The effect of S427F mutation on RXRα activity depends on its dimeric partner

RXRs are nuclear receptors acting as transcription regulators that control key cellular processes in all tissues. All type II nuclear receptors require RXRs for transcriptional activity by forming heterodimeric complexes. Recent whole-exome sequencing studies have identified the RXRα S427F hotspot mutation in 5% of the bladder cancer patients, which is always located at the interface of RXRα with its obligatory dimerization partners. Here, we show that mutation of S427 deregulates transcriptional activity of RXRα dimers, albeit with diverse allosteric mechanisms of action depending on its dimeric partner. S427F acts by allosteric mechanisms, which range from inducing the collapse of the binding pocket to allosteric stabilization of active co-activator competent RXRα states. Unexpectedly, RXR S427F heterodimerization leads to either loss- or gain-of-function complexes, in both cases likely compromising its tumor suppressor activity. This is the first report of a cancer-associated single amino acid substitution that affects the function of the mutant protein variably depending on its dimerization partner.


Introduction
Bladder cancer is the second most common urogenital cancer with 550 000 new cases every year worldwide. 1 Whole-exome sequencing of bladder cancer patients identied a hotspot mutation within the gene encoding the retinoid X receptor a (RXRa) involving Ser427 changing to Phe and less frequently to Tyr (RXRa S427F/Y ). 2 RXRa is a nuclear receptor of the vitamin A metabolite 9-cis retinoic acid (9-cis RA), and serves as an obligate homo-and hetero-dimerization partner for many subfamily 1 nuclear receptors, including the retinoic acid receptor a (RARa) and peroxisome proliferator-activated receptor g (PPARg). 3 These protein dimers can then bind to target response elements consisting of a direct repetition of a half-site motif (5 0 -AGGTCA-3 0 ) with an intervening spacer of 1-5 bp (DR1-5) leading to regulation of gene expression for various biological processes. 4 Thus, RXRa, as well as RARa and PPARg, function as ligandactivated transcriptional factors consisting of a ligand-binding domain (LBD) linked through a hinge region to a DNA binding domain (DBD). The LBD of these three receptors exhibits the canonical folding of nuclear receptors in which a single b-turn and 12 a-helices (H1-H12) enclose a hydrophobic binding pocket (Fig. S1 †).
Ligands control allosterically the interactions of nuclear receptors with coactivators and corepressors by altering the conformation of the short helix H12, also referred to as AF2activation function 2, at the carboxy-terminal end of the LBD. 5 In the absence of a ligand, H12 is in an open conformation that enables binding of co-repressors. Upon agonist binding, H12 undergoes a conformational change that leads to the formation of a novel surface to which coactivator proteins can dock through a short helix that contains the consensus LxxLL motif. 6 Recent studies have indicated that RXRa S427F allosterically activates the PPARg H12 helix in the absence of 9-cis RA. 7,8 This could explain the strong PPARg transcriptional signature in bladder tumors bearing RXRa S427F/Y mutations. 2 However, it is yet unknown how the S427F mutation on RXRa exerts its biological actions on other heterodimers, such as the RXRa-RARa as well as the homodimer RXRa-RXRa.
Examination of several crystal structures that exist in the literature reveals that the S427F mutation lies at the interface between homo-and heterodimer RXRa complexes. Thus, the obvious hypothesis for the S427F mutant mechanism of action is the alteration of its dimerization capability. In order to assess the effect of the S427F mutation on the ability of RXRa to dimerize with its obligate partners, we used a combination of biochemical experiments and Molecular Dynamics (MD) simulations to further elucidate the effect of the S427F mutation on the structure and dynamics of RXRa in its homodimeric and heterodimeric form with RARa and PPARg. Our results indicate a differential allosteric effect of the S427F mutation on the RXRa homo/heterodimers, which depends on the dimeric partner.

Mutation of Ser 427 of RXRa does not affect its DNA binding activity
The crystal structures of the full-length nuclear receptor complexes of RARb-RXRa and PPARg-RXRa show that the DBD and LBD regions of RARb and PPARg are physically connected through a DBD-LBD linker. Therefore, we rst asked whether the S427F mutation affects DNA binding. To investigate this, we expressed wild-type (WT) and mutant (S427F) RXRa in bacteria Escherichia coli tagged with glutathione S-transferase (GST) to facilitate protein purication. Puried recombinant proteins (RXRa WT and RXRa S427F ) were used in electrophoretic mobility shi assays (EMSA). In EMSA experiments, DNA binding proteins are co-incubated in vitro with a radiolabeled oligonucleotide containing cognate DNA binding sites, while the binding reaction is electrophoresed under non-denaturing conditions in a polyacrylamide gel. The protein-bound oligonucleotide migrates slower than the free one, generating a "gel shi" when exposed to an autoradiograph. For our experiments, as DNA target we used an oligonucleotide containing a DR1 site, which can be bound mainly by RXR homodimers (and RXR-PPARg heterodimers). As Fig. S2 † indicates, both RXRa WT and RXRa S427F bind equally well their cognate target sequence (lanes 1 and 6), indicating that the S427F mutation does not affect the DNA binding capacity of RXRa. The binding is specic, because it can be outcompeted by an excess (5Â and 25Â) of unlabeled DR1-containing oligonucleotides (lanes 2-3 and 7-8). Competition with oligonucleotides containing DR1 half-sites, which cannot be bound by RXR dimers, fail to compete for binding by RXRa WT (lanes 4-5). However, we cannot exclude the possibility of RXRa S427F binding as a monomer, as indicated by a moderate competition with unlabeled half sites (lanes 9-10). Mixtures of RXRa WT and RXRa S427F at different ratios do not affect binding efficiency, again supporting the conclusion that the DNA binding capacity is not affected by this point mutation.
