DNA binding alters ARv7 dimer interactions

Androgen receptor (AR) signaling is important at all stages of prostate cancer (PCa) growth and development (Helsen et al., 2014; Lonergan and Tindall, 2011). Given this critical role, metastatic and recurrent PCa is commonly treated by androgen deprivation therapy (ADT) (Acar et al., 2013; Daneshmand and Ahmadi, 2014). Yet, although treatment is initially effective, the cancer almost always develops resistance and progresses to a castration-resistant state (Karantanos et al., 2013; Wadosky and Koochekpour, 2016; Zong and Goldstein, 2013). Common mechanisms of resistance to ADT include AR overexpression, altered co-activator recruitment, AR point mutations, epigenetic modifications and the expression of AR splice variants (Chen et al., 2004, 2008; Karantanos et al., 2013; Metzger et al., 2005; Saraç et al., 2020; Sun et al., 2010; Xu et al., 2012). These splice variants generate truncated AR isoforms that completely or partially lack the ligand-binding domain (LBD) (Hu et al., 2009; Uo et al., 2017). These constitutively active AR variants no longer require androgen to initiate transcription and are therefore intrinsically resistant to ADT. The most commonly expressed AR variant is ARv7 (Hörnberg et al., 2011; Qu et al., 2015; Sharp et al., 2019; Zhang et al., 2020), consisting of exons 1-3, and a cryptic exon (CE3) (Hu et al., 2009). Lacking an LBD, ARv7 does not require androgens for nuclear localization and can induce gene expression in the absence of androgens (Hu et al., 2009, 2012; Li et al., 2013; Yu et al., 2014). ARv7 is expressed at very low levels in primary PCa (Chen et al., 2018a; Li et al., 2018) and then exponentially increases when the patient develops castration-resistant prostate cancer (CRPC) (Sharp et al., 2019; Welti et al., 2016; Zhu et al., 2018). Various clinical studies have shown that ARv7 expression in circulating tumor cells correlates with resistance to abiraterone and enzalutamide in CRPC (Antonarakis et al., 2014; Okegawa et al., 2018). Taken together, these and other findings demonstrate that ARv7 plays a significant role in ADT resistance. However, AR splice variants are not found in isolation. In clinical CRPC samples, the median transcript ratio of the ARv7 to the full-length AR (ARfl) is ∼1:5 (Antonarakis et al., 2014). Although there is vastly more ARfl than ARv7, there is conflicting evidence as to what role the highly expressed full-length protein plays in ARv7-driven ADT resistance (Kounatidou et al., 2019; Watson et al., 2010).

Dimerization is important for ARfl and ARv7 transcriptional activity (van Royen et al., 2011; Xu et al., 2015). In the canonical ARfl pathway, binding of androgens induces a monomeric intramolecular N/C interaction between the ARfl N-terminal domain (NTD) and the activation factor 2 (AF2) co-activator binding groove in the LBD (Dubbink et al., 2004; Hur et al., 2004; van de Wijngaart et al., 2006). This interaction exposes a nuclear localization signal that drives translocation of the ARfl from the cytoplasm to the nucleus (Ni et al., 2013). Once in the nucleus, the ARfl forms a homodimer via the serine (S597), alanine (A596) and threonine residues (T602) (Shaffer et al., 2004) in the dimerization box (D-box) of the DNA-binding domain (DBD) (van Royen et al., 2011). This D-box interaction induces a transition from an intramolecular N/C interaction within an individual protein to a head-to-tail intermolecular N/C interaction between two ARs (Schaufele et al., 2005; van Royen et al., 2007, 2011). Fluorescence recovery after photobleaching (FRAP) studies have shown that the ARfl homodimer forms stable interactions with DNA at both high and low affinity androgen response elements (ARE) (Farla et al., 2005; van Royen et al., 2011, 2014). Demonstrating the importance of ARfl dimerization, mutations to the D-box reduces the formation of stable DNA interactions, and these mutants can bind only to high affinity sites in the genome (Farla et al., 2005; van Royen et al., 2011, 2014). Similar to ARfl homodimers, ARv7 and ARfl can also form heterodimers through N/C and D-box interactions (Xu et al., 2015). In contrast, ARv7 only forms homodimers through D-box interactions as they lack an LBD (Xu et al., 2015). D-box mutations to ARv7 significantly reduce the variant enhancer activity, showing that dimerization is essential for inducing transcription (Xu et al., 2015). Yet, although dimerization is critical to ARv7 heterodimer and homodimer activity, their spatiotemporal and structural organization are unclear.

