Thoughts from the forest floor: a review of cognition in the slime mould Physarum polycephalum

Sensing, communication, navigation, decision-making, memory and learning are key components in a standard cognitive tool-kit that enhance an animal’s ability to successfully survive and reproduce. However, these tools are not only useful for, or accessible to, animals—they evolved long ago in simpler organisms using mechanisms which may be either unique or widely conserved across diverse taxa. In this article, I review the recent research that demonstrates these key cognitive abilities in the plasmodial slime mould Physarum polycephalum, which has emerged as a model for non-animal cognition. I discuss the benefits and limitations of comparisons drawn between neural and non-neural systems, and the implications of common mechanisms across wide taxonomic divisions. I conclude by discussing future avenues of research that will draw the most benefit from a closer integration of Physarum and animal cognition research.

Keywords: Basal cognition, Minimal cognition, Extended cognition, Embodied cognition, Non-neural, Collective behaviour

Animals can acquire information from their environment through their senses, remember it, use it to solve problems and learn, anticipate and communicate, all of which serves to increase their fitness in a world that is complex, dynamic and dangerous. Animals (at least those traditionally associated with being ‘cognitive’) perform these feats using the highly sophisticated information-processing hardware that we call a brain. These facts are well established, and few schoolchildren (and still fewer cognitive scientists) would challenge them. Similarly, few would argue that organisms without brains—forming the vast majority of current and past life on this planet—live in some parallel world that is homogeneous, static and safe. Since all organisms face similar environmental challenges, and thus stand equally to benefit from the adaptive advantages of detecting, storing, and learning from environmental information, it is strange that the word most often used to describe this function—cognition—has traditionally been strictly reserved for the historically recent, neurally equipped minority.

The belief that cognition requires a nervous system remains pervasive, but faces continued and growing challenge (Allen 2017; Lyon et al. 2021; Solé et al. 2019). This groundswell has led to a formal framework called basal cognition for reframing the definition of cognition, based on the premise that complex cognitive function had to evolve from earlier, simpler systems, rather than beginning with assumptions of (typically human) uniqueness and working ‘down’ the evolutionary tree (Lyon et al. 2021). This method, the authors argue, would simply bring the cognitive sciences into alignment with all other life sciences in taking a bottom-up approach that begins with looking at the smallest, simplest, earliest-evolved organisms to display phenomena of interest, identifying the principles of function, and from there scaling up to more complex organisms, comparing similarities and differences.

One of the difficulties of the bottom-up approach is that there is an awful lot of bottom to choose from, with small, simple organisms occupying every conceivable niche that could support life. While the majority may face similar fundamental environmental challenges, peculiarities of their evolutionary past, life history, and physiology make it likely that their evolved mechanisms for responding to their world are as diverse as the organisms themselves. Naturally, research has been limited to those particular ‘basal’ organisms that are easy to culture in the lab, and can be rigorously tested with behavioural experiments to which they unambiguously and reliably respond. This has included bacteria (Lyon 2015; Shapiro 2007), plants (reviewed in Segundo‐Ortin and Calvo 2022), protists (e.g. Paramecium (Armus et al. 2006; Gelber 1957; Gershman et al. 2021), Amoeba (De la Fuente et al. 2019) and Stentor (Bennett and Francis 1972; Rajan et al. 2022)), and simple neural animals such as planaria (Prados et al. 2013, 2020) and Cnidaria (Jennings 1905; Logan 1975). One non-neural model system in particular has recently shown consistent and unambiguous success in demonstrating a range of cognitive functions: the slime mould Physarum polycephalum.

The purpose of this review is to introduce Physarum polycephalum to the general readership of animal cognition research, highlighting recent and historical work that demonstrates basal cognitive capacities in this fascinating organism. One key goal is to elucidate the hypothetical mechanisms of various cognitive functions in this single-celled protist, to guide future comparative approaches that span a greater breadth of the phylogenetic tree. Hopefully, this will lead to a greater understanding of the evolution of cognition; it will at the very least reveal some of the great diversity of biological approaches to making sense of the world.

