Abstract
During tissue invasion individual tumor cells exhibit two interconvertible migration modes, namely mesenchymal and amoeboid migration. The cellular microenvironment triggers the switch between both modes, thereby allowing adaptation to dynamic conditions. It is, however, unclear if this amoeboidmesenchymal migration plasticity contributes to a more effective tumor invasion. We address this question with a mathematical model, where the amoeboidmesenchymal migration plasticity is regulated in response to local extracellular matrix resistance. Our numerical analysis reveals that extracellular matrix structure and presence of a chemotactic gradient are key determinants of the model behavior. Only in complex microenvironments, if the extracellular matrix is highly heterogeneous and a chemotactic gradient directs migration, the amoeboidmesenchymal migration plasticity allows a more widespread invasion compared to the nonswitching amoeboid and mesenchymal modes. Importantly, these specific conditions are characteristic for in vivo tumor invasion. Thus, our study suggests that in vitro systems aiming at unraveling the underlying molecular mechanisms of tumor invasion should take into account the complexity of the microenvironment by considering the combined effects of structural heterogeneities and chemical gradients on cell migration.
Introduction
Solid tumors become invasive if cells migrate away from their initial primary location. The tumor cell microenvironment with its variety of biomechanical and molecular cues plays a critical role in the localized invasion throughout the tissue. For example, tumor cells are known to react to soluble factors, such as chemokines and growth factors, by directional movement towards the extracellular gradient of chemicals^{1}. The importance of the extracellular matrix (ECM) in tumor invasion has recently received particular attention^{2,3}. The ECM, which fills the space between cells through a complex organization of proteins and polysaccharides, imposes a biomechanical resistance that moving cells need to overcome. To migrate, tumor cells might either degrade the ECM to pass through, or modify their shape and squeeze through the ECM pores^{4}. These two distinct migration modes are commonly termed “pathgenerating” mesenchymal and “pathfinding” amoeboid mode^{5,6}. The mesenchymal migration mode is characterized by an elongated cell morphology, adherence to the surrounding ECM mediated by integrins and ECM degradation by proteases^{7}. In contrast, during amoeboid migration, cells are highly deformable, their adhesion to the ECM is rather weak, and proteolytic activity is reduced or absent. The low adhesion of cells in the amoeboid migration mode enables the cells to move comparatively faster than those migrating in mesenchymal migration mode^{5,8}. Remarkably, tumor cells are able to adapt their migration mode to changing microenvironmental conditions^{3,4,7,9,10}, a feature called migration plasticity. In particular, it has been observed that ECM parameters like density or stiffness, regulate the transition between amoeboid and mesenchymal migration modes, which is very dynamic and comprises intermediate states, where cells display properties of both migratory phenotypes^{3,9,11}. At the subcellular to cellular level, the impact of ECM properties on molecular mechanisms of individual cell motility has been studied using both experimental^{7,10,12} and theoretical^{13,14,15,16,17,18} approaches. However, it remains unclear how the adaptation responses of amoeboid and mesenchymal migration modes contribute to the tumor invasion process. In particular, it is not known if and how amoeboidmesenchymal plasticity allows a more effective invasion compared to the nonadaptive amoeboid or mesenchymal modes. So far, only the impact of interactions between nonswitching moving cells and the ECM on tumor invasion has been studied^{4,6,19}. Hecht et al.^{6} considered an agentbased model for fixed proportions of amoeboid and mesenchymal cells migrating in a mazelike environment with directional cue, and showed that an amoeboid cell population can benefit from the presence of a small number of path creating mesenchymal cells.
In this work, we develop a mathematical model to study the consequences of amoeboidmesenchymal migration plasticity on tumor invasion in a switching population of cells that adopt their migration mode in response to the ECM conditions. We are interested in the question under what environmental conditions the plasticity of amoeboid and mesenchymal migration modes provides an advantage for tumor invasion. We formulate and analyze a cellular automaton model, where cells are able to switch between a slow mesenchymal migration mode with ECM degradation, and a fast amoeboid migration mode without ECM degradation, depending on the biomechanical resistance imposed by the ECM. With computer simulations of the mathematical model, we compare the invasion behavior of a population of cells where each cell is allowed to switch between ameoboid and mesenchymal migration modes with the behavior of a nonswitching cell population, under different environmental conditions. In particular, we distinguished spatially homogeneous and heterogeneous ECM resistance conditions, ranging from weakly structured to highly structured, in combination with different responses towards a chemotactic gradient which directs cell migration. Our numerical analysis reveals that ECM structure and presence of a chemotactic gradient are key determinants of the model behavior. The amoeboidmesenchymal migration plasticity leads to a more widespread invasion compared to the nonswitching situation only in complex microenvironments, more specifically, if the ECM is strongly heterogeneous and a chemotactic gradient directs migration. The latter conditions are characteristic for in vivo tumor invasion. This suggests that experimental studies on tumor invasion should represent this complexity of the microenvironment.
