Modulation of nongenomic activation of PI3K signalling by tetramerization of N-terminally-cleaved RXRα

Retinoid X receptor-alpha (RXRα) binds to DNA either as homodimers or heterodimers, but it also forms homotetramers whose function is poorly defined. We previously discovered that an N-terminally-cleaved form of RXRα (tRXRα), produced in tumour cells, activates phosphoinositide 3-kinase (PI3K) signalling by binding to the p85α subunit of PI3K and that K-80003, an anti-cancer agent, inhibits this process. Here, we report through crystallographic and biochemical studies that K-80003 binds to and stabilizes tRXRα tetramers via a ‘three-pronged’ combination of canonical and non-canonical mechanisms. K-80003 binding has no effect on tetramerization of RXRα, owing to the head–tail interaction that is absent in tRXRα. We also identify an LxxLL motif in p85α, which binds to the coactivator-binding groove on tRXRα and dissociates from tRXRα upon tRXRα tetramerization. These results identify conformational selection as the mechanism for inhibiting the nongenomic action of tRXRα and provide molecular insights into the development of RXRα cancer therapeutics.

In addition to acting as heterodimeric partners of some nuclear receptors [4][5][6][7][8][9][10][11]17 , RXRa can form homodimers 18 that trigger specific signalling pathways [19][20][21] . A unique property of RXRa is an auto-repressive mechanism involving homotetramerization, in which a 'dimer-of-dimers' pack in a bottom-to-bottom manner, with H12 from each monomer invading (and invaded by) a neighbouring domain across the tetramer interface 22 . This pairwise 'exchange of arms' enables each monomer to bind to and occlude the coregulator-binding groove of its neighbour 22 . Some molecules have been shown to promote the auto-repressed state and repress gene transcription 23 . The equilibrium between inactive homotetramers and active homodimers is tightly regulated by ligand binding 18,22,24 . Thus, binding of agonists such as 9-cis-retinoic acid (9-cis-RA) results in dissociation of homotetramers into homodimers, which has been proposed as a regulatory mechanism for RXRa transactivation 22,25 . However, compounds that act to promote RXRa homotetramerization remain to be identified and characterized.
Recent advances have revealed important nongenomic functions of RXRa and its non-canonical modulators 7,26-28 . We previously found that RXRa is abnormally cleaved in many cancer cells, resulting in a truncated RXRa (tRXRa) that lacks a portion of its N-terminal A/B domain 29 . Unlike the wild-type RXRa, which normally resides in the nucleus, tRXRa is cytoplasmic, and we showed that it interacts with the p85a regulatory subunit of the phosphatidylinositol 3-kinase (PI3K), a critical player in a wide range of cellular processes including cell growth, proliferation, survival, motility, metabolism, protein synthesis and migration 30,31 . The interaction with p85a promotes PI3K/AKT activation and enhances tumour cell growth, revealing an oncogenic effect of tRXRa in tumour cells. We also reported that K-80003, a promising anti-cancer agent derived from the non-steroidal anti-inflammatory drug (NSAID) Sulindac, binds to tRXRa and inhibits its interaction with p85a and activation of PI3K signalling 29 . However, the molecular mechanism of this potent inhibition remained to be determined.
Here, we report the crystal structure of RXRa-LBD in complex with K-80003, and the characterization of the role of K-80003mediated tRXRa tetramerization in regulating its interaction with p85a and the nongenomic activation of PI3K signalling. Our results reveal a previously unrecognized role of RXRa tetramers in modulating subcellular localization and nongenomic interaction with cytoplasmic signalling proteins. Moreover, we show that K-80003 inhibits tRXRa interaction with p85a by stabilizing a tetrameric form of tRXRa through a 'three-pronged' mechanism involving both canonical and non-canonical binding. This work opens up new avenues for developing novel RXRa-based therapeutics, by selectively stabilizing a particular oligomeric state.

Results
Crystal structure of K-80003-bound RXRa-LBD. To understand how K-80003 ( Supplementary Fig. 1a) modulates the biological activity of tRXRa, we determined the crystal structure of RXRa-LBD in complex with K-80003 to a resolution of 2.6 Å ( Table 1). We found that the RXRa-LBD/K-80003 complex adopts a tetrameric structure, similar to those observed in crystals of apo-RXRa or RXRa-LBD/tRA-isomer complex 22 ( Supplementary Fig. 1b), in which two RXRa-LBD canonical homodimers (labelled A1/B1 and A2/B2) pack in a bottom-tobottom manner (Fig. 1a) with point group symmetry D2. Besides the canonical dimer interface 32 , there are two symmetry-related interfaces involving subunits A1/B2 and B1/A2. These 'tetramer' interfaces comprise three sub-regions: parallel packing between symmetry-related H3 helices; 'end-to-end' packing between H11 that reduces their length by two helical turns (compared with the agonist-bound structure); and the invasion of each H12 helix into its apposing domain, where it binds to the coregulator-binding groove, consisting of elements of H3 and H4 (Fig. 1a). The apposing domain reciprocates the process in a pairwise 'exchange of C-terminal arms'. An LMEML motif near the C terminus of H12 binds to the coregulator groove by mimicking the LxxLL/ LxxIL motif of coregulators ( Supplementary Fig. 2). The tetramer interfaces create two substantial symmetry-related interfacial cavities that are readily accessible to solvent and small molecules.
