. E , 74:031910, 9 2006.[5] Paul C. Bressloff and Ethan Levien. Synaptic democracy and vesicular transport in axons.
Phys. Rev. Lett. , 114:168101, 4 2015.[6] M. Burger, M. Di Francesco, J.F. Pietschmann, and B. Schlake. Nonlinear cross-diffusion withsize exclusion.
SIAM J. Math. Anal. , 42(6):2842 – 2871, 2010.[7] Martin Burger, Marco Di Francesco, Jan-Frederik Pietschmann, and B¨arbel Schlake. Nonlinearcross-diffusion with size exclusion.
SIAM J. Math. Anal. , 42(6):2842–2871, 2010.[8] Martin Burger and Jan-Frederik Pietschmann. Flow characteristics in a crowded transportmodel.
Nonlinearity , 29(11):3528, 2016.[9] Jonathan A. Cooper. Cell biology in neuroscience: mechanisms of cell migration in the nervoussystem.
The Journal of Cell Biology , 202(5):725–734, September 2013.[10] B. Derrida. An exactly soluble non-equilibrium system: The asymmetric simple exclusionprocess.
Physics Reports , 301(1):65 – 83, 1998.[11] Carlos G. Dotti and Gary A. Banker. Experimentally induced alteration in the polarity ofdeveloping neurons.
Nature , 330(6145):254–256, November 1987.[12] Virginie Ehrlacher and Athmane Bakhta. Cross-diffusion systems with non-zero flux andmoving boundary conditions.
Mathematical Modelling and Numerical Analysis , 2018.[13] MR Evans, DP Foster, C Godreche, and D Mukamel. Asymmetric exclusion model with twospecies: spontaneous symmetry breaking.
Journal of statistical physics , 80(1-2):69–102, 1995.[14] Santos J. Franco, Isabel Martinez-Garay, Cristina Gil-Sanz, Sarah R. Harkins-Perry, and UlrichM¨uller. Reelin regulates cadherin function via Dab1/Rap1 to control neuronal migration andlamination in the neocortex.
Neuron , 69(3):482–497, February 2011.[15] K. Goslin and G. Banker. Experimental observations on the development of polarity byhippocampal neurons in culture.
The Journal of Cell Biology , 108(4):1507–1516, April 1989.[16] Laura F. Gumy, Eugene A. Katrukha, Ilya Grigoriev, Dick Jaarsma, Lukas C. Kapitein,Anna Akhmanova, and Casper C. Hoogenraad. MAP2 Defines a Pre-axonal Filtering Zoneto Regulate KIF1- versus KIF5-Dependent Cargo Transport in Sensory Neurons.
Neuron ,94(2):347–362.e7, April 2017.[17] Stephanie L. Gupton and Frank B. Gertler. Integrin Signaling Switches the Cytoskeletal andExocytic Machinery that Drives Neuritogenesis.
Developmental Cell , 18(5):725–736, May 2010.[18] Yumiko Hatanaka and Kenta Yamauchi. Excitatory cortical neurons with multipolar shapeestablish neuronal polarity by forming a tangentially oriented axon in the intermediate zone.
Cerebral Cortex (New York, N.Y.: 1991) , 23(1):105–113, January 2013.[19] Catherine Jacobson, Bruce Schnapp, and Gary A. Banker. A change in the selective translo-cation of the kinesin-1 motor domain marks the initial specification of the axon.
Neuron ,49(6):797 – 804, 2006.
20] Ansgar J¨ungel. The boundedness-by-entropy method for cross-diffusion systems.
Nonlinearity ,28(6):1963, 2015.[21] Lemont B Kier, Robert Tombes, Lowell H Hall, and Chao-Kun Cheng. A cellular automatamodel of proton hopping down a channel.
Chemistry & biodiversity , 10(3):338–342, 2013.[22] Zofia M. Lasiecka and Bettina Winckler. Mechanisms of polarized membrane trafficking inneurons – focusing in on endosomes.
Molecular and Cellular Neurosciences , 48(4):278–287,December 2011.[23] Pierre Mangeol, Bram Prevo, and Erwin J. G. Peterman. KymographClear and KymographDi-rect: two tools for the automated quantitative analysis of molecular and cellular dynamicsusing kymographs.
Molecular Biology of the Cell , 27(12):1948–1957, April 2016.[24] Sonia Martinez-Arca, Rachel Rudge, Marcella Vacca, Gra¸ca Raposo, Jacques Camonis,V´eronique Proux-Gillardeaux, Laurent Daviet, Etienne Formstecher, Alexandre Hamburger,Francesco Filippini, Maurizio D’Esposito, and Thierry Galli. A dual mechanism controllingthe localization and function of exocytic v-snares.
Proceedings of the National Academy ofSciences , 100(15):9011–9016, 2003.[25] Takashi Namba, Yasuhiro Funahashi, Shinichi Nakamuta, Chundi Xu, Tetsuya Takano, andKozo Kaibuchi.
Extracellular and Intracellular Signaling for Neuronal Polarity , volume 95.July 2015.[26] Sylvia Neumann, Romain Chassefeyre, George E. Campbell, and Sandra E. Encalada. Kymo-Analyzer: a software tool for the quantitative analysis of intracellular transport in neurons.
