Learning sans neurons in a unicellular organism

Cognition in general, and learning in particular, are usually associated with animals that have complex nervous systems. However, even single celled organisms can demonstrate a wide range of behaviors normally associated with cognition, despite not having a brain. For example, the giant single-celled ciliate Stentor coeruleus displays habituation, a fundamental type of learning shared among all animals. In habituation, an organism gradually stops responding to a stimulus when that stimulus is repeated over and over. How does this happen in a single cell with no nervous system? We have used a computer-controlled device to provide programmable stimulus sequences to individual Stentor cells and query their responses. This work showed that while a population of cells shows a graded response, at the single-cell level, the response appears to be step-like. Based on this result, we proposed a two-state Markov model in which cells switch between a responsive and a non-responsive state. But this model was unable to account for a number of features having to do with response recovery after cessation of the stimulus. The two-state model was also extremely abstract and did not explain how such switch-like behavior would be synchronized at the molecular level. We have now developed an alternative "bottom up" model, in which we represent the expected molecular features of the cellular response, including internalization, degradation, and recycling of the putative mechanoreceptor molecule. This model is characterized by processes having two different time scales: internalization/recycling of receptors, which is rapid, and protein-level degradation and synthesis, which is slow. The resulting separation of timescales allows the model to explain a wide range of known phenomena relating to habituation in Stentor, and also suggested new experiments, in the form of various sequences of strong and weak stimuli, which we are now using to probe the model and estimate parameters. The model also provides a way to characterize the learning mechanism in terms of signal processing, which we are now applying to better understand what it is that the cell is "trying" to learn, as well as to devise new ways to probe the learning machinery using frequency responses and random input sequences.