Active fluid networks excite visco-elastic modes for efficient transport

Abstract

Active fluid transport is a hallmark of many biological transport networks. While animal circulatory systems generally rely on a single heart to drive flows, other organisms employ decentralized local pumps to distribute fluids and nutrients. Here, we study the decentralized pumping mechanism in the slime mold Physarum polycephalum which is locally triggered by active release, uptake, and transport of a chemical solute within the organism’s vascular network to drive global oscillations. Based on a conceptual network model combining active elasticity and fluid transport we identify a set of contractile modes specific to each network and show that modes corresponding to large-scale oscillations are preferentially and robustly excited both in model simulations and in experimental data obtained from living Physarum plasmodia. These dominant modes are computed explicitly and shown to drive large-scale flows within the organism. Furthermore, Physarum must transport nutrients over long distances. As each mode corresponds to pure shuttle flow, long-range, directed transport must rely on a non-linear coupling beyond harmonic dynamics. Using simulations, we demonstrate that the network’s transport capability is optimized when two dominant modes are excited at a phase shift of $\pi/2$, resulting in contractile excitations similar to those observed in real Physarum. Our results provide a conceptual framework for understanding active decentralized transport in Physarum and other contractile biological networks, such as brain vasculature, as well as decentralized transportation networks more generally.

I Introduction

Active transport networks are ubiquitous in nature and engineering. Gas and sewage networks are engineered for efficient delivery of fluids in large-scale production [1] and infrastructure [2] contexts, and may be centrally driven. The animal vascular system is actively driven by one central heart in the case of mammals [3] or several in the case of insects [4] and is used for long-range transport of oxygen and nutrients. Plants employ decentral actively maintained concentration gradients for sugar transport in the phloem network [5, 6, 7], and contractile muscles surrounding the brain vasculature are recruited by surrounding neurons for localized oxygen delivery [8]. Thus, while engineered networks are often centrally driven, decentralized local driving appears to be an effective strategy in biological transport. Furthermore, both biological and human-made networks can often be described in the language of synchronizing coupled oscillators [9, 10], and many examples are optimized for a particular functionality [11, 12, 13] and can adapt to their environment [14, 15, 16] by harnessing the combined effects of fluids and elasticity [17].

Here, we focus on the well-studied biological model organism Physarum polycephalum [18, 19, 20], a slime mold which in its plasmodium phase consists of a complex network of contractile tubes filled with cytoplasm [21]. The organism is able to actively contract and expand its tubular network using a chemical trigger, producing large scale oscillatory waves and excitations that drive flows and transport nutrients across its body [22, 23, 24, 25, 26]. It has been shown that Physarum is able to self-organize and react to external stimuli [27, 28], solve complex computational problems [29, 30, 31, 32], optimize its transportation network [30, 33, 34, 35, 36] and store memories based on available food [37]. An important emerging class of models to describe biological dynamics, empirical mode decompositions [38, 39, 40, 41], have been applied to Physarum data, revealing a continuous spectrum of empirical modes [42]. While significant effort has been made to study the local properties of Physarum’s chemically driven visco-elastic active oscillations [22, 43] as well as the phenomenology of the network oscillations [22, 33], a conceptual understanding of the entire network dynamics and its relationship to functionality is still missing.

Figure 1: Our visco-elastic flow network model. a, A cartoon of the model which depicts pressure driven flow (white arrows) travelling through discretized cylindrical vessels. A contraction triggering chemical travels through the vessels, and is released when vessels are stretched past their equilibrium. b, The model implemented on a $n=12$ path graph showcases a peristaltic wave of contractions. Still frames taken from Video 1, where the red arrows depict flow velocity. c, The model implemented on a real Physarum polycephalum network, corresponding to the microscopy image. The experimentally observed large-scale organization of contraction amplitudes are reproduced by the active network model, which is based on contraction-triggering solutes. Microscopy and model network still frames taken from Videos 2 and 3 respectively.

Here, we propose a network model of visco-elastic active oscillations and fluid flow in Physarum based on physical principles and show using simulations that it can reproduce large-scale coordinated contraction waves similar to those observed in real Physarum networks. Based on analysis of our model, we then show that it is possible to compute a set of linear contractile modes based on network topology alone. We demonstrate that only a few of the largest-scale modes are sufficient to explain the network dynamics in both our simulations and in experimental data obtained from analysis of brightfield microscopy time series of Physarum contractions. We further show that the model dynamics corresponds to an attempt to maximize order and synchrony among the individual pumping vessel elements, subject to physical constraints. Finally, we employ the conceptual knowledge gained to study the Physarum network’s functionality: long-range transport of nutrients. Using a simple model based on advection of particles on top of the network fluid flows, we demonstrate in real Physarum network topologies that long-range transport is maximized when the two largest modes are excited at a phase shift of approximately $\pi/2$. This phase shift had been measured but not fully explained in previous experiments [35].

