Identifying CNB and GAF Domains in Plants

In order to identify the proteins which contain CNB or GAF domains, we initially used the Simple Modular Architecture Research Tool (SMART at smart.embl.heidelberg.de; [60 (link)-62 (link)]) to scan all predicted Arabidopsis proteins for CNB and GAF domains in the EMBL, TIGR or NCBI databases. Once redundancies were removed, a list of proteins was generated [see additional file 2]. In order to ensure broad coverage of possible variants, we also examined the Interpro collection of protein sequence analysis algorithms, all of which use slightly different methods [63 (link)]. As an additional method, the predicted proteins of the Arabidopsis genome were searched using the BLAST algorithm [64 (link)]. As search bait, we used several known cyclic nucleotide binding domains including those from GAF domains (human PDE2A [Swiss-Prot: O00408] and Anabaena cyaB1 [Trembl: P94181]) as well as CNB domains (human CGK2 [Swiss-Prot: Q13237], human RIIβ [Swiss-Prot: P31323], human Epac2 [Swiss-Prot: Q8WZA2], human rod CNGC [Swiss-Prot: P29973] and E. coli CAP [pir: E86000]). This yielded no new inclusions to our list of proteins, but did confirm each of our previous entries. For examination of the Oryza sativa spp. Japonica genome we performed BLAST searches using the aforementioned baits, as well as each of the Arabidopsis proteins. This search was performed using the Blast utility of the TIGR rice database [40 (link)]. The criterion for inclusion was that the CNB or GAF domain had to match the consensus motif with an E-value of less than 0.5 over the entire domain as determined by SMART. For newly identified proteins from the Orzya sativa, we named them so that they agreed best with the nomenclature of Maser et al. [37 (link)] [see additional files 1, 2, 3]. Sequence alignments were performed using the ClustalX [65 (link)] or T-COFFEE algorithms [66 (link)]and then inspected visually. Neighbor-joining trees were generated by ClustalX or PHYLIP [67 (link)], then were visualized with TreeView [68 (link)]. Trees generated using a variety of analysis methods (parsimony, distance and maximum likelihood) yielded similar results to the neighbor-joining trees.