AR binding displays three distinct modes of androgen receptor-DNA interaction that are specific to ligand-activated androgen receptor. (a) K-means clustering of LNCaP-induced DNase-seq signal into three consistent clusters within AR binding sites. (b) K-means clustering (k = 3) was repeated 100 times on both LNCaP and LNCaP-induced DNase-seq data around all DHS sites with a full-site canonical AR motif. Shown is the distribution of correlations between cluster centers for each run. The asterisk denotes the statistically significant difference between the correlation distributions (Mann-Whitney P < 2.2e-16). (c) Motif analysis of the entire 25 bp span up- and downstream from AR motif matches for each cluster. MEME motifs identified within this interval (E < 0.1, E-value shown below logo) are shown in logo format. Motifs that significantly match a known motif (E < 0.05, by TomTom) are marked with an asterisk. The name of the most significant match according to TomTom is indicated next to the logo, as is the percentage of regions that contain the enriched motif. For matches resembling FOX family factors, we note that these motifs are very similar to each other. DNase-seq signal is shown as the aggregate signal from all cluster members with the dotted lines marking the location of the AR motif within the plot. AR: androgen receptor; bp: base pairs; DHS: DNase I hypersensitive; DNase-seq: DNase I hypersensitivity analysis coupled with high-throughput sequencing; FOX: Forkhead box; NF1C: nuclear factor 1 C-type.