In this study, we present the genome-wide profiling of FXR-binding sites in mouse liver chromatin. This ChIP-seq analysis revealed 1656 FXR genomic-binding sites with a high degree of confidence (P
< 1 × 10−5
, FDR < 5%). Most of the identified FXR-binding sites are located in distal intergenic regions (44%) or introns (32%), with fewer sites localizing to more proximal promoter regions (10%; B). The high distribution of binding sites within intergenic and intron regions are consistent with similar reports for other nuclear receptor transcription factors, including PPARγ (26
), estrogen receptor α (28
), and the androgen receptor (29
However, it should be emphasized that although only 10% of the binding sites were localized to within 2 kb 5′ of a TSS, this is significantly higher than expected based on random localization (B). Thus, even though the FXR-binding sites show a broad genome-wide distribution, the 10% indicates significant preference for promoter proximity as well. Additionally, even though most sites were located intragenically or within introns, they do localize close to known TSS ().
While there is evidence for agonist dependent changes in binding of some hormone receptors to DNA such as estrogen receptor (ER) in cultured cell models (30
), the influences of agonist binding on DNA occupancy for nuclear receptors in vivo
in general and for those where endogenous metabolically derived compounds function as agonists in particular is complicated and not clearly understood. In the course of our studies, we first compared the genome-wide association of FXR in livers of a control group fed normal chow versus a group fed chow supplemented with GW4064, currently the synthetic agonist of choice for FXR. While there was a mild induction of gene expression for a handful of known FXR target genes in this analysis, there was no statistically significant difference in genome-wide binding of FXR revealed by this comparison (data not shown). Because the GW4064 agonist is not very potent at stimulating endogenous target genes in vivo
, it is not clear if the lack of difference in DNA binding is meaningful. This is an important area of study that we will investigate in the future when more potent in vivo
FXR agonists become available.
Our analysis showed that 76% of the peaks contain a stringently identified IR-1 element (P
< 0.001; A). This discovery both confirms and extends the data from in vitro
DNA-binding site selection and the small number of individual gene analyses that have been reported (9
). Additionally, the Weblogo obtained from the position weight matrix emphasizes that the bases in one of the half-sites are highly preferred whereas there is relaxed flexibility and less preference observed for the second half site. A similar half-site asymmetry preference was also revealed in the Weblogo for PPARγ/RXR (26
) where a conserved DR-1 site showed a higher preference for bases in the 3′ half site, which is known to be specifically occupied by the RXR monomer. An explanation for this was likely revealed by the crystal structure for the PPARγ/RXR dimer bound to a DR-1 site (31
). The structure revealed that the zinc finger region in the RXR DNA-binding domain makes more half site-specific base contacts compared to the zinc-finger region of PPARγ.
Because the FXR recognition site is a palindrome, the rotational symmetry makes it impossible to assign a specific monomer to each half-site without any more information. However, based on the information above, it is likely that the more highly conserved half site of the IR-1 is bound by RXR.
It is interesting that there are over 1.7 million sites in the mouse genome that match our IR-1 PWM with P < 0.001. The fact that only 1656 (0.09%) were occupied by FXR in our analysis indicates that other local genomic features such as hepatic nucleosome positioning and epigenetic markers that alter chromatin architecture and genomic access influence FXR site occupancy. Additionally, co-occupancy by neighboring DNA-binding factors like LRH-1 likely play a significant role in addition to the primary sequence of the IR-1 motif in defining where FXR is localized in hepatic chromatin.
We compared the list of genes within 20 kb of a FXR peak to a list of genes that were rank-ordered by significance for differential mRNA expression in primary hepatocytes infected with an adenovirus that expresses a constitutively active FXRα–VP16 hybrid protein relative to a control adenovirus. This gene set enrichment analysis (GSEA) was displayed by a KS plot (23
) which tests for how well the two data sets correlate with each other. This analysis showed a high degree of correlation between FXR binding and FXR-dependent gene activation (B; P
). Thus, the FXR-binding peaks likely represent functional FXR response elements. It should be noted that the FXR–VP16 fusion protein would activate through genomic sites where wild-type FXR might repress gene expression and there have been reports that FXR binding may repress gene expression (32
). Thus, our GSEA provides a good correlation between FXR binding by ChIP-seq and target gene identification but it does not allow us to differentiate between genes that are normally activated or repressed directly through FXR response elements.
