Allosteric Conversation in the Androgen Receptor Ligand-Binding Domain Surfaces

Joint BSC-IRB Research Programme in Computational Biology (S.G., J.F.-R.), Life Sciences Department, Barcelona Supercomputing Center, 08034 Barcelona, Spain; Institute of Biomedicine of the University of Barcelona (IBUB) (L.R.C., V.B., E.E.-P.), University of Barcelona (UB), Barcelona Science Park, 08028 Barcelona, Spain; Genomic Medicine (P.N., J.D.B., P.W.), The Methodist Hospital Research Institute, Houston, TX; and Androgen Signalling Laboratory (G.B.,C.B.), Department of Surgery and Cancer, Imperial College London, London W12 0NN, United Kingdom

Androgen binding is known to trigger widespread structural and dynamic alterations within the AR LBD, although detailed structural data are missing.By analogy with other NR LBDs, unliganded (apo-) AR LBD may feature a dislodged helix 12 (H12) adopting an unstructured molten globule organization.Upon admission of the hormone into the core of the LBD, the overall stability of the domain is increased, achieving a more defined structure (21).The best-described allosteric rearrangement in NRs upon ligand binding takes place with the conformational change of H12 that completes AF-2 (22).
AR LBD is subject to mutations in advanced PCa and AIS (Androgen Receptor Gene Mutations Database: http://androgendb.mcgill.ca).PCa mutations often result in increased transactivation or expanded ligand binding preference (5,7,8,10,23,24).Conversely, AIS mutations usually reduce AR activity and cause varying degrees of fertility problems and undervirilization (4,6,9,13,25).Whereas AR mutations that arise in both diseases commonly affect known functional regions of the protein, including the ligand-binding pocket and AF-2 surface, many others affect regions of the AR surface with no assigned function, implying that they disrupt as yet undefined aspects of AR activity.
AR is the pharmacological target for antiandrogens used in PCa treatment.Current PCa clinical treatments involve combinations of androgen-deprivation therapy and antiandrogens (e.g.bicalutamide) that inhibit AR action by competing for androgen binding and displacing H12 to prevent formation of a productive AF-2 pocket.Unfortunately, prolonged antiandrogen treatment results in emergence of hormone-refractory PCa with poor prognosis; incompletely defined mechanisms result in reactivation of AR in the absence of androgens and presence of antiandrogens (12, 26 -29).
We have proposed that the AR surface may harbor attractive sites for intervention with small molecules (30 -33).AR AF-2 undergoes subtle induced fit rearrangements upon coactivator binding and several residue side chains (e.g., K720, M734, and M894) move to create hydrophobic subpockets that bind apolar side chains of coactivator NR boxes and can deepen further to accommodate bulky hydrophobic side chains of W/FxxLF motifs that characterize AR LBD binding peptides within the NTD-and AR-specific coactivators (19,20).These surface cavities are attractive targets for small molecules and, because AR is reactivated in recurrent PCa, such small molecules could inhibit growth of both early-stage PCa and late-stage hormone-refractory forms of the disease (30 -33).In a previous study, we screened chemical libraries for small molecules that inhibit coactivator binding to AF-2 (30).Surprisingly, several compounds, including Triac and members of the fenamic acid series of antiinflammatories, inhibited AR/LxxLL peptide interactions but preferentially were localized by x-ray crystallography to a distinct surface, binding function 3 (BF-3) (30); BF-3 which is topologically adjacent to, but distinct from, AF-2, displays characteristics of a protein-binding site, and is target for PCa and AIS mutations (30 -32).Site-TABLE 1. Summary of the experimental results obtained in vitro using AF-2 transcriptional activation and mammalian two-hybrid assays with AR LBD WT and mutant variants, and computational parameters from their corresponding MD simulations.directed mutagenesis of BF-3-lining residues confirmed its modulating role in AR activity (30).Furthermore, comparisons of AR LBD crystal structures with and without Triac at BF-3 suggested that compound binding triggers allosteric alterations that propagate to AF-2 and inhibit coregulator binding (30).Thus, x-ray structures suggest that BF-3 and AF-2 pockets are structurally coupled and that allosteric communication between them exists, but the succession of conformational changes and function of these effects are not clear.