Refinement of the structure and careful inspection of difference Fourier maps revealed six molecules of K-80003-bound per tetramer, with three molecules per A1/B2 or A2/B1 interfacial cavity ( Fig. 1a; Supplementary Fig. 3). This stoichiometry is  Fig. 5a). The 2 A1/B2 and A2/B1 interfacial cavities are symmetry-related and the three bound molecules in each cavity (K-80003A, K-80003B and K-80003C, respectively, Fig. 1b) appear to play distinct but complementary roles in stabilizing the tetramer. Thus, K-80003A and K-80003B are arranged about the A1/B2 pseudo-dyad, making similar but distinct interactions (Figs 1b and 2b). They bind in a region where the H12 is located in the agonist-bound RXRa-LBD structure (Fig. 2a) and thus distinct from the canonical ligandbinding region. The indene ring of K-80003A makes many interactions with monomer B2, including parallel aromatic stacking with W305, and hydrophobic binding with L276 and L309, and R302. Its isopropylphenyl ring contacts K-80003C. On the other face of K-80003A, there are several hydrophobic/ aromatic interactions, including with the side-chains of F439, I447 and L451 from the A1 monomer (Fig. 2b). On the fourth side, to complete the 'cage' around K-80003C, there are numbers of polar and ionic interactions. Notably, the carboxylate of K-80003A makes a bifurcated salt-bridge with K440 from A1 and R302 from B2, as well as an H-bond with the indole N-H of W305 (B2) (Fig. 2b). Thus, K-80003A is firmly encased on all sides at a unique location: by side-chains from A1 and B2 at the top and bottom, by molecule K-80003C on one side, and by elements of the invading H11-H12 turn and H12 helix on the other. For K-80003B, the symmetry-related packing is similar but distinct with fewer interactions. The ionic interactions with R302 of A1 and K440 of B2 are much weaker, which may explain why K-80003B is less ordered than K-80003A, and with less welldefined electron density. Nevertheless, it makes similar aromatic stacking interactions-in this case, parallel with F439, but a lessoptimal (45°stacking angle) with W305 (Fig. 2), as well as hydrophobic interactions with L433 of A1 and L436 of B2.
K-80003C binds in the pocket of monomer B2, in a mode that resembles a canonical ligand (Fig. 2a). Unlike K-80003A and K-80003B, it makes strong contacts only with the B2 subunit ( Fig. 2b). Its carboxyl motif makes a salt-bridge with R316, while the indene ring sits in a broad hydrophobic cavity, and both are well-defined in the electron density map. However, the isopropylbenzene moiety sticks out of the pocket and into the cavity, making only weak contacts with other residues; it presumably adopts multiple conformations, consistent with the weak electron density ( Supplementary Fig. 3).
Taken together, the interactions between the three bound K-80003 molecules and the protein suggest that K-80003 stabilizes RXRa-LBD tetramer by a unique combination of distinct and canonical-binding mechanisms acting like a '3-pronged' binding mode: K-80003A is tightly packed in a hydrophobic, aromatic and polar cage that strengthens the tetramer and provides additional glue to hold H12 in its place to prevent coregulator binding. K-80003B also contributes to tetramer stabilization, but appears less tightly packed; significantly, however, it blocks the entrance to the unoccupied pocket in monomer A1, which might be otherwise favored by ligands such as 9-cis-RA. K-80003C binds in the pocket in monomer B2 interacting with K-80003A and B2.    To understand why K-80003 binds asymmetrically to the tetramer, we closely examined the shape of the tetramer's interfacial cavity. Superposition of A1/B2 subunit on to itself reveals that A1/B2 is asymmetric ( Supplementary Fig. 4a) with a smaller pocket in monomer A1 than in the B2 monomer ( Supplementary Fig. 4b). This asymmetry seems to be resulted from the shifts of H3 and H11 into the pocket in A1 ( Supplementary Fig. 4c) and provides an explanation for why small molecules bind asymmetrically to the tetramer 22,33 . In the previously reported RXRa-LBD/tRA-isomer tetramer structure 22 , tRA-isomer binds to the narrower pocket in monomer A1 where K-80003 molecule is too big to fit in (Supplementary Fig. 4b). In our structure, the large size of the K-80003C molecule presumably selects the wider LBP pocket of the A1/B2 dimer. Despite these differences in ligand binding, the two tetramers have similar overall structures (r.m.s. deviation for Ca atoms is 1.20 Å), and the largest changes are observed at the tetramer interface ( Supplementary Fig. 1b), presumably induced by the non-canonical binders, K-80003A and K-80003B.
K-80003 promotes tetramerization of tRXRa but not RXRa. The unique binding of K-80003 observed in the tetrameric form of K-80003-bound RXRa-LBD crystal structure prompted us to determine whether K-80003 binding could promote RXRa-LBD tetramerization. In non-denaturing polyacrylamide gel electrophoresis, purified RXRa-LBD protein existed as two distinct bands corresponding to homodimer and homotetramer, respectively (Fig. 3a). As expected, incubation of RXRa with 9-cis-RA known to induce homodimerization 18 resulted in a shift of RXRa-LBD from tetramers to dimers. In contrast, incubation with K-80003 induced an accumulation of tetramers. L433 near the C terminus of H10 packs directly against the indene ring of K-80003B, and its mutation to D would destabilize key interactions with K-80003 due to loss of existing hydrophobic interaction and introduction of repulsive chargecharge interaction with the carboxyl motif of K-80003B. Indeed, K-80003 failed to bind to RXRa-LBD/L433D ( Supplementary Fig. 5a) and promote its tetramerization (Fig. 3a). Substitution of Q275 in H3 with E, or F439 in H11 with A, which are involved in the binding of K-80003 ( Fig. 2), impaired the tetramerization of respective mutants by K-80003, confirming the role of H3 and H11 in the K-80003-induced stabilization of RXRa-LBD tetramers. Interestingly, mutating R316 in LBP, which is essential for 9-cis-RA binding 12 and involved in K-80003C binding (Fig. 2b), with E, resulted in a mutant (RXRa-LBD/R316E) that exhibited mainly as a tetramer independent of the presence of either 9-cis-RA or K-80003, in agreement with a previous report 25 . The effect of K-80003 on stabilizing RXRa-LBD tetramers appeared overpower the 9-cis-RA-induced homodimerization as RXRa-LBD was mainly found as a tetramer in the presence of both K-80003 and 9-cis-RA (Fig. 3a).