Here, we conduct detailed biophysical studies to characterize the structural organization of ARv7 homodimer and ARfl/ARv7 heterodimer dimerization. We demonstrate that ARfl/ARv7 heterodimers readily form while freely diffusing in the nucleus with a similar affinity to ARfl homodimers. Of all possible heterodimer orientations, an intermolecular N/C interaction is the most dominant. Once bound to DNA, both the ARv7 homodimer and heterodimer have a very short residence time compared to ARfl that is independent of dimerization. Similar to ARfl homodimers (van Royen et al., 2007), head-to-tail intermolecular N/C interaction between ARfl and ARv7 is lost upon DNA binding.

ARv7 induces transcription as either a variant homodimer or heterodimers with ARfl (Cato et al., 2019; Xu et al., 2015). To investigate the structural organization of ARv7 in homodimer and heterodimers, we first analyzed the protein-protein interactions using acceptor photobleaching FRET (abFRET) of ARv7 and ARfl in Hep3B cells. Nuclear receptors were labeled with either EYFP or ECFP fluorophores attached to a fixed linker at the N or C terminus. In this approach, an increase in donor (ECFP) intensity after photobleaching of the acceptor (EYFP) indicates that the fluorophores, and therefore the proteins, are in close proximity (van Royen et al., 2009). To evaluate the role of different domains in ARfl and ARv7 homodimerization and heterodimerization, we introduced point mutations to disrupt sites that are important for ARfl homodimerization, including the DBD D-box required for dimerization (A569T/S597T indicated as D), the NTD FQNLF motif required for N/C interactions (AQNAA indicated as F in Fig. 1A) and the DBD P-box required for DNA binding (R585K indicated as P) (Fig. 1A) (Cato et al., 2019; van Royen et al., 2007). Western blot showed that all fusion proteins are of the expected size and are expressed at a similar level (Fig. S1A). Further, all fusion proteins were transcriptionally active in an AR reporter assay except the ARv7 D-box mutant, suggesting that ARv7 dimerization was required for enhancer activity (Fig. S1B). AbFRET analysis of ARfl/ARv7 heterodimers demonstrated that heterodimerization was facilitated through both D-box and N/C interactions similar to ARfl homodimers (Fig. 1B,C,E,F,H) (van Royen et al., 2011). However, abFRET efficiency is significantly higher in YARv7/ARC, suggesting that the intermolecular N/C interaction (head-to-tail) is more dominant than N/N or C/C proximities (Fig. 1B). Point mutations in the D-box dramatically decreased but did not completely abolish heterodimer formation (Fig. 1E). Mutating the critical FQNLF site in the ARv7 NTD prevented the variant from forming a head-to-tail intermolecular N/C interaction with ARfl (Fig. 1H). As the same FQNLF mutation in ARfl did not affect heterodimerization, this suggested that the ARv7 NTD can interact with ARfl LBD (Fig. 1H). Together, this indicates that both the D-box and the remaining intermolecular N/C interaction facilitate heterodimerization that brings the N and C termini of both proteins in close proximity (Fig. 1B). Of note, intermolecular N/C interaction in heterodimers was not observed in the absence of androgen stimulation (Fig. S1C). In contrast, the ARv7 homodimer does not form head-to-tail intermolecular N/C interactions as it lacks an LBD (Fig. 1D) (Cato et al., 2019). This was corroborated by abFRET, in which a signal is only observed in a head-to-head (YARv7/CARv7) or tail-to-tail (ARv7Y/ARv7C) orientation, suggesting a more dominant role for the D-box in dimerization (Fig. 1D). Supporting this, ARv7 homodimerization was completely abolished by the D-box mutation (Fig. 1G). Overall, ARv7 homodimerization occurs through the D-box (Fig. 1D,G), and both D-box and N/C interactions are critical for ARfl/ARv7 heterodimers.

Having demonstrated that heterodimerization occurs in a similar mechanism to ARfl homodimerization, we then tested the relative affinity of ARfl with either ARfl (homodimer) or ARv7 (heterodimer) using a cell line stably expressing both double-labeled ARfl (YARC) and a doxycycline-inducible unlabeled ARfl or ARv7. Using this model, we could therefore alter the expression of either unlabeled ARfl or ARv7 to disrupt double-labeled ARfl abFRET in a competition assay (Fig. 2A). Similar to published work, we observed that increasing concentrations of the unlabeled ARfl decreased abFRET efficiency as the YARC homodimers were disrupted (van Royen et al., 2011). The abFRET was not totally abolished as intramolecular N/C interactions were not inhibited by either ARv7 or ARfl dimerization. Interestingly, the expression of the ARv7 induced a very similar reduction of abFRET compared to the homodimers (YARC/AR), indicating that the relative affinity of ARfl to ARv7 (heterodimer) and ARfl (homodimer) is similar (Fig. 2B).