Luciferase
Here, we employed a combination of in vitro transactivation assays, mammalian two-hybrid assays with AR LBD, and computational molecular dynamics (MD) simulations to understand how mutations in residues in or near BF-3 may influence AR function and dynamics and how allosteric communication between BF-3 by AF-2 may take place.Our data show that mutations in BF-3 act as allosteric elicitors of conformational changes that are transmitted towards AF-2, and that this allosteric communication affects AR LBD function as experimentally shown in vitro.Furthermore, a series of residues from BF-3, the boundary of BF-3/AF-2, and AF-2 are structurally interconnected and allosterically coupled.Moreover, our data indicate that several residues belonging to BF-3 and AF-2 surface pockets are key players of an allosteric network that may influence multiple aspects of AR LBD function.

Western blot analysis
For AR Western blot analysis, HeLa cells were transfected with 1 g of AR LBD expression constructs as described above.Cells were washed with cold PBS 24 h after hormone treatment, lysed with cold RIPA buffer (150 mM NaCl, 1% Triton X-100, 0.1% sodium dodecyl sulfate, 0.5% sodium deoxycholate and 50 mM Tris, pH 7.5) supplemented with complete protease inhibitors (Roche Diagnostics, Indianapolis, IN) and centrifuged for 10 min at 14,000 rpm.Lysates were boiled and loaded onto a NuPAGE 4 -12% Bis-Tris gel (Invitrogen, Carlsbad, CA), after which proteins were transferred to a polyvinylidene difluoride membrane.Western immunoblotting was performed using a polyclonal antibody against the C-terminus of AR (C-19, Santa Cruz Biotechnology, Inc., Santa Cruz, CA), followed by incubation with a horseradish peroxidase-conjugated mouse antirabbit antibody (Abcam, Cambridge, MA).AR LBDs were visualized by a reaction with Luminata Forte Western HRP substrate (Millipore).

Preparation of AR LBD input structures
We used the x-ray structure of AR LBD in complex with the hormone dihydrotestosterone (DHT) (PDB code 1T7T; 1.7 Å resolution) (19) as starting template to model all the mutants described herein, using the AMBER module LEAP.The PDB file was converted into an AMBER structure file using LEAP.The preparation of the files for the hormone DHT was done with the AMBER modules ANTECHAMBER and LEAP.
As a first preparation step, the crystallographic water molecules were removed from the structure.Then a quick minimization step was performed to optimize the structure within the force field in vacuum.Later on, each AR LBD structure to study was placed in a periodic truncated octahedron box of TIP3P water molecules (the minimum distance between protein and walls of the box was set at 12 Å).Na ϩ and Cl Ϫ counterions were then added to the solvent bulk of protein/water with LEAP to neutralize the system and achieve 150 mM NaCl concentration.

System setup and simulation protocol
We performed the MD simulations using the force field AMBER parm99 of the AMBER10 package (34,35).Our equilibration protocol for each MD simulation started by 1 psec minimization with a 50 kcal/(mol ⅐ Å 2 ) harmonic potential to restrain the protein atoms to minimize the solvent molecules.Then, we applied 40 psec of MD simulation at constant pressure (1 atm) with a 12 Å nonbonding cut-off distance during which the temperature was raised from 50 to 300K using Langevin dynamics, applying restraints of 25 kcal/(mol ⅐ Å 2 ) to all the protein atoms.After this, we ran 40 psec MD simulation with restraints of 10 kcal/(mol ⅐ Å 2 ), and another 40 psec MD simulation with restraints of 5 kcal/(mol ⅐ Å 2 ) on all the protein atoms.Then, we applied 20 psec MD simulation with 5 kcal/ (mol ⅐ Å 2 ) restraints only to the protein backbone atoms, and another 20 psec MD simulation with 1 kcal/(mol ⅐ Å 2 ) restraints on the protein backbone atoms as well.Finally, we ran 60 psec MD simulations in which all the atoms of the system were able to move freely.After the above-mentioned system equilibration step, 40 nsec of MD simulation were run for each mutant/complex at constant pressure (1 atm) and a temperature of 300 K.The particle mesh Ewald method was used, with a collision frequency of 0.2 psec Ϫ1 excluding bonds involving hydrogen atoms.

