Class III adenylyl cyclases usually possess six highly conserved catalytic residues. Deviations in these canonical amino acids are observed in several putative adenylyl cyclase genes as apparent in several bacterial genomes. This suggests that a variety of catalytic mechanisms may actually exist. The gene Rv0386 from Mycobacterium tuberculosis codes for an adenylyl cyclase catalytic domain fused to an AAA-ATPase and a helix-turn-helix DNA-binding domain. In Rv0386, the standard substrate, adenine-defining lysine-aspartate couple is replaced by glutamine-asparagine. The recombinant adenylyl cyclase domain was active with a V(max) of 8 nmol cAMP.mg(-1).min(-1). Unusual for adenylyl cyclases, Rv0386 displayed 20% guanylyl cyclase side-activity with GTP as a substrate. Mutation of the glutamine-asparagine pair either to alanine residues or to the canonical lysine-aspartate consensus abolished activity. This argues for a novel mechanism of substrate selection which depends on two non-canonical residues. Data from individual and coordinated point mutations suggest a model for purine definition based on an amide switch related to that previously identified in cyclic nucleotide phosphodiesterases.