Physarum polycephalum (hereafter Physarum) is a unicellular eukaryote currently placed in the kingdom Protista. The vegetative stage of Physarum’s life cycle (Fig. 1a, b) is called a plasmodium, which is a single cell containing millions of nuclei flowing freely inside a motile network of protoplasm. The multinucleate nature of Physarum means that individual fragments removed from a plasmodium will self-organise into independent clones of the parent cell, a process that takes only minutes (Kobayashi et al. 2006). Under ideal conditions, a plasmodium can grow as large as 900 cm2 in size. It moves in an amoeboid fashion, by extending pseudopods, at speeds of up to 5 cm/h (Kessler 1982) to explore its environment and engulf its prey of bacteria, yeasts and micro-particles. The exploratory ‘search front’ of the organism is a dense fan shape (Rusch 1980). Behind the search front, however, cell morphology is organised into a network of intersecting tubules (Fig. 1a). Within the tubules, protoplasm circulates back and forth rhythmically, with a period of around 1–5 min depending on internal state (Wohlfarth-Bottermann 1979). This constant internal flow transports biomass, nutrients, chemical signals and information (Collins and Haskins 1972), and is coordinated by oscillatory and biochemical mechanisms, which have been the focus of much of the Physarum literature.

As a free-living organism, the plasmodium is equipped with a range of surface receptors and internal mechanisms for sensing and responding to the world around it. One of the key response mechanisms, underpinning the majority of cognitive, locomotive and homeostatic functions, is oscillation (Durham and Ridgway 1976). The plasmodium can be considered a connected mass of multitudes of tiny oscillating units, capable of expanding and contracting in response to local sensory stimuli via actin-myosin interactions—the same contractile mechanism in human muscle tissue. When cell surface receptors detect attractants such as food and moisture, oscillation frequency increases in the local area, which decreases cell surface tension, making the plasmodium more fluid (Ueda et al. 1980). This causes protoplasm to flow towards the stimulus area, and directs the movement of the entire cell. Repellents such as light and certain salts induce the opposite response, increasing the local stiffness of the cell and restricting further flow into the area.

Much of the cognitive behaviour observed in Physarum is fundamentally a consequence of communication between the myriad contractile units. While each unit senses and responds to the environment around it, physical coupling between adjacent oscillators entrains them to each other’s frequencies (Nakagaki et al. 1999). This means they can respond to and influence the behaviour of their neighbours, and transfer information about the quality of local environments to distant parts of the cell. For instance, in the construction of efficient tubule networks between multiple food sources (discussed in further detail below), attractants such as food and moisture are locally sensed, resulting in increased oscillation frequency in nearby contractile units. Physical coupling results in waves of contraction that propagate outwards from the stimulus area and communicate information between proximal and distal parts of the organism. Network tubules that lie perpendicular to the direction of contractile propagation receive a lower flux of protoplasm and thus begin to decay, while parallel tubes receive more protoplasm and are reinforced and thickened. The tubules able to accommodate the highest flux are those that link the network via the shortest path; network length is thereby optimised via a positive feedback loop (Nakagaki et al. 2000; Tero et al. 2006). This simple method of local communication illustrates a form of distributed collective behaviour that leads to the emergence of sophisticated properties at the organism level (Reid and Latty 2016).

More recent work has focussed on the evidence for a biochemical signal propagated through the organism by the peristaltic waves themselves (Alim et al. 2017). A signalling molecule (later called a ‘softening agent’ (Kramar and Alim 2021)) was found to travel along with the internal flow, increasing contraction amplitude as it travels, and thereby facilitating its own transport via positive feedback. The signalling molecule is unknown but is likely to be either ATP or calcium ions, which are known to be vital for actomyosin interactions. It is important to note that a self-reinforcing signalling molecule could only work for attractant responses in Physarum, which induce fluidity, and not for responses to repellents, which increase stiffness. There is no reason that biochemical and coupled-oscillator models should be mutually exclusive, however, and most likely the two mechanisms are both exploited in Physarum’s toolkit for sensing and responding to the world.

Due to Physarum’s early popularity as a model for cell motility, there is a host of previous studies into positive and negative taxes in plasmodia responding to gradients of carbohydrates, proteins, amino acids, free nucleotides, volatile organic chemicals, salts, pH, light, humidity and temperature (Fig. 1c; (de Lacy Costello and Adamatzky 2014; Chet et al. 1977; Kincaid and Mansour 1978; Knowles and Carlile 1978; Ueda et al. 1980). There is also evidence that Physarum can sense and respond to the direction of gravity (geotaxis (Wolke et al. 1987)), magnetic fields (magnetotaxis (Shirakawa et al. 2012)) and even use mechanosensation to detect heavy masses at long-range (Murugan et al. 2021). Thus, a broad base of literature supports Physarum’s ability to sense, and adaptively respond to, diverse information acquired from its environment.

