Based on the in vitro data on the role of the second zinc finger in selective ARE binding, and the fact that this receptor fragment is encoded by a separate exon in the AR as well as in the GR gene, a transgenic model was developed in which this exon in the AR gene was swapped by that of the GR gene. The resulting model, called SPARKI for ‘SPecificity affecting AR Knock In’ expresses an AR that still binds classical AREs with high affinity but has lost high affinity for selective AREs (Schauwaers et al. 2007) . In effect, this model can be considered a knockout of selective AREs . These mice have smaller reproductive organs and reduced fertility, while no differences are observed in other androgen target tissues like bone, muscle, kidney or lacrimal glands. Therefore, it seems that selective AREs are not involved in the anabolic effects of androgens, but have a specific role in reproduction.

The reduced fertility observed in SPARKI is mainly explained at two sites: in the testis, the number of Sertoli cells is reduced and the spermatogenic process seems to be affected at the second meiotic division; in the epididymis, the sperm maturation is impaired and this correlates with the reduced expression of a subset of the androgen-regulated genes in this tissue. Several of these genes have a known role in sperm maturation, and selective AREs were identified in two of them which is proof for the fact that the second zinc finger of the AR has a crucial role in the recognition of selective AREs (Kerkhofs et al. 2012) .

Several of the AREs described in AR-binding segments found in human prostate cancer cell lines are selective AREs (Denayer et al. 2010; Wang et al. 2007) . The fact that the prostate of SPARKI mice is smaller (Schauwaers et al. 2007) indicates that selective AREs have a role in the development of normal prostates, but it still is unclear whether this type of AREs is also involved in the etiology or evolution of prostate cancer.

As discussed above, the second zinc finger plays an essential role in the binding of AR and the non-binding of GR to selective AREs. This suggests that the discrimination between the direct repeat motifs and the classical palindromes is not determined by a difference in monomeric sequence recognition, but rather by a difference in dimerization of the receptor DBDs (Helsen et al. 2012b; Verrijdt et al. 2006) . Compared to that of the GR, the AR second zinc finger differs at four positions. Two of them have been demonstrated to be involved in the stronger dimerization of AR compared to GR by creating additional interactions that stabilize the AR dimer interface (Shaffer et al. 2004) , but in functional assays, swapping these residues between AR and GR DBDs had no effect on their selectivity (Verrijdt et al. 2006) . However, removal of the carboxy-terminal part of the CTE containing these two residues abolishes AR binding to selective AREs but has no effect on AR-DBD interaction with classical AREs (Schoenmakers et al. 2000) . Deletions and point mutations in the full size receptor have similar effects on DNA binding (Haelens et al. 2007; Tanner et al. 2010) . Thus, the stronger dimerization function in the AR-DBD seems to allow the AR to dimerize on response elements that are divergent from the ARE consensus, e.g. organized as direct repeats instead of the canonical inverted repeats.

A recent study addressed the role of the DBD in ensuring specificity of AR binding to chromatin in vivo . In that study, AR binding events were compared between wild type and SPARKI mice (Sahu et al. 2014) . As discussed above, SPARKI mice express an AR that still binds classical AREs but has lost high affinity for selective AREs. Motif search analysis indicated that, as expected, an inverted palindromic repeat resembling the ARE consensus is highly enriched among the AR binding sites preferred by SPARKI AR, with a highly conserved threonine at position 12 (Sahu et al. 2014). On the other hand, analysis of the binding sites preferred by wild type AR uncovered a different element in which the 5’ hexamer is almost identical to the 5′-AGAACA-3′ consensus but followed by a second hexamer with weak sequence conservation (Sahu et al. 2014). Interestingly, the very highly conserved thymidine at position 12 present at the SPARKI AR-preferred sites is dispensable among the elements for wild type AR binding (Sahu et al. 2014). These data indicate that AR selectivity is achieved by relaxed rather than increased element stringency at chromatin binding sites which, in turn, attenuates binding of other steroid receptors to these elements. In line with this, genome-wide ChIP-seq experiments on several cell lines and tissues have shown that response elements for GR contain a highly conserved thymidine at position 12 (Grontved et al. 2013; John et al. 2011) . Altogether, these findings support the abovementioned model that the stronger dimerization of AR-DBD allows binding of a second AR molecule on 3′ hexamers that show less resemblance with the 5′-AGAACA-3′ consensus, in contrast to the GR. Interestingly, the PR also seems to have more relaxed DNA binding selectivity compared to GR (Denayer et al. 2010) .