The ability of K-80003 to stabilize RXRa-LBD tetramer was also illustrated by gel filtration chromatography showing that purified RXRa-LBD protein mainly existed as a tetramer in the presence of K-80003, while it displayed predominantly as a homodimeric complex in the presence of 9-cis-RA (Fig. 3b). Evaluation of a panel of chemical crosslinkers identified homobifunctional N-hydroxysuccimide-based chemical crosslinker BS3 as the most efficient one to crosslink the RXRa-LBD tetramer ( Supplementary Fig. 5b). Thus, BS3 was subsequently used to study the effect of K-80003 on RXRa tetramerization. Extracts from cells transfected with RXRa-LBD and treated with either K-80003 or 9-cis-RA were prepared and subsequently incubated with BS3. Figure 3c showed that the treatment of cells with K-80003 produced crosslinked species on SDS-PAGE gels of 25, 50 and 100 kDa, corresponding to monomer, dimer and tetramer of RXRa-LBD, respectively. For comparison, treatment with 9-cis-RA resulted in only monomer and homodimer. Interestingly, RXRa or mutants expressed in cells mainly existed as a monomer even after crosslinking, which is different from purified RXRa proteins. Examination of the effect of K-80003 on tetramerization of tRXRa and RXRa revealed that K-80003 effectively promoted the formation of tRXRa tetramers. Unexpectedly, K-80003 failed to promote the tetramerization of the full-length RXRa, indicating that the N-terminal A/B domain interferes with its tetramerization.
The RXRa N/C interaction inhibits its tetramerization. The above observation prompted us to determine how the N-terminal A/B domain of RXRa inhibited its tetramerization (Fig. 1). Thus, the interaction between RXRa-A/B domain and RXRa-LBD ( Fig. 3d) was studied by cell-based coimmunoprecipitation (coIP) assays. Immunoprecipitation of the RXRa-A/B protein resulted in a strong coIP of the RXRa-LBD protein, demonstrating their interaction. The interaction was inhibited by 9-cis-RA in a dosedependent manner (Fig. 3e), and was confirmed by immunostaining showing extensive colocalization of transfected RXRa-A/B with RXRa-LBD but not RXRa in the cytoplasm of cells ( Supplementary Fig. 6a). Cotransfection of the RXRa A/B domain completely suppressed the effect of K-80003 on promoting tRXRa tetramerization (Fig. 3f). These data reveal an extensive intramolecular interaction between the N terminus and the C terminus (N/C) in RXRa and its critical role in regulating RXRa tetramerization, and also provides a molecular explanation for the differential effect of K-80003 on the tetramerization of tRXRa and RXRa.
To further study the N/C interaction, we first conducted deletion analysis of the N-terminal A/B region to narrow down the region required for the interaction. Deletion of either the N-terminal 40 or 60 amino acids could not confer the ability of the resulting mutants to interact with the A/B domain (Fig. 3g). Deletion of additional 20 amino acids resulted in a mutant (tRXRa), which strongly interacted with the RXRa-A/B protein. RXRa-D60, like tRXRa, interacted strongly with RARg ( Supplementary Fig. 6b), demonstrating that RXRa-D60 is still active in heterodimerization with RARg. These results suggested that amino acids from 60 to 80 are critical for the N/C interaction. We next determined region in RXRa-LBD required for binding A/B domain. RXRa-A/B could interact with RXRa mutants lacking N-terminal sequences (tRXRa, RXRa-D100 and RXRa-LBD) but not with mutant lacking C-terminal LBD (RXRa-1-235) or mutants lacking AF2/H12 region (RXRa-DAF2 and RXRa-DA/BDAF2) (Fig. 3h). tRXRa without AF2/ H12 also failed to interact with RXRa-A/B ( Supplementary  Fig. 6c). Thus, the N/C intramolecular interaction involves the N-terminal A/B domain and the C-terminal AF2/H12. The conclusion was supported by data showing the inability of the A/B domain to bind to the full-length RXRa, likely due to the unavailability of the C-terminal binding site masked by its own N-terminal A/B domain. The AF2/H12 is involved in the formation of the hydrophobic coactivator-binding groove that was shown to mediate the N/C interaction of some nuclear receptors 34,35 . To determine whether the N/C interaction involves the coactivator-binding groove, we tested whether W305, which is located in H5 and was shown to play a critical role in the formation of the coactivator-binding groove 12 , was involved in the N/C interaction. Our results showed that RXRa-LBD/W305Q with W305 mutated to Q, which failed to bind to an LxxLLcontaining protein (see below), could still bind to the RXRa-A/B protein in a 9-cis-RA sensitive manner similar to the wild-type RXRa-LBD (Fig. 3i). These results preclude the involvement of the coactivator-binding groove in the N/C interaction.
An LxxLL motif in p85a mediates p85a interaction with RXRa. tRXRa differs from RXRa in its ability to reside in the cytoplasm and interact with cytoplasmic p85a protein, an event that is inhibited by K-80003 (ref. 29). To address the role of tetramerization in modulating tRXRa interaction with p85a, we first determined how p85a binds to tRXRa. Thus, several p85a mutants ( Fig. 4a) were constructed and analysed for their interaction with tRXRa. CoIP experiments showed that the N-terminal region of p85a, including the N-terminal SH3 and BCR domains, is sufficient for interacting with tRXRa, whereas the C-terminal region including NSH2, iSH2 and CSH2 domains was dispensable (Fig. 4b). We further observed that a p85a mutant encompassing only the BCR domain could interact with tRXRa and RXRa-LBD but not RXRa-A/B (Fig. 4c).
In an effort to understand the molecular basis for BCR interaction with tRXRa, we noticed the presence of two LxxLL motifs, 161 . 4a) was therefore synthesized and examined for its binding to RXRa-LBD by Biacore assay. The peptide binds strongly to RXRa-LBD in the presence of 9-cis-RA with a K d of 320 nM (Fig. 4e), which is in the range of coregulator peptide binding to nuclear receptor 37 .