Aer we conrmed that the S427F mutation does not affect DNA binding of RXRa, we focused on the LBDs of RXRa complexed with RARa or PPARg and of the RXRa homodimer. Through extensive, unbiased MD simulations spanning the ms timescale for both the wild type and mutant RXRa in its homodimeric and heterodimeric form with RARa and PPARg, we were able to determine mechanistic details of the effect of the S427F mutation on the activity of the different dimers and verify these with functional experiments.

S427F substitution leads to less active RXRa-RARa complexes
To assess the effect of the S427F mutation in the presence of an agonist, we ran the simulations in the absence (apo) and presence of 9-cis RA (holo) bound to the binding site of each monomer (Table S1 †). For these simulations, we used the crystal structure of the RARa in its agonist conformation in complex with RXRa in an inactive conformation (resembling its homotetrameric state), where H12 points to the solvent (PDB ID 3A9E). 25 Over the course of the apo RXRa WT -RARa simulations, the side chain of Ser427 engages predominantly in a hydrogen bond with the backbone of Pro423 ( Fig. 1A(a)), which lies on the RXRa-RARa interface, while it also forms hydrogen bonds occasionally with Arg348 of H7 (Fig. S3 †). The removal of the cocrystalized ligand leads to a gradual relaxation of the side chains of the binding-site residues in the void that the ligand leaves. Aer 300 ns, the conformation of the apo RXRa WT converges towards its auto-inhibitory conformation, in which H12 binds to the co-factor binding-site resembling PDB ID:1DKF (Fig. S4 †). Interestingly, Leu435 and Phe436 of H11 rotate with respect to the initial conformation in such a way that they pack themselves between H10 and H3, maintaining the relative position of the two helices ( Fig. 1A(c)). This movement underlines the regulatory role of H11, which can be positioned differentially by distinct ligands or the absence of them thereby, controlling the position of H3 and the packing of H12 in the different conformers.
Because Ser427 is located at the dimer interface and outside the ligand-binding pocket, it has been suggested that mutations of this residue are unlikely to alter the ligand binding properties of RXRa. 8 However, according to all our apo RXRa S427F -RARa simulations, H11 almost immediately rejoins with H10 ( Fig. 1A(b)) allowing H3 to collapse into the hydrophobic ligandbinding site (Fig. 1A(d)). Specically, the N-terminal part of H3 bends around H5, causing the binding site to be no longer ligand accessible. Then, H12 rotates gradually from the solvent back against the main body of the LBD and adopts a conformation similar to the one observed in the WT simulations. The closure of the binding site of RXRa due to the movement of H3 is reected in the decrease of the binding site volume (Fig. 1B), which can prevent the RXRa ligand 9-cis RA from binding to the cognate pocket under physiological conditions. The inability of RXRa S427F to bind 9-cis-RA was also conrmed using docking calculations (Table S2 and  Proline residues usually cause kinks in a-helices, which leave the i+4 backbone carbonyl without its typical hydrogen bond donor and have been reported to play functional roles in proteins. 9,10 In RXRa WT , the loss of the i+4 backbone hydrogen bond between Pro423 and Ser427 is compensated by a hydrogen bond with the side chain of Ser427 (Fig. S3 †), which stabilizes the kinked H10. However, the loss of Ser427-Pro423 interaction upon mutation causes a subtle tilt in the curvature of H10 of around 3 (Fig. S6 †), which triggers the rejoining of H11 with H10. Over the course of the RXRa S427F -RARa simulations, the side chain of Phe427 participates in a p-cation interaction with Arg348 of H7 ( Fig. S3 †), which increases the coupling of the motions of H10 and H7.