To characterize the chromatin binding dynamics of ARfl and ARv7 homodimers and ARfl/ARv7 heterodimers, we performed strip-FRAP in Hep3B cells expressing EYFP-labeled ARfl and ARv7. Western blot analysis showed that ARfl and ARv7 R585K mutant EYFP fusion proteins were expressed at the expected size (Fig. S1D). In FRAP analysis, a small strip spanning the nucleus was bleached and the recovery of fluorescence intensity reflecting the dynamic behavior (e.g. diffusion coefficient, the immobile fraction and the average binding time to DNA) of the labeled AR is detected in time (van Royen et al., 2009). Previous quantitative FRAP analysis with agonist-bound ARfl revealed two types of DNA binding: a relatively long (multiple seconds) residence time, due to stable ARfl DNA binding, and one with a relatively short (<1 s) residence time that is also observed in the DNA-binding mutant ARfl R585K. Short immobilization is suggested to be a phenomenon that allows nuclear proteins to find their target binding sites via random DNA interaction (van Royen et al., 2014). Here, we quantified ARv7 and ARfl DNA-binding time by fitting the experimental FRAP curves with a computer simulation, as previously described (Geverts et al., 2014). We calculated the short and long immobilized fractions of the ARv7 or ARfl, and their residence time using a fixed diffusion rate obtained from previous studies (D=1.4 µm² s⁻¹) and DNA-binding parameters described in Materials and Methods (Fig. 3A) (van Royen et al., 2014). Interestingly, only ARfl was found to form a substantial long-term immobilized fraction (25%), with a binding time of 4.48±0.75 s. In contrast, ARv7 only had a very small long-term immobilized fraction (3%), similar to both the R585K ARfl and ARv7 DNA-binding mutants. Instead, a substantial percentage (42%) of both ARfl and ARv7 fractions had short immobilizations, with millisecond binding times (Fig. 3A). Supporting these FRAP results, only wild-type ARfl had a speckled distribution in the nucleus, whereas both the ARv7 and DNA binding mutants were homogenously distributed (Fig. 3B) (Cato et al., 2019; van Royen et al., 2011). Given that ARv7 is predominantly bound to chromatin via short-term interactions, we wanted to determine whether this led to a difference in the type of binding sites. When we characterized endogenously expressed ARv7 and ARfl chromatin binding sites from publicly available ChIP-seq (Cato et al., 2019), we found that ARv7 preferentially bound to those sites with significantly higher ARfl occupancy (Fig. 3C). Next, we co-expressed ARfl or ARv7 in a FRAP experiment to determine whether heterodimerization would affect DNA binding of either the short-term binding ARv7 or the long-term binding ARfl (Fig. 3D,E). Surprisingly, the co-expression of ARfl did not significantly change the short-term immobilization of ARv7 (Fig. 3E). These data suggest that although ARfl and ARv7 do dimerize and bind to the same genomic locations, their chromatin binding dynamics are independent of each other. Together, the combined results of FRET, FRAP and ChIP-seq suggest that although ARfl and ARv7 bind to DNA as dimers they do not influence the DNA binding behavior of each other, suggesting that individual ARs can exchange within the DNA bound complex.

Although FRAP measures the mobility of a total pool of molecules, it does not provide specific information about the interacting population. Therefore, to analyze the protein interaction of DNA-bound ARfl and ARv7 dimers we performed combined FRET and FRAP in Hep3B cells. In this approach, both donor (ECFP) and acceptor (EYFP) fluorescence is measured simultaneously (Fig. 4A) (van Royen et al., 2007). The recovery of the ECFP signal provides information about the mobility of only the interacting proteins, whereas, the acceptor FRAP curve (EYFP) represents the DNA-binding dynamics of the total pool of labeled AR proteins, similar to a conventional FRAP (Fig. 4B). Previously, FRET/FRAP with double-labeled ARfl (EYFP-ARfl-ECFP) showed that N/C interactions were lost when the ARfl was bound to DNA (van Royen et al., 2007). However, this approach did not distinguish between intramolecular and intermolecular N/C interactions. To analyze the intermolecular N/C interaction of ARfl dimers, we performed FRET/FRAP with YAR/ARC. We observed that the ARfl interacting population had a faster recovery compared to the total population. Based on previous models, we propose that this was due to a disruption of the ARfl homodimer intermolecular N/C interaction following DNA binding, thereby causing a loss of FRET, and the ECFP curve represents the freely diffusing ARfl dimers. To test this, we used DNA-binding mutants and observed that the mutant ARfl total population had a similar recovery as the interacting population, suggesting that DNA binding disrupts the intermolecular FRET (Fig. 4C,D). The DNA-binding deficient mutants (R585K) showed lower abFRET efficiency compared to ARfl homodimers and heterodimers (Fig. S3A-C). Nevertheless, abFRET efficiency of R585K mutants was sufficiently high to perform FRET/FRAP analysis. With the YARv7/ARC heterodimers, the N/C interacting molecules showed a faster recovery compared with the total population of ARv7, suggesting that the heterodimer intermolecular N/C interaction was also lost following DNA binding (Fig. 4E). In support of this model, the DNA-binding R585K mutants did not show a difference between the unbound N/C interacting YARv7/ARC dimers (Fig. 4F). To test the effects of the fluorophore position, we also analyzed YAR/ARv7C heterodimers and observed a similar fast distribution of N/C interacting proteins, demonstrating that the location of the fluorophore did not alter FRET/FRAP response (Fig. S2A,B). Similar to our earlier experiments, we did observe that the total pool of molecules (EYFP) showed different curves due to the long-term immobilization of ARfl on chromatin that is independent of ARv7 dimerization.