Methods
The model
We develop a mathematical model to study the effects of amoeboidmesenchymal migration plasticity on tumor invasion. To determine the specific impact of migration plasticity of individual cells on overall cell population invasion dynamics, we coarsegrain to a cellbased model, namely a probabilistic cellular automaton (CA), which is analyzed at the population level. Probabilistic cellular automata are a class of spatially and temporally discrete mathematical models which allow to (i) model cellcell and cellECM interactions, as well as cell migration, and (ii) to analyze emergent behavior at the cell population level^{20,21,22,23,24,25,26}.
We consider the ECM as a physical barrier which imposes a resistance against the moving cell body. A widely studied parameter which mechanically impedes cell movement is the ECM network density. Other physical properties of the ECM, such as porosity, as well as biomechanical properties like ECM stiffness, have been observed to either enable or restrict cell migration. Importantly, the different ECM properties are not independent but rather connected^{2}. Thus, for instance, the density of fibrillar ECM is positively interconnected with stiffness and inversely proportional to pore size, such that alterations of either property impact the overall ECM structure^{12}. In view of this, we incorporate ECM properties like density and stiffness into a lumped parameter called “resistance”.
We distinguish two migration modes, namely amoeboidlike (A) and mesenchymallike (M) migration. We assume that amoeboidlike cells migrate in a protease and integrinindependent way, driven by Rho/ROCKmediated contractility. In contrast, mesenchymallike cells migrate protease and integrindependently, where Rho/ROCK signaling is inhibited. Thus, the amoeboid and mesenchymal migration phenotypes in our model represent two ends in a broad spectrum of individual cell migration modes^{11}.
We assume that cells change their migration phenotype in dependence of local ECM resistance. This assumption is supported by several studies. In particular, it has been reported that under soft ECM conditions, cells are roundshaped and their migration is independent of protease activity but rather driven by Rho/ROCKmediated contractility. In contrast, under higher ECM stiffness conditions, Rho/ROCK signaling is inhibited, cells show elongated morphology and their migration relies on protease activity^{4}. In another recent study it has been shown that in the absence of focal adhesions and under conditions of confinement, mesenchymal cells can spontaneously switch to a fast amoeboid migration phenotype^{8}. Accordingly, we assume in our model that amoeboidlike cells move relatively fast and are not able to degrade ECM, while mesenchymallike cells move slower and locally degrade the ECM. We further assume that cell migration speed is highest for low ECM resistance but decreases with increasing ECM resistance. This assumption is based on findings by Wolf et al.^{12} migration speed gradually diminishes as a function of decreasing pore size (which is associated with increasing ECM resistance) with a steeper slope when cells migrate in an amoeboid fashion. An optimal migration speed has been observed during migration through tissue of sufficient to high porosity (which corresponds to low ECM resistance). In our coarsegrained model, we do not explicitly account for cell shape changes or Rho/ROCK and integrindependent signaling pathways.
Furthermore, we assume that the direction of cell movement is determined by a constant chemotactic gradient. Cells respond to this external gradient with a certain intensity, independent of their migration phenotype.
Our CA model is described on a twodimensional regular lattice \({\rm{S}}\subset {{\rm{Z}}}^{{\rm{2}}}\). Each lattice site is in one of a set of states (η, μ) ∈ {0, 1, 2} × [0,1], with η denoting the cell state value and μ the ECM state value. We interpret the cell state value 0 as an unoccupied lattice site, cell state value 1 as an Acell and cell state value 2 as an Mcell. The ECM state value represents the physical resistance of the ECM.
The time evolution of our model is defined by the following rules:

Phenotypic switch. A cell changes its phenotype depending on the local ECM resistance μ. In particular, an Acell changes its migratory phenotype with rate αμ and an Mcell with rate β(1–μ), respectively, see Fig. 1(a).

ECM degradation. Mcells locally degrade the ECM with constant rate δ>0.

Cell migration. Acells migrate with rate λ _{ A }(μ), Mcells with rate λ _{ M }(μ), where both λ _{ A }(μ) and λ _{ M } (μ) depend on the local ECM resistance μ, see Fig. 1(b). The direction of movement is determined by a chemotactic gradient. Cells respond to this gradient with an intensity κ > 0. If κ = 0, cells perform independent random walks, which is equivalent to the case when no gradient field is present.
At each discrete model time step, one cell is selected at random and updated according to the rules (R1)–(R3). On average, each cell is updated once per n consecutive time steps if there are n cells on the lattice. The sequence of n time steps corresponds to one Monte Carlo Step, which is the time unit of our system.
Cell migration, rule (R3), is implemented as an exclusion process^{27}, which means that cells move to neighboring nodes if the target node is empty, otherwise the movement is aborted. Thus, the model accounts for steric interactions between cells due to volume exclusion. A detailed description of the model update rules (R1)–(R3) can be found in Supplementary Section 1. A schematic illustration of the update rules is depicted in Fig. 2.
In each simulation, we track individual cells and measure the migration distance (d) from the initial position for each individual cell. At the population level, the model observable characterizing invasion dynamics is the migration distance from the initial position averaged over the entire cell population (d _{ p }). In addition, we measure the maximum migration distance (d _{max}) of the cells in the population in each simulation. We consider different simulations to compare (d _{ p })values (respective (d _{max})values) between populations of cells where each cell is allowed to switch between ameoboid and mesenchymal migration modes and those of nonswitching cell populations (Δd _{ p }, respective Δd _{max}). Calculation details are given in Supplementary Section 2.