Communication between organisms is generally defined as requiring both a signaller and a receiver, where a signal is “any act or structure that alters the behaviour of other organisms, which evolved because of that effect, and which is effective because the receiver’s response has also evolved” ((Maynard Smith and Harper 2003), p. 3). A far less stringent definition of communication might also include the use of ‘cues’, which are essentially any useful bits of information in the environment, even those originating from other organisms, but which need not have evolved specifically to alter the behaviour of others (Maynard Smith and Harper 2003). There are relatively few studies of communication between slime moulds, either directly or indirectly (let alone making the distinction between ‘clonemates’ originating from the same culture, or distinct individuals), and these few studies have all been framed in a cooperation/competition context.

In the absence of evidence for signalling, studies have focussed on Physarum’s response to cues left behind in the extracellular slime (ECS) trails of clonemates, conspecifics (originating from different cultures or strains), or other species of plasmodial slime moulds. The results indicate a rich repertoire of responses to multiple sources of information about the presence and state of other individuals. Physarum will generally avoid the ECS of clonemates, which thereby acts as a source of externalised spatial memory (see “Memory” section below) and prevents it from wasting energy foraging in overexploited areas (Reid et al. 2012). This avoidance response, however, is context-specific, and can be overridden in certain scenarios, for instance when food source cues are detected in high concentration (Reid et al. 2013). Similarly, while Physarum plasmodia will actively avoid environments previously explored by starved or stressed clonemates, they are attracted to environments previously explored by well-fed clonemates (Briard et al. 2020). Hence, information about the potential foraging value of an environment can pass between individuals. Physarum rigidum can recognise conspecific plasmodia by cues within the extracellular slime and chooses to fuse with individuals collected from nearby geographic regions (hence more likely to be allogeneic, and characterised as ‘self’), while avoiding individuals originating from more distant locations (Masui et al. 2018). With the option to fuse its body with that of other individuals, and the complex fitness consequences of doing so, Physarum has clearly evolved a range of mechanisms for sensing and responding to conspecifics.

Physarum can also recognise the presence of ECS originating from other species of slime mould and exhibits a similar contextual response to this information. When encountering the ECS of Didymium bahiense, Physarum will prefer to explore elsewhere, unless the only other available choice is over terrain marked with clonemate ECS (Reid et al. 2013). This can be interpreted as the slime mould making a choice between an area that is likely to have been stripped of preferred resources (by either itself or conspecifics), or another that may contain more resources, due to slight differences in heterospecifics’ food preferences. This ‘eavesdropping’ behaviour is similar to that observed in animals such as bumblebees, which not only recognise and reject exploited flowers marked by their own pheromone, but also those marked by other bumblebee species (Goulson et al. 1998).

Only a few studies have placed multiple plasmodia together in the same space at the same time. In one recent study two clonemates placed in the same petri dish containing a single food source had a shorter latency to begin foraging in the presence of clonemates (Stirrup and Lusseau 2019). Foraging began soonest when the focal plasmodium was starved and the clonemate was satiated. While these results were interpreted in a competitive context, they align with the results of a subsequent study (Briard et al. 2020), which found an attraction towards environments previously explored by well-fed clonemates. These results suggest that rather than exhibiting purely competitive responses to conspecifics, Physarum has the potential to exhibit facilitative responses that hint at an as-yet hidden world of social interactions with clonemates.

While taxis responses (such as chemotaxis) and navigation both involve an organism moving from one place to another, the taxis examples above require only the ability to orient in space. For a macroscopic organism such as Physarum, the ability to sense an environmental gradient and move along it is a relatively simple feat. Many organisms are faced with the cognitive challenge of moving along a route between multiple points, and Physarum can achieve this in two ways. The first is conventional navigation, where an entire plasmodium must migrate through space to reach a distant point. I would argue, however, that the slime mould’s method of exploratory network construction, followed by optimisation of the path between points of interest, is analogous to path-planning strategies utilised by autonomous systems, and hence can be classified as navigation.