Recent crystallographic work has demonstrated that the ARfl can dimerize through its LBD (Nadal et al., 2017). When we conducted ARfl FRET/FRAP with both fluorescent tags on the C-terminus (ARY/ARC), the recovery kinetics were similar to N/C interacting ARfl, suggesting that ARfl homodimer C termini are not in close proximity when bound to chromatin (Fig. 4G,H). Together, this clearly demonstrates that the ARfl/ARv7 heterodimers and ARfl homodimers undergo a structural modification following DNA binding.

Unlike the ARfl dimers, the ARv7 homodimers do not form intermolecular N/C interactions but instead solely interact through the DBD (Fig. 1D,G). To investigate variant homodimerization on DNA, ARv7 was labeled either at the C termini (ARv7Y/ARv7C) or N termini (YARv7/CARv7) and subjected to FRET/FRAP. ARv7 DNA-binding deficient mutant dimers (R585K) showed similar abFRET efficiency with ARv7 homodimers (Fig. S3D,E). Interestingly, the C-termini-labeled ARv7 homodimers acceptor and donor-FRAP curve recovery rates were very similar, indicating that the ARv7 DBDs are in close proximity when bound to chromatin (Fig. 5A,B). Next, we performed FRET/FRAP analysis with N-termini-labeled ARv7 dimers to determine whether the NTD proximity is sustained with the DNA-bound ARv7 homodimer. The difference in recovery between interacting and pooled YARv7/CARv7 indicates that N termini are not in close proximity when the dimer is DNA bound (Fig. 5C). In agreement, DNA-binding mutations showed similar recovery in both the total pool and interacting ARv7 (Fig. 5D). Overall, ARv7 homodimers are in close proximity while they are off DNA at both the NTD and DBD. Yet, when the ARv7 homodimer binds to chromatin, a structural change occurs and leaves only the DBDs in close proximity during their short-term immobilization.

Expression of ARv7 is a common mechanism of ADT resistance and strongly correlates with development of CRPC (Sharp et al., 2019). However, AR splice variants are not found in isolation, and cancer cells typically also express high levels of ARfl. In clinical samples, the ARv7 is expressed at a median 1:5 ratio with the ARfl (Antonarakis et al., 2014). Given the critical role of AR dimerization for proper function, this study characterized the structural and biophysical properties of ARv7 homodimers and heterodimers, and delineated the spatiotemporal and structural organization on chromatin.

Interestingly, we observed that ARfl did not have a preference for homodimerization over ARv7 heterodimerization, suggesting that dimerization is more dependent on protein availability than binding affinity (Fig. 2B). During the ARfl/ARv7 dimerization, both head-to-tail intermolecular N/C and DBD interactions contributed to heterodimerization. However, the N/C interactions were the most dominant interaction (Fig. 1B), corroborating previous work (Xu et al., 2015). In this heterodimer, all four termini of the ARv7 and ARfl are kept in close proximity before binding DNA (Fig. 6A). Once bound to chromatin, the ARfl/ARv7 heterodimer complex dramatically changes, causing both the N and C termini to be separated while maintaining the DBDs in close proximity. In support of this, we demonstrate that the ARfl/ARv7 heterodimer undergoes structural modifications following DNA binding that disrupts intermolecular N/C interactions between the NTD of ARv7 and LBD of ARfl (Fig. 4E). When bound to DNA, the residence times of ARfl and ARv7 are different; ARv7, unlike ARfl, only formed short-term interactions preferentially at high-occupancy ARfl sites. Although speculative, this suggests that the LBD is involved in the recruitment of co-activators needed to form stable transcriptional hubs at weak AR binding sites. Surprisingly, heterodimerization with ARfl does not significantly increase the ARv7 immobilized fraction on chromatin, suggesting that although dimerization is critical for binding, DNA residence is determined by the individual protein. Therefore, when this heterodimer binds to DNA the ARfl is immobilized on DNA longer (seconds), whereas ARv7 is released relatively rapidly (milliseconds) (Fig. 6A).