Simulation study
The CA model is simulated on a rectangular lattice with dimensions S_{1} × S_{2}. The vertical length of the lattice is given by the r _{1} coordinate, \(1\le {r}_{1}\le {S}_{1}\) and the horizontal length of the lattice is given by the r _{2} coordinate, \(1\le {r}_{2}\le {S}_{2}\). For the simulations we chose initial and boundary conditions such that we can analyze cell migration using onedimensional, vertically averaged, cell density profiles. Periodic boundary conditions are applied along the horizontal boundaries (r _{1} = 1 and r _{1} = S _{1}), and reflecting boundary conditions on the vertical boundaries (r _{2} = 1 and r _{2} = S _{2}). The initial conditions of the model are illustrated in Fig. 3. At the initial time, 50% Mcells and 50% Acells are placed at random along the left border at r _{2} = 1, as depicted in Fig. 3(a). Notice that the results do not depend on the initial fractions of M and Acells but on the phenotypic switch ratio α/β (data shown in Supplementary Fig. S2). Figure 3(b) shows the chemotactic gradient field, which controls the direction of cell movement. The chemotactic gradient is chosen such that the onedimensional cell density profile advances along the S _{2}axis. In particular, the gradient is specified by a linear function
which describes an attractive signal from the target location r _{2} = S _{2}, see Fig. 3(b). We consider different chemotactic responsiveness parameters κ ≥ 0, reflecting a range from undirected cell movement (κ = 0) to a high response to the presence of a chemotactic gradient (κ ≫ 0).
For the ECM resistance, different initial scenarios are implemented. Homogeneous ECM is modeled by a constant ECM resistance, as illustrated in Fig. 3(c). Heterogeneous ECM is modeled by a sinusoidal function
where ξ~ (0, 1) is a fixed uniformly distributed random variable and θ ∈ [0,1] is a heterogeneity parameter. On average, the ECM resistance is μ(r _{1},r _{2}) ≈ 0.5. The heterogeneity parameter controls the level of ECM structure, ranging from completely random (θ = 0) to completely coherent structures (θ = 1). An example of a less structured ECM is shown in Fig. 3(d), a highly structured ECM is illustrated in Fig. 3(e). We have also tested other functional representations of the ECM resistance distribution. The results we describe below remain valid for other functional forms of μ, as long as μ is not a constant but modeled by a spatial function μ: S _{1} × S _{2}→ [0, 1] that depends on the spatial position (r _{1}, r _{2}) and on a random component, where the latter adds local irregularities to an otherwise regular spatial structure. See Supplementary Fig. S1 for further examples.
The A and Mcell migration rates are modeled by sigmoidal functions
where C _{ A } and C _{ A } are constants, chosen such that the migration speed of Acells is higher than that of Mcells for low to moderate ECM resistance, see Fig. 1(b). For μ = 0.5, the slopes of the migration rate functions are steepest. The sigmoidal shape of the migration rate functions accounts for three levels of matrix resistance on cell migration: low, middle and high. Cell migration is unhindered for low ECM resistance, linearly decreases for medium ECM resistance and is impeded for high ECM resistance.
The focus of our study is on the impact of the switch parameters α and β on the migration behavior of the cell population. We refer to a switching cell population if cells are allowed to change their phenotype (α, β > 0), and to a nonswitching cell population if cells do not adapt their phenotype (α = β = 0). For switching populations, we change the phenotypic switch parameters α = 1/10, …, 1, β = 0, β = 1 and α = 1, β = 1/10,…,1, such that the phenotypic switch ratio is α/β ∈ {1/10, 2/10, …, 1, 2, …, 10}. A low switch ratio means that cells predominantly switch to Atype, while a high switch ratio leads to a preferred switch to Mtype cells. We measure how far the switching population migrates from the start position. Notice that this migration distance does not depend on the particular choice of α and β as long as the switch ratio α/β remains fixed (simulations not shown). In order to identify novel behaviors under the influence of the phenotypic switching, we compare our results with the nonswitching situation α = β = 0, that is when A and Mcells do not change their phenotype. For the nonswitching population, we simulate the same number of cells as in the switching scenario, with changing ratio of A and Mcells. The fraction of Mcells in a nonswitching population is denoted by γ. Notice that the nonswitching situation reflects a modeling scenario similar to the one studied by Hecht et al.^{6}.
In our simulations, we compare the migration behavior of switching and nonswitching cell populations under different environmental conditions: homogeneous or heterogeneous ECM structure in combination with different chemotactic responsiveness. A summary of all possible simulation scenarios is given in Table 1. In each scenario, we compare the average migration distance d _{ p } and the maximum migration distance d _{max} of switching and nonswitching populations of size n = 50. For the switching population, we change the phenotypic switch ratio α/β as described above. For the nonswitching population, we vary the fraction of Mcells: γ = 1 (pure Mcell population), γ = 0.7 (70% Mcells, 30% Acells), γ = 0.3 (30% Mcells, 70% Acells) and γ = 0 (pure Acell population). Each simulation is performed for 200 Monte Carlo steps (see Supplementary Section 2). Throughout the study, we change the migration rate ratio c _{ M }/c _{ A } by varying the Mcell migration rate (\(0\le {c}_{M}\le 1\)), while keeping the Acell migration rate fixed (c _{ A } = 1). A small ratio c _{ M }/c _{ A } indicates a large difference between the two migration rates, if the ratio c _{ M }/c _{ A } is close to 1, Mcells migrate nearly as fast as Acells. The ECM degradation rate is set to δ = 0.1 ensuring partial but not complete degradation. An overview of the model parameters is given in Table 2.