II Results and Discussion

Figure 2: Both the observed and modeled temporal dynamics of Physarum can be decomposed by the network’s visco-elastic decay modes. a, The observed Physarum dynamics is projected onto each mode, and the root-mean-squared signal of that projection is plotted over time. Each mode’s corresponding curve is colored based on its mode number. The mode with the maximum projection is plotted in red, and the time where this maximum is achieved is denoted by the diamond. b, A plot of each mode’s maximum projection, where the lowest modes are colored in yellow and the highest in purple. The mode with the largest projection is denoted by the red diamond. c, The mode with the largest projection is displayed on the right, and the observed Physarum behavior when the maximum is achieved is on the left. d-f, The same analysis as a-c but for modeled Physarum dynamics.

In its plasmodium stage, Physarum consists of a densely connected, hierarchically organized network of cytoplasm-filled tubes which are able to rhythmically contract, leading to shuttle flow of the cytoplasm (Fig. 1 (a, b)). The mechanism by which the contractions synchronized across the entire organism is through fluid advection of a triggering chemical, generally believed to be Calcium ions [25]. The walls of Physarum’s tubes are embedded with an active actomyosin cortex that reacts to the presence of Calcium ions by contracting, and with Calcium channels that release Ca${}^{2+}$ ions upon mechanical extension (Ref. [44], Fig. 1). Together, these two processes lead to self-sustained sinusoidal oscillations of the tube walls and with that to fluid flow inside the network. The flow is able to transport the triggering chemical, leading to long-range organization across the organism (Fig. 1 (c)). The active elasto-chemical oscillations have been studied in models [44, 25]. We now consider a model that extends to an entire complex network topology, based on the one-dimensional model from Ref. [43]. It turns out to be useful to proceed in two steps. We first construct a passive model coupling fluid transport and visco-elastic forces within the vessel walls. This model already yields a useful modal description of the network dynamics. In a second step, we include solute-triggered contractions and solute dynamics, capturing the active, self-sustained oscillations observed in Physarum plasmodia.

We model the biological vascular network as a graph, with each edge $(ij)$ directed from node $j$ to node $i$ representing one entire tube, or a piece of a tube of volume $V_{ij}$. Because the tube volume is changing at a rate of $\dot{V}_{ij}$, the volume flow rates into and out of each vessel are related by

$$Q^{\mathrm{(out)}}_{ij}=Q^{\mathrm{(in)}}_{ij}-\dot{V}_{ij}.$$ (1)

Here, the volume flow rates $Q_{ij}$ are positive if flow is directed from node $j$ to node $i$, and negative otherwise. At each junction of the network, the total flow must be conserved,

$$\sum_{j}Q^{\mathrm{(in)}}_{ij}=0,$$ (2)

and over the time scales we consider, the total volume of the network is approximately constant,

$$\sum_{(ij)}\dot{V}_{ij}=0.$$ (3)

Following Ref. [43], we choose a visco-elastic model for the mechanical properties of the network. Each vessel acts as an overdamped spring following

$$\tau_{ij}\dot{V}_{ij}=E\varepsilon_{ij}+\sigma_{ij},$$ (4)

where $\tau_{ij}$ is the damping constant, $E$ is the Young’s modulus, the strain $\varepsilon_{ij}=(R_{ij}-R^{(0)}_{ij})/R^{(0)}_{ij}$ with the vessel radius $R_{ij}$ and equilibrium radius $R^{(0)}_{ij}$, and $\sigma_{ij}$ is an additional external stress. The constraints Eq.  (1), (2), and (3) can be combined with the visco-elastic force law Eq. (4) using a generalization of Thompson’s principle for electrical networks (Supporting Information, Ref. [45]). In vector notation, the resulting equations of motion read

$$M\,\dot{\mathbf{V}}=-E\bm{\varepsilon}+\bm{\sigma}-\mu\mathbf{1},$$ (5)

