Protein engineering to alter recognition underlying ligand binding and activity has enormous potential. Here, ligand binding for E. coli phosphoenolpyruvate carboxykinase (PEPCK), which converts oxaloacetate into CO2 and phosphoenolpyruvate as the first committed step in gluconeogenesis, was engineered to accommodate alternative ligands as an exemplary system with structural information. From our identification of bicarbonate binding in the PEPCK active site at the supposed CO2 binding site, we probed binding of nonnative ligands with three oxygen atoms arranged to resemble bicarbonate geometry. Crystal structures of PEPCK and point mutants with bound nonnative ligands thiosulfate and methanesulfonate along with strained ATP plus reoriented oxaloacetate intermediates and unexpected bicarbonate were solved and analyzed. The mutations successfully altered the bound ligand position and orientation, as well as its specificity: mutated PEPCKs bound either thiosulfate or methanesulfonate, but never both. Computational calculations predicted a methanesulfonate binding mutant and revealed that release of active site ordered solvent exerts a strong influence on ligand binding. Besides nonnative ligand binding, one mutant altered the Mn2+ coordination sphere: instead of the canonical octahedral ligand arrangement, the mutant in question only had a five-coordinate arrangement. From this work, critical features of ligand binding, position, and metal ion co-factor geometry required for all downstream events can be engineered with small numbers of mutations to provide insights into fundamental underpinnings of protein-ligand recognition. Through structural and computational knowledge, the combination of designed and random mutations aids robust design of predetermined changes to ligand binding and activity in order to engineer protein function.