As a final step, we study if cooperative effects in a switching population contribute to an efficient migration of the entire cell population. To this end, we compare the migration behavior of a single cell moving alone with that of an individual cell moving in a cell population of size n = 500. In particular, we first simulate n repetitions of the single cell scenario, each time measuring the migration distance d of the single cell, and at the end determine the maximum migration distance d. Secondly, we simulate a cell population of size n, using the same model parameters as in the singlecell scenario. We measure the migration distance d of each individual cell of the population and determine the maximum migration distance d Both simulations are repeated 100 times to account for stochastic fluctuations.
Results
Advantage of nonswitching behavior under homogeneous ECM conditions
We first study if phenotypic switching provides an advantage in terms of cell population migration distance under homogeneous ECM conditions. Exemplary cell trajectories are shown in Supplementary Figs S3 and S4. Here, we compare the average and maximum migration distances of switching and nonswitching populations under homogeneous ECM conditions, varying the level of μ(r _{ 1 }, r _{2}) = constant initially, in combination with a high chemotactic responsiveness. Figure 4 illustrates simulation results under different homogeneous ECM resistance conditions: homogeneous, low ECM resistance (Fig. 4(a)) and homogeneous, high ECM resistance (Fig. 4(b)). The figure shows the average migration distance d _{ p } of the switching cell population as a function of the phenotypic switch ratio α/β. Also shown is the average migration distance d _{ p } of nonswitching populations with different Mcell fraction γ ∈ {0, 0.3, 0.7, 1}, which remains constant as a function of α/β. In Fig. 4(a), for homogeneous, low ECM resistance, we observe that the average migration distance d _{ p } of the nonswitching population depends monotonously on the Mcell fraction γ: the higher γ, the smaller the average migration distance d _{ p }. Similarly, the average migration distance d _{ p } of the switching population is highest for low switch ratio α/β, which means that cells predominantly switch to Atype. As the phenotypic switch ratio α/β increases, that is fewer Acells and more Mcells are present, the average migration distance d _{ p } of the switching population decreases as the phenotypic switch ratio α/β increases. In contrast, under homogeneous, high ECM conditions, the average migration distance d _{ p } of a nonswitching population increases monotonously with increasing Mcell fraction γ, see Fig. 4(b). Likewise, for the switching population we observe that increasing the switch ratio α/β, which leads to a preferred switch to Mtype cells, increases the average migration distance d _{ p }. Under both homogeneous ECM conditions, low and high, we find that the possibility to switch the phenotype does not constitute a benefit in terms of migration distance of the entire cell population. Additional simulation data with the maximum migration distance d _{max}, shown in Supplementary Fig. S5, confirms our finding. The critical ECM resistance value above which the greatest migration distance is no longer observed for the pure Acell population but for the pure Mcell population, is approximately μ(r _{1}, r _{2}) = 0.5.
The observed behavior can be understood by considering the cell migration characteristics. On the one hand, for low to moderate, homogeneous ECM resistance, cell migration is not hindered and Acells migrate with higher rate than Mcells. Thus, the higher the fraction of Acells is, the higher the migration distance d _{ p } of the entire cell population. On the other hand, sufficiently high homogeneous ECM resistance presents a barrier for Acells and only Mcells are able to create space to migrate through the ECM. Therefore, a high Mcells fraction increases the migration distance d _{ p }.
Migration plasticity can be advantageous under heterogeneous ECM conditions
In a next step, we determine if a heterogeneous ECM structure provides an environmental condition under which the switching population of cells migrates the greatest distance from the initial position. To this end, we compare the average and maximum migration distance of switching and nonswitching populations under heterogeneous ECM conditions (μ as defined in equation (2)), varying the levels of ECM heterogeneity θ, in combination with a high chemotactic responsiveness. Under heterogeneous, weakly structured ECM conditions, we observe that the average and the maximum migration distance is obtained by the nonswitching population. In contrast, under heterogeneous, highly structured ECM conditions, we find that the average and the maximum migration distance of the switching population may be lower or higher than of the nonswitching population, depending on the migration rate ratio c _{ M } /c _{ A }. In the following, we present simulation results about the average migration distance d _{ p }. Qualitatively similar results obtained by considering the maximum migration distance d _{max} are illustrated in Supplementary Fig. S6. Figure 5(a) shows the average migration distance d _{ p } of switching and nonswitching populations as a function of the switch ratio α/β under heterogeneous, highly structured ECM conditions and a large migration rate ratio (c _{ M }/c _{ A } ≈ 1). We observe that the higher the portion of Mcells, the higher the average migration distance d _{ p } of the nonswitching population. Similarly, for the switching population, the higher the switch ratio α/β, that is cells are predominantly Mtype, the higher the average migration distance d _{ p }. The nonswitching cell population with γ = 1 (pure Mcell population) migrates the greatest distance from the initial position. The situation is different if the migration rate ratio is small (c _{ M }/c _{ A } ≫ 1): while the average migration distance d _{ p } of the nonswitching population is higher, the more Mcells are present, the maximum average migration distance d _{ p } of the switching population is obtained for α/β = 1, that is when both cell types are equally present, see Fig. 5(b). Under the latter conditions, that is the ECM is heterogeneous, highly structured and the difference between A and Mcell migration speeds is large, we observe that the switching behavior provides an advantage in terms of cell population migration distance.
