The cognitive cell: bacterial behavior reconsidered
Research on how bacteria adapt to changing environments underlies the contemporary biological understanding of signal transduction (ST), and ST provides the foundation of the information-processing approach that is the hallmark of the ‘cognitive revolution,’ which began in the mid-20th century. Yet cognitive scientists largely remain oblivious to research into microbial behavior that might provide insights into problems in their own domains, while microbiologists seem equally unaware of the potential importance of their work to understanding cognitive capacities in multicellular organisms, including vertebrates. Evidence in bacteria for capacities encompassed by the concept of cognition is reviewed. Parallels exist not only at the heuristic level of functional analogue, but also at the level of molecular mechanism, evolution and ecology, which is where fruitful cross-fertilization among disciplines might be found.
Keywords: cognition, evolution, signal transduction, memory, information-processing, valence, learning, communication
The idea that microbes might know their world, not merely bump into it, is controversial but unoriginal. At the beginning of the 20th century, Jennings (1905/1962) claimed the behavior of the ‘lower animals,’ notably paramecia but including bacteria, could tell us much about the behavior of the ‘higher animals,’ including humans. The Animal Mind (Washburn, 1936), the first US textbook on comparative psychology (first edition 1908), begins with amoeba. Biochemist Daniel Koshland provided the modern scientific equivalent of Jennings’ argument in an under-appreciated monograph on bacterial chemotaxis (CT) as a model system for the study of behavior, and remarked (following Pope) that “the proper study of mankind is the bacterium” (Koshland, 1980b). Philosopher of science Karl Popper went further and argued (only slightly tongue-in-cheek) that in the evolution of problem-solving “from the amoeba to Einstein is just one step” (Popper, 1999).
More recently, in view of tremendous advances in methods for studying individual cells as well as population-based microbial behavior, the bacterium has been compared explicitly to a parallel distributed processing (PDP) network (Bray, 2009) that displays ‘minimal cognition’ (Lengeler et al., 2000; van Duijn et al., 2006; Shapiro, 2007). Arguments concerning bacterial ‘intelligence’ (Jacob et al., 2004; Hellingwerf, 2005; Marijuán et al., 2010) and even cells ‘thinking’ (Ramanathan and Broach, 2007) are appearing in mainstream journals, including the special series in this one. British psychologist Richardson (2012), who has been researching human intelligence (sometimes despairingly) since the early 1970s, recently concluded in an extraordinary article in EMBO that the nascent study of unicellular intelligence might provide the key to understanding intelligence in complex vertebrates, including humans. Unknown to Richardson (2012), a microbiologist specializing in computational biology has introduced a plausible formula for establishing ‘bacterial IQ,’ based on genome size and proportion of DNA segments coding for signal transduction (ST) proteins, as well as a rough gage of ‘introversion’ or ‘extroversion’ based on the relative proportion of environment-contacting ST systems (Galperin, 2005). Finally, neuroscientists and neurobiologists tracing the evolution of complex human, brain-based behavior increasingly locate its origins in the microbial realm (Allman, 1999; Damasio, 1999; Greenspan, 2007).
Yet, while work on bacterial adaptation underlies the contemporary biological understanding of sensory ST (Wadhams and Armitage, 2004), and ST in turn provides the foundation of the information-processing approach to cognition that (rightly or wrongly) remains the hallmark of the ‘cognitive revolution’ (Miller, 2003), cognitive scientists are largely oblivious to research in this and other areas of microbial behavior that might provide insights into problems in their own domains. For example, some flagellated bacterial species, paradigmatically Escherichia coli, have been found to possess large clusters of interacting sensory receptors, typically at the leading pole of the cell, which function as complex information-processing units linked to motor activity (Hazelbauer et al., 2007) and have been dubbed by some a ‘nanobrain’ (Webre et al., 2003). Meanwhile, microbiologists seem equally unaware of the potential importance of their work to understanding cognitive capacities in metazoans including vertebrates.
