(A) Phylogenetic relationships of modern-day vertebrate SRs are shown, with ancestral proteins AncSR1 and AncSR2 marked. Each protein's preferred response element (RE) is shown: estrogen RE (ERE; purple) or steroid REs (SRE1, SRE2; light and dark green, respectively), with the half-site sequence of each. Gray box indicates evolutionary interval in which SRE specificity evolved (McKeown et al., 2014). (B) Interface of steroid hormone receptor DNA-binding domains (DBDs) with their preferred RE half-sites. X-ray crystal structures of AncSR1 with ERE (left, 4OLN) and AncSR2 with SRE1 (right, 4OOR). The RH is shown as a colored cylinder; sticks, side chains that differ between AncSR1 and AncSR2. Colored surface, nucleotides that differ between REs. (C) Close-up of protein-DNA interface for AncSR1:ERE (left) and AncSR2:SRE1 (right). In the DBD, the RH is shown as ribbon, with side chains of variable amino acids shown as sticks and Cα as spheres. In the RE, variable nucleotides are shown as sticks with backbone as cartoon. Atoms are colored by element. Dashed lines, polar interactions between variable amino acids and nucleotides. (D) Historical RH replacements change AncSR1's affinity for REs. Binding energies of AncSR1 (left) and AncSR1+RH replacements were measured using fluorescence polarization to single half-site REs containing all possible combinations of nucleotides at the sites that vary between ERE and SREs. ERE, SRE1 and SRE2 are highlighted in purple, light green and dark green, respectively. ΔGdissociation is the free energy of dissociation, calculated from dissociation constant (Kd). Technical replicates (dots) with mean and SEM (lines) are shown. (E) RH replacements change the genetic determinants of affinity within the RE. Energy logos for AncSR1 (left) and AncSR1+RH (right) show the effects of nucleotide states on binding energy relative to the average across all REs tested; states with ΔΔGdissociation > 0 are associated with higher affinity binding. Main effects of nucleotides at variable positions 3 and 4 are shown, as is the epistatic effect of nucleotide combinations, defined as the excess effect beyond that predicted under additivity. The height of each state indicates the magnitude of their effect on binding energy; states are ranked from top to bottom by the magnitude of its effect. Each column's width shows the portion of variation in binding energy attributable to the effects of states in that column, calculated as the increase in the adjusted R2 when terms corresponding to those states are added to a linear regression model and fit to the experimental binding data. *, significant improvement in model fit (likelihood ratio test, p < 0.05 Bonferroni-corrected). For complete explanation of linear modeling approach, see ‘Materials and methods’.