The diversity of Type-II toxin–antitoxin (TA) systems in bacterial genomes requires tightly controlled interaction specificity to ensure protection of the cell, and potentially to limit cross-talk between toxin–antitoxin pairs of the same family of TA systems. Further, there is a redundant use of toxin folds for different cellular targets and complexation with different classes of antitoxins, increasing the apparent requirement for the insulation of interactions. The presence of Type II TA systems has remained enigmatic with respect to potential benefits imparted to the host cells. In some cases, they play clear roles in survival associated with unfavorable growth conditions. More generally, they can also serve as a “cure” against acquisition of highly similar TA systems such as those found on plasmids or invading genetic elements that frequently carry virulence and resistance genes. The latter model is predicated on the ability of these highly specific cognate antitoxin–toxin interactions to form cross-reactions between chromosomal antitoxins and invading toxins. This review summarizes advances in the Type II TA system models with an emphasis on antitoxin cross-reactivity, including with invading genetic elements and cases where toxin proteins share a common fold yet interact with different families of antitoxins.

Toxin-antitoxin (TA) systems pair a protein toxin with a protein antitoxin through an extensive interface, typically spanning more than 1200 Å2, accounting for more than 20% of the available surface of each protein. These pairings are highly co-evolved, such that cognate interactions are strictly maintained even when multiple types of the same TA family are present in the same cell. Physiological functions of chromosomal TA systems remain unclear, but an "anti-addiction" model posits a chromosomal antitoxin cross-reacting and neutralizing an exogenouslyderived toxin, such as encoded by a phage. While plausible, specific examples are scarce. Investigations of the potential for cross-interactions are limited by the relatively small pool of available structures as compared to the varied complexity of the interacting interfaces. We would like to apply advances in prediction models, especially AlphaFold2, to broaden the pool of available highly reliable interfaces. We initiated this using seven available crystal structures of the ParDE type of TA system and comparing these to AlphaFold2 predictions. Remarkable, the predictions match exceptionally well despite the divergent sequence identity between complexes. This highlights how structural conservation can dominate predictions to overcome sequence diversity. This workflow provides a viable means of larger scale Type II TA system protein-protein interface predictions, facilitating studies of crossreactivity and other sequence co-evolution models.