Similar to the ARfl/ARv7 heterodimer, both intermolecular N/C and DBD interaction contribute to ARfl homodimerization (Fig. 1C). We demonstrated that the head-to-tail intermolecular N/C interaction of ARfl homodimer is disrupted following chromatin binding. This corroborates earlier work that demonstrated loss of both intermolecular and intramolecular N/C interaction on chromatin (van Royen et al., 2007). In freely diffusing ARfl homodimers, the individual LBDs are also in close proximity (Nadal et al., 2017). Interestingly, we also show with C-termini-labeled proteins that the LBDs of ARfl do not remain close to each other on chromatin in living cells (Fig. 4G, Fig. 6B). The loss of N/C interaction and LBD proximity in ARfl homodimers following DNA binding is similar to the ARfl/ARv7 heterodimer in which all four termini of chromatin-bound ARfl homodimers are separated (Fig. 6B). Thus, this separation of ARfl dimers is likely a necessary step in transcriptional regulation and potentially occurs due to co-activator binding (van Royen et al., 2007; Yu et al., 2020). In a recent cryo-electron microscopy (cryo-EM) structure of ARfl homodimers bound to DNA, the DBDs were found to be at the core of the dimer-DNA interaction, whereas the NTDs were solvent exposed (Yu et al., 2020). In this, all three domains created a head-to-tail interaction interface. Interestingly, the NTD orientations were asymmetrical with one maintaining an N/C interaction with the LBD. This conflicts previous FRET/FRAP analysis of ARfl homodimerization, which showed that the N/C interaction is disrupted when bound to DNA (van Royen et al., 2007). This discrepancy between cellular biophysical measurements and protein cryo-EM could be caused by numerous factors, including the recruitment of large co-activator complexes, DNA occupancy times, differences in static versus dynamic measurements and many others. Overall, our results suggest that following binding to chromatin the DBDs of ARfl or ARv7 remain in close proximity, and the orientation change of NTD and LBD cause these two domains to move apart.

As the ARv7 homodimer cannot form intermolecular N/C interactions due to the absence of an LBD, homodimerization is therefore primarily initiated through DBD interactions. Unlike ARfl homodimers or ARfl/ARv7 heterodimers, this complex interacts in a head-to-head orientation with both the NTDs and DBDs in close proximity (Fig. 1D, Fig. 6C). Once bound to chromatin, the ARv7 homodimer DBDs remain in close proximity while the NTDs separate, as observed in ARfl/ARv7 heterodimers. However, these homodimers do not form long-term immobilized complexes and selectively bind to high affinity AR binding sites. Yet, despite these different orientations the presence of NTD and DBD is sufficient for transcriptional activity of ARv7 (Chan and Dehm, 2014).

This study identified the structural organization of ARfl and ARv7 dimers in two spatial forms: when they are freely diffusing (in the nucleus) and DNA bound (long or short immobilization). Overall, we found that homodimers and heterodimers have markedly different conformations when they diffuse in the nucleus or interact with DNA to promote transcription of the target genes. Freely diffusing heterodimers form a structure in which each protein terminus is close to the other. When bound to DNA, this intermolecular N/C interaction, as well as N/N and C/C proximity, is disrupted. Loss of this interaction may potentially disrupt heterodimer stabilization, leading to the individual protein determining immobilization time on chromatin. In contrast, freely diffusing ARv7 homodimer only shows N/N or C/C proximity, and does not form a head-to-tail intermolecular interaction. When binding DNA, this NTD proximity is lost, whereas the C/C proximity (DBD interaction) is maintained.

Our biophysical analysis provides a mechanistic insight into heterodimer-DNA interaction in which ARv7 rapidly binds DNA whereas ARfl forms longer immobilized complexes. The relative ARfl affinity for ARv7 and fast turnover on chromatin indicate that ARv7 might be important for facilitating ARfl-chromatin interaction, suggesting heterodimers may potentially be therapeutic targets for CRPC treatment.