NMR resonances corresponding to residues within the apo-RXRa WT ligand-binding site and H12 surface have indicated that these regions exist as a dynamic ensemble of conformations, and binding of 9-cis RA to RXRa stabilizes the ligand-binding pocket and H12 surface. 11 Our holo RXRa/ LG100754-RARa/atRA simulations are in line with this observation, as the overall structure of the receptor stays intact throughout the simulations. In the case of RXRa WT -RARa, where the antagonist LG100754 is bound to RXRa WT , H12  (Table S1 †). The dashed line in the middle indicates the median of the distribution, while the other two dashed lines represent the interquartile ranges. Each RXRa WT monomer is colored with green, each RXRa S427F monomer is colored in grey and each RARa monomer in pink. (C) and (D) Boxplot indicating relative luciferase activity of the WT and mutant RXRa in heterodimers with RARa in response to 9-cis RA (C) and LG1002681 (D). In the Y axis is shown the normalized activity relatively to cells transfected with the DR5 reporter alone. * indicates p value < 0.05; ** indicates p value < 0.01; **** indicates p value < 0.0001. Student's t-test was used.
assumes an autoinhibitory position between H3 and H4 in the co-factor binding-site (Fig. S7 †). In the case of RXRa S427F -RARa, Leu455 of H12 interacts with Leu276 and Val280 of H3 (Fig. S7 †) and the position of H12 resembles an autoinhibitory position, even though it has not converged to the exact conformation during the simulation. Unlike the apo RXRa S427F -RARa, where residues H10 and H3 form direct interactions, in the holo state, the propoxy group of LG100754 extends between the two helices and maintains their relative position.
To explore the functional implications of the predictions that arose from the simulations, we employed a cell-based functional assay which relies on luciferase reporters with DR1 or DR5 elements controlling the rey luciferase gene expression. The activity in response to combinations of RARa, RXRa and PPARg can be quantitated in cell extracts in the form of luminescence (see below).
The RXRa-RARa heterodimer is considered nonpermissive, 12 indicating that the RARa binding of its cognate ligand, all-trans retinoic acid (atRA), is required for dimer activity. Unlike other nonpermissive dimers however, subsequent 9-cis RA binding to RXRa boosts the activity of RXRa-RARa complex by 20-30%. 13 If 9-cis RA cannot be bound by the RXRa in RXRa S427F -RARa heterodimers, then the expectation is that the transcriptional activity of the RXRa S427F -RARa heterodimeric complex in the presence of both ligands will be 20-30% lower than the RXRa WT -RARa one. Indeed, luciferase assays show that following concomitant treatment with atRA and 9-cis RA, the RXRa S427F -RARa heterodimeric complex leads to 40% lower luciferase activity (2.4Â vs. 4Â) in comparison to the RXRa WT -RARa complex (Fig. 1C). Because 9-cis RA can also bind RARa, we repeated the experiment using the synthetic rexinoid LG1002681, which binds exclusively to RXRs. Similar to 9-cis RA, LG1002681 also induced the activity of the RXRa WT -RARa complex by approximately 40% (7.5Â vs. 10.5Â). Contrary to this, LG1002681 failed to induce the transcriptional activity of the RXRa S427F -RARa complex beyond the activation achieved by atRA alone, strongly indicating that indeed, RXRa S427F cannot bind its ligand, at least not as efficiently (Fig. 1D). It should be pointed out that in a recent study, the activity of RXRa S427F -RARa was also shown to be lower than that of RXRa WT -RARa heterodimer; however, the authors did not report it as a statistically signicant nding. 7 S427F substitution leads to more stable but less active RXRa homodimers Because RXRa mutations are not homozygous while RXRa can form homodimers, it is expected that all three combinations of mutant and WT homodimers will coexist in mutant cells. To address the effect of the S427F mutation in RXRa homodimers, in our MD simulations we used the crystal structure of the LBD of RXRa bound to the rexinoid agonist BMS649 (PDB ID 1MVC) 24 and introduced the S427F mutation in one or both monomers.
It has been reported that the apo RXRa WT is found predominately as a tetramer, which dissociates to its monomers upon addition of 9-cis RA. 14 In a recent report, the RXRa S427F was found predominately in its monomeric form. 8 The absence of RXRa S427F homodimers was rationalized based on the crystal structure of the RXRa WT , which indicates that the bulky phenylalanine would disrupt the RXRa S427F homodimer interface. Our EMSA experiments (Fig. S2 †), however, suggest that this is not the case. Not only do we observe binding, but also the existence of two clearly distinct shis indicates the presence of two different DNA complexes (Fig. S2 †). It has been previously shown that, in the absence of its ligand, RXRa binds DNA preferentially as a tetramer and then as a dimer. 15 Thus, although we cannot exclude the possibility that RXRa S427F deviates from this pattern, we speculate that the observed shis can only correspond to RXRa homodimers and homotetramers.
