We have used in vitro selection to isolate adenosine/ATP-binding DNA s
equences from a pool of approximate to 2 x 10(14) different random-seq
uence single-stranded DNA molecules. One of these aptamers has been ch
aracterized and binds adenosine in solution with a dissociation consta
nt of 6 +/- 3 mu M. Experiments with ATP analogs indicate that functio
nal groups on both the base and the sugar of ATP are involved in the l
igand/aptamer interaction. The binding domain of this aptamer was loca
lized to a 42 base sequence by deletion analysis. A pool of mutagenize
d versions of this sequence was then synthesized and screened for func
tional adenosine binding sequences; comparison of the selected variant
s revealed two highly conserved guanosine-rich regions, two invariant
adenosine residues, and two regions of predominantly Watson-Crick cova
riation. This data led us to propose a model of the ATP-binding DNA st
ructure which is based on a stable framework composed of two stacked G
-quartets. The two highly conserved adenosine residues may stack betwe
en the top G-quartet and the two short stems, forming a pocket in whic
h the adenosine or ATP ligand binds. Site-directed mutagenesis, base a
nalog substitution studies, and the design of highly divergent but fun
ctional sequences provide support for this model.