The analysis was based on lymphoblastoid lines, for which genotypes were available from the 1000 Genomes Pilot Project [21]. We focused on the CTCF-binding data from McDaniell et al. [16] (Gm12892, Gm19239, Gm19238 and Gm19240) and confirmed the results using an independently generated dataset (Gm12872, Gm12873 and Gm12874) [44] processed through quantile normalization using the R/Bioconductor package preprocessCore. The remaining two datasets from [16] (Gm12878 and Gm12891) were excluded due to highly inconsistent overall binding score distributions. Global major allele data were from [75,76]; assuming all reference alleles as major gave consistent results (not shown). Interaction models were plotted using the R package effects [77]. The sequences, PWM scores and ChIP binding signals for all TFBSs included in these analyses are listed in Additional files 5 (individuals from [16]) and 6 (individuals from [44]).
