Teaching Bacteria to Learn

Press release (1 October 2008)

European researchers have devised a method to implement Pavlovian learning as a molecular circuit in bacteria. Plasmids with the genetic model of a Hebbian learner are inserted in an E. coli bacterium, enabling the cell to learn correlations among specific chemicals in its environment and then report these back to the researcher in the lab. The study will be published as an open access article in the Journal of the Royal Society Interface.

Researchers from the European Union's ESIGNET project design molecular circuits that allow for associative learning in single-celled organisms

Learning is considered an essential property of our human intelligence and our abilities to adapt to an ever-changing environment. It is traditionally assumed that learning is an ability that is confined to higher organisms with nervous systems, as neurons are intuitively accepted as the basis for the processes that govern learning in individuals. Indeed, artificial neural networks (ANNs) -- mathematical models of such neurons -- have provided many explanations on how a brain is capable of learning or memorizing in organisms with these nervous systems. As an example, the behavior of Pavlov's dogs -- who have been shown to be able to learn associations between the sound of a bell (a "conditioned stimulus") and the smell of food (an "unconditioned stimulus") -- can be captured in an elementary mathematical model of a biologically workable system based on neurons. Recent work in systems biology reveals molecular circuits that are rather similar to neural networks within individual cells are also plausible, opening the doors for implementations of higher level processes and behaviors, such as associative learning, in these molecular circuits.

In their forthcoming paper in the Journal of the Royal Society Interface (titled "Molecular circuits for associative learning in single-celled organisms", to be published October, 1st 2008), collaborating researchers from the United Kingdom, the Netherlands and Germany demonstrate that these mathematical models are not restricted to neural systems, but that intracellular molecular interaction networks in a single-celled organism could implement these models of associative learning. Instead of neurons that learn associations and act upon these, the study has shown that similar behaviors can be constructed within the single-celled organisms, grounding the neural behavior in concentrations of specific genes, their products and their promotors. The authors show various ways on how such an intracellular network can be constructed in the lab, using artificial gene constructs ("plasmids"). As such, it is feasible to construct artificially adapted E. coli bacteria that are capable to learn associations between chemicals in their environment, and then report such correlations back to the researcher in the lab. In the paper, a mathematical model of such learning molecular circuits is developed, and preliminary designs for implementing the model in genetic and molecular signaling networks are proposed.

It is interesting to speculate about the potential medical applications of such molecular circuits capable of learning. One idea is to use them as intelligent bio markers for reporting on existing associations between cellular components. To do this one constructs learning plasmids as described in the paper, learning is then used to train the resulting gene regulatory perceptron to classify the observed input vector. Secondly, although subject to difficulties and challenges, one might envisage that an implanted bacterial system might adaptively tailor the anticipatory release of a drug to predict antecedents to a toxic chemical that vary from patient to patient, and even within the lifetime of one patient. The circuit described in the paper provides a potential basis for such systems to learn to release a drug to suit the patient in which the bacteria resides for the maximum benefit of that patient.

Molecular circuits for associative learning in single-celled organisms

Journal of the Royal Society Interface, 1 October 2008
by Chrisantha T. Fernando (1,5), Anthony M.L. Liekens (2), Lewis E.H. Bingle (1), Christian Beck (4), Thorsten Lenser (4), Dov J. Stekel (1), Jonathan E. Rowe (3)
(1) Systems Biology Centre, University of Birmingham, Birmingham, B15 2TT, United Kingdom
(2) TU/e Technische Universiteit Eindhoven, the Netherlands
(3) School of Computer Science, University of Birmingham, Birmingham, B15 2TT, United Kingdom
(4) Bio Systems Analysis Group, Friedrich Schiller University Jena, Germany
(5) MRC National Institute for Medical Research, Mill Hill, London, London NW7 1AA, United Kingdom

Abstract: We demonstrate how a single-celled organism could undertake associative learning. Although to date only one previous study has found experimental evidence for such learning, there is no reason in principle why it should not occur.We propose a gene regulatory network that is capable of associative learning between any pre-specified set of chemical signals, in a Hebbian manner, within a single cell. A mathematical model is developed, and simulations show a clear learned response. A preliminary design for implementing this model using plasmids within Escherichia coli is presented, along with an alternative approach, based on double-phosphorylated protein kinases.

(a) The neural network implementation of Hebbian learning for two inputs u1 and u2. The orange circles represent presynaptic neurons that project onto a single post-synaptic neuron (blue). The simultaneous firing of the input neurons causes the synaptic weights w1 and w2 to increase, reinforcing their association. The blue curved lines show how this Hebbian positive feedback works, e.g. the weight w1 increases as a product of the output firing rate p and the input firing rate u1. (b) The equivalent gene circuit implementation using three genes is shown. The two input molecules (enhancers) are shown as orange circles, u1 and u2. They bind to the repressors (red circles) r1 and r2 and this results in activation of transcription of w1 and w2 molecules (in conjunction with transcription factor p) and activation of transcription of the p molecule (in conjunction with w1 and w2). To correspond to (a), the output molecule p is shown in blue. (c) Plasmid structures that could implement one half of the circuit. The first plasmid contains fnr and tetR. The second plasmid contains orfP (cI ) and gfp, see text for details. (d ) Alternative implementation using phosphorylation cycles. The inputs are again shown as orange circles u1 and u2, here they represent kinases that do one of two phosphorylation steps on the weight molecules w1 and w2 again shown in grey. The first phosphorylation step is done by a double phosphorylated output molecule p. Phosphorylation state is represented as yellow stars, one star means single phosphorylated, and two stars means double phosphorylated. Reversible and irreversible reactions are shown. The dotted arrow from w**1 to w*1 and w**2 to w*2 indicates that this reaction is slow, i.e. that memory persists in the form of double phosphorylated w**1 .

About the ESIGNET project: The ESIGNET project (Evolving Cell Signaling Networks in Silico) is a Specific Targeted Research Project funded by the European Commission under the Sixth Framework Programme. The goal of this project is to study the computational properties of Cell Signaling Networks (CSNs) by evolving them using methods from evolutionary computation, and to re-apply this understanding in developing new ways to model and predict real CSNs. The project is highly interdisciplinary. Its completion requires insight into the subject from many points of views. The research will be at the interface of (at least) Biology, Computer Science, and Control Engineering. It also utilises a plethora of approaches and methods. The high potential of the proposal is largely due to the co-ordinated and concerted multi-disciplinary and methodological approaches. This is reflected in the composition of the consortium. All researchers in this consortium have previously been involved in research at the interface between Computer Science and Biology and have a strong ability to integrate insights from those fields.

Correspondence: For further questions and correspondence, contact Chrisantha T. Fernando, Mathematical Biology, National Institute for Medical Research, Mill Hill, London NW7 1AA, United Kingdom, ctf20@sussex.ac.uk