The cognitive cell: bacterial behavior reconsidered

Tight coupling of the two pathways in combination with susceptibility to alteration in the sequence of expression suggests a high probably of crosstalk between the signaling pathways. In the second series of experiments, crosstalk between two metabolic pathways in E. coli were not merely decoupled but completed dissociated (Mitchell et al., 2009). These discoveries would have been impossible without taking the organism’s habitat and ecology into account (Tagkopoulos et al., 2008).

In the mammalian GI tract E. coli is exposed to nutrient sugars in a sequence: first lactose, then maltose. Mitchell et al. (2009) confirmed that in the presence of lactose, transcription of the pathways for metabolizing maltose begins before maltose is detected. The dependency is asymmetrical, however. Exposure to maltose does not induce lactose utilization. Furthermore, in evolutionary experiments with a conditioning regime in which maltose does not follow lactose, over 500 generations E. coli loses the ability to predict the ecologically important sequence, but not the ability to utilize maltose. When maltose is detected, physiology is still modified for its metabolism in the evolved strains. What was lost during laboratory evolution, the researchers conclude, is “the asymmetrical cross talk between the two pathways” (p.221).

The third set of predictive experiments tested responses to QS in three species of proteobacteria with radically different lifestyles: Burkholderia glumae, a rice pathogen; Burkholderia pseudomallei, a human pathogen; and Burkholderia thailandensis, non-pathogenic saprophyte (Goo et al., 2012). QS signaling is an indicator of increasing cell density but could also predict the limits of population carrying capacity. In all three species, QS induced physiological preparation for stationary phase, survival of which requires the alteration of global metabolic processes. In contrast to QS-intact strains, QS mutants showed “massive and rapid population crashes,” which began shortly after commencement of the stationary phase. The cause of the population crash proved to be ammonia-related alkaline toxicity due to amino acid catabolism associated with stationary phase.

While both QS-competent and QS-mutant cells produced ammonia at roughly similar rates, the QS-competent strains were protected from catastrophe by QS-mediated production of oxalate, which counteracts base toxicity. Oxalate is produced at neutral pH, which suggests that QS-dependent “anticipatory production” of oxalate is not induced by a pH-dependent regulatory mechanism. The authors conclude the results “show that QS can function to allow anticipation of overcrowding and promote bacterial survival at maximum population carrying capacity, and provide insight into how QS bacteria have evolved to control both public and private goods,” namely, oxalate production and cell survival (Goo et al., 2012, p. 19778).

Based on these and other findings, Freddolino and Tavazoie (2012) assert that the homeostatic paradigm can no longer sustain an appropriate understanding of cellular behavior, whereas a ‘predictive-dynamic framework’ is more explanatory. They conclude (correctly, in my view) that regulatory networks in microbes and neural networks in metazoans have essentially the same function, and that microbiologists are now moving into behavioral territory previously occupied by animals with nervous systems. Whether these also share mechanisms with similar design principles remains to be seen.

A cautionary note must be sounded, however. Researchers in two of the studies described here (Tagkopoulos et al., 2008; Mitchell et al., 2009) claim their discoveries are demonstrations of associative learning. However, conditioning in these cases is clearly epigenetic. Whether this will count as ‘genuine conditioning’ remains an open question. Also, Tagkopoulos et al. (2008) claim that one of the implications of their research is that bacteria possess internal models or representations of the environment. The existence of representations and models, except in a metaphorical sense, is still open to debate even in humans (see, for example, Bechtel, 1998; van Gelder, 1998; Haselager et al., 2003), so microbiologists need to be careful.

So how does an individual bacterium integrate the information from a dizzying array of signaling pathways—sensorimotor, physiological, chemosensory, and communicative—into a coherent adaptative response? The short answer is we don’t know. This is perhaps the greatest challenge facing work in this field. We have already seen how cross talk can link substantially different signaling pathways (heat shock and respiration) to enable cells to predict regular changes in their habitat, but the mechanism is unknown. Mathematical or computational frameworks for modeling specific signaling pathways are proliferating, well, like bacteria in exponential growth (see, for example, Di Paola et al., 2004; Mehta et al., 2009; Liebal et al., 2010; Mitchell and Pilpel, 2011). How successful these proposals will prove remains to be seen.

At present prokaryotes offer very few examples of specialized information-processing ‘organs’ for investigation. The chemosensing receptor clusters at the leading pole of flagellated bacteria such as E. coli have already been suggested as analogous in important respects to neural clusters in metazoans with nervous systems (Stock et al., 2002). With an estimated 10,000 receptors in each cluster, with five different sensory targets and multiple binding sites, the computational complexity of the nanobrain is not especially tractable, however, (see Box 2). A second candidate involved in sensorimotor coordination are the FACs on the sides of M. xanthus together with the protein complex for transducing signals from the cell surface to the oscillator-regulator located in the cytosol. A third candidate organ for specialized information-processing in bacteria is the ‘stressosome,’ a large (multimillion dalton) protein complex associated with the general stress response that integrates multiple signals into a single outcome (σB activation), and is best-characterized in B. subtilis, although it is found many microbial phyla (Marles-Wright et al., 2008)5. Interestingly, the stressosome of B. subtilis recently has been found to be activated by a blue-light photoreceptor (van der Steen et al., 2013).