The role of the 161 LRQLL 165 motif was also illustrated by the enhancing effect of 9-cis-RA on p85a-BCR interaction with either RXRa-LBD or tRXRa (Fig. 4f). Furthermore, 9-cis-RA-induced RXRa-LBD interaction with p85a-BCR was inhibited by the LxxLL-containing BCR peptide conjugated with the cellpenetrating peptide derived from trans-activator of transcription (TAT), similar to the effect of K-80003, but not by the corresponding mutant peptide (Fig. 4g). Substitution of L164 and L165 in 161 LRQLL 165 motif with A also abolished the interaction of BCR with RXRa-LBD (Fig. 4h). Mutating W305 critical for the formation of the coactivator-binding groove 12 to Q impaired the binding of RXRa-LBD with p85a-BCR either in the presence or absence of 9-cis-RA ( Fig. 4i), even though the same mutation had no effect on RXRa-LBD interaction with RXRa-A/ B (Fig. 3i). The interaction of the LxxLL-like motif in p85a with the coactivator-binding groove of tRXRa is biologically relevant as TNFa-induced activation of AKT in cells transfected with tRXRa and p85a was inhibited by cotransfection of BCR but not BCR mutant, similar to the inhibitory effect of K-80003 ( Supplementary Fig. 7a). Exposure of cells to BCR peptide also resulted in a similar inhibition ( Supplementary Fig. 7b). Altogether, these results demonstrate that the LxxLL motif in p85a can bind to the coactivator-binding groove of RXRa in analogy to the binding of transcriptional coactivators.
To further characterize the 'coactivator-like' binding of p85a, we studied the requirement of AF2/H12 in tRXRa, which is essential for the formation of the coactivator-binding groove 12,38 , for tRXRa interaction with p85a. As previously reported 29 , tRXRa interacted with the full-length p85a in the presence of TNFa. Removing AF2/H12 from tRXRa (tRXRa/DAF2) abolished its interaction with p85a (Fig. 5a). Interfering AF2/H12 activity by transfecting RXRa-A/B capable of binding AF2/H12 also inhibited 9-cis-RA-induced interaction of tRXRa with p85a-DiSH2 (Fig. 5b). These results further support the 'coactivatorlike' binding of p85a. Because the coactivator-binding groove of RXRa was shielded in the tetrameric form of RXRa-LBD ( Fig. 1), our observation that the LxxLL-like motif in p85a interacted with the coactivatorbinding groove of RXRa implied that the p85a-binding activity of tRXRa is impaired in its tetrameric form. To address this, we evaluated the binding of p85a by two tRXRa mutants, tRXRa/ F313A and tRXRa/R316E, which exist in different oligomeric forms 25 . RXRa/F313A is a transcriptionally constitutively active mutant receptor with H12 adopting an active conformation to form the coactivator-binding groove 39 . The mutant does not exist as a tetramer as its H12 is not available to contribute to the formation of tetramers 25 . In contrast, tRXRa/R316E, a transcriptionally inactive mutant, exhibited predominantly as a tetramer 25 . In agreement with previous observations 25 , our crosslink experiments detected a strong tetrameric species of tRXRa/ R316E but not tRXRa/F313A independent of the presence of 9-cis-RA or K-80003 (Fig. 5c). When both mutants were analysed for their interaction with p85a by coIP experiments, we found that tRXRa/R316E could not bind to p85a either in the absence or presence of TNFa or 9-cis-RA, whereas tRXRa/F313A showed a constitutive and 9-cis-RA-independent interaction with p85a compared to tRXRa (Fig. 5d). K-80003 also failed to modulate the interaction of both mutants with p85a (Fig. 5e). Immunostaining revealed lack of colocalization of the tRXRa/ R316E mutant with p85a in the absence or presence of TNFa (Fig. 5f). To further determine the inhibitory effect of tRXRa tetramerization, we examined the interaction of p85a-DiSH2 with tRXRa tetramers crosslinked by BS3. Cells transfected with HA-p85a-DiSH2 and Myc-tRXRa were treated with 9-cis-RA and/or K-80003, and subsequently exposed to BS3. 9-cis-RAinduced tRXRa interaction with p85a-DiSH2 was potently inhibited by BS3 that stabilized tRXRa tetramers (Fig. 5g). Furthermore, the monomeric form but not the tetrameric form of RXRa-LBD interacted with p85a-BCR (Fig. 5h). These results together with our structural information demonstrate that K-80003 inhibits tRXRa interaction with p85a by promoting tRXRa tetramerization that masks the p85a-binding region on tRXRa.
Tetramerization regulates tRXRa subcellular localization. The interaction of tRXRa with p85a occurs in the cytoplasm 29 . We next determined whether K-80003-induced stabilization of tRXRa tetramers could modulate its subcellular localization. When transfected into cells, both tRXRa and RXRa resided mainly (480%) in the nucleus. However, upon TNFa treatment Figure 3 | Induction of RXRa tetramerization by K-80003 and its regulation by the N-C intramolecular interaction. (a) Equal amounts of purified RXRa-LBD or mutant protein were incubated with DMSO, 9-cis-RA, and/or K-80003, and separated by non-denaturing polyacrylamide gel electrophoresis followed by Coomassie Bright Blue staining. The percentage of tetramer and dimer of RXRa-LBD or mutants was quantitated by densitometric analysis of the corresponding blots. One of four similar experiments is shown. (b) RXRa-LBD incubated with K-80003 or 9-cis-RA was subject to gel filtration chromatogram assay. Results showed that 9-cis-RA-induced RXRa-LBD was mostly in dimer (D), while K-80003-induced RXRa-LBD was mostly in tetramer (T). One of three similar experiments is shown. (c) HepG2 cells transfected with RXRa, tRXRa or RXRa-LBD were treated with 9-cis-RA or K-80003. Cell lysates prepared were then subjected to BS3 crosslinking, and analysed by western blotting using DN197 anti-RXRa antibody. One of more than five similar experiments is shown.   tRXRa was found in the cytoplasm of cells (465%), colocalizing extensively with p85a. In contrast, TNFa had no effect on the nuclear localization of RXRa. When cells were cotreated with K-80003, TNFa-induced colocalization of tRXRa with p85a in the cytoplasm was inhibited, resulting in tRXRa nuclear localization (Fig. 6a). The effect of K-80003 was likely due to its binding to tRXRa as the cytoplasmic colocalization of p85a with tRXRa/L433D mutant defective in K-80003 binding was not affected by K-80003. Thus, TNFa-induced cytoplasmic localization of tRXRa was likely due to its cytoplasmic retention by TNFa-activated p85a through protein/protein interaction, suggesting that tRXRa tetramers incapable of binding p85a might reside in the nucleus. To address this, extracts prepared from cells transfected with tRXRa were subjected to crosslinking by BS3. Nuclear and cytoplasmic fractions were then prepared and analysed. Western blotting showed that K-80003-stabilized tetrameric form of tRXRa was found exclusively in the nuclear fraction, while tRXRa monomer was distributed both in the nuclear and cytoplasmic fractions (Fig. 6b) nuclear, likely resulted from tRXRa dissociation from p85a or other cytoplasmic proteins.