B-Factors derived from MD simulation
We calculated the temperature factors (B-factors) from the MD simulation, using the ptraj AMBER tool (35).B-Factors were computed as atomic positional fluctuations multiplied by 8/3 2 and then mass weighted and averaged for each residue.For the sake of comparison, the original B-factors derived from x-ray diffraction data in the AR LBD PDB 1T7T were also mass weighted and averaged by residue.

Analysis of cross-correlation matrices
Correlation matrices, representing all the residue-residue pair-wise root mean square deviation correlations along the complete MD trajectory, were computed using the ptraj AM-BER tool (35).The standard convention was used with a positive value between 0 and 1 reflecting correlated motion and a negative coefficient between 0 and Ϫ1 reflecting anticorrelated motion.Atoms with correlated motions move in phase whereas atoms with perfect anticorrelated motion move in antiphase.

Accessible surface area calculations
The hormone DHT accessible surface area (ASA) along the simulations was computed with the ICM-Browser program (www.molsoft.com),based on the center of a spherical probe of 1.4 Å radius rolling over the structure of AR bound to the hormone (36).This parameter gives a measure of the solvent accessibility of DHT, which is buried within the LBP.

Analysis of surface cavities along MD simulations
To identify the evolution of surface cavities or ligand pockets during the MD simulations, we used the fpocket program and its module MDpocket developed to track the persistence of pockets within MD trajectories (37,38).MDpocket ran fpocket iteratively on 100 trajectory snapshots (extracted every 400 psec) to compute ␣-spheres (defined as those in contact with four atoms without containing any internal atom inside).Then the density of the conserved positions of the ␣-spheres during the trajectory was calculated.High-density regions corresponded to stable and well-defined cavities whereas lower densities indicated transient pockets.To visualize all volumes detected by MDpocket for each trajectory, we used an isovalue of 2.19, which selects the top 1% of the detected volume for the WT density data.

Contacting pairs and distance calculations
We calculated all contact pairs (residue-residue minimal interatomic distance Ͻ4 Å) formed by residues in the peptidebinding region (i.e.those at a distance Ͻ4 Å from the ARA70 peptide in the structure 1T5Z after superimposing AR coordinates) (20), in the snapshots generated every 400 psec from the MD trajectories.We compared the frequency of such contact pairs in the mutants with respect to the WT.

BF-3 mutations alter AR AF-2 activity
To address functional effects of different BF-3 mutations, we tested in vitro the transactivation activity of agonist-bound WT AR LBD (WT AR LBD) and nine AR LBD mutants (Fig. 1) and performed in parallel an exhaustive comparison of dynamics of WT AR and mutant ARs using MD simulation (see Figs. 2-5 and SI).The chosen mutated residues are located either lining the BF-3 pocket (I672, F826, N833, R840) or at the boundary between AF-2 and BF-3, hence part of both pockets (R726, N727) (Fig. 1, A and B).Additionally, we chose for control purposes a mutation (V757A) located at the end of H5, hence distant from both studied pockets (Fig. 1A).The selected dataset of mutations studied herein includes: I672R (http://androgendb.mcgill.ca),V757A (39 -40), and R726L (41-46), associated with PCa; F826L (47) and N727K (48), found in AIS patients, as well as mutations that have not been associated with pathology (F826R, R840A, R840E, and N833R) (Fig. 1, A  and B, and Table 1).We have not studied mutations F826L, N727K, and R840A before (Fig. 1 and Table 1).WT and all mutant AR LBDs investigated exhibit comparable levels of expressed protein as assessed by Western blot (Fig. 1C).
We observe that mutants I672R and N833R behave as super-AR variants, as earlier reported (30), as well as the new mutant R840A.Mutants N727K and F826L also moderately enhance AR AF-2 activity although without exhibiting super-AR behavior.Also along previous observations, F826R and R726L marginally reduce, and R840E totally abolishes AR LBD activity in vitro (Fig. 1D and Table 1).