Figure 6 illustrates exemplary individual cell trajectories for switching and nonswitching populations under heterogeneous, highly structured ECM conditions with small migration rate ratio (c _{ M }/c _{ A } ≪ 1). Figure 6(a) shows that Acells are locally impeded by high ECM resistance regions and move around such regions. In contrast, Mcells do not make a detour and move through highly structured areas, see Fig. 6(b). The combined benefits of fast Acell migration in low ECM resistance regions and the ability of Mcells to locally degrade the ECM, lead to a high average and maximum migration distance for switching populations, see Fig. 6(c). The advantage of the switching behavior is no longer observed if the migration rate ratio is large (c _{ M }/c _{ A } ≈ 1), as shown in the trajectory representation in Supplementary Fig. S7. Notice that the trajectory figures point out that the variance of the average migration distance d _{ p } within the simulations is quite large (see also Supplementary Fig. S6). Such a large “withinsimulation variances” is explained by the stochastic effects of the switching mechanism and environmental conditions. The “betweensimulation variances”, however, are almost negligible (coefficient of variation below 5%, see Fig. 5). This can be expected since there are 50 cells observed in each simulation such that the law of large numbers applies.
Figure 7 shows the observed dependency of the average migration distance d _{ p } on the level of ECM heterogeneity and the cell migration rate ratio. In the phase diagram, the difference between the switching population with maximum migration distance d _{ p } with respect to varied switch ratio α/β and the nonswitching population with maximum dp is illustrated. We observe that the switching behavior is favorable if the ECM is highly structured and the migration rate ratio is small. Otherwise the maximum average migration distance dp is reached by the nonswitching population. We conclude that, provided the difference between A and Mcells is sufficiently large enough, a high chemotactic responsiveness together with a pronounced physical resistance imposed by the ECM, represents an environmental situation under which phenotypic adaptation of cell migration modes is beneficial. We hypothesize that cells are faced with a tradeoff between high ECM resistance and chemotactic responsiveness: the higher the chemotactic responsiveness, the stronger the attempt of cells to move in a directional manner towards a target location. The higher the ECM resistance, the harder it is for the cells to move forward and the locally favorable direction may not be the overall best choice. While rather fast Acell movement is impeded by high ECM resistance, slower moving Mcells create their own short paths. Thus, if cells in the population are allowed to adapt their phenotype to the local environmental conditions, they are able to follow the shortest path towards the target location and to overcome local ECM constraints.
Chemotactic responsiveness influences the efficiency of cell movement
We now compare the migration behavior of switching and nonswitching populations under different chemotactic responsiveness. To determine if the response towards the chemotactic gradient alters the simulation results described in the previous sections, we repeat the simulation study where we explored homogeneous and heterogeneous ECM conditions, this time however changing the chemotactic responsiveness. The result of this sensitivity analysis is that under homogeneous ECM conditions, as well as under heterogeneous, weakly structured ECM conditions, the migration distance of the nonswitching population is higher than of the switching population, independent of the chemotactic responsiveness, as shown in Supplementary Figs S8 and S9. Under heterogeneous, highly structured ECM conditions, we find that the advantage of phenotypic switching in terms of cell population migration distance depends on the chemotactic responsiveness, such that the advantage of the switching behavior becomes more pronounced the higher the chemotactic responsiveness. In particular, if the ECM is highly structured and the migration rate ratio c _{ M }/c _{ A } is small, the difference between the migration distance of switching and nonswitching populations increases with increasing chemotactic responsiveness, see Supplementary Fig. S10.
Cooperative effects increase the migration distance
To study if cooperative effects in a switching population influence the efficiency of cell movement, we measure the maximum migration distance d of the switching population for different population sizes. In particular, the maximum distance d of a single switching cell migrates among 500 simulation repetitions is compared to the migration distance d of the farthest individual cell in a switching cell population of size n = 500, as described in the Methods section. Figure 8 illustrates simulation results under heterogeneous, highly structured ECM conditions. We observe that the maximum migration distanced is higher for a switching cell population than for a single cell which adapts its migratory phenotype. For comparison, we repeat the simulation procedure for the nonswitching situation, with results also shown in Fig. 8. We find that the maximum migration distance d of the nonswitching population is lower than the switching situation, and lowest in the case of a single nonswitching cell. We conclude that cooperative effects are present in a switching population. We hypothesize that Mcells benefit Acells by creating space in an otherwise impenetrable ECM. Our argument is in agreement with the findings of Hecht et al.^{6}, who showed that in a nonswitchingpopulation, paths created by mesenchymal cells are used by amoeboid cells. In fact, our results show that cooperative effects are present in a switching population and allow the cells to migrate longer distances compared to the nonswitching situation.