where $\mathbf{V}$ is the vector of vessel volumes, the damping matrix $M=\tau-\frac{1}{4}B^{\top}L^{\dagger}B$ with $B$ the graph’s unweighted incidence matrix and $L$ the graph’s weighted Laplacian matrix (Supporting Information), and $\mu$ is a Lagrange multiplier ensuring total volume conservation and playing the role of an external hydrostatic pressure. The Laplacian encodes the radii of all vessels, assumed to be constant in a linear approximation, by way of modeling the vessels as cylindrical tubes subject to Poiseuille flow (Supporting Information). The dagger represents the Moore-Penrose pseudoinverse, and we introduced the vector of all ones, $\mathbf{1}=(1,\dots,1)^{\top}$. The damping matrix $M$ encodes the visco-elasticity of the vessel walls in addition to viscous damping in the internal fluid. Here, the the pressures were eliminated. This equation of motion for the passive network Eq. (5) admits solutions that can be decomposed into modes. The volume rate of change of cylindrical vessels with length $L$ can be related to their radii $R$ by $\dot{V}=2\pi LR\dot{R}$. Then assuming small excitations around the equilibrium radii of the form $\mathbf{R}(t)=\bar{\mathbf{{R}}}+\mathbf{u}\,e^{-\lambda t}$ leads the to generalized eigenproblem

$$A\mathbf{u}=-\lambda M\mathbf{u},$$ (6)

where the matrix $A$ encodes vessel geometry (Supporting Information).

Equation (6) provides a powerful framework for thinking about the dynamics of Physarum and similar systems. Due to its mathematical properties (Supporting Information), Eq. (6) is subject to a generalization of Courant’s nodal domain theorem [46, 47] which states that, when arranged in ascending order of the eigenvalue, the eigenvectors corresponding to the $n$’th eigenvalue partition the network into no more than $n$ connected domains according to the sign of the entries of $\mathbf{u}$. Thus, we expect that large scale excitations of the biological network can be expressed economically using a few of the lowest modes encoded by Eq. (6).

We now proceed to apply the modal decomposition to measured Physarum dynamics. Plasmodia of Physarum were cultured on agar plates based on the protocol in Ref. [43], and microscopically imaged over the span of approximately two hours to capture many periods of shuttle flow oscillation while avoiding behaviors such as growth that occur over longer time spans (Supporting Information). Images were then discretized using custom software, and a graph representation of the vascular network including the radii of all vessels over time (Fig. 1) was constructed from which the modes Eq. (6) were calculated using the equilibrium radii (Supporting Information). The time series of the $i$’th mode amplitude was computed by projecting the vector of all vessel radii $\mathbf{R}(t)$ onto the mode, $a_{i}(t)=\mathbf{R}(t)\cdot\mathbf{u}_{i}$. With this, the dynamics of the network is expanded as $\mathbf{R}(t)=\sum_{i}a_{i}(t)\mathbf{u}_{i}$. It can be seen that over the duration of our observations of several specimens, a few modes dominate the dynamics, with the bulk making only small contributions (Fig. 2 a-b, d-e). Visualizing these dominant modes, we see that they are large-scale excitations that span the entire network and that they correspond to the largest eigenvalues of Eq. (6) (Fig. 2 c,f).

While much insight can be gleaned from the passive features of the organism, Physarum’s network actively self-organizes its dynamics. The basic model from Eq. (5) can be augmented with active contraction triggering by chemical solutes as well as solute transport. We employ a discrete version of the advection-diffusion equation to model transport, adsorption, and decay of a solute with total amount $\mathcal{C}_{ij}$ in edge $(ij)$ of the network (Supporting Information). We again follow Ref. [43] for the active, solute-dependent stress $\bm{\sigma}(\mathcal{C})$ (Supporting Information). Large solute concentrations within a vessel trigger active contractions, which in turn wash solute away. At the same time, relaxation of the vessel triggers enhanced solute production. Locally, this leads to a self-sustained oscillation of the vessel but on large scales, transport of the solute by advection and diffusion organizes and synchronizes the network [22]. Similar non-linear wave propagation phenomena have been described recently in generic models of active transport networks [48, 49, 50], and are known from the theory of excitable media [51]. Most physical parameters of the model have been determined before, but for the two parameters describing the nonlinear excitations, no experiments exist. We therefore performed a detailed numerical parameter study to identify the regions in parameter space where the non-linear equations yield stable oscillatory dynamics (Supporting Information). There exist parameters where the stable dynamics reproduces amplitudes and time scales of our Physarum specimens (Supporting Information, Fig. 1), suggesting that the model can match the real organism in relevant metrics. In this biological regime, the model predicts both contractile waves and chemical waves of solute concentration that propagate across the organism (Fig. 1).

Figure 3: Nutrient transport through a Physarum network is optimized by exciting the two dominant visco-elastic decay modes at a $\phi=\pi/2$ phase shift. a, An illustration of our nutrient transport model. We simulate advection dominated particles travelling at their vessel’s average velocity $v_{\text{avg}}$. Junction transitions are determined stochastically, such that flow direction is preserved and transition rates are proportional to the volumetric inflow $Q$ of the target vessel: $p(i\rightarrow j)=|Q_{j}|/Z$, where $Z=\sum_{\text{valid }j}|Q_{j}|$ is the normalization. b, An illustration of nutrient particles advected on a network, taken from Video 4. After a long time, each particle (identified by color) is transported away from its initial location. c, We simulate transport for every pair of twenty modes with the highest excitation, and plot the optimal phase between the two modes. When overlaying two different modes, we find that $\phi=\pi/2$ optimizes transport. When overlaying the same mode, transport is optimized by no phase shift. d, We plot the average displacement at the end of each simulation for every mode pair phase shifted at the optimal shift. We see that overlaying large scale modes yields the best transport. e, Overlaying the two largest scale modes at a $\phi=\pi/2$ phase shift results in a circularly propagating contraction wave.

