This should probably have gone in the last post, but I just read it:
Ali Zarrinpar, Sang-Hyun Park and Wendell A. Lim, "Optimization of Specificity in a Cellular Protein Interaction Network by Negative Selection", Nature 426 (2003): 676--680 [link]
Abstract: Most proteins that participate in cellular signalling networks contain modular protein-interaction domains. Multiple versions of such domains are present within a given organism: the yeast proteome, for example, contains 27 different Src homology 3 (SH3) domains. This raises the potential problem of cross-reaction. It is generally thought that isolated domain–ligand pairs lack sufficient information to encode biologically unique interactions, and that specificity is instead encoded by the context in which the interaction pairs are presented. Here we show that an isolated peptide ligand from the yeast protein Pbs2 recognizes its biological partner, the SH3 domain from Sho1, with near-absolute specificity—no other SH3 domain present in the yeast genome cross-reacts with the Pbs2 peptide, in vivo or in vitro. Such high specificity, however, is not observed in a set of non-yeast SH3 domains, and Pbs2 motif variants that cross-react with other SH3 domains confer a fitness defect, indicating that the Pbs2 motif might have been optimized to minimize interaction with competing domains specifically found in yeast. System-wide negative selection is a subtle but powerful evolutionary mechanism to optimize specificity within an interaction network composed of overlapping recognition elements.
There are lots of fascinating problems related to signal transduction networks. One of the most intriguing, to my mind, is how the cell avoids cross-talk --- different signalling pathways with common or even just similar components interfering with each other. This paper does a very nice job of showing one way in which cells evade the problem, namely by exploiting the power of chemistry and negative selection. (There is also a good "news and views" commentary on this paper, by Drew Endy and Michael B. Yaffee, on pp. 614--615 of the same issue.) This is potentially very important for (among others) people in complex systems working on protein interaction networks, especially their evolution.