Mazes have a long history of use in animal cognition experiments, and these are framed almost exclusively in a learning context (reviewed in Kabadayi et al. 2018). Physarum too has been tested in classic labyrinth mazes requiring the whole organism to move or extend through the maze towards a goal. However, these have never been framed in terms of learning (individuals are never challenged with solving the same maze more than once) and are always investigations of Physarum’s navigational ability. For instance, Adamatzky (2012) placed a food source at the centre of a circular labyrinth maze with an agar floor, challenging the plasmodium to navigate towards the goal. In this case, the walls of the maze extended below the level of the agar, creating a distinct channel of agar along which chemoattractants can diffuse from the centre to the plasmodium inoculation site. It is therefore no surprise that the slime mould simply follows this gradient, and in doing so follows the shortest path through the maze.

Other maze-navigation studies have relied on a subtly different technique whereby the agar substrate is undisturbed by the maze barriers, which are constructed of thin acetate sheets laid on top of the agar surface. As Physarum avoids the hydrophobic surface of the acetate, these barriers prevent Physarum movement, while allowing food cues to diffuse in a continuous radial gradient from the food goal. Thus, the organism must make local navigational choices based on a more-or-less global directional preference, in some cases having to override local chemotactic preference to reach the goal. These problems can be more difficult for Physarum to solve, yet they remain tests of navigation by chemoattraction (Smith-Ferguson et al. 2017, see “Memory” section below discussing the results of this study).

The more commonly researched example of navigation is experimentally framed around Physarum’s excellent ability to construct efficient networks between multiple points of interest. The plasmodium spreads itself out in search of food (exploration), often finding multiple food sources simultaneously some distance apart. This stage is analogous to the slime mould building a map of its surroundings, with the plasmodial network itself forming the many potential routes between food sources A and B. Having found multiple food sources, the plasmodium now seeks to engulf them with biomass, while also staying connected as a single entity (exploitation). This fundamental trade-off has resulted in strong selection for shortest-path-finding strategies in Physarum, which has been a major focus of slime mould behavioural research within the last two decades.

The landmark study which single-handedly spurred this flurry of research activity is a brief communication in Nature by Nakagaki and colleagues (2000), in which Physarum plasmodia were spread through a labyrinth maze connecting two food sources, forming a single cell in the shape of the maze (Fig. 1d). Over time, the plasmodia retracted biomass from the dead ends and longer paths through the maze, until eventually a single tubule remained, tracing out the single solution. As stated by the authors, “This remarkable process of cellular computation implies that cellular materials can show a primitive intelligence.” (Nakagaki et al. 2000). These humble words sparked a revolution, and a host of studies followed, exploring the capabilities of Physarum in network optimisation (Nakagaki et al. 2001, 2004; Reid and Beekman 2013), and building mathematical models based on empirical insights (Tero et al. 2006, 2007).

These studies have been of particular interest to network engineers seeking new methods for designing optimisation algorithms. Human-designed networks, such as those for telecommunication or transport, tend to place a high priority on shortest paths, as these provide the quickest travel time through the network, and the lowest construction cost. These networks are, however, the most at risk of catastrophic failure after even the slightest disruption, so natural selection has favoured biological networks that find a trade-off between path efficiency and the robustness of additional redundant links (Middleton and Latty 2016). A landmark study produced by Nakagaki’s lab (Tero et al. 2010) sought to extract network design rules from Physarum that specifically encoded an optimal trade-off between shortest-path efficiency and robustness. The researchers allowed Physarum to explore an agar plate in the shape of the Tokyo district, with food sources placed in the locations of railway stations (Fig. 1e). As the slime mould pruned its network to connect these ‘stations’ (over the course of approximately 26 h), the resulting networks had comparable efficiency, fault tolerance and cost to the existing Tokyo Railway infrastructure. Moreover, the authors were able to use the behavioural observations of Physarum network construction to build mathematical models for network growth, with tuneable parameters for adjusting features such as fault tolerance and transport efficiency. These could be useful for engineers designing future transport networks, or for guiding the development of self-organised networks such as remote sensor arrays, wireless mesh networks, or the Internet-of-Things (Tero et al. 2010; see also Gao et al. 2019 for an extensive review of Physarum-inspired models and computations, and their impacts).