To study whether the anti-tumour effect of K-80003 could be attributed to its effect on the subcellular localization of tRXRa in vivo, we stably transfected tRXRa or RXRa into MCF-7 cells, and the resulting stable clones (Supplementary Fig. 8a) were inoculated into nude mice. Overexpression of tRXRa but not RXRa in MCF-7 breast cancer cells enhanced AKT activation in vitro (Supplementary Fig. 8b) and promoted the growth of MCF-7 tumour in animals, which was suppressed when animal were treated with K-80003 ( Fig. 6c; Supplementary Fig. 8c). Inhibition of the growth of MCF-7/tRXRa tumour by K-80003 was accompanied with reduced AKT activation and enhanced PARP cleavage (Fig. 6d). To determine whether the anti-cancer activity of K-80003 was associated with its modulation of tRXRa subcellular localization, tumour specimens from nude mice were analysed by RXRa immunostaining (Fig. 6e). While RXRa was nuclear, tRXRa was predominantly cytoplasmic. However, a significant amount of tRXRa was found in the nucleus when animals were administered with K-80003. We also used polyomavirus middle T antigen (PyMT) transgenic mice 40 to study the anti-cancer effect of K-80003 and its modulation of the subcellular localization of tRXRa, which was highly expressed in PyMT mammary tumour developed in these mice ( Supplementary Fig. 9a). K-80003 potently inhibited the growth of PyMT mammary tumour in this animal model (Fig. 6f), accompanied with induction of PARP cleavage and inhibition of cyclin D expression (Fig. 6g) as well as inhibition of tumour cell proliferation ( Supplementary Fig. 9b). Immunostaining of PyMT tumour specimens using D197 anti-RXRa antibody that recognizes both RXRa and tRXRa revealed a predominant cytoplasmic RXRa staining (Fig. 6h). In contrast, RXRa staining was mainly found in the nucleus when mice were dosed with K-80003. Thus, K-80003 induction of tRXRa nuclear localization through its modulation of tRXRa tetramerization likely represents a major mechanism by which the compound exerts its potent therapeutic effect.

Discussion
RXRa is unique in that it can form not only homodimers and heterodimers but also homotetramers [4][5][6][7][8][9][10][11]17,25,33,38 , suggesting that the equilibrium between these different oligomeric states plays a role in regulating RXRa functions. While the function and mechanism of RXRa homodimer and heterodimers are well documented, the role of RXRa homotetramers and its regulation remains poorly defined. We report here that K-80003 represents a unique RXRa ligand capable of shifting the oligomeric equilibrium by selectively binding to and stabilizing the tetrameric conformation. Our crystal structure of RXRa-LBD in complex with K-80003 shows that six molecules of K-80003 bind in a tetramer by a unique combination of distinct and canonical binding mechanisms acting like a '3-pronged' binding mode. The comprehensive interactions between the bound K-80003 molecules and the protein provide a plausible explanation for how K-80003 binding can stabilize the tRXRa tetrameric state. Biochemical studies confirmed that K-80003 binding could promote tetramerization of tRXRa but not full-length RXRa. K-80003-induced stabilization of tRXRa tetramer was associated with inhibition of its interaction with p85a and activation of tRXRa-dependent PI3K/AKT signalling. The tetramerization also represents a crucial event in determining the subcellular localization of tRXRa. Collectively, these results unravel a critical role of tRXRa tetramerization in regulating its subcellular localization and nongenomic interaction with cytoplasmic signalling proteins.
Although how tRXRa binds to p85a remains to be precisely defined, our data demonstrated that an LxxLL motif in p85a and the amino acid residues at the coregulator-binding groove of RXRa are critical for their interaction (Fig. 4). These data in combination with information from comparing the RXRa-LBD dimer and RXRa-LBD/K-80003 tetramer crystal structures demonstrate that the potential p85a-binding site on tRXRa resides in the coregulator-binding groove. In the K-80003stabilized RXRa-LBD tetramer, H12 protrudes into the coregulator-binding groove of the neighbouring monomer ( Fig. 1), making the groove inaccessible for p85a binding ( Supplementary Fig. 2). Modulation of gene transcription by nuclear receptors is mediated through their interaction with coactivators or corepressors via the LxxLL motif and the L/IxxI/ VI motif, respectively 10,13-16 . Thus, our identification of a functional LxxLL motif in p85a suggests that p85a may serve as a nongenomic coregulator of tRXRa in analogy with the regulation of nuclear receptor transactivation by coregulators in the nucleus 10,[13][14][15][16] . The genomic and nongenomic coregulators of RXRa may act coordinately to mediate the crosstalk between RXRa nuclear and cytoplasmic signallings. It is noteworthy that several nuclear receptors, including oestrogen receptor, glucocorticoid receptor, peroxisome proliferator-activated receptors, thyroid receptor and retinoic acid receptor could interact with p85a [41][42][43][44][45][46][47][48][49] . Whether p85a serves as a common coregulator for different nuclear receptors to modulate PI3K signalling remains to be seen. The concept of cytoplasmic nuclear receptor coregulators will expand our conventional view of coregulators and may lead to the identification of a class of cytoplasmic signalling proteins that mediate the nongenomic action of RXRa and perhaps other nuclear receptors.