In the hope of stimulating interest among microbiologists in this area, this review brings together contemporary evidence for cognition in eubacteria in core areas of cognitive research: sensory ST, valence, communication, sensorimotor coordination, memory, learning, anticipation, and decision making in complex and changing circumstances. The article will show that parallels exist not only at the heuristic level of functional analog, but also at the level of molecular mechanism, evolution and ecology, which is where fruitful cross-fertilization among disciplines might be found. Archaea, whose behavior is less well documented and appear (so far) to possess far fewer ST systems than bacteria (Ulrich and Zhulin, 2010), will not be discussed here.
The review is organized into two parts. Part 1 provides an introduction to the study of cognition. Part 2 sets out a proposed bacterial cognitive toolkit. Because there is a great deal of ground to cover, and not a lot of space to cover it in, the mechanisms of how systems work will be sketched in broad outline, not in detail, except in those cases where the details are either not widely known or not easily accessible in the literature. The aim here is to stimulate microbiologists to think about their work in the broader context of a cognitive biology (Kovác, 2000; Auletta, 2011; Bechtel, 2014; Fitch, 2014) and to consider designing experiments to explicitly enhance this understanding.
Whether or not bacterial behavior ultimately should be described as ‘cognitive’ is a matter for the scientific community to decide, when consensus finally forms around what cognition is (Bechtel et al., 1998). The absence in psychology and elsewhere in the cognitive sciences of agreed definitions for key concepts, such as cognition, has been decried for well over a century – James (1984) arguably setting the benchmark for despair – and continues to be a cause for concern (Staats, 1983; Sternberg, 2005). As it stands, cognition, like ‘intelligence,’ is a highly polysemic theoretical construct that can be defined quite broadly (La Cerra and Bingham, 2002) or very narrowly (Wynne, 2004), depending on the investigator’s personal intuitions and the ground in which those intuitions are embedded, whether in the principles of biology or the human case (Lyon, 2006; see Table 1). Needless to say, this state of affairs is far from ideal.
The study of cognition as a biological function may be unique in the life sciences in the extent to which findings ultimately are calibrated against a single species, Homo sapiens (Shettleworth, 1993). The study of respiration and other biological functions, for example, are not so calibrated, although scientific investigation doubtless began with concern for the human case. (We stop breathing, we die – so what is breath?) Rather, evidence is followed wherever it leads, and it can lead to unexpected places. We know now that the oxidation of nicotinamid dinucleotide (NADH), the molecular substrate of cellular respiration, is a process universally shared, and in animals is relevantly similar across phyla. By contrast, prokaryotic memory was discovered more than four decades ago (Macnab and Koshland, 1972), yet is far from being accepted – even by microbiologists – as anything relevantly like memory in complex vertebrates, to say nothing of humans. Cognitive scientists might be similarly dismissive were they to give it any thought at all.
Memory, needless to say, is critical to cognition. Without memory, present circumstances have no context; the detection of change is impossible. Without the ability to detect change, the decision to alter behavior can only be random, haphazard. Without memory, learning of any kind is impossible. While cognitive scientists now accept that discoveries concerning the molecular basis of memory in the marine invertebrate Aplysia are relevant to the study of human memory (Kandel, 2006), they (to say nothing of microbiologists) have yet to connect the dots with memory processes in prokaryotes.
Nevertheless, similar ST mechanisms appear to be at work, albeit with different leading biochemical actors and different degrees of regulatory complexity. For example, ST in bacterial two-component systems (2CSs) relies principally on histidine kinases for protein phospho-transfer and DNA methylation, while genomic analysis shows that serine/threonine kinases, their analog in eukaryotes, comprise roughly 1 in 4 of the more than 500 protein kinases involved in ST in humans (Capra et al., 2006)1. However, analysis of bacterial genomes increasingly reveals serine/threonine kinases in prokaryotes, particularly those that live in changing environments and display complex social behavior, such as myxobacteria (Munoz-Dorado et al., 1993; Schneiker et al., 2007), Bacillus subtilis (Macek et al., 2007), and Paenibacillus vortex (Sirota-Madi et al., 2010). Calcium/calmodulin-dependent kinase II (CaMKII) is a serine/threonine kinase critical for memory formation in humans (Giese and Mizuno, 2013). Does it have a prokaryotic homologue? To my knowledge, no one is looking.