Following DNA recognition, conformational changes take place in the second zinc finger that favor receptor dimerization (Freedman 1992) . As discussed above, multiple AREs are often found in the regulatory regions of androgen-responsive genes. The study of their respective role has led to the concept of cooperativity due to allosterically-induced receptor changes (Reid et al. 2001; Scheller et al. 1998) . Indeed, studies with the Slp gene enhancer show that increased AR activity is due to the interplay of several weak AREs, and that N/C-interactions contribute to this effect (Scheller et al. 1998). Cooperative interactions of multiple AREs have also been reported in the Prostate-Specific Antigen (PSA) and probasin gene promotors (Reid et al. 2001). The demonstration of cooperativity between response elements suggests the existence of binding proteins able to recognize synergy control motifs and to bridge receptor dimers bound to adjacent DNA elements. A specific amino acid motif has been postulated to be responsible for the synergy control mechanism in the GR (Iniguez-Lluhi et al. 2000) . Later on, they were identified as sumoylation sites. For the AR, the sumoylation sites in the AR-NTD have been proposed as synergy control motifs (Callewaert et al. 2004; Poukka et al. 2000) , but the exact nature of the bridging factor is still unknown. How exactly the sumoylation controls synergism is unclear, but for the AR it seems less pronounced on selective AREs (Callewaert et al. 2004), indicating a role of DNA binding.

Many AR mutations have been found in patients with complete or partial androgen insensitivity as well as in biopsies of castration-resistant metastatic prostate cancer (Tilley et al. 1996) . Most of these mutations affect the function of the domain they are situated in. However, some DBD mutations do not affect DNA binding and some LBD mutations do not affect ligand binding. Surprisingly, a DBD mutation can affect ligand binding and, vice versa, an LBD mutation can affect DNA binding (Helsen et al. 2012a) . The mutated residues are situated at the surface of these domains pointing away from the DNA or the ligand. Based on modeling of the AR domains on the DBD-LBD coordinates of the crystal structure of the PPARγ-RXRα, as well as on docking AR-DBD against AR-LBD, a functional interface has been proposed between these domains, allowing signals from the DNA to reach the LBD and signals from the ligand to reach the DBD (Helsen et al. 2012a).

The NTD is structurally important for AR-dependent gene expression by mediating multiple protein-protein interactions (McEwan 2004) . In contrast to the LBD and the DBD, the NTD has little stable secondary structure (Reid et al. 2002) . However, in vitro experiments with a purified AR-NTD-DBD polypeptide show that binding to an ARE results in changes in the intrinsic fluorescence emission spectrum of the NTD, indicating a conformational change in this domain upon DNA binding (Brodie et al. 2005) . Moreover, protease-resistance of the polypeptide is increased in the presence of the DNA element, adding evidence for a conformational change in the NTD upon binding to the ARE (Brodie et al. 2005). Taken together, these findings indicate that DNA binding results in an allosterically-induced conformational change within the NTD, suggesting a communication between the DBD and the NTD.

Several features of the AR have been studied by monitoring the effect of point mutations on the functionality of the receptor in reporter assays. The effects of disrupting the N/C interaction, sumoylation of the NTD and the role of the polyglutamine stretch in the control of the overall activity of the human AR have initially all been described on reporter genes controlled by classical AREs (Claessens et al. 2008) . However, the same analyses performed with reporter genes based on selective AREs gave much less pronounced or no effects. Indeed, deletion of the FQNLF motif in the NTD, which blunts the N/C-interaction, strongly reduced the AR capacity to transactivate through classical AREs but did not affect its activity on selective elements (Callewaert et al. 2003b) . Mutation of lysine 385 in the NTD, blocking sumoylation at that site, clearly affects the cooperativity of the receptor on multiple classical AREs. However, when selective AREs where tested, the lysine 385 mutation did not increase the androgen response (Callewaert et al. 2004). Deletion of the polyglutamine tract results in an increase in the transactivation through classical AREs, whereas this effect is not observed on selective elements (Callewaert et al. 2003a). Finally, modulation of AR activity by coactivators, AR sensitivity to proteolysis, and the effect of mutations in the AR dimerization interface of the DBD also vary with the nature of the ARE, the observed effect being different when using classical AREs compared to selective ones (Geserick et al. 2003) .