BF-3 mutations affect AR LBD activation by GRIP1
We additionally investigated the functionality of the studied AR LBDs by measuring AR AF-2 transactivation activity in vitro in the presence of coactivator and by addressing AR LBD interaction either with its N-terminal domain or corepressors using a mammalian two-hybrid experiment.We first determined effects of the coactivator GRIP1 on WT and mutant AR activity (Fig. 1E).GRIP1 is the mouse orthologue of the human protein transcription intermediary factor 2, which is known to interact with the AR AF-2 pocket in a hormone-dependent manner (49).GRIP1 enhances activity of all the mutants, but there are variations in the extent of GRIP1 potentiation (Fig. 1E and Table 1).GRIP1 enhances activity of super-mutants I672R and R840A similarly to WT, but fold increase in activity of supermutant N833R by GRIP1 is lower.GRIP1 also enhances activity of PCa mutants V757A and R726L, AIS-associated mutations F826L and N727K, although to a lesser degree and rescues activity of the medium-inhibiting mutant F826R, resulting in the highest fold increase among all the studied AR LBD mutants.Even the very weak activity of the R840E mutant is enhanced by GRIP1.Thus, some BF-3 mutations (notably N833R and F826R) alter the extent of GRIP1 coactivation, but none abolish GRIP1 interaction.

BF-3 mutations affect the N/C interdomain interaction
To assess whether mutations at the BF-3 pocket affect AR LBD/NTD (N/C) interaction, a mammalian two-hybrid was performed (Fig. 1F and Table 1).Again, some of the BF-3 mutants altered AR N/C interaction, but the pattern of effects was different from that with GRIP1.All three supermutant AR LBDs, I672R, N833R, and R840A, show a decreased interaction with the NTD as compared with WT.Mutants R726L and V757A, associated with PCa, display a WT or moderately higher increase in their interaction with the NTD.Medium-inhibiting mutant F826R, which exhibited the highest increase in activity in the presence of coactivator, also exhibits the largest fold induction with the NTD.Mutants N727K and F826L, associated with AIS, show mild impairment in NTD interaction, superactivating mutant N833R exhibits the lowest capacity for N/C interaction, and activity of the severely impaired mutant R840E is not enhanced by NTD overexpression, unlike the case with GRIP1.

BF-3 mutations alter AR interaction with N-CoR and SMRT
Because DHT-liganded AR also interacts weakly with NR corepressors (50), we used a mammalian two-hybrid assay to assess how BF-3 mutants affect this interaction.The chosen corepressors were the silencing mediator of retinoid and thyroid receptors (SMRT) and the nuclear receptor corepressor (N-CoR) (Fig. 1, G and H).
As seen with GRIP1 and the AR NTD, BF-3 mutants broadly affect the capacity of AR to bind corepressors (Fig. 1, G and H, and Table 1).Both supermutants I672R and R840A exhibit a significant decrease in their ability to interact with the receptor-interacting domains of corepressors SMRT and N-CoR, being R840 the one that disrupts such interactions the most among all the AR LBD herein studied.V757A, F826L, and the severe mutant R840E also reduce corepressor interaction.Supermutant N833R and mutant F826R show impaired interaction with SMRT, but maintain a WTlike interaction with N-CoR, whereas N727K decreases interaction with N-CoR but leaves interaction with SMRT unaffected.Finally, R726L has little effect on corepressor interaction in these assays.