In short, in contrast to other life scientists, as well as to their forebears of yesteryear, contemporary cognitive scientists have been less inclined to follow the Darwinian principle – that biological functions differ substantially ‘in form but not in kind’ (Darwin, 1874/1909) – very far ‘down’ the phylogenetic yellow brick road in the search for clues as to the what and why of cognitive function (Bekoff et al., 2002).
Behavior generation lies at the heart of all cognitive inquiry, and while unicellular organisms are simpler than metazoans in structure, they are not always behaviorally less complex. The well-described social predation style, rippling behavior and fruiting body formation of Myxococcus xanthus arguably are more complex than the activities of most Porifera and even some worms with simple nervous systems. Extensive experimental evidence shows that microbial behavior is guided by processes that, in other contexts, are readily regarded as part of biological cognition, capacities which together encompass an organism’s ability to navigate, become familiar with, value, learn from and solve critical existential problems within its world of experience, including coordinating action with conspecifics. This may explain why, for the past several decades, microbial researchers increasingly have helped themselves to cognitive terminology (i.e., ‘decide,’ ‘talk,’ ‘listen,’ ‘cheat,’ ‘eavesdrop,’ ‘lure,’ ‘vote’) to describe complex bacterial behavior, often without caveat (e.g., Adler and Tso, 1974; Fuqua and Greenberg, 1998; Ben-Jacob et al., 2004; Bassler and Losick, 2006; Baker and Stock, 2007; Williams et al., 2007).
In the following pages I hope to show that such linguistic usage is not wholly metaphorical. I believe there is something going on at the microscopic level that doesn’t just ‘look’ cognitive, it is cognitive, or, more accurately, it is typically considered cognitive when studied in animals more like us.
Discussions of cognition across phyla often degenerate into qualms about definition that typically begin with ‘yes, but what do you mean by…’ The concept that guides this investigation will be specified therefore. Because we are not concerned with the special capacities of a single mammalian species (i.e., Homo sapiens), a general definition from comparative psychology will serve, with some modification. Shettleworth (1998, p. 5) defines cognition as “the mechanisms by which animals acquire, process, store, and act on information from the environment. These include perception, learning, memory, and decision making.” Cognition thus is comprised of the total suite of mechanisms that underwrite information acquisition, storage, processing, and use. Although this definition can be applied to phyla Shettleworth (1998) may not have had in mind, e.g., microbes, it is (for the most part) uncontroversial within the domain of comparative psychology and cognitive ethology.
Shettleworth’s definition, and others like it in the cognitive scientific literature, emphasizes the psychologically delineated elements of the function. Others concerned with the specifically biological manifestation of cognitive function – which historically includes William James, H.S. Jennings, Jacob von Uexküll, Konrad Lorenz, James J. Gibson, and Humberto Maturana—emphasize the ecological context of information ‘use,’ what cognition does for the organism in the business of staying alive in a particular ecological niche. With this in mind, I extend the definition as follows:
Biological cognition is the complex of sensory and other information-processing mechanisms an organism has for becoming familiar with, valuing, and [interacting with] its environment in order to meet existential goals, the most basic of which are survival, [growth or thriving], and reproduction. (Lyon, 2014, p. 174)
While they have several forms of metabolism not found in any eukaryote (e.g., nitrogen fixation, methanogenesis), prokaryotes possess most of the biological functions shared by all animals, either by virtue of their own structure or those of their symbionts. Bacteria possess processes for breaking down and transforming nutrients into usable forms of energy for self-production or generating behavior, storing energy in molecules that can be catabolized when needed, and eliminating molecular waste generated during these processes. To stay alive, bacteria also appear to possess a ‘basic toolkit’ (Godfrey-Smith, 1996), which in other forms of life—especially mammalian life – would be termed cognitive. In the case of well-studied bacterial species (e.g., E. coli, B. subtilis, M. xanthus, Pseudomonas aeruginosa), the basic toolkit includes at least eight capacities found in different but recognizable forms in complex metazoans (see Box 1).