Object sensing. James Shapiro described as “the greatest paper nobody has read.”∗ Dworkin’s (1983) report of the directed movement of M. xanthus swarms toward three-dimensional objects in multiple experiments. The swarms moved toward biologically salient objects (a clump of prey cells) but also toward chemically inert objects (sterilized glass beads). These experiments should be replicated, and, if the findings are confirmed, the mechanism investigated. It would be interesting to see whether previously uncharacterized photoreceptors are involved.

Bistability. Given the critical importance of bistability to cognition in more complex animals, what light can bistable regulatory networks in bacteria throw on the phenomenon, including its role in perception, learning and memory?

Endogenous activity. One of the greatest recent challenges to the classical computational view of cognition has been the discovery of endogenous, often oscillatory activity in the brain independent of external stimuli (Bechtel, 2012). Can oscillatory activity in regulatory circuits governing prokaryotic behavior tell us anything about the processes involved in endogenous brain activity?

Nanobrain. The cluster of chemoreceptors that governs motility (and other behaviors?) in E. coli provides a relatively tractable model for understanding how signals from multiple sources can be integrated into a behavioral response but still presents a monumental challenge.

Valence. Virtually nothing is known, in bacteria or humans, about how combinations of external stimuli of differing valence are integrated with interoceptive cues into a coherent behavioral response. Even more basically, hardly any information exists on the power spectrum (i.e. intensity and frequency) of signals to which any microorganism in any natural environment is exposed, except for circadian (24-h) frequency. Well-characterized stress responses in prokaryotes, particularly those in which multiple behavioral options exist, provide potentially productive platforms for investigation.

Memory and learning. Bacterial memory is yet to be properly investigated, and the mechanisms by which it occurs definitively characterized in different species. The possibility of associative learning in prokaryotes also remains terra incognita. M. xanthus may prove a good candidate for such exploration, given the unusually large number of genes (97) involved in producing serine/threonine protein kinases (STPK) in this diverse predator (Goldman et al., 2006). CaMKII is an STPK capable of autophosphorylation that is involved in multiple signaling cascades involved in homeostasis in animals, and also is thought to be important in human memory and learning. It would be interesting to see if there is any sequence homology between CaMKII and any of the myxobacterial STPKs.

Communication. The ability of QS molecules to participate in modifying the behavior of dispersed cells, particularly in relation to stress, has led to the suggestion that they may be precursors of hormonal messengers in animals, and indeed some homologies have been identified, for example, between AHLs and ghrelin (Tizzano and Sbarbati, 2006). Hormones, particularly those associated with the HPA axis, are major mediators of cognitive, affective and motive function in humans, (Vedder, 2008). The HPA axis, in turn, is the major mediator of the stress response in humans. Is there an evolutionary relation between QS molecules and the stress-induced molecular messengers of the mammalian brain, including not only hormones and neuropeptides but also immune elements such as cytokines?

*Shapiro made the remark in a workshop presentation attended by the author at the University of Exeter, in July 2006.

While Jacques Monod’s famous assertion that “anything…true of E. coli must also be true of an elephant” (Friedmann, 2004) is an overstatement for many reasons, the Darwinian assumption that biological functions differ substantially ‘in form but not in kind’ continues to be born out in surprising ways, particularly since the advent of comparative genomics. Phylogenetic snobbery in the study of biological cognition is misguided, and for the same reasons such a bias would be in genetics or the study of respiration: there is much to be learned about the basics of biological functions and processes in ‘simple’ systems.

It is always satisfying, and the life sciences are seen to advance, when connections can be drawn between widely differing phyla with respect to important biological functions and processes. The study of cognition, for much of its existence, has proven resistant to this impulse. However, modern biology’s tools of genetic manipulation, comparative genomics, and molecular imaging techniques that enable the study of single behaving cells represent the means by which a continuum might be established, if it exists in fact.

Although knowledge of ST in both prokaryotes and eukaryotes has increased spectacularly in the interim due to technological and methodological innovations, interest in exploring the parallels between prokaryotic and neuronally based behavior has waned substantially since its heyday in the 1970s and 1980s. It is time to revive the project. The technical means available are better than they have ever been.

The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The author would like to thank the editors of this special topic on microbial intelligence, for the opportunity to develop these ideas; the two reviewers for comments that improved the manuscript immeasurably; the Australian Research Council (DP0880559) for funding the initial research into this topic; Jon Opie for important discussions in developing these ideas; Jeffry Stock for providing valuable feedback on an early draft that forms the basis of this article; and Richard Bradshaw, without whom none of this would have been possible.