After first masking the IR-1 sites we repeated the de novo
motif search around the peak summits (±150 bp) to search for putative enriched FXR co-regulatory DNA-binding partners. This analysis revealed that an additional nuclear half site, 5′-AGGTCA-3′, was present close to 71% of the IR-1 containing FXR peaks (). It is unlikely that FXR interacts directly with this additional half site because this would have been revealed as a peak summit when the sequence reads were mapped to the genome. We also analyzed directly whether the half-sites might correspond to ‘weak’ IR-1 sites that fell below our statistical threshold. This analysis revealed that at least 80% of the identified half-sites are true half sites (Supplementary Figure S5
). It should be noted that FXR has been shown to possibly interact with DNA as a monomer (33
) and we cannot rule out with certainty that some of the half-sites might bind a monomeric form of FXR.
Because monomeric nuclear receptors bind to isolated half-sites and LRH-1 is a liver enriched monomeric nuclear receptor, we proposed that LRH-1 would be a good candidate for an FXR co-regulatory protein. Because most FXR-binding sites were localized to introns or intergenic regions it would be difficult to construct reporter genes that retain the native spacing for the proximal promoter together with the intronic/intergenic FXR/RXR response element. Therefore, to analyze LRH-1 as a putative FXR co-regulatory partner we performed gene specific ChIP analysis of 12 peaks predicted to contain LRH-1-binding sites (C) and we chose four genes where the FXR/RXR and associated extra half-sites were located within the proximal promoter region for promoter activation assays. In this analysis, all four promoter–reporters confirmed that FXR/RXR activation was significantly enhanced by the addition of an LRH-1 expression vector in the transfection assay (). We also showed that FXR and LRH-1 associate with each other directly by co-immunoprecipitation (). These three approaches strongly support our hypothesis that FXR and LRH-1 function together to activate hepatic gene expression.
It is interesting to note that there are additional monomeric nuclear receptors that are expressed in the liver such as the reverbs and RORs which are regulated by heme and sterol agonists respectively (34
) and SF-1 which is very similar to LRH-1 (36
). Reverbs and RORs are expressed in reciprocal diurnal patterns in the liver. It will be interesting to analyze these additional monomeric nuclear receptors in future studies for their potential roles as FXR co-regulatory factors that might be associated with a diurnally regulated pattern of FXR activity.
In a GO analysis, the broad category of lipid and fatty acid metabolism was the most significant gene cluster linked to the FXR peak associated genes as expected. However, genes of glycolysis also showed a significant enrichment as well. This is interesting because FXR has been shown to be involved in modulating glucose metabolism associatead with diabetes in mice (3–5
). In addition, a number of other broad-based gene clusters related to metabolism, transport and signaling were also significantly enriched. These latter results suggest that FXR may have a much wider role in regulating cellular metabolism than has been proposed to date. Indeed, identification of new FXR targets within these categories may explain the pleiotropic effects on metabolism and cellular physiology noted in both animal studies and patients with bile acid disorders (1
While we were completing the analyses for our study, a comparative evaluation of the genome-wide pattern of FXR binding to hepatic and intestinal chromatin was reported (38
). Overall, the binding results are very comparable but the two approaches have some differences in that the synthetic FXR agonist GW4064 was added a few hours before sacrifice by Thomas et al and the sequence mapping and analysis were performed by different methods. Interestingly, Thomas et al also identified an IR-1 element with a co-enriched additional nuclear receptor half site close to the FXR peaks. As mentioned above, we did not observe a consistent difference in FXR binding by the addition of the GW4064 agonist. Thus, we were not surprised that overall the results are similar to ours. In addition to defining the genomic sites for FXR binding, our study goes further and also provides functional evidence that FXR is likely to affect expression of the genes associated with the peaks. We also provide evidence that LRH-1 is an important monomeric nuclear receptor partner for FXR that binds to the co-enriched nuclear receptor half site to co-activate gene expression.