The AR is a transcription factor and a primary driver of prostate cancer. Understanding the key determinants of its transcriptional specificity remains a critical issue. By integrating analysis of DNase-seq data with AR ChIP-seq and mRNA-seq, we showed that AR activation induced genome-wide changes in chromatin structure that were associated with AR binding and transcriptional response. We also uncovered multiple modes of AR utilization of its DNA recognition motif. Although a subset of AR binding occurs in qualitatively poised chromatin exhibiting nucleosome depletion prior to hormone treatment, we demonstrated that AR binding is consistently associated with a quantitatively significant increase in DNase-seq signal, suggesting stabilization of nucleosome depletion and chromatin remodeling.
Several prior reports also support AR-induced chromatin remodeling [16
], including a very recently published study utilizing DNase-seq by He et al.
]. Our data combined with these prior reports suggest a different model for nuclear receptor interaction with the genome than that proposed by John et al.
for the GR [29
], where almost all GR binding occurred in poised DHS sites. The AR and GR, though possessing similar DNA-response elements, seem to display fundamentally different interactions with chromatin and DNA. Our data represent a significant additional resource for understanding the association between chromatin accessibility and nuclear receptor function for several reasons. First, our DNase-seq experiments were sequenced very deeply (approximately 130 million reads), which is similar to the depth of sequencing with which John et al.
observed GR binding to poised chromatin. Second, we utilized a different AR ligand (R1881) and time point of 12 hours as compared with 4 hours by He et al.
and 1 hour by John et al.
Similar to He et al.
, who also utilized a quantitative measure of change in DNase-seq signal, we observed that less than half of AR binding targets poised chromatin and we were also able to associate AR-induced chromatin remodeling with AR-induced transcriptional changes, suggesting that the mechanism of chromatin remodeling and its phenotypically relevant association with differential transcription requires longer periods of receptor activation. Importantly, we used a different statistical measurement of quantitative change in DNase-seq signal to reach the same result and conclusion. In our study and those by He et al.
and John et al.
, we note that the degree of nuclear receptor binding within regions of poised chromatin decreases with increased hormone treatment time (37% in He et al.
, 88% in John et al.
and 20% to 30% in our study). Although this observation is confounded by differences in receptor, receptor ligand, sequencing depth and DNase-seq protocol among the mentioned studies, these data suggest that more extensive comparative analyses over a full time course of ligand stimulation of both AR and GR are needed to fully understand the similarities and differences of different hormone receptors with respect to their interaction with chromatin.
While the majority of high confidence AR binding occurred in regions sensitive to DNase I cleavage, a substantial proportion of AR binding events occurred in regions of low DNase-seq signal. It is possible that inconsistent and/or intermittent nucleosome depletion at these genomic regions decreases DNA accessibility and limits detection by our assay; this attribute of nucleosome depletion appears to be associated with a slightly different AR motif. Consistently, we also found that AR binding (as measured by AR ChIP-seq signal intensity) is significantly lower in non-DHS regions than in DHS regions. Thus, it is plausible that regions that are identified with weaker AR binding and lower DNase-seq signal may experience a dynamic equilibrium of nucleosome and nuclear receptor binding, as has been previously proposed [14
]. Loci with reduced DNase I cleavage and AR binding could reflect low levels of AR binding at linker regions of non-displaced nucleosomes or residual nucleosome occupancy, limiting accessibility to DNase I cleavage in the cell population.
AR footprinting analysis further revealed the complexity of the AR-DNA interaction. The aggregate DNase-seq signal around AR motifs demonstrated a relatively weak but consistent pattern of protection that corresponds to the expected binding pattern, consistent with other DNase I footprinting studies [26
]. In addition, we found three distinct patterns of DNase I protection significantly associated with LNCaP cells treated with androgen. The footprint patterns suggest that either AR binds to the full AR consensus motif as a dimer (cluster 3) or only binds to half of the motif (clusters 1 and 2). We also cannot exclude the possibility that clusters 1 and 2 represent AR dimers with only one AR molecule binding to half of the consensus motif. AR binding to either half site did not appear to be random, as evidenced by reproducible detection of distinct clusters. In other words, random binding to either half site in a population of cells would not show consistent half-site protection. Intriguingly, clusters 1 and 2 may provide the first in vivo
and endogenous evidence of functional AR monomers that have been suggested to exist as a stable subpopulation of AR molecules [45
]. Only the AR binding sites that displayed a full-site dimer protection pattern (cluster 3) were enriched for the NF1C motif, which is a known co-factor of AR. Therefore, there appears to be multiple modes that AR binds to canonical DNA motifs in vivo
, and these modes are associated with different co-factors. These observations are consistent with a recently proposed model of a transient interaction between nuclear receptors such as the AR and DNA rather than a stronger and more stable AR-DNA interaction [46
]. Our analysis also provides the first evidence of substructure within a nuclear receptor footprint
The dynamics of AR-DNA binding are likely impacted by additional co-factors that may facilitate AR binding directly or indirectly. Distal regulatory elements identified by DNase-seq displayed an enrichment of SP1 and E2A/TCF3 motifs within DHS specifically accessible in LNCaP cells compared with 113 independent cell lines. TCF3, a basic helix loop helix factor involved in Wnt/β-catenin signaling [47
], represents a new putative co-factor for the AR that warrants further investigation to understand its role in AR-mediated chromatin dynamics as well as the crosstalk between AR and β-catenin signaling. SP1 is especially interesting both because its motif was enriched in ΔDNase regions and also in light of a recent report that identified SP1 as necessary for the expression of a variety of chromatin modifying enzymes, such as the histone deacetylases 1 to 4 in LNCaP cells [49
]. Additionally, small molecule inhibitors of histone deacetylases have been shown to decrease the growth rate of AR-positive prostate cancer cell lines [50
] and disrupt AR-induced expression of its target genes [52
]. Our relative enrichment score of less than one for the SP1 motif and an observation that SP1 motifs often co-localize with AR binding suggest complexity in the interplay between SP1 and the AR.