ARfl and ARv7 were separated from the fluorescent tag by (Gly-Ala)₆ spacer in all constructs expressing ARfl or ARv7 fusion proteins (Farla et al., 2004). N-terminal EYFP-AR and ECFP-AR, C-terminal AR-EYFP and AR-ECFP, R585K mutants of EYFP-AR and AR-ECFP, A596T/S597T mutants of EYFP-AR, AR-EYFP and AR-ECFP, and F23,27A/L26A mutant of AR-ECFP were constructed as described previously (van Royen et al., 2011). R585K and A596T/S597T mutants of ECFP-AR were generated, and replaced wild-type AR with a mutant AR.

ARv7 fusion proteins EYFP-ARv7 and ARv7-ECFP were generated by replacing ARfl with ARv7 in EYFP-AR and AR-ECFP constructs, respectively. For ARv7-EYFP subcloning, EYFP from the EYFP-ARv7 construct was PCR amplified with a stop codon. ECFP was removed from the ARv7-ECFP construct via PCR amplification. ARv7 construct without ECFP was combined with a stop codon-containing EYFP to generate the ARv7-EYFP construct. For ECFP-ARv7 subcloning, ECFP from the ARv7-ECFP construct was PCR amplified and the stop codon was removed. Plasmid containing ARv7 was PCR amplified without EYFP from the EYFP-ARv7 construct. ARv7 construct without EYFP was combined with the stop codon-removed ECFP to generate ECFP-ARv7 construct.

Most of the ARv7 R585K, A596T/S597T and F23,27A/L26A mutants were generated via site-directed mutagenesis. R585K forward primer 5′-GTCTTCTTCAAAAAAGCCGCTGAAGGG-3′, reverse primer 5′-CTTGCAGCTTCCACATGTGAGAGCTC-3′, A596T/S597T forward primer 5′-GTACCTGTGCACCACCAGAAATGATTG-3′, reverse primer 5′-TTCTGTTTCCCTTCAGCGGCTCTTTT-3′; F23,27A/L26A forward primer 5′-ACCTACCGAGGAGCTGCACAGAATGCTGCACAG-3′ and reverse primer 5′-AGGTCTTGGACGGCGGCCGAGGGTAGAC-3′ were used for PCR amplification. Following PCR amplification, the product was incubated with DpnI at 37°C for 1 h to digest methylated template. Enzyme was inactivated at 80°C for 20 min. The 5′ ends of the PCR product were phosphorylated with T4 PNK enzyme (NEB, M0236S). Subsequently, the PCR product was ligated with T4 DNA ligase (NEB, B0202) and transformed into competent bacteria. Once EYFP-ARv7 and ARv7-ECFP mutants were generated, they were used to replace wild-type ARv7 in other constructs with a mutant ARv7. The (ARE)2-TATA Luc reporter includes two high-affinity AREs. They are underlined in the following sequence: 5′-CCGGGAGCTTGTACAGGATGTTCTGCATGCTCTAGATGTACAGGATGTTCTGGTA-3′. It was a gift from G. Jenster (Rotterdam, Netherlands). To generate doxycycline-inducible ARfl and ARv7 constructs, gateway cloning (Invitrogen, 11791-020-Gateway LR Clonase II) was performed between pENTR1A AR or pENTR1A ARv7 plasmid with pLenti CMV tight hygro DEST expression plasmid.

Hep3B cells (human, ATCCHB-8064) were used for the experiments. Importantly, Hep3B does not express endogenous AR, and was selected as cellular context to avoid potential competition within the AR dimerization experiments. Two days before the experiments, Hep3B cells were grown on glass coverslips in six-well plates for imaging experiments. Hep3B cells were cultured in Dulbecco's modified Eagle medium (DMEM) without phenol red (Lonza, Belgium, BE12-604F) supplemented with 10% fetal calf serum (FCS, Gibco, 10270-106), 1% penicillin-streptomycin (P/S) (Lonza, Belgium, DE17-602E) and 1% L-glutamine (Lonza, Belgium, BE17-605E). One hour before transfection, medium on Hep3B cells on glass coverslips was replaced with medium containing 1 nM dihydrotestosterone (DHT) and 5% FCS stripped with dextran-coated charcoal (DCC-FCS). For FRET/FRAP experiments only, cells were treated with 100 nM DHT. Transfections were performed with 2 µg per well of ARfl or ARv7 expressing fluorescent fusion constructs in FuGENE6 (Promega)-medium without FCS. FuGENE6 was used at a 1:3 [plasmid (µg):FuGENE6 (µl)] ratio. After overnight incubation, transfection medium was replaced with medium containing 1 nM DHT and 5% DCC-FCS. The Hep3B cell line was regularly tested for mycoplasma contamination.