Like all organisms, bacteria adapt to changes in their environment by modifying their metabolism and behavior. The traditional view is that this process depends on a limited number of highly canalized, inflexible mechanisms, which bear little resemblance to their behavioral counterparts in animals with nervous systems. However, evidence increasingly points to the idea that bacteria—particularly ecological ‘generalists’ as distinct from ‘specialists’—are highly social, flexible responders, responders (however limited their actual response styles) that rely on a sophisticated suite of sensory and information-processing mechanisms (Shapiro, 1998). Such mechanisms are characterized by non-linear responsivity, integration of multiple information channels, and (in the case of CT receptors, at least) habituation and adaptation phenomena. These structural and dynamical features are among the defining features of neuronal sensory processing in animals, including humans.
Keep in mind that a great deal of complexity will be obscured in the simplifications that follow. In each cell of E. coli, for example, which is not the smartest proteobacterium on the block (but no dummy, either), there are upward of 10,000 chemoreceptors per cell, each with multiple binding sites, whose output interacts with several flagella, on each of which are about 40 binding sites for CheY-P, the protein that modifies the direction of rotation. The behavioral output for a single cell is thus hugely complex, and much more refined (tumbles, twiddles, etc.) than normally described. When we simplify to get just two outputs, run and tumble, and then bundle behaviors to get population averages, a lot of the cognitive output is likely to be occluded. Consider an example with humans: if we assumed just two behaviors, walking and standing, and measured population averages using different stimuli, more would be hidden in the results than is revealed2.
As do animals and plants, bacteria monitor the biotic and abiotic features of their surrounding milieu using a wide variety of sensory systems, and adapt their behavior and physiology in response to what they perceive in order to stay alive and reproduce. While the ecological niche of some bacteria is fairly stable, if sometimes extreme, many inhabit highly mutable environments and conduct highly complex lifestyles within them, which present considerable challenges, some of which are capable of being anticipated and others not. Like all organisms, bacteria do not gather information in a vacuum, but are able to assess the significance of the signals they receive relative to their own functioning, their internal milieu—although it is worth noting that the power spectrum of signals remains unknown for any bacterium3.
In the psychological literature, assignment of an existential value to the sum of information concerning current circumstances, internal and external, is called valence, which has been described as ‘a basic building block of emotional life’ (Barrett, 2006). Valence refers to the attractiveness, acceptability or tolerability of a stimulus (Lyon, 2014). Usually such stimuli are rendered in terms of ‘positive’ or ‘negative,’ ‘attractant’ or ‘repellant,’ in addition to neutral stimuli to which the organism is indifferent, but these characterizations are not always stable. Aversive conditioning depends on the capacity of previously neutral or even positive stimuli to become negative for an organism under particular circumstances. Thus, all sensory signals have a context-dependent valence—positive, negative or neutral—which influences the available response.
In mammals specialized brain-based systems exist for this purpose, notably the hypothalamic-pituitary-adrenal (HPA) axis and the limbic system, which together appraise for a given signal not only what it is and where it is but also what it means ‘for me’ (Vedder, 2008). This involves brain structures such as the amygdala and the orbitofrontal cortex. Although bacteria lack such specialized organs, the valence of a signal is implicit in the processes that coordinate, say, the rotation bias of flagellar motors or the complex developmental sequence of signaling, genetic transcription and protein expression that leads to sporulation.
One of the most resilient obstacles to thinking about microbes as potentially cognitive organisms is the lack of specialized sensory and information-processing systems that exist in vertebrates and especially mammals. In bacteria “sensory, regulatory, and metabolic networks must all drive environmental perception and corresponding action” within the boundaries of a single cell (Freddolino and Tavazoie, 2012; p. 364). In actual fact, while cognitive scientists for many years have tried to draw principled distinctions between ‘metabolic’ and ‘cognitive’ function (even in microbes—see, for example, van Duijn et al., 2006), this dividing line is increasingly difficult to defend even in human beings, given increasing understanding of immune system involvement in normal memory and learning (Yirmiya and Goshen, 2011) and the surprising effects of psychological stressors on metabolic physiology (Maier and Watkins, 1998; Yirmiya et al., 2006).
Bacteria are capable of sensing and responding to an astonishing variety of environmental signals: amino acids, sugars, oxygen, pH, osmolarity, temperature, light, secondary metabolites, molecular waste products (e.g., ammonia), environmental DNA, and other microbes (both conspecifics and other species)—even physical objects such as tiny, chemically inert beads (Dworkin, 1983).
