Glycinamide ribonucleotide transformylase (GAR Tfase) has been the target of anti-neoplastic intervention for almost two decades. Here, we use a structure-based approach to design a novel folate analogue, 10-(trifluoroacetyl)-5,10-dideazaacyclic-5,6,7,8-tetrahydrofolic acid (10-CF(3)CO-DDACTHF, 1), which specifically inhibits recombinant human GAR Tfase (K(i) = 15 nM), but is inactive (K(i) > 100 microM) against other folate-dependent enzymes that have been examined. Moreover, compound 1 is a potent inhibitor of tumor cell proliferation (IC(50) = 16 nM, CCRF-CEM), which represents a 10-fold improvement over Lometrexol, a GAR Tfase inhibitor that has been in clinical trials. Thus, this folate analogue 1 is among the most potent and selective inhibitors known toward GAR Tfase. Contributing to its efficacious activity, compound 1 is effectively transported into the cell by the reduced folate carrier and intracellularly sequestered by polyglutamation. The crystal structure of human GAR Tfase with folate analogue 1 at 1.98 A resolution represents the first structure of any GAR Tfase to be determined with a cofactor or cofactor analogue without the presence of substrate. The folate-binding loop of residues 141-146, which is highly flexible in both Escherichia coli and unliganded human GAR Tfase structures, becomes highly ordered upon binding 1 in the folate-binding site. Computational docking of the natural cofactor into this and other apo or complexed structures provides a rational basis for modeling how the natural cofactor 10-formyltetrahydrofolic acid interacts with GAR Tfase, and suggests that this folate analogue-bound conformation represents the best template to date for inhibitor design.