For the ARfl and ARv7 abFRET competition experiment, stably YARC-expressing Hep3B cells were, initially, stably transduced with pLenti CMV rtTA3 Blast containing lentiviruses to generate a doxycycline inducible system. After blasticidin (8 µg ml⁻¹) selection for 5 days, the survivor cells were transduced with either pLenti CMV tight ARfl or ARv7 hygro containing viruses. They were selected with hygromycin (200 µg ml⁻¹) for 1 week and then propagated. Two days before an abFRET competition experiment, doxycycline-inducible ARfl and ARv7 expressing YARC Hep3B cells were seeded on coverslips in six-well plates. The following day, the medium was replaced with 5% DCC-FCS and 1 nM DHT containing medium supplemented with variable doxycycline concentrations (0, 5, 10 and 20 ng µl⁻¹) to activate unlabeled ARfl or ARv7 expression.

For viral packaging, HEK 293T cells (human, ATCCCRL-3216) were cultured in DMEM with 10% FCS and 1% P/S. A third generation viral packaging system was used. HEK 293T cells were transfected with 2250 ng of pMDLG/pRRE, 2000 ng of pRSV/Rev and 2500 ng of transfer plasmid. Viruses were harvested 48 h and 72 h following transfection.

For transactivation assays, Hep3B cells were seeded in 24-well plates in DMEM without phenol red supplemented with 10% FCS, 1% P/S and 1% L-glutamine. The next day, medium was replaced with DMEM supplemented with 5% DCC-FCS in the presence of EtOH or 1 nM DHT. Hep3B cells were transfected with 50 ng of ARfl or ARv7 fusion construct and 100 ng of luciferase reporter construct (ARE2TATA-luc) per well. The medium was replaced after overnight incubation. Next, 48 h after transfection, cells were lysed, and luciferase activity was measured using a GloMax Microplate Luminometer (Promega).

Hep3B cells were seeded and transfected with ARfl or ARv7 fluorescently labeled constructs in six-well plates. Then, 48 h after transfection, cells were washed in ice-cold PBS and lysed. The lysis buffer contained 50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 1% NP40, 0.5% Na-deoxychelate and protease inhibitor (Roche) (one tablet for 10 ml of buffer). A bicinchoninic acid assay (Pierce BCA Protein Assay kit, 23225) was used for protein quantification. Equal amounts of protein were combined with 10% β-mercaptoethanol (Bio-Rad) and 4× Laemmli buffer (Bio-Rad), and denatured at 95°C for 10 min. Equal amounts of the samples were run on a 10% SDS-polyacrylamide gel and then transferred to a PVDF membrane (Bio-Rad). Blots were incubated with custom mouse monoclonal anti-AR (F39.4.1, 1:2000) (van Royen et al., 2007, 2011) or mouse monoclonal AR441 (Abcam, ab9474, 1:1000) (Sha et al., 2020) and rabbit polyclonal GAPDH (Santa Cruz Biotechnology, SC-25778, 1:2000) (Saraç et al., 2020) primary antibodies. Horseradish peroxidase-conjugated polyclonal goat anti-mouse IgG (Abcam, ab97023, 1:4000) (Saraç et al., 2020) or goat anti-rabbit IgG antibody (Abcam, ab97051, 1:4000) (Saraç et al., 2020) were used as secondary antibodies. Proteins were detected using Super Signal West Pico Luminol solution and imaged using an Amersham (AI600) Chemiluminescent Imager or LI-COR Odyssey Fc.

abFRET analysis was performed to identify interactions between ARfl and ARv7. EYFP- and ECFP-labeled ARfl and ARv7 constructs were imaged 1 day after transient transfection. A Zeiss LSM 510 confocal laser scanning microscope equipped with a Plan-Neofluar 40×/1.3 NA oil objective (Carl Zeiss, Microimaging, Inc.) was used to perform abFRET experiments at 100 nm lateral resolution. EYFP- or ECFP-labeled ARfl and ARv7 with stoichiometric and physiologically relevant co-expression levels were selected in all quantitative imaging experiments. ECFP was excited at 458 nm at moderate laser power, and a 470 nm-500 nm band-pass emission filter was used for detection. EYFP was excited at 514 nm at moderate laser power, and a 560 nm long-pass emission filter was used for detection. Following EYFP and ECFP image collection, acceptor (EYFP) in the nucleus (100 µm² area covering the nucleus) was bleached 25 times at 100% laser power at 514 nm. After photobleaching, second images of EYFP and ECFP were collected. Background was subtracted from EYFP and ECFP fluorescence intensities. abFRET efficiency was calculated using the following equation: abFRET=((ECFP_after–ECFP_before)×EYFP_before)/((ECFP_after×EYFP_before)–(EYFP_after×ECFP_before)). ECFP_before and EYFP_before represent the pre-bleach intensities (after background substraction), whereas ECFP_after and EYFP_after represent the post-bleach intensities in the bleached nucleus (van Royen et al., 2009). Finally, apparent abFRET efficiency was expressed as normalized abFRET between abFRET₀ (abFRET between free EYFP and ECFP proteins) and abFRET_CYFP (abFRET in EYFP-ECFP fusion protein) using the following equation: apparent abFRET efficiency=(abFRET–abFRET₀)/(abFRET_CYFP−abFRET₀).