Moreover, according to our simulations, Phe427 can be accommodated on the interface and also is in contact with Leu430 of the other monomer introducing a network of stabilizing hydrophobic and p-cation interactions between Phe427, Leu430, and Arg426 of the two monomers ( Fig. 2 and S8 †). Interestingly, Leu430 of RXRa is substituted with Ala in the case of RARa and Thr in the case of PPARg, neither of which have a side chain long enough to interact effectively with Phe427. Similar to the case of RXRa-RARa, the side chain of S427 in RXRa WT -RXRa WT and RXRa S427F -RXRa WT forms a hydrogen bond with the backbone of Pro423 and the side chain of Arg426 (Fig. S8 †). In fact, we observe that the interaction energy between the monomers of the RXRa WT -RXRa S427F or RXRa S427F -RXRa S427F dimers is stronger than that of the RXRa WT -RXRa WT (Fig. S9 †). The reported interaction energy between the monomers suggests that the mutation stabilizes the homodimer. Given that the mutation does not affect DNA binding efficiency (see above), we hypothesize that genomic RXR bindings sites will be more frequently occupied by the more stable RXRa WT -RXRa S427F and RXRa S427F -RXRa S427F dimers leading to reduced expression of the respective target gene.
Over the course of the simulations, we observe that the volume of the binding site in the RXRa S427F -RXRa S427F and RXRa WT -RXRa S427F dimers is affected (Fig. 2B), possibly impacting the binding of 9-cis RA in the pocket. The average volume of the binding pocket of each RXRa monomer for the three replica MD simulations of the RXRa-RXRa dimers (Table  S3 †) as well as the independent volume of each RXRa monomer in all replica MD simulations of the RXRa-RXRa dimers (Table  S4 †) can be found in the ESI. † Docking calculations (Table S5 and Fig. S10 †) performed on the RXRa homodimers show that ligand binding demonstrates different behaviour depending on the presence of the S427F mutation. Docking scores in RXRa WT -RXRa S427F showed that 9-cis RA can bind weakly to the RXRa WT binding pocket (À3.76 kcal mol À1 on average for the three replicas, Table S5 †), but cannot bind to the RXRa S427F counterpart at all (predicted DG > 0 kcal mol À1 ). On the contrary, in the RXRa WT -RXRa WT homodimer, 9-cis RA demonstrated negative docking scores for both monomers (À7.24 kcal mol À1 on average for the three replicas). Finally, 9cis RA binds only slightly to the binding pocket of one of the two RXRa S427F -RXRa S427F dimers (À1.88 kcal mol À1 on average, Table S5 †), while it does not bind to the other dimeric partner (predicted DG > 0 kcal mol À1 ). In summary, in both RXRa WT -RXRa S427F and RXRa S427F -RXRa S427F dimers, only one of the monomers can bind 9-cis RA, due to binding pocket volume changes. This implies that both dimers are partially active at comparable levels.
This is corroborated by our luciferase assays. As Based on the stoichiometry, in the wt and mutant co-transfection, the contribution of the wt : wt and mut : mut dimers represent each 25% of the observed luciferase activity, while the remaining 50% comes from the wt:mut heterodimer. Based on the activity in the wt and mut alone transfections, the wt : wt dimer would contribute 1.12 fold increase (25% of 4.48-fold), the mut : mut would add 0.48-fold (25% of 1.92), while the remaining 0.94-fold from the 2.54-fold in the wt : mut co-transfection would come from the 50% of wt-mut pair (1.88-fold increase for the 100%). This implies that the mut : mut and wt : mut dimers are equally active (1.92 vs. 1.88-fold). Collectively, the molecular  (Table S1 †). The dashed line in the middle indicates the median of the distribution, while the other two dashed lines represent the interquartile ranges. Each RXRa WT monomer is colored with green and each RXRa S427F monomer is colored in grey. (C) Boxplot indicating relative luciferase activity of the WT and mutant RXRa homodimers in response to 9-cis RA. In the Y axis is shown the normalized activity relatively to cells transfected with the DR1 reporter alone. * indicates p value < 0.05; ** indicates p value < 0.01. Student's t-test was used.
simulations and the luciferase experiments indicate that RXRa WT -RXRa S427F dimers show comparable activity with the RXRa S427F -RXRa S427F ones.
The different binding activity of 9-cis RA between RXRa WT -RXRa WT , RXRa S427F -RXRa S427F and RXRa WT -RXRa S427F homodimers could also be explained from a network of residues that connects Trp305 of both monomers. Trp305 belongs to helix H5 of RXRa and it has been previously shown to mediate contacts with the bound ligand in the active RXR WT homodimer. Moreover, in PPRAg/RXRa heterodimers, helix Η5 of RXRa is an important part of the LBD core that transmits PPARg allosteric signals from the dimer interface, that are induced from PPARg agonists to the RXRa ligand-binding pocket. Ser427 of the RXRa WT -RXRa WT homodimer was a part of the network of residues connecting the two tryptophans, while this was not the case for Ser427 of the RXRa S427F -RXRa WT dimer or for Phe427 of the RXRa S427F -RXRa S427F homodimer (Fig. S12 †). Thus, the disruption of this network in RXRa WT -RXRa S427F and RXRa S427F -RXRa S427F could elucidate a possible allosteric mechanism that affects the binding of 9-cis RA in the RXRa pocket.
