The developmental control of expression of the human epsilon-globin gene appears to be mediated, at least in part, by a transcriptional silencer in the DNA 5' to the cap site of this gene. We have used site-directed mutagenesis and DNA-protein binding assays to define the active motifs of this epsilon-globin silencer. DNase I foot-printing of the silencer region with K562 cell nuclear extracts defined a sequence, which we designate as the epsilon-globin silencer motif or epsilon GSM (epsilon -278 to -258 base pairs (bp)) containing a region (epsilon -270 to -258) with 90% homology to the yeast mating type silencer, ABF-1 (autonomous replicating sequence binding factor one) and which also overlaps at (epsilon -269 to -262) with the human YY1 consensus sequence, an element which mediates transcription repression and activation of viral, mouse, and human genes. The DNase I footprint extended 5' in the silencer region to include an inverted repeat of a six-nucleotide motif (epsilon -267 to -278 bp) which shares 5 of 6 bases with the GATA-1 consensus sequence. In gel mobility shift assays, two specific proteins (A and B) in nuclear extracts from erythroleukemia K562 cells bound to the DNase I-footprinted region. Protein B, associated with epsilon-globin silencer activity in vitro, required an intact epsilon GSM sequence for binding. Mutation of 5 bases within the epsilon GSM in an epsilon-globin promoter-containing fragment extending upstream to 1400 bp in transient transfection assays increased activity by 3.0-fold compared with the native sequence, suggesting that the silencer activity was mediated by the epsilon GSM sequence. We found that protein A could be displaced by a competitor containing the GATA-1 consensus sequence, suggesting that protein A is a GATA-like protein. The region from -267 to -271 within the epsilon GSM and GATA-1 homology region was important for binding of both proteins A and B. These data suggest that protein binding to the epsilon GSM and GATA motifs mediate the negative effect of the silencer on transcription, possibly via direct competition for binding to this DNA region. Recombinant yeast ABF-1 and human YY1 bound to the epsilon GSM. Mutating three bases (epsilon -259, -262, -264) in the epsilon GSM decreased the binding affinity of protein B and recombinant human YY1 but increased the binding affinity of recombinant yeast ABF-1. Furthermore, competitor containing the YY1 consensus sequence competed for protein B binding, whereas competitor containing a perfect yeast ABF-1 consensus sequence did not.(ABSTRACT TRUNCATED AT 400 WORDS)