FRAP was used to study the DNA-binding dynamics of the nuclear receptors (van Royen et al., 2011). A Zeiss LSM 510 confocal laser scanning microscope was used for FRAP experiments. A strip spanning the nucleus was excited at 514 nm with a low laser power and detected with a 560-nm-long pass filter. After 40 scans, a high-intensity laser pulse at 514 nm (100%) was applied for two iterations to photobleach EYFP inside the strip. The bleached strip was continuously scanned for 360 scans, and the scan was completed at 40 s. The curves were normalized using the following equation: I_norm=(I_raw−I₀)/(I_after−I₀). I_after and I₀ represent the average intensity of the last 40 scans and intensity just after photobleaching (van Royen et al., 2008). Quantitative FRAP analysis was performed using a Monte Carlo simulation environment for modeling complex biological molecular interaction networks (Geverts et al., 2014). Experimental curves were fitted to nearly 50.000 simulated curves with a previously identified diffusion coefficient (D=1.4 µm² s⁻¹) (van Royen et al., 2014), up to 50% shortly immobilized ARs with binding times ranging from 0 to 2 s, and up to 40% long immobilized ARs with binding times ranging from 2 to 15 s, based on our previous findings. The bound fractions and binding times were determined as the average of the ten best fitting simulated curves.

FRET/FRAP was used to study the mobility of interacting molecules (van Royen et al., 2007, 2009). A Leica TCS SP8 laser scanning confocal microscope equipped with Plan-apochromat (HC PL APO CS2) 40×1.3 NA oil objective (Leica Microsystems) was used for FRET/FRAP experiments. An argon laser was adjusted to 50% power. A strip spanning the nucleus was excited at 458 nm with low laser power (1%) (because of simultaneous EYFP and ECFP scanning in FRET/FRAP) (van Royen et al., 2007). ECFP intensity was detected with a photomultiplier tube (PMT) at 465 nm-500 nm. EYFP intensity was detected using a PMT at 560-670 nm. A bi-directional scan was used to scan the strip. After 40 scans with 100 ms time intervals, a high-intensity laser pulse (100%) at 514 nm was applied for four iterations to photobleach EYFP inside the strip. ECFP and EYFP intensities were recorded for 360 scans following photobleaching. A FRAP booster was used for FRET/FRAP. EYFP and ECFP curves were separately normalized using the following equation: I_norm=(I_raw−I₀)/(I_after−I₀). I_after and I₀ represent the average intensity of the last 40 scans and intensity just after photobleaching.

We obtained raw sequencing files of GSM2842700 (ChIP_LN95_parental_ETOH_ARfl), GSM2842704 (ChIP_LN95_parental_ETOH_ARv7) and GSM2842702 (ChIP_LN95_parental_ETOH_Input) samples from GEO series GSE106559, and mapped these samples with Burrows–Wheeler Aligner-MEM (0.7.17) to the hg19 reference genome. From the mapped reads, ChIP peaks were called using MACS2 callpeaks (2.1.2) (Zhang et al., 2008) for each antibody sample (ChIP_LN95_parental_ETOH_ARfl antibodyChIP_LN95_parental_ETOH_ARv7) with respect to input library sample (ChIP_LN95_parental_ETOH_Input). Parallel to this, we generated bigwig files for these samples using the deepTools (3.1.3) (Ramírez et al., 2016) bamcoverage function using ‘--blackListFileName --normalizeUsing RPKM’ options [for blacklist, we used the ENCODE ENCFF001TDO blacklist bed file (Dunham et al., 2012)]. After that we overlapped ChIP_LN95_parental_ETOH_ARfl and antibody ChIP_LN95_parental_ETOH_ARv7 peaks using the venn function in Intervene (Khan and Mathelier, 2017). For those ARfl only and ARfl/ARv7 common regions, we calculated average read signal intensity using the deepTools computematrix function with reads per kilobase of transcript per million mapped reads normalized (RKPM) bigwigs. We then used deepTools profilePlot to visualize density plots.

We thank Halil Bayraktar for his helpful advice. F.Ö. and N.A.L. gratefully acknowledge the Erasmus MC Optical Imaging Centre for confocal microscopes and technical support.

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.258332