Finally, data from the apo MD simulations show that at least 1.5 ms are required for equilibrating the RXRa homodimers. RMSD plots as a function of time show that RXRa homodimers as a whole remain stable over the course of the MD simulations both in the presence and absence of 9-cis RA (Fig. S12 †), however, the binding pockets of the apo RXRa S427F -RXRa WT homodimer sample different conformations over the rst 1.5 ms of the MD simulations (Fig. S13 †), indicating local instability. Indeed, in a recent study by Yang et al., 16 it is shown that the apo-RXRa LBD homodimer displays a single thermal unfolding transition much lower than the average value of small globular proteins. On the other hand, when 9-cis RA is bound, RXRa homodimers retain more stable binding sites (Fig. S14 †), also conrmed by differential scanning calorimetry and differential scanning uorimetry experiments showing an increase in the free energy of the complex as a result of a more favorable entropic change due to interactions between the rexinoid and hydrophobic residues in the binding pocket. 16 Finally, RXRa heterodimers with PPARg and RARa are more stable, with respect to the homodimer, as indicated by interaction energy plots (Fig. S15 †), where the homodimer interaction energy is signicantly smaller. As a result, PPARg and RARa might be RXRa preferential partners while waiting for the ligand.

S427F substitution leads to constitutively active RXRa-PPARg complexes
It has been demonstrated that the hotspot mutations S427F/Y in RXRa induce the activation of the PPRAg/RXRa pathway in bladder cancer, leading to suppression of cytokine secretion from cancer cells. 8 The crystal structure of the PPARg/RXRa S427F bound to 9-cis RA (PDB ID 5JI0) 1 suggests that the mutation does not affect the organization of the residues around the site of the mutation, but the aromatic interaction between RXRa S427F and the terminal tyrosine Tyr477 found in all PPARs is responsible for the activation of PPARg and stabilization of the heterodimer. 7,8 During the rst 80 ns of our RXRa S427F -PPARg simulation, PPARg Tyr477 forms a p-p interaction with Phe427 of RXRa S427F (Fig. S16 †), consistent with what has been observed in previous simulations of the same length. 7 Aer that and until the end of the simulation, the side chain of PPARg Tyr477 rotates and starts interacting with residues of H10 of PPARg. This rotation brings Asp475 of PPARg close to Arg348 of RXRa, which in turn forms a p-cation interaction with Phe427 (Fig.  S16 †). The interplay of these two states stabilizes the active conformation of H12 of PPARg.
H12 is a crucial helical component of the LBD of nuclear receptors because its ligand-induced repositioning creates the necessary surface for co-activator interaction and thereby generates the transcriptional activity. Therefore, the stabilization of its position in the active, co-activator-competent conformation through the direct interaction of Phe427 with Ty477 or through an indirect interaction of Phe427 with Asp475 via Arg348, is likely to tip the equilibrium of PPARg towards active conformations rather than autoinhibitory or corepressor-competent ones.
Sequence alignment of RARA with PPARG ( Fig. S17 †) shows that the C-terminal residue Tyr of PPARG corresponds to Glu in RARA, which would be unable to interact with Phe427 of RXRa. It is, therefore, not surprising that the effect of the S427F is not the same as the one seen when RXRa S427F is in complex with RARa. Moreover, unlike RXRa S427F -RARa and RXRa S427F -RXRa S427F , the introduction of S427F mutation does not alter the size or shape of the ligand binding pockets of either RXRa or PPARg within the same timescale of simulation (Fig. 3), highlighting yet another difference in the effect of the mutation with respect to different dimerization partners. It should be noted that the S427F does not seem to have a statistically signicant effect on the curvature of RXRa S427F H10 with respect to the WT (Fig. S6 †). However, RXRa S427F/WT H10 of RXRa-PPARg exhibits the smallest curvature among all examined dimers. We reasoned that Gln444 and especially Gln451 of PPARg H10, which interact with Arg426 and Glu434, respectively (Fig. S18 †), bring H10 of the two monomers together and suppress any effect of the mutation on the curvature (Gln444 is replaced by Ser and Gln451 is replaced by Glu in both RXRa and RARa).
In agreement with published 7 and our own simulations, our luciferase assays indeed support the hypothesis that the RXRa S427F -PPARg heterodimer is constitutively active. As Fig. 3C indicates, the RXRa S427F -PPARg heterodimer is transcriptionally at least 3Â stronger than the RXRa WT -PPARg one. Moreover, while 9-cis RA induces the activity of the RXRa WT -PPARg by a factor of 3, its effect on the mutant counterpart is negligible. This ligand-independent activity of the RXRa S427F -PPARg heterodimer might account for its presumptive oncogenic activity in human cancer. mutations in various cancer types, but more commonly in bladder cancer. 17 Recent studies have focused on the hotspot mutation RXRa S427 and have elucidated its putative oncogenic role through heterodimerization with PPARa. 7,8 RXRs, however, are obligatory partners for a number of nuclear receptors and can also form homodimers. By extending our studies to additional homodimeric and heterodimeric complexes (with RARa and PPARg), we uncovered that the functional consequences of the S427F mutation are not universal.
