Transcription Regulatory Network: A primitive brain in bacteria

Source: Tagkopoulos et al (2008). Predictive Behavior Within Microbial Genetic Networks. Science 320:1313-17.

Whole-genome fitness assays don't match gene expression microarrays. For example, a heat shock in E. coli results in up- and down-regulation in a range of genes; however, many of these changes are dispensable. Our lab makes the case that the transcription network makes many changes not based on necessity but based on anticipation.

Artificial activation of the envelop stress response in turn activates cytoplasmic heat shock response, not because it contributes to protein folding in periplasm but because higher temperature outside the cell is coupled with higher temperature in cytoplasm. E. coli uses the data from periplasm to anticipate a future change in its cytoplasm. This is a very basic associative learning (or classical conditioning). A graduate student and a postdoc in our lab have elegantly demonstrated the ability of transcription networks to incorporate environmental signals into effective anticipations. This study has both computational and experimental facets. First, Ilias developed a simulation framework called Evolution in Variable Environments (EVE) to show that transcription network models are very well capable of learning simple associations (AND, OR, XOR and etc.). In parallel, Yirchung Liu studied the negative correlation between the oxygen level and tempertaure in the natural habitat of E. coli as it passes through the mammalian GI tract to the outside world and vice versa. She showed that increasing the temperature results in a shift towards anaerobic respiration in wild-type E. coli. Now if you evolve E. coli in an environment in which the correlation between temperature and oxygen level is de-coupled, the organism forgets this correlation and does not switch to anaerobic respiration in higher temperatures.

This case portrays one of the weak points of gene-expression microarray analysis.