Interaction between the N-and the C-terminal segments of a protein appears to be evolutionarily selected for some functional Figure 5 | Tetramerization of tRXRa prevents its interaction with p85a. (a) AF2/H12 is required for binding p85a. HEK293T cells transfected with HA-p85a and Myc-tRXRa or mutants were treated with TNFa (10 ng ml À 1 ) for 1 h, and analysed by coIP assay using anti-Myc antibody. (b) Regulation of tRXRa interaction with p85a by ligand and RXRa-A/B region. HEK293T cells transfected with the indicated HA-p85a-DiSH2, Myc-tRXRa and RXRa-A/B were treated with or without 9-cis-RA (10 À 7 M) for 6 h, and analysed by coIP assay using anti-Myc antibody. (c) Tetramerization of tRXRa/R316E and tRXRa/F313A was analysed in MCF-7 cells transfected with tRXRa/R316E or tRXRa/F313A, which were then treated with 9-cis-RA (10 À 7 M) or K-80003 (5 Â 10 À 6 M) for 6 h. Cell lysates were subjected to BS3 crosslinking and analysed by western blotting using DN197 anti-RXRa antibody. (d) Interaction of tRXRa/R316E and tRXRa/F313A with p85a. HEK293T cells transfected with the indicated expression plasmids were treated with TNFa (10 ng ml À 1 ) or 9-cis-RA (10 À 7 M) for 1 h, and analysed by coIP assay using anti-Myc antibody. (e) Characterization of p85a interaction with tRXRa/R316E or tRXRa/F313A. HEK293T cells transfected with the indicated expression plasmids were treated with 9-cis-RA (10 À 7 M) or K-80003 (5 Â 10 À 6 M) for 6 h., and analysed by coIP assay using anti-HA antibody. (f) Mutation of R316 impairs tRXRa cytoplasmic localization. HEK293T cells cotransfected with Myc-tRXRa/R316E and HA-p85a were pretreated with K-80003 (5 Â 10 À 6 M) for 3 h before exposed to TNFa (10 ng ml À 1 ) for 30 min. Cells were immunostained with anti-Myc and anti-p85a antibody, and visualized by confocal microscopy. Scale bar, 10 mm. (g) Tetramerization of tRXRa impairs its interaction with p85a. HepG2 cells transfected with Myc-tRXRa together with HA-p85a-DiSH2 were treated with K-80003 and or 9-cis-RA. Cell lysates were then subjected to BS3 crosslinking, and analysed by western blotting using anti-Myc antibody. (h) Tetramerization of RXRa-LBD impairs its interaction with p85a-BCR. HEK293T cells transfected with Myc-RXRa-LBD together with HA-p85a-BCR were treated with K-80003. Cell lysates were subjected to BS3 crosslinking, and analysed by western blotting using anti-HA antibody. For western blotting, one of three or four similar experiments is shown. advantages 50 . Although the N-terminal A/B domains of nuclear receptors are intrinsically disordered, studies have now shown that they are capable of undergoing a disorder-to-order transition upon binding specific target molecules 3 . Thus far, the N/C intramolecular interaction has been described for several nuclear receptors, including oestrogen receptor 51 , progesterone receptor 52 , peroxisome proliferator-activated receptors 53 and androgen receptor 34 . We report here that N/C interaction also occurs in RXRa and acts to mediate RXRa interaction with p85a, defining a regulatory paradigm for RXRa action. Although how the N/C interaction in RXRa is mediated remains unclear, our inspection of the sequences ( 61 mgppfsvisspmgphsmsvp 80 ) at the N-terminal A/B domain of RXRa (Fig. 3g) revealed the existence of potential polyproline II (PPII) domain characterized by a PxxP core motif known to bind to SH3, WW and EVH1 domain 54 . This is reminiscent of previous reports that proline-rich nuclear receptor co-regulatory protein 2 (PNRC2) uses a SH3 domainbinding motif (  nuclear receptors including oestrogen receptor, glucocorticoid receptor, progesterone receptor, thyroid receptor, retinoic acid receptor and RXR 55 , and that a unique interaction motif exists in the ligand-binding domain of oestrogen receptor for binding PPII-like motif 56 . Although AF2/H12 at the C-terminal end of RXRa is required for the N/C interaction (Fig. 3h), the coactivator-binding groove involving the AF2/H12 was not required (Fig. 3i). This is supported by the result that the N/C interaction was inhibited by 9-cis-RA (Fig. 3e). How the AF2/H12 engages the N/C interaction remains to be determined. The N-terminal A/B domain through intramolecular interaction with the C-terminal domain may allosterically affect the function of distant domains to ensure that RXRa activity is appropriately achieved. As the disruption of the N/C interaction has been implicated in the regulation of nuclear receptor activity, ligand sensitivity and subcellular localization 34,35,[51][52][53] , our illustration of the existence of the N/C intramolecular interaction in RXRa provides a molecular explanation for the oncogenic effect of tRXRa in tumour cells 29 . It is noteworthy that in addition to various proteolytically cleaved RXRa products [57][58][59][60][61][62][63][64] , RXRa N-terminal splicing variants have been identified 65,66 . Furthermore, the N-terminal A/B domain is enriched with phosphorylation sites and several kinases have been identified to phosphorylate the region 67,68 . RXRa migrates from the nucleus to the cytoplasm in response to differentiation 69 , survival 29,70 , apoptosis 71 and inflammation 29,70,72 . It remains to be determined if and how the N/C interaction in RXRa plays a role in mediating the crosstalk between RXRa and signal transduction pathways under both physiological and pathophysiological conditions. In conclusion, our results elucidate a previously unrecognized role of RXRa tetramers and demonstrate that conformational selection plays a critical role in the regulation of the nongenomic function of RXRa. We showed that the tetramerization of RXRa could be regulated by several mechanisms including ligand binding, intra domain-domain interaction and nongenomic interaction with cytoplasmic signalling proteins. This study opens an opportunity to develop novel RXRa-based therapeutics by selectively stabilizing a particular oligomeric state. The selective induction by K-80003 of the tetramerization of tRXRa but not RXRa suggests that this class of compounds may preferentially affect tRXRa activity, which is therapeutically desirable since tRXRa is often overproduced in cancer cells 29 .