To our knowledge, this is the rst description of a naturallyoccurring cancer-associated single amino acid substitution that affects the function of the mutant protein variably, depending on its dimerization partner. Our experimental data show that the RXRa S427F mutation leads to responsiveness to RAR but not RXR ligands in RXRa S427F -RARa heterodimers, implying a loss-of-function effect of RXRa S427F . On the other hand, the same mutation is gain-of-function independently of RXRa or PPARg ligands in RXRa S427F -PPARg heterodimers. Finally, RXRa S427F behaves as a loss-of-function in response to RXR ligands in RXRa S427 -RXRa S427 homodimers.
Our simulations rationalize the experimental ndings for the three different dimers: (1) for the RXRa S427F -RARa heterodimer, the position of H11 of RXRa S427F and consequently of H3 can be affected by the presence of the S427F mutation. The mutation-driven rotation of H3 towards the ligand-binding pocket abolishes, in turn, the ability of RXRa S427 to bind 9-cis RA or other ligands. (2) The gain of function phenotype of the mutant RXRa S427F -PPARg heterodimer is manifested through the cooperative effect of two states that stabilize PPARg H12 in the active conformation. In one state, the terminal Tyr477 of PPARg interacts directly with Phe427 of RXRa S427 , while in the other Asp475 of PPARg H12 interacts with Arg348 of RXRa S427 , which interacts with Phe427. In the RXRa-RXRa case, the stronger interactions of the RXRa S427F -RXRa S427F and RXRa WT -RXRa S427F complexes indicate that these dimers are in a more stable conformation compared to the RXRa WT -RXRa WT one. Moreover, the volume change of the binding pocket in the RXRa S427F -containing complexes may affect favorable interactions with 9-cis RA rendering the complex less active, as indicated by functional assays. This was also conrmed from the docking calculations where one RXRa S427F monomer was able  (Table S1 †). The dashed line in the middle indicates the median of the distribution, while the other two dashed lines represent the interquartile ranges. Each RXRa WT monomer is colored with green, each RXRa S427F monomer is colored in grey and each PPARg monomer in yellow. (C) Boxplot indicating relative luciferase activity of the WT and mutant RXRa heterodimers with PPARg in response to 9-cis RA. In the Y axis is shown the normalized activity relatively to cells transfected with the DR1 reporter alone. * indicates p value < 0.05; ** indicates p value < 0.01. Student's t-test was used.
to partially bind 9-cis RA in both RXRa S427F -RXRa S427F and RXRa WT -RXRa S427F dimers. Thus, the RXRa-RXRa homodimer demonstrates a completely different behaviour, concerning the binding of 9-cis RA, compared to the RXRa-PPARg and RXRa-RARa heterodimers. These ndings indicate that the S427F substitution, contrary to being thought to disrupt the dimer interface where it resides, rather affects allosterically the binding pocket.
A rather unexpected experimental nding was that RXRa S427F -RXRa S427F and RXRa WT -RXRa S427F dimers show comparable transcriptional activity implying that the mutation is detrimental in RXRa WT -RXRa S427F dimers, which is not the case in the RXRa S427F -RARa heterodimers. It has been reported that the LBD of RXRa forms homodimers which are less stable than heterodimers with RARa, while mutations in the LBD region of RXRa close to S427 almost completely abolish RXRa homodimerization with no effect on heterodimers with RARa. 18 Moreover, ligand binding is crucial for the stability of RXRa LBD homodimers. 16,39 Therefore, the impact of the S427F mutation on the RXRa homodimers could be the result of both structural instability and reduced affinity to ligand binding due to pocket size changes.
RXRs are obligatory partners for a large number of nuclear receptors in various tissues. 17 The S427F mutation is more frequently identied in bladder cancer, in which retinoic acid signaling plays important roles in tissue homeostasis. 19 On the other hand, S427 RXRa mutations are also found in hepatocellular carcinoma and pancreatic adenocarcinoma 20 in which the Liver X receptor (LXR) is the preferential partner of RXRs. It would be thus interesting to investigate how the same amino acid substitution affects the activity of RXRa heterodimers with LXR or any other partner, and the functional implications of this in tissue homeostasis and disease. Retinoids have been long used as chemotherapeutics and chemopreventives in several tissues, including the urinary bladder. 21 This implies that activation of retinoic acid signaling suppresses tumorigenesis. On the other hand, expression data from bladder cancer tumors harboring RXRa mutations revealed increased expression of genes involved in adipogenesis and lipid metabolism, implying that S427 mutations cause constitutive activation. Moreover, genes upregulated in RXRa mutant, and particularly the RXRa S427 mutant tumors, are identied as PPARg targets. 22 Overall, these ndings imply that RXRa S427F mutant acts as an oncogene, a notion that contradicts the established role of retinoic signaling as a tumor suppressor. This apparent contradiction highlights the complex nature of nuclear receptor function which relies upon heterodimer formation. The RXR partner denes the target gene. Promoters containing DR1 sites can be bound by heterodimers with RARs, PPARs, HNF4, COUP-TF1 and RXR homodimers; DR2 is recognized by PPARs, RARs and RXRs; DR3 is bound by RXR heterodimers with vitamin D receptor (VDR); DR4 is targeted by RXR heterodimers with thyroid receptor (TR), LXR and RARs; and DR5 is recognized by heterodimers with RARs and Nurr77. 23 It is safe to assume that the protein abundance of RXRs as well as of all candidate partners within cells will likely affect whether an RXRa mutation will confer a selective advantage in the process of neoplastic transformation. Moreover, the net functional outcome of RXRa S427F could also be tuned by the effect of the mutation in the stability of each heterodimer. Our MD simulations showed that RXRa S427F -containing homodimers are more stable than the respective wt ones. This implies that the S427F mutation might entrap wt RXRa molecules in nonfunctional or partially functional dimers, thus depriving other nuclear receptors, such as PPARg, of their obligatory dimerization partner.