Traffic (Copenhagen, Denmark) , 18(1):71–88, 2017.[27] Karl H. Pfenninger. Plasma membrane expansion: a neuron’s Herculean task.
Nature Reviews.Neuroscience , 10(4):251–261, April 2009.[28] Karl H. Pfenninger, Lisandro Laurino, Diego Peretti, Xiaoxin Wang, Silvana Rosso, GerardoMorfini, Alfredo C´aceres, and Santiago Quiroga. Regulation of membrane expansion at thenerve growth cone.
Journal of Cell Science , 116(Pt 7):1209–1217, April 2003.[29] Ekaterina Pronina and Anatoly B Kolomeisky. Spontaneous symmetry breaking in two-channelasymmetric exclusion processes with narrow entrances.
Journal of Physics A: Mathematicaland Theoretical , 40(10):2275–2286, 2 2007.[30] Santiago Quiroga, Mariano Bisbal, and Alfredo C´aceres. Regulation of plasma membraneexpansion during axon formation.
Developmental Neurobiology , 78(3):170–180, 2018.[31] Ida Rishal, Naaman Kam, Rotem Ben-Tov Perry, Vera Shinder, Elizabeth M. C. Fisher,Giampietro Schiavo, and Mike Fainzilber. A motor-driven mechanism for cell-length sensing.
Cell Reports , 1(6):608–616, June 2012.[32] Max Schelski and Frank Bradke. Neuronal polarization: From spatiotemporal signaling tocytoskeletal dynamics.
Molecular and Cellular Neurosciences , 84:11–28, 2017.[33] Max A. Schlager, Andrea Serra-Marques, Ilya Grigoriev, Laura F. Gumy, Marta Esteves daSilva, Phebe S. Wulf, Anna Akhmanova, and Casper C. Hoogenraad. Bicaudal d familyadaptor proteins control the velocity of Dynein-based movements.
Cell Reports , 8(5):1248–1256, September 2014.[34] B.A. Schlake.
Mathematical Models for Particle Transport: Crowded Motion . PhD thesis,Westf¨alische Wilhelms-Universit¨at M¨unster, 2011.
35] Bhavin Shah, Daniela Lutter, Yaroslav Tsytsyura, Natalia Glyvuk, Akira Sakakibara, J¨urgenKlingauf, and Andreas W. P¨uschel. Rap1 GTPases Are Master Regulators of Neural CellPolarity in the Developing Neocortex.
Cerebral Cortex (New York, N.Y.: 1991) , 27(2):1253–1269, 2017.[36] M J Simpson, B D Hughes, and K A Landman. Diffusion populations: Ghosts or folks.
Australasian Journal of Engineering Education , 15(2100):59–68, 2009.[37] Takuro Tojima and Hiroyuki Kamiguchi. Exocytic and endocytic membrane trafficking inaxon development.
Development, Growth & Differentiation , 57(4):291–304, May 2015.[38] Michinori Toriyama, Yuichi Sakumura, Tadayuki Shimada, Shin Ishii, and Naoyuki Inagaki.A diffusion-based neurite length-sensing mechanism involved in neuronal symmetry breaking.
Molecular Systems Biology , 6(1):394, 2010.[39] Krasimira Tsaneva-Atanasova, Andrea Burgo, Thierry Galli, and David Holcman. Quantifyingneurite growth mediated by interactions among secretory vesicles, microtubules, and actinnetworks.
Biophysical Journal , 96(3):840–857, 2 2009.[40] Fabio L. Urbina, Shawn M. Gomez, and Stephanie L. Gupton. Spatiotemporal organization ofexocytosis emerges during neuronal shape change.
The Journal of Cell Biology , 217(3):1113–1128, 2018.[41] Atul Kumar Verma, Natasha Sharma, and Arvind Kumar Gupta. Far-from-equilibriumbidirectional transport system with constrained entrances competing for pool of limitedresources.
Phys. Rev. E , 97:022105, 2 2018.[42] Tong Wang, Yang Liu, Xiao-Hui Xu, Cai-Yun Deng, Kong-Yan Wu, Ji Zhu, Xiu-Qing Fu,Miao He, and Zhen-Ge Luo. Lgl1 activation of rab10 promotes axonal membrane traffickingunderlying neuronal polarization.
Developmental Cell , 21(3):431–444, September 2011.[43] Alex H Williams, Cian O’Donnell, Terrence J Sejnowski, and Timothy O’Leary. Dendritictrafficking faces physiologically critical speed-precision tradeoffs. eLife , 5:e20556, 12 2016.[44] Amy M. Winans, Sean R. Collins, and Tobias Meyer. Waves of actin and microtubulepolymerization drive microtubule-based transport and neurite growth before single axonformation. eLife , 5:e12387, 2 2016.[45] Amy M Winans, Sean R Collins, and Tobias Meyer. Waves of actin and microtubule polymer-ization drive microtubule-based transport and neurite growth before single axon formation. eLife , 5:e12387, 2 2016.[46] Hideaki Yamamoto, Takanori Demura, Mayu Morita, Gary A. Banker, Takashi Tanii, andShun Nakamura. Differential neurite outgrowth is required for axon specification by culturedhippocampal neurons.
Journal of Neurochemistry , 123(6):904–910, December 2012., 123(6):904–910, December 2012.