Methods
Protein expression and purification. The human RXRa LBD (residues Thr223 to Thr462) cloned as an N-terminal histidine-tagged fusion protein in pET15b expression vector was expressed in Escherichia coli BL21 strain (Stratagene). After sonication, cell extract was incubated with the His60 Ni Superflow resin, and Histagged RXRa-LBD-resin complexes were washed, eluted and concentrated to 5 mg ml À 1 for subsequent trails 73,74 . For crystallization experiment, bovine thrombin (Sigma) was used to cleave the His tag from the purified His-tagged RXRa-LBD. The resulting His tag was removed on the Resource-Q column (GE) using 0.1-1 M NaCl gradient and the Tris-Cl pH 8 buffer. RXRa-LDB was further purified using gel filtration on a Superdex 200 2660 column (GE) pre-equilibrated with the 75 mM NaCl, 20 mM Tris-Cl buffer (pH 8).
Crystallization and structure solution. Crystallization conditions were initially determined using the sitting-drop vapor-diffusion method and the crystallization screens Index and PEG-Ion solutions from Hampton Research. Final crystals were obtained by cocrystallizing the protein and the ligand. The protein-ligand complex contained 0.35 mM of RXRa-LBD, 0.5-0.7 mM of a ligand, 1% DMSO, 100 mM NaCl and 20 mM Tris-Cl buffer (pH 7.6). 0.2 ml of the protein-ligand complex was mixed with 0.2 ml of the well solution (20% PEG3330 and 0.2 M Na Acetate) and incubated at 20°C. Crystals appeared in 5-10 days and grew into 0.2 Â 0.1 Â 0.1 mm prisms. The crystals were flash-cooled in mother liquor containing 1.5 mM of K-80003 and 20% glycerol as a cryoprotectant. Diffraction data was collected from the cryo-cooled crystal (100 K) on an in-house diffractometer (Rigaku X-ray generator with a rotating anode, VariMax optic and R-AXIS HTC detector). Data was processed using the programme iMOSFLM, part of the CCP4 suite 75 . The b-factor of the data set according to the Wilson plot was 60 Å 2 .
The crystal structures were determined using the molecular replacement program Phaser 76 with pdb entry 4N8R as an initial model. Coot 77 and the program suite Phenix 78 were used for the model rebuilding and refinement. Parameter files and initial models for the ligands were prepared by eLBOW of Phenix. The data collection and refinement statistics are summarized in Table 1. Overall, 96.5% of residues in the refined structure are in favored regions of the Ramachandran plot, and no residues fall into disallowed regions. Coordinate errors, as estimated by Phenix, is 0.37 (Å).  Supplementary Table 1 for primer sequences).
Peptide synthesis. Peptides were synthesized on MBHA resin using Fmoc synthesis and DIC/HOBt coupling with an Advanced Chem Tech 350 and 396 multiple peptide synthesizer 79 . Cell-penetrating peptide derived from TAT (GRKKRRQRRRPPQ) was conjugated with p85a-peptide 79 . All peptides were acetylated on their N termini, and all were amidated on their C termini. Standard deprotection conditions were used for all peptides. Peptides were purified by HPLC on C18 columns and confirmed by MALDI mass analysis.
Gel filtration. Gel filtration was carried out using an AKTA FPLC system with HiLoad 16/600 Superdex 200 pg (GE Healthcare Life) pre-equilibrated with Binding buffer (50 mM Tris-Cl (pH 7.4), 200 mM NaCl) at a flow rate of 1 ml min À 1 . Purified RXRa-LBD incubated with K-80003 or 9-cis-RA for 2 h was Figure 6 | Effect of tetramerization on the subcellular localization of tRXRa. (a) Effect of TNFa and K-80003. MCF-7 cells cotransfected with Myc-RXRa, Myc-tRXRa, tRXRa/L433D and p85a were pretreated with or without K-80003 (5 Â 10 À 6 M) for 3 h before exposed to TNFa (10 ng ml À 1 ) for 30 min. Cells were immunostained with anti-Myc and anti-p85a antibody, and their subcellular localization revealed by confocal microscopy. (b) Tetrameric tRXRa resides in the nucleus. HEK293T cells cotransfected with Myc-tRXRa were treated with or without K-80003 (5 Â 10 À 6 M) for 6 h. Nuclear (N) and cytoplasmic (C) fractions were prepared, subjected to BS3 crosslinking, and analysed by western blotting using anti-Myc antibody. The purity of fractions was examined by analysing the expression of nuclear PARP and cytoplasmic a-tubulin in non-crosslinked fractions. One of three similar experiments is shown. (c) Effect of K-80003 on the growth of MCF-7 cells in mice. Nude mice injected with MCF-7 cells stably transfected with control vector, Myc-RXRa, or Myc-tRXRa were administered with K-80003 (20 mg kg À 1 ) for 12 days. *Po0.05; **Po0.01. (d) Effect of K-80003 on PARP cleavage and AKT activation in MCF-7 xenograft tumours. Lysates prepared from tumours from nude mice treated with vehicle or K-80003 were analysed by western blotting. (e) K-80003 alters the subcellular localization of tRXRa in MCF-7 xenograft tumour cells. Tumour sections prepared from nude mice treated with vehicle or K-80003 were immunostained with anti-Myc antibody. (f) Effect of K-80003 on the growth of MMTV-PyMT mammary tumour. Four-week old MMTV-PyMT mice were fed with or without diet containing K-80003 (100 mg kg À 1 ) for 4 weeks, and the appearance of tumour was determined. **Po0.01. (g) Effect of K-80003 on PARP cleavage and cyclin D1 expression in MMTV-PyMT mammary tumour cells. Lysates prepared from tumours from MMTV-PyMT mice fed with vehicle or K-80003 were analysed by western blotting. (h) K-80003 induces RXRa nuclear localization in MMTV-PyMT tumour cells. Tumour sections from MMTV-PyMT mice fed with or without K-80003 (100 mg kg À 1 ) for 4 weeks were immunostained with DN197 anti-RXRa antibody. Scale bar, 10 mm. then subject to gel filtration chromatogram assay. Elutes of RXRa-LBD protein were separated by 8% non-denaturing PAGE followed by silver staining.