Discussion
We developed a cellular automaton model to study the impact of amoeboidmesenchymal migration plasticity on tumor invasion. The switch between the two migration modes was assumed to depend on the local microenvironment through the biomechanical resistance imposed by the ECM. Cells in the model are able to switch between a slow mesenchymal migration mode with ECM degradation, and a fast amoeboid migration mode without ECM degradation, depending on ECM resistance. With computer simulations of the mathematical model we analyzed invasion dynamics, characterized by the migration distance of the cell population, under different environmental conditions. In particular, we distinguished spatially homogeneous and heterogeneous ECM resistance conditions, ranging from weakly structured to highly structured, in combination with different responses towards a chemotactic gradient which directs cell migration. Provided that the difference between amoeboid and mesenchymal migration speeds is sufficiently large, as observed in experiments^{8}, we found that an amoeboidmesenchymal migration plasticity provides an advantage in terms of migration distance if the spatial structure of ECM resistance is strongly heterogeneous and if migration is directed.
In our model, we only distinguish two migration modes, which we term amoeboid and mesenchymal. The mesenchymal mode is characterized by slow migration speed and ECM degradation, while cells in amoeboid mode move fast and do not degrade the ECM. It is clear that this is a strong simplification of biological reality. Cells may display features of both migration modes, in particular concerning parameters that are not explicitly incorporated in our model, such as Rho/ROCK signaling and cell shape changes. However, our coarsegrained model captures the extreme cases of plasticity behavior. Therefore, we expect that our predictions remain valid also for intermediate situations.
In this paper, we assumed that physical and biomechanical properties of the ECM, such as density and stiffness, can be combined in a single parameter called ECM resistance. We are aware that this simplification does not account for the structural and molecular complexity of the ECM. However, we coarsegrain the extra and intracellular details into a simplified model to provide a basic understanding of emerging plasticity effects. Our model is a first step to analyze consequences of migration plasticity on tumor invasion.
The chemical gradient which directs cell migration in the model is assumed to be independent of the ECM resistance. It is known that chemotaxis is important for tumor invasion^{1}. Other types of directed cell migration have been observed in tumor cells, such as haptotaxis (movement along gradients of ECM ligands) or durotaxis (gradient of ECM stiffness). Our choice to consider ECMindependent chemotaxis serves as a first attempt to incorporate the effects of directed migration into an amoeboidmesenchymal plasticity model for tumor invasion. We expect that our results hold for other than chemotactic gradients.
In the model, we assumed that amoeboid and mesenchymal cell migration speeds decrease monotonically with increasing ECM resistance. We based our assumption on findings of Wolf et al.^{12}, where it was shown, by using breast cancer and fibrosarcoma cell lines, that cell migration speed does decrease monotonically with decreasing ECM pore size but does not depend on ECM stiffness. Both pore size and ECM stiffness contribute to the biomechanical ECM resistance against cell migration. However, experimental findings on the relation between cell migration speed and the ECM environment are controversial. For example, experiments with brain tumor cell lines have shown that an increase of ECM stiffness induces an increased cell migration speed^{28}. A possible explanation for different experimental findings might be the type of tumor studied. Furthermore, it is wellknown that while ECM porosity and stiffness can be independently controlled under in vitro conditions, in vivo, stiffening of the ECM is highly correlated with alterations of the ECM fiber structure^{29}. Clarifying the particular contributions of different ECM properties to cell migration is primarily an experimental task.
In our model, we indirectly account for energetic costs of migration plasticity. Different cellular tasks as migration and matrix degradation consume energy and we assume a tradeoff between the energies invested in these tasks. Consequently, cells in mesenchymal migration mode which degrade the ECM move slower than amoeboid cells which do not spend energy for proteolysis. This assumed difference in migration speeds is supported by several biological findings^{8,30}. Our results on the existence of cooperative effects in switching cell populations support previous findings of Hecht et al.^{6}. They showed that an amoeboid cell population can benefit from the presence of a small number of path creating mesenchymal cells in a model where metabolic cost for ECM degradation is modeled directly as tradeoff with migration speed.
To our knowledge, our theoretical study is the first that highlights the effects of plasticity and heterogeneity in cell migration modes on tumor dissemination. Previous models studied the role of different types of cellular heterogeneity and plasticity for tumor growth, dissemination, and invasion^{21,31,32,33}.
Our model is a cellbased, discrete cellular automaton model which we have analyzed with computer simulations. In addition, we have derived an approximate macroscopic description of the automaton dynamics (see Supplementary Section 3). It is given by a system of coupled nonlinear driftdiffusion differential equations. This system offers the potential to relate our discrete model to a continuum model and exploit the analytical tools available for the differential equation model. However, this analysis is beyond the scope of the current study.