This study opens up new avenues to fully elucidate the role of S427 RXRa mutations in human cancer and the mechanistic aspects of its involvement in tumorigenesis. This work is a signicant step towards this goal because our results introduce for the rst time the concept of pleiotropy of an individual single amino acid substitution depending on the structural and possibly the cellular context. Given the fact that all transcription factors act as homodimeric, heterodimeric or even supramolecular complexes with other transcriptional regulators, this work raises the intriguing possibility that these mechanistic insights apply to other transcription factors as well.

Expression of recombinant human RXRa protein in Escherichia coli (E. coli)
The coding region of the human RXRa WT or RXRa 427F was subcloned into the pGEM-5X-1 vector. The BL21 + E. coli strain was used for protein expression. Single colonies were inoculated in 5 mL LB containing 100 mg mL À1 ampicillin overnight at 37 C and used as inoculum for 500 mL of LB containing 100 mg mL À1 ampicillin. Bacterial cultures were incubated with shaking at 37 C until they reached the mid-log phase of growth (A550 ¼ 0.5-1.0). Protein expression was induced with 1 mM IPTG for 3 h. Bacteria were harvested with centrifugation at 5000g and cell pellets were further processed according to the protein purication protocol or kept at À20 C until further use. GST-fused RXRa WT and RXRa S427F proteins were puried by affinity chromatography on Glutathione agarose using Protino® Glutathione Agarose 4B (Macherey-Nagel, REF. 745500.10) according to the manufacturer's protocol.

Electrophoretic mobility shi assay (EMSA)
One hundred pmol of radiolabeled probe and 1 mL of 1 mg mL À1 poly(dI-dC) were mixed with recombinant RXRa proteins in binding buffer (10 mM Tris HCl pH 8.0, 150 mM KCl, 0.5 mM EDTA, 0.1% Triton X-100, 12.5% glycerol, 0.2 mM DTT). The nal volume of the reaction was 20 mL. The binding reactions were incubated for 10 minutes at RT and 5 minutes at 37 C. The DNA-protein complexes were run on 5% polyacrylamide gels in 0.5Â TBE at 200 V. Polyacrylamide gels were pre-run at 200 V for 2 hours. Both pre-running and DNA-protein complexes electrophoresis were performed at 4 C. Aer the electrophoresis was completed, the gel was transferred to Whatman paper, covered with a plastic membrane and dried at 80 C in a vacuum dryer for 1-2 hours. The dried gel was exposed to an autoradiograph lm (Fujilm) at À80 C for 1-2 hours.

EMSA probe generation and labeling
To generate DR1-containing and control (half-site) probes for EMSA, we annealed the following oligonucleotides: Name sequence DR1_F: TCGAGGGTAGGGGTCAGAGGTCACTCGTCGA DR1_R: TCGACGAGTGACCTCTGACCCCTACCCTCGA Half-site_F: AGCTTGGCGCCAGGGGTCAGGTCAGAATT Half-site_R: AATTCTGACCTGACCCCTGGCGCCAAGCT Prior to annealing, the forward (F) oligonucleotide was radiolabeled with T4 polynucleotide kinase (New England Biolabs) according to the vendor's instructions.

Luciferase reporter assays
For luciferase assays, 3 Â 10 5 HEK293T were transfected with 500 ng of a reporter plasmid containing the luciferase gene under the control of DR1 or DR5 response elements. In experiments using the DR1-luc reporter, cells were co-transfected with a modied pLKO.1/IRESegfp vector expressing either RXRa WT or RXRa S427F under the control of a CMV promoter. In experiments using the DR5-luc reporter, cells were cotransfected with equal amounts of a pLKO.1/IRESegfp expressing either RXRa WT or RXRa S427F and RARa. In all experiments a plasmid expressing b-galactosidase was co-transfected in tracer amounts. Twelve hours aer transfection fresh medium supplemented with RXRa or RARa agonists was added. Cells were harvested 48 hours post transfection for luciferase assays and b-galactosidase colorimetric assays.

