In terms of the full length proteins

In terms of the full-length proteins, it was noted early on that ACs have two cellular localizations: as integral proteins of the plasma membrane and as soluble proteins in the cytosol. Based on the distribution in mammalian sequences, it was initially proposed that subclass IIIa contains only membrane-bound forms and IIIb only soluble ones. With the increasing availability of AC sequences, it became apparent that transmembrane domains were only one aspect of AC architectural cftr [24,25] which in fact did not correlate with the IIIa-IIIb division in bacteria, where each subclass contains both forms. Several studies have sought to catalogue the domain diversity of ACs (e.g. [26,27]) and derive from it information on general principles of regulation [28,29]. While the observed domain architectures did not lend themselves to a meaningful classification and their overall number kept growing with the number of sequenced genomes, it was found that the constituent AC domains could be grouped functionally into: (i) input domains, which receive direct stimulation (e.g. by the intracellular pH, by the binding of a ligand, by phosphorylation, or by light); (ii) transducer domains, which propagate signals from upstream input domains or other transducers; and (iii) the AC catalytic domain, which generates the second-messenger signal. Much of the observed architectural diversity can be explained by the modular recombination of input and transducer domains within this layout, arguably reflecting the physiological role of ACs as converters of a large diversity of stimuli into a uniform intracellular cAMP signal. As a general rule, input domains and transducers constitute the N-terminal part of the proteins, with the catalytic domain almost invariably located at the C-terminus. This closely parallels the arrangement in other families of signaling proteins, such as histidine kinases and diguanylate cyclases. All these proteins frequently form a coiled-coil backbone in the dimer, along which the intramolecular signal is propagated (see e.g. the structures of the light-activated AC bPAC [30], the chemoreceptor Tsr [31,32], and the histidine kinases VicK and NarQ [33,34]). Correspondingly, all input and transducer domains in these proteins are compatible with a coiled-coil structure, both for receiving and emitting conformational changes – an observation which readily explains their suitability for modular recombination into diverse architectures.
The phylogenetic distribution of class III ACs To get an up-to-date view of domain architectures, we extracted class III ACs from the Uniprot reference proteomes (release 2016_11, consisting of 4058 genomes from bacteria, 229 from archaea, and 793 from eukaryotes) in an iterative HMMer3 search [35], using the catalytic domain of Rv1625c from Mycobacterium tuberculosis as a query (Uniprot identifier P9WQ35). Bacterial class III ACs were predicted in practically all major lineages, although frequently not in every species thereof. Only eight bacterial clades containing three or more fully sequenced species in different genera had no AC predicted: Coriobacteria and Rubrobacteria of the phylum Actinobacteria; Erysipelotrichia, Tissierellia, and Negativicutes of the Firmicutes phylum; Elusimicrobia; Synergistia; and Thermodesulfobacteria. Almost two thirds of the predicted bacterial ACs belong to subclass IIIb (3949 proteins in our dataset), followed in decreasing abundance by IIIc/d (1633 proteins, of which 422 belong to the IIId branch), and IIIa (669 proteins). While most bacteria encode only one or a few ACs, some species possess many predicted AC genes, such as M. tuberculosis, which encodes a total of 15 ACs from all subclasses, several of which have been studied experimentally [36,37]. Even more ACs are found in Mycobacterium sp. 1164966.3, with 52 AC genes, or Turneriella parva, with 42 AC genes, for which, however, so far only bioinformatic evidence exists.