Armed with the non-linear model from above and an understanding of the importance of large scale visco-elastic modes for the dynamics of Physarum, we now proceed to study the functionality of the network. Presumably, an important function of the transport network is to take up nutrients from food sources and distribute them throughout the organism. To reveal how dynamical modes and transport may play together to enhance nutrient dispersion, we simulate a simple model of transport on one of our realistic networks. We assume that transport is dominated by the advection of particles that move at the velocity of the surrounding fluid. Given that the distribution of flow velocities in our Poiseuille approximation is parabolic as a function of distance from the vessel walls (Supporting Information), for simplicity we consider only particles that move with the mean velocity. While the position $x(t)$ of a particle along each vessel can be found by simply integrating $\dot{x}=v(t)$, where $v(t)$ is the local flow velocity, at each junction a decision must be made as to which vessel the particle will continue in. Since a full fluid-dynamical simulation of the junction flows is beyond the scope of our analysis, we resort to the simple rule that the probability of choosing a particular next vessel is proportional to the flow rate in it (Fig. 3 a). We then place three particles on each edge of the network and simulate a combination of two elastic modes $i$ and $j$ such that the flow rates follow $\mathbf{Q}(t)=\mathbf{Q}_{i}(t)+\mathbf{Q}_{j}(t-\phi\,T/(2\pi))$, where $T$ is the period and $\phi$ is a phase shift. We choose to consider two modes because there have been indications that such combinations may be relevant in Physarum organisms [35].

For each combination of modes and phase shifts we then computed the mean square displacement $\sqrt{\langle x^{2}\rangle}$ averaged over both a simulation time of one hour and over all particles (Supporting Information). The mean square displacement quantifies the ability of the network to transport nutrients quickly (Fig. 3 b). For each combination of modes we typically find that the displacement is maximized for a range of phase differences centered around $\pm\pi/2$, independent of the particular pair of modes we considered (Fig. 3 b). This phase lag between the dominant harmonics has been observed in real Physarum [35]. The optimal phase lag generally corresponds to contractile waves that move circularly around the network (Fig. 3 e). Finally, we find that the optimal combination of modes for transport consists of the two largest modes (Fig. 3 d). Thus, the observed organism-spanning contractions in real networks are consistent with an adaptive optimization for efficient transport.

III Conclusions

Here we demonstrated that the dynamics of living organisms that transport fluid and nutrients by rhythmic contraction of tubes can be understood in terms of dynamical modes that are based on a physical model of the organism’s transport network. Based on experimental observations of Physarum networks over several hours, we showed that indeed, the slime mold network excites few large scale visco-elastic modes during its activity. The visco-elastic modes can be computed from network topology, geometry, and elasticity alone, without recourse to a detailed model of the active motion. Thus, they may be applicable to more organisms than Physarum alone, as long as the passive physics are known. We then proceeded to further elucidate Physarum dynamics by constructing a network model of its active dynamics. We were able to reproduce not only crucial qualitative properties of the real dynamics like large-scale excitations, but also matched amplitudes and periods of radial oscillations, as well as an order parameter of the network. With this complete model, we proceeded to study the functionality of fluid-carrying networks. By considering the motion of particles that are advected by the flows inside the network, such as particles of food taken up by the slime mold, we demonstrated that the most effective transport is achieved by combining the two largest-scale dynamical modes at at relative phase of $\pi/2$. This had been seen in real Physarum before, suggesting that the slime mold may indeed optimize its rhythmic contractions for nutrient transport.

While we showed that the dynamics of Physarum can be understood in terms of an interplay between a small number of modes determined by passive physics and network topology, active contractions, and optimization for transport, we anticipate that our results will have wider-ranging consequences. Specifically, our passive model applies to all elastic, fluid filled networks such as the vasculature of mammals or insects wings [52]. The visco-elastic modes we computed may play a role in their passive dynamics as well. Furthermore, there exist active fluid-pumping networks such as the brain vasculature, where nerve cells actively control vascular contractions to direct oxygen transport [53]. Here, our model may have even more direct applicability to the study of brain hemodynamics, bridging the gap between disparate biological systems.