Chemical crosslinking. Cells were transfected with RXRa or mutant expression vectors for 24 h and then exposed to 9-cis-RA (10 À 7 M) or K-80003 (5 Â 10 À 6 M) for 6 h. Cell lysates were then prepared and incubated with bis[sulfosuccinimidyl] suberate (BS3, Pierce) dissolved in dimethyl sulfoxide to a final concentration of 1 mM. The reactions were stopped with protein gel loading buffer. Samples were resolved by SDS/PAGE in 8% acrylamide gels and protein bands were visualized by western blotting.
Cell culture and transfection. MCF-7 breast cancer cells, HepG2 liver cancer cells, A549 lung cancer cells, and HEK293T human embryonic kidney cells (from ATCC) were cultured in Dulbecco's modified Eagle's medium containing 10% fetal bovine serum. The cells were maintained at 5% CO 2 at 37°C. Subconfluent cells with exponential growth were used throughout the experiments. Cell transfections were carried out by using Lipofectamine 2000 (Invitrogen) according to the instructions of the manufacturer.
Compound binding. Binding of compounds to RXRa-LBD or mutant was studied in phosphate buffer at 25°C using isothermal titration calorimetry (ITC) assay that measures the heat either released or absorbed during binding. The purified His-tagged RXRa-LBD or mutant protein (50 mM in 25 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.1% DMSO) was placed in the sample cell of MicroCal VP-ITC titration calorimeter. The compounds to be studied were diluted to a concentration of 1 mM in the same buffer. All solutions were degassed before the titrations. Titration was carried out using a 2-ml compound with injection time 4 s and a 120 s delay between each injection. The heat of dilution was obtained by injecting compounds into the same buffer and subtracted from the reaction before the fitting process. Calorimetric data were analysed using MicroCal Origin software (version 7.0).
Coimmunoprecipitation assay. For coIP assay 29 , HEK293T cells grown in 10 cm dishes were transfected with various plasmids. Cells were lysed in 1 ml of lysis buffer (20 mM Tris (pH 7.5), 150 mM NaCl, 1% Triton X-100, 1 mM EDTA, 30 mM NaF, 2 mM sodium pyrophosphate) with a cocktail of proteinase inhibitors (Roche). Lysates were incubated with the appropriate antibody for 12 h at 4°C and subsequently incubated with protein A-Sepharose beads for 1 h. The immunoprecipitates were collected and washed three times with lysis buffer. The protein-antibody complexes recovered on beads were subjected to immunoblotting using appropriate antibodies after separation by SDS-PAGE. Input represents 5% of cell lysates used for coIP assays.
Confocal microscopy. Cells mounted on glass slides or tumour tissue frozen sections (5-mm-thick) were washed with PBS and fixed in 4% paraformaldehyde (PFA) for 15 min and permeabilized with PBS containing 0.1% Triton X-100 for 15 min. Fixed cells or tumour tissues were blocked with 5 mg ml À 1 BSA in PBS for 30 min at room temperature, followed with incubation with primary antibodies for 3 h and secondary antibodies at room temperature for 1 h and co-stained with 4 0 6 0 -diamidino-2-phenylindole (DAPI) to visualize nuclei. The images were taken under an LSM-510 confocal laser scanning microscope system (Carl Zeiss).
Animal studies. For nude mice xenograft study, Myc-RXRa and Myc-tRXRa cloned into pcdna3.0 vector were stably transfected into MCF-7 cells, and the resulting stable clones and control clone (2 Â 10 6 cells in 100 ml) were injected subcutaneously into nude mice (BALB/c, SPF grade, 16-18 g, 4-5-week old). For drug treatment, mice were intraperitoneally injected (i.p.) daily with K-80003 (20 mg kg À 1 ) diluted in Tween-80 or vehicle (tween-80). Body weight and tumour size were measured every 2 days. Mice were killed after 12-day drug treatment and the tumours removed for various assessments. For MMTV-PyMT transgenic mouse study, 4-week old female mice of a transgenic mouse on the FVB/N genetic background expressing the PyMT oncogene under the control of MMTV LTR promoter 40 were fed with diet containing with or without K-80003 (100 mg kg À 1 ) for 4 weeks. Estimation of tumour appearance was performed by palpation once every day. Tumour tissues of left thoracic mammary glands were excised after the treatment. Portion of tissues was used to prepare extracts for western blotting, while the rest was fixed in 4% phosphate-buffered paraformaldehyde, and used for immunohistochemical staining using anti-Ki67 antibody and immunostaining using D197 anti-RXRa antibody. All experimentations and animal usage were performed and approved by the Animal Care and Use Committee of Xiamen University.
Data analyses. Data were expressed as means±s.d. from three or more experiments. Statistical analysis was performed using Student's t-test. Differences were considered statistically significant with Po0.05.
Data availability. Coordinates and structure factors for the complex of RXRa-LBD/K-80003 have been deposited in the Protein Data Bank under accession code PDB 5TBP. The rest of the data that support the conclusions of this study are available from the corresponding authors upon reasonable request.