DNA gyrase, a type II topoisomerase, regulates DNA topology by creating a double-stranded break in one DNA duplex and transporting another DNA duplex [T-DNA (transported DNA)] through this break. The ATPase domains dimerize, in the presence of ATP, to trap the T-DNA segment. Hydrolysis of only one of the two ATPs, and release of the resulting Pi, is rate-limiting in DNA strand passage. A long unresolved puzzle is how the non-hydrolysable ATP analogue AMP-PNP (adenosine 5'-[β,γ-imido]triphosphate) can catalyse one round of DNA strand passage without Pi release. In the present paper we discuss two crystal structures of the Mycobacterium tuberculosis DNA gyrase ATPase domain: one complexed with AMP-PCP (adenosine 5'-[β,γ-methylene]triphosphate) was unexpectedly monomeric, the other, an AMP-PNP complex, crystallized as a dimer. In the AMP-PNP structure, the unprotonated nitrogen (P-N=P imino) accepts hydrogen bonds from a well-ordered 'ATP lid', which is known to be required for dimerization. The equivalent CH2 group, in AMP-PCP, cannot accept hydrogen bonds, leaving the 'ATP lid' region disordered. Further analysis suggested that AMP-PNP can be converted from the imino (P-N=P) form into the imido form (P-NH-P) during the catalytic cycle. A main-chain NH is proposed to move to either protonate AMP-P-N=P to AMP-P-NH-P, or to protonate ATP to initiate ATP hydrolysis. This suggests a novel dissociative mechanism for ATP hydrolysis that could be applicable not only to GHKL phosphotransferases, but also to unrelated ATPases and GTPases such as Ras. On the basis of the domain orientation in our AMP-PCP structure we propose a mechanochemical scheme to explain how ATP hydrolysis is coupled to domain motion.