DNA binding properties of RXRa S427F
To assess the effect of the S427F mutation on DNA binding, we expressed RXRa 427F and RXRa WT in E. coli bacteria and puried the recombinant proteins. We used those in EMSA experiments with radiolabeled DR1-containing oligonucleotides. As Fig. S2 † indicates, both mutant and WT proteins bind the DR1 site with the same efficiency. In fact, we observe two different band shis (shis 1 and 2) possibly corresponding to dimeric and tetrameric complexes of RXRa. Bands disappear upon competition with unlabeled DR-containing oligonucleotides implying that the observed binding is absolutely specic.

MD simulations
We performed a series of molecular dynamics (MD) simulations under different mutation and ligated conditions. A list of all the performed simulations, along with the simulation time of each is given in Table S1. † The atomistic model of the LBD-LBD of the RXRa-RXRa homodimer was based on the crystal structure of the human RXRa LBD bound to the rexinoid agonist BMS649 (PDB ID 1MVC), 24 of the RXRa-RARa heterodimer on the crystal structure of the human RARa LBD bound to its natural agonist all-trans retinoic acid (atRA), and the mouse RXRa LBD bound to the rexinoid antagonist LG10074 (PDB ID 3A9E). 25 Modeling of LBD-LBD of the RXRa-PPARg heterodimer was based on the crystal structure of the human RXRa LBD bound to its natural agonist 9-cis RA, and the human PPARg LBD bound to the agonist rosiglitazone (PDB ID 1FM6). 26 Full details of protein modeling, protonation states, etc., can be found in the ESI. ‡ Atomic charges for the at-RA, 9-cis RA, and LG100754 ligands were calculated by restrained electrostatic potential (RESP) tting (RESP-A1A mode) at the Hartree-Fock level with a 6-31G* basis set, as applied in the R.E.D. IV server version 3.0. [27][28][29] The general amber force eld (GAFF) 30 was used for the bonded and non-bonded interactions of the ligands. The GROMACS v5.0.7 MD engine 31 was used for the simulations of the RXRa-RARa complexes, while GROMACS v.2018.6 was used for the simulations of the RXRa-PPARg complexes and GROMACS v.2020.4 was used for the simulations of the RXRa-RXRa complexes. We used the amber 99SB*-ILDN force eld 32 to describe the protein dynamics of the RXRa-RXRa and RXRa-RARa complexes, and the Amber03 force eld 33 for the RXRa-PPARg in accordance with the force eld that was used by Halstead et al. 7 Prior to MD simulations, all structures were subjected to 10 000 steps of energy minimization using the steepest descent algorithm, followed by position restraint equilibration rst in the NVT and then in the NPT ensemble for 150 ps, respectively. Once equilibrated at constant pressure, unbiased MD simulations were carried out in the canonical ensemble (NPT) with the atomic coordinates of the system saved every 10 ps. Production runs range from 370 ns to 2 ms and were performed in three replicas each for the apo systems and one replica for the holo systems. Long-range electrostatic interactions were treated using the particle-mesh Ewald scheme 33 with a grid spacing of 1.6Å, while a cut-off 10Å was applied for the van der Waals interactions. All bonds were constrained using the LINCS algorithm allowing for a time-step of 2 fs. The non-bonded potential energy functions were switched, with forces decaying between 0.8 and 1.0 nm. The Parrinello-Rahman barostat 34 maintained a target pressure of 1 bar isotropically with a time constant of s P ¼ 2 ps and compressibility of 4.5 Â 10 À5 bar À1 , while the Nosé-Hoover thermostat 35 was applied throughout all the simulations to keep the temperature at 310 K using a coupling constant of s T ¼ 0.5 ps. To obtain the representative structure of the equilibrated RXRa-RARa and RXRa-PPARg dimeric complexes, we clustered the conformations of each complex during the last 100 ns of each simulation, while we clustered the conformations of each RXRa-RXRa complex during the last 500 ns of each simulation. For this purpose, the GROMOS algorithm 36 of the gmx_cluster routine (GROMACS) was used. The natural RXRa agonist 9-cis RA was docked into both receptors of the RXRa-RARa and RXRa-RXRa representative structures using the default protocol of the Glide -Ligand Docking tool of the Schrödinger suite (Schrödinger, LLC, New York, NY, 2020). The natural RARa agonist atRA was docked into RARa receptor of the RXRa-RARa representative structures using the same docking protocol. The Epock VMD plug-in ref. 37 was used to calculate the accessible volume of the RXRa, RARa and PPARg binding sites. Kink Finder 38 was used to measure angles in helices. The Network-View VMD plug-in was used to perform the Dynamical Network Analysis method in order to construct network models obtained from our MD simulations. 37

Author contributions
I. G., M. P., P. K. and G. S. performed computational experiments. V. B. and P. V. performed molecular biology experiments. I. G., V. B., M. P., Z. C. and A. K. wrote and revised the manuscript. Z. C. and A. K. conceived the study and supervised the research.

Conflicts of interest
There are no conicts of interest to declare.