While these feats of optimisation are impressive, they present to the organism an ideal situation of a static problem in which the environment never changes. Biological systems have been selected to survive in dynamic environments, and so must have evolved mechanisms for solving this difficult class of problems. In my own previous work (Reid and Beekman 2013), we used a well-known shortest-path problem—The Towers of Hanoi maze—to test Physarum’s dynamic problem solving in a maze-navigation scenario (Fig. 1f). When allowed to build a network through the Towers of Hanoi maze, Physarum occasionally constructed minimal length paths, but often built longer networks, in contrast to the results of Nakagaki’s first labyrinth maze. This may be because the Towers of Hanoi maze was much more difficult, having 32,678 unique paths through the maze, compared to the Nakagaki maze’s 4 possible solutions. However, when the Towers of Hanoi maze shape was changed following Physarum solving it, the plasmodium often retracted fully from the maze and then flowed through it anew, building new connections and pruning its network. Upon solving the new maze, Physarum was much more efficient, always building networks with minimal-length paths. In response to these dynamic changes, Physarum was able to utilise some sections of its original network, and construct large sections of the network from scratch, perhaps being guided by its aversion to previously laid ECS to find the optimal solution space. We compared Physarum’s solution strategy to another distributed biological system that has been tested with the same maze—ant colonies (Reid et al.

2011)—finding that the stronger reliance on positive feedback ensures that ants are more likely to converge quickly on the optimal solution to a maze, but at the cost of being less adaptable to dynamic changes than Physarum, which likely relies less on positive feedback (Reid and Beekman 2013).

Decision-making is defined here following Reid et al. (2015) as “the action by an entity of selecting an option from a set of alternatives, based on characteristics of the alternatives that the entity can perceive.” The utility of this definition is that it relies only on the observable actions of the organism in question, and makes no assumption of underlying mechanisms. As Physarum can clearly perceive a vast array of information about its environment, it is no surprise that plasmodia consistently choose the better of two presented options when they differ in only a single attribute, such as caloric concentration (Latty and Beekman 2010), temperature (Durham and Ridgway 1976) or light levels (Latty and Beekman 2011b). Indeed, most of these decisions can be made using simple taxis responses. An exception is a study of length discrimination (Mori and Koaze 2013). When presented with the option of connecting two food sources via either a short or long route in a circular arena, Physarum predictably chose the shorter route. This choice was consistent despite changing the diameter of the arena, indicating that Physarum bases its decision on the ratio of the two lengths, rather than any absolute difference. This pattern of cognition of difference in stimulus magnitude (constrained by Weber’s law (Fechner 1948)) is consistently observed in decision-making systems from humans (Deco et al. 2007) to other mammals (Yoshioka 1929), birds (Dixit et al. 2022) and insects (Perna et al. 2012).

Similarly, when examining the decision to remain within a food patch or explore elsewhere, both Physarum and another plasmodial slime mould (Didymium bahiense) were found to use incremental patch-departure heuristics, just as insects and humans do (Latty and Beekman 2015). Plasmodia were inoculated inside a grid of circular food discs of either high or low quality, and the time taken to leave the patch was recorded as a function of the number and quality of food sites sampled. Importantly, chemosensory cues of patch quality were precluded by placing each of the food sources on non-permeable plastic circles. Furthermore, the experimental plates were designated as either ‘safe’ (darkened) or ‘risky’ (well lit) environments. Physarum tended to stay longer within a patch if they had recently experienced high-quality food, and within darkened, ‘safe’ patches. D. bahiense tended to remain within a patch if it had recently encountered food of any quality and did not alter its strategy in dark or lit environments. Studies such as this highlight the utility of cognitive paradigms applied across (and within) broad taxa, as well as outlining a rich future avenue of research into why and how these diverse strategies exist, even between two species of plasmodial slime mould.

In the interests of understanding cognition, it is necessary to explore more difficult decision-making scenarios, such as when multiple attributes per option can be evaluated independently, and when several of these attributes may conflict with each other. These so-called ‘multi-attribute compensatory problems’ are the gold standard, because they require the organism to compare options based on their relative differences, integrating information along multiple axes of ‘quality’, rather than simply whether one attribute exceeds a desired threshold (see Reid et al. (2015) for a detailed discussion of decision-making in Physarum and other non-neural organisms).