Previous theoretical works have studied amoeboid and mesenchymal cell migration under different environmental conditions^{6,18}. Here, for the first time, the impact of plasticity of amoeboid and mesenchymal migration modes on tumor invasion is investigated. Contrary to our initial expectation, amoeboidmesenchymal migration plasticity is not advantageous per se. Interestingly, it is only with the combined influence of ECM heterogeneity and chemotactic responsiveness that the switching behavior provides an advantage in terms of migration distance. So far, experimental studies of amoeboid and mesenchymal migration either focus on ECM structure or on varying chemotactic responsiveness, but not on the combined influence of both environmental factors. There are currently no experimental findings we can compare our results with, although it is wellknown that in real tumors the ECM is characterized by structural heterogeneity^{19,34} and the presence of chemotactic gradients at the same time^{1,35}. Our results emphasize an important aspect of future experimental studies on tumor invasion. In particular, our study suggests that in vitro systems aiming at unraveling the underlying molecular mechanisms of tumor invasion should take into account the complexity of the microenvironment by considering the combined effects of structural heterogeneities and chemical gradients on cell migration.
References
 1.
Roussos, E. T., Condeelis, J. S. & Patsialou, A. Chemotaxis in cancer. Nat Rev Cancer 11, 573–87 (2011).
 2.
Lu, P., Weaver, V. M. & Werb, Z. The extracellular matrix: a dynamic niche in cancer progression. J Cell Biol 196, 395–406 (2012).
 3.
Taddei, M. L., Giannoni, E., Comito, G. & Chiarugi, P. Microenvironment and tumor cell plasticity: an easy way out. Cancer Lett 341, 80–96 (2013).
 4.
Brábek, J., Mierke, C., Rösel, D., Veselý, P. & Fabry, B. The role of the tissue microenvironment in the regulation of cancer cell motility and invasion. Cell Commun Signal 8, 22 (2010).
 5.
Panková, K., Rösel, D., Novotný, M. & Brábek, J. The molecular mechanisms of transition between mesenchymal and amoeboid invasiveness in tumor cells. Cell Mol Life Sci 67, 63–71 (2010).
 6.
Hecht, I. et al. Tumor invasion optimization by mesenchymalamoeboid heterogeneity. Sci Rep 5, 10622 (2015).
 7.
Friedl, P. & Wolf, K. Plasticity of cell migration: a multiscale tuning model. J Cell Biol 188, 11–9 (2010).
 8.
Liu, J. Y. et al. Confinement and low adhesion induce fast amoeboid migration of slow mesenchymal cells. Cell 160, 659–72 (2015).
 9.
Friedl, P. & Alexander, S. Cancer invasion and the microenvironment: plasticity and reciprocity. Cell 147, 992–1009 (2011).
 10.
Te Boekhorst, V., Preziosi, L. & Friedl, P. Plasticity of cell migration in vivo and in silico. Annu Rev Cell Dev Biol 6, 491–526 (2016).
 11.
Huttenlocher, A. & Horwitz, A. R. Integrins in cell migration. Cold Spring Harb Perspect Biol 3, 1–16 (2011).
 12.
Wolf, K. et al. Physical limits of cell migration: control by ECM space and nuclear deformation and tuning by proteolysis and traction force. J Cell Biol 201, 1069–84 (2013).
 13.
Giverso, C., Scianna, M., Preziosi, L., Buono, N. L. & Funaro, A. Individual cellbased model for invitro mesothelial invasion of ovarian cancer. Math Model Nat Phenom 5, 203–23 (2010).
 14.
Hecht, I., Levine, H., Rappel, W. J. & BenJacob, E. “Selfassisted” amoeboid navigation in complex environments. PLoS One 6, e21955 (2011).
 15.
Sakamoto, Y., Prudhomme, S. & Zaman, M. H. Viscoelastic gelstrip model for the simulation of migrating cells. Ann Biomed Eng 39, 2735–49 (2011).
 16.
Scianna, M., Preziosi, L. & Wolf, K. A cellular potts model simulating cell migration on and in matrix environments. Math Biosci Eng 10, 235–61 (2013).
 17.
Tozluoglu, M. et al. Matrix geometry determines optimal cancer cell migration strategy and modulates response to interventions. Nat Cell Biol 15, 762 (2013).
 18.
Tozluoglu, M., Mao, Y., Bates, P. A. & Sahai, E. Costbenefit analysis of the mechanisms that enable migrating cells to sustain motility upon changes in matrix environments. J R Soc Interface 12, 20141355 (2015).
 19.
Polacheck, W. J., Zervantonakis, I. K. & Kamm, R. D. Tumor cell migration in complex microenvironments. Cell Mol Life Sci 70, 1335–56 (2013).
 20.
Böttger, K., Hatzikirou, H., Chauvière, A. & Deutsch, A. Investigation of the migration/proliferation dichotomy and its impact on avascular glioma invasion. Math Model Nat Phenom 7, 105–135 (2012).
 21.
Böttger, K. et al. An emerging Allee effect is critical for tumor initiation and persistence. PLoS Comput Biol 11, e1004366 (2015).
 22.
Deutsch, A. & Dormann, S. Cellular automaton modeling of biological pattern formation (Birkhäuser, 2005). (2nd ed. 2017).
