Definition and prediction of the full range of transcription factor binding sites--the hepatocyte nuclear factor 1 dimeric site

In animals, transcription factor binding sites are hard to recognize because of their extensive variation. We therefore characterized the general relationship between a specific protein-binding site and its DNA sequence and used this relationship to generate a predictive algorithm for searching other DNA sequences. The experimental process was defined by studying hepatocyte nuclear factor 1 (HNF1), which binds DNA as a dimer on two inverted-repeat 7-bp half sites separated by one base. The binding model was based on the equivalence of the two half sites, which was confirmed in examples where specific modified sites were compared. Binding competition analysis was used to determine the effects of substitution of all four bases at each position in the half site. From these data, a weighted half-site matrix was generated and the full site was evaluated as the sum of two half-site scores. This process accurately predicted even weak binding sites that were significantly different from the consensus sequence. The predictions also showed a direct correlation with measured protein binding.