Physarum is capable of making trade-offs between exploitation of food and exposure to danger. For instance, when choosing between high-quality food (positive stimulus) that is illuminated with strong light (negative stimulus), and an alternative option of low-quality food in the dark, Physarum will choose the safer, low-reward option. However, if the risky option is at least five-fold higher in food concentration than the safe option, Physarum will gamble on the illuminated food source (Latty and Beekman 2010). When forced to build a network connecting two food sources that passes through an intensely lit region, Physarum will make the optimum trade-off of path efficiency and exposure to light (Nakagaki et al. 2007). The single tubule connecting the food sources travels along a deflected path that occupies more space within the dark region when the lit region has a higher photo-intensity—a geometric feature similar to the path of light travelling through two materials with different refractive indices (Nakagaki et al. 2007). Importantly, the tubule path did not differ between totally dark and totally lit controls, indicating the amount of deflection is determined entirely by Physarum’s computation of the ratio of light intensity between the two regions.

For reasons of simplicity, most studies of decision-making focus on binary decisions (choosing between only two available options), where the decision to select one option is indistinguishable from the decision to reject the other. Few studies examine ternary or higher-order decision-making. In a notable exception, Physarum plasmodia were given a choice between 3 food sources that were all identical in value (Marshall et al. 2022). When the sum of the quality of all the food sources was low (low ‘magnitude’), plasmodia took significantly longer to make a decision than when the options had high magnitude; an identical result was recorded by Dussutour and colleagues (2019) in a binary-choice scenario. This is evidence for the widespread phenomenon of magnitude-sensitivity (also called value-sensitivity), where decision-makers show faster, less accurate responses when the value of all available options is high (Pirrone et al. 2022). Presented with a simultaneous choice of 11 different diet options, each of which differed in their ratio of protein (P) to carbohydrate (C) although none comprised the P:C ratio known to be Physarum’s ideal (Simpson and Raubenheimer 1993), the plasmodia spread their biomass over multiple options to select a combination that together was closest to their ideal intake target (Dussutour et al. (2010), Fig. 1 h). This is analogous to solving the ‘Knapsack problem’ of combinatorial optimisation, a problem relevant to many real-world processes where, given a set of items of known weight and value, an operator must choose the combination of items that maximises the value of the collection, but remains below a weight limit (Martello and Toth 1990).

While most animal cognition studies focus on adaptive decision-making, where the outcome of the decision-making process is a beneficial response that increases the organism’s fitness, it can be useful to examine maladaptive cases, such as irrational decision-making. Previous explanations of irrationality have centred around neurological mechanistic explanations, such as dopamine-reward systems (Anselme and Güntürkün 2019; Cocker et al. 2012), or specific brain regions such as the ventromedial prefrontal cortex (Koenigs et al. 2007). However, evidence from Physarum suggests that even non-neural organisms can be irrational. When Physarum plasmodia choose between two options that vary equally in two competing attributes—food concentration and light intensity—they show no preference for one option over the other. However, when a third, inferior option is introduced, plasmodia change their preference (Latty and Beekman 2011a). This is irrational because the decoy option should not affect the organism’s choice. In a related study, the same researchers showed that Physarum plasmodia are subject to speed/accuracy trade-offs when faced with difficult decisions (Latty and Beekman 2011b), another phenomenon that has been traditionally explained using a neuroscientific lens (Bogacz et al. 2010; Chittka et al. 2009).

It is clear (though often overlooked) that decision-making in non-neural and neural organisms is based on at least some shared underlying principles. We cannot hope to understand shared mechanisms, however, until we learn more about how decision-making works outside of a brain. To this end, some recent research has focussed on elucidating the mechanisms of decision-making in Physarum. A classic model for examining decision-making in animals is the multi-armed bandit problem, which explores how animals respond to the exploration–exploitation trade-off: should I exploit familiar but potentially sub-optimal options, or risk further exploration to potentially obtain superior ones? The bandit problem is inspired by casino slot machines (one-armed bandits), where gamblers decide which machine to play to maximise net payoff (Gittins 1979). The paradigm has been applied empirically to humans, birds, fish and social insects (Krebs et al. 1978; Keasar et al. 2002; Shettleworth and Plowright 1989; Thomas et al. 1985; Toyokawa et al. 2014) and also to Physarum (Reid et al. 2016).