 23.
Hatzikirou, H. & Deutsch, A. Cellular automata as microscopic models of cell migration in heterogeneous environments. Curr Top Dev Biol 81, 401–34 (2008).
 24.
Hatzikirou, H., Böttger, K. & Deutsch, A. Modelbased comparison of cell densitydependent cell migration strategies. Math Model Nat Phenom 10, 94–107 (2015).
 25.
Peruani, F., Klauss, T., Deutsch, A. & VossBöhme, A. Traffic jams, gliders, and bands in the quest for collective motion of selfpropelled particles. Phys Rev Lett 106, 128101 (2011).
 26.
VossBöhme, A. & Deutsch, A. The cellular basis of cell sorting kinetics. J Theor Biol 263, 419–436 (2010).
 27.
Simpson, M. J., Landman, K. & Hughes, B. D. Multispecies simple exclusion processes. Physica A 388, 399–406 (2009).
 28.
Ulrich, T. A., de Juan Pardo, E. M. & Kumar, S. The mechanical rigidity of the extracellular matrix regulates the structure, motility, and proliferation of glioma cells. Cancer Res 69, 4167–74 (2009).
 29.
Das, A., Kapoor, A., Mehta, G. D., Ghosh, S. K. & Sen, S. Extracellular matrix density regulates extracellular proteolysis via modulation of cellular contractility. J Carcinogene Mutagene 2013 (2013).
 30.
Wolf, K. et al. Compensation mechanism in tumor cell migration: mesenchymalamoeboid transition after blocking of pericellular proteolysis. J Cell Biol 160, 267–77 (2003).
 31.
Anderson, A. R., Weaver, A. M., Cummings, P. T. & Quaranta, V. Tumor morphology and phenotypic evolution driven by selective pressure from the microenvironment. Cell 127, 905–15 (2006).
 32.
Basanta, D. et al. Investigating prostate cancer tumourstroma interactions: clinical and biological insights from an evolutionary game. Br J Cancer 106, 174–81 (2012).
 33.
Reher, D., Klink, B., Deutsch, A. & VossBöhme, A. Cell adhesion heterogeneity reinforces tumour cell dissemination: novel insights from a mathematical model. (under review) (2017).
 34.
Erler, J. T. & Weaver, V. M. Threedimensional context regulation of metastasis. Clinical & Experimental Metastasis 26, 35–49 (2009).
 35.
MuinonenMartin, A. J. et al. Melanoma cells break down lpa to establish local gradients that drive chemotactic dispersal. PLoS Biol 12, e1001966 (2014).
Acknowledgements
The authors would like to thank Katarina Wolf (Nijmegen, Netherlands) and Miguel A. Herrero (Madrid, Spain) for helpful discussions. AD acknowledges support by Deutsche Krebshilfe. AD and EACA were financially supported by SFBTRR79 projects M8 and B5 (Deutsche Forschungsgemeinschaft). AVB acknowledges support by Sächsisches Staatsministerium für Wissenschaft und Kultur (SMWK) in the framework of INTERDIS2. The authors thank the Center for Information Services and High Performance Computing at TU Dresden for providing an excellent infrastructure.We acknowledge support by the German Research Foundation and the Open Access Publication Funds of the TU Dresden.
Author information
Affiliations
Contributions
Wrote the paper and discussed results: K.T., E.A.C.A., A.D., A.V.B. Supervised the study and gave substantial input to the manuscript: A.D., A.V.B. Conceived and designed the model and performed simulations: KT. Analyzed the model: K.T., A.V.B. All authors reviewed the manuscript.
Corresponding author
Ethics declarations
Competing Interests
The authors declare that they have no competing interests.
Additional information
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Electronic supplementary material
41598_2017_9300_MOESM3_ESM.mov
Nonswitching population (pure Acell) under heterogeneous, highly structured ECM conditions with small migration rate ratio.
41598_2017_9300_MOESM4_ESM.mov
Nonswitching population (pure Mcell) under heterogeneous, highly structured ECM conditions with small migration rate ratio.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Talkenberger, K., CavalcantiAdam, E.A., VossBöhme, A. et al. Amoeboidmesenchymal migration plasticity promotes invasion only in complex heterogeneous microenvironments. Sci Rep 7, 9237 (2017). https://doi.org/10.1038/s41598017093003
Received:
Accepted:
Published:
Further reading

Electrospun nanofiber scaffolds for the propagation and analysis of breast cancer stem cells in vitro
Biomedical Materials (2021)

Contractility, focal adhesion orientation, and stress fiber orientation drive cancer cell polarity and migration along wavy ECM substrates
Proceedings of the National Academy of Sciences (2021)

Modeling the Mechanobiology of Cancer Cell Migration Using 3D Biomimetic Hydrogels
Gels (2021)

Engineering T cells to enhance 3D migration through structurally and mechanically complex tumor microenvironments
Nature Communications (2021)

Inhomogeneities in 3D Collagen Matrices Impact Matrix Mechanics and Cancer Cell Migration
Frontiers in Cell and Developmental Biology (2020)
Comments
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.