Signal transduction systems in bacteria come in several basic forms that perform a wide range of behavioral and physiological roles (Ulrich and Zhulin, 2010): one-component systems (1CSs); 2CSs, which also come in hybrid forms; CT systems, a special sub-class of 2CSs; extracytoplasmic function (ECF) sigma factor proteins, the so-called ‘third pillar of bacterial ST’ (Staroń et al., 2009), as well as non-canonical systems that involve three and more components (Marijuán et al., 2010).
One-component systems consist of a single polypeptide chain containing both a sensor domain that registers a stimulus/signal and an output domain that elicits a response, for example, binding to DNA or interacting with other effector proteins. Thought to be evolutionarily older, 1CSs monitor the level of intracellular metabolites or respond to concentrations of membrane-permeable extracellular ligands (Galperin, 2005; Ulrich et al., 2005). Ulrich and Zhulin (2010) conclude on the basis of genomic data from thousands of species of bacteria and archaea that 1CSs are the most numerous by a considerable margin (see Table 2).
Two-component systems segregate the sensor/input and effector/output functions between a sensory kinase (SK), which upon stimulation autophosphorylates, typically at a highly conserved histidine residue, and a response regulator (RR) to which the SK binds and transfers phosphoryl groups, sometimes at multiple sites. Phosphorylation and dephosphorylation, which occur at different rates at different sites, may affect transcription directly or as one step in a signaling sequence (Krell et al., 2010). The high-energy state created by modification of the RR constitutes a marker of environmental conditions at a particular time, which lasts until the chemical group is removed, thus constituting the basis of molecular memory (Koshland, 1977). Although 2CSs are commonly comprised of membrane-bound sensors interacting with the surrounding milieu and with one or more intracellular RRs, in some systems both components are located entirely within the cytosol (Krell et al., 2009).
Because 2CSs have two inherent advantages—multiple possibilities for control and graded signal amplification—they appear to form ‘phospho-relay systems’ with other regulatory components that employ transient phosphorylation as a means of transducing signals within the cell (Hellingwerf et al., 1995). Processes that support the creation of phospho-relay systems include cross-talk between (e.g., Mitchell et al., 2009) and/or constitutive convergence of pathways (e.g., Tagkopoulos et al., 2008). Thus a phospho-relay system may incorporate a single accessory protein, which recognizes the signal/stimulus and transmits the molecular information to the sensor kinase, making a three-component system (Krell et al., 2010). Other systems comprise several components. For example, E. coli has a seven-component signaling system for detecting orthophosphate (Pi) and regulating the genes of the phosphate (Pho) regulon through the PhoR/PhoB 2CS, a paradigm in bacteria of a membrane-bound receptor controlling cytoplasmic gene expression (Hsieh and Wanner, 2010). Pi is the bacterium’s preferred source of phosphorus, which is essential for physiological functioning, not least due to its ubiquitous role in ST. The seven protein components include the PhoR histidine kinase and its cognate PhoB RR; four components of the ABC transporter Pst; and PhoU, a chaperone-like protein that inhibits PhoR/PhoB (Hsieh and Wanner, 2010).
The third most abundant ST elements in bacteria belong to the family of sigma (σ) factors, which bind to RNA polymerase to promote genetic transcription, the basic ‘housekeeping’ form of which is present in all RNA polymerase holoenzyme complexes (Mascher, 2013). ECFs belong to the σ70 subfamily of alternative σ factors. The average bacterial genome contains six (Staroń et al., 2009), although some species contain none (e.g., Streptococcus aureus) while others contain many (e.g., 41 in M. xanthus; Ulrich and Zhulin, 2010). Alternative sigma factors possess different promoter-recognition properties, which increase the options for transcription under challenging conditions. As Gruber and Gross (2003, p. 441) put it: “The cell can choose from its repertoire of sigmas to alter its transcriptional program in response to stress.” This suggests significant heterogeneity may exist in sigma factor utilization between cells, based on individual history as well as growth stage, which may provide a source of selectable variability. Mascher (2013, p. 153) observes that the majority of ECF groups are phylum-specific, “reflecting the similar life style and physiology of a given group of related organisms.” The activity of many sigma is typically controlled, and normally suppressed, by a cognate anti-sigma.