AR LBD MD simulations
To understand the structural and dynamic effects of the above-described mutations, we first analyzed WT AR LBD by running MD simulation for 40 nsec in explicit solvent, after an initial step to allow equilibration of AR LBD.The WT AR LBD structure appeared stable along the dynamics, and there was no evidence of large-scale reorganization or denaturation processes (Fig. 2, A  and B).
The overall apparent mobility of WT AR LBD amino acids along the MD trajectory resembled that of the AR LBD x-ray structure PDB 1T7T (19).We estimated residue B-factor values by computing fluctuations of each residue along the MD simulation.Figure 2C shows these values against B-factor values derived from electron density uncertainties in the AR LBD structure (19).In general, B-factors derived from MD simulation were similar to the crystallographic ones, implying that our simulation produces a reasonably accurate representation of the molecular motions that are detected in the crystal although the most N-terminal helix of AR LBD (H1) and amino acids 687-695 (H1-H3 loop) appear more flexible in the simulation whereas two highly flexible regions [amino acids 844 -850 (H9-H10) and 880 -886 (H11 and H11/12 loop)] appear more rigid than equivalent regions in the x-ray structure.Interestingly, the B-factor measurements flag the H3-H4/5 loop and the S3-H9 loop as relatively flexible compared with other regions of the AR LBD; this was seen both in MD simulations and in B-factors derived from the x-ray structure (Fig. 2C and Supplemental Fig. 1 for mutant B-factors fluctuations).Both regions are located in the boundary of AF-2 and BF-3 pockets, suggesting a possible dynamic link between these binding surfaces (Fig. 2D).
Next, we studied the effects of mutations on AR LBD dynamics, by obtaining 40 nsec MD trajectories for each of the mutants (Supplemental Fig. S2).Interestingly, consideration of mobility of the entire LBD reveals that mutations I672R, F826L, and N727K, which enhance AR LBD activity, exhibit larger average B-factors (Table 1) indicative of higher mobility.This is not true for supermutants N833R and R840A.On the contrary, inhibitory AR mutations generally display smaller average B-factors, indicative of reduced mobility (Table 1).Correlation of B-factors with activity is even more striking when mobility of only the BF-3 residues is considered (again, excluding N833R and R840A).There is no similar correlation between mobility of the AF-2 residues and activity (Table 1).

AR mutations reveal allosteric coupling between BF-3 and AF-2
To analyze how specific AR mutations affect dynamics of BF-3 and other regions of the LBD and pinpoint possible allosteric effects, we computed the motion correlation of all AR residues against each other along the MD trajectories, shown as correlation matrices (see Supplemental Fig. 3 and Materials and Methods).The crosscorrelation plots for WT activating (I672R, F826L) and inhibiting (R840E) mutations are shown in Supplemental Fig. 3.The values for WT AR LBD (Supplemental Fig. 3) show regions that are correlated (in red) or anticorrelated (in blue) along the MD trajectory.There is little obvious coupling between different regions of the WT AR LBD.Interestingly, the correlation matrix for AR I672R shows significant differences: there is stronger correlation between motion of residues 672-673 (H1, the mutation site), residues 710 -740 (H3 and loop 3/4, mostly AF-2 and the boundary region of AF-2/BF-3) and residues 820 -840 (the BF-3 lining loop S3/H9 and H9).In the case of F826L, we observe, in general, less motion correlation between regions than with I672R, but there is significant correlation between the region of the mutated residue (H9, close to BF-3) and the adjacent AF-2 pocket and this is stronger than WT.Increased correlation of mobility of BF-3 and AF-2 is not seen with the N833R and R840A mutants.In the case of R840E mutant, there is less motion correlation and no clearly correlated regions that can be highlighted.Reductions in correlation between mobility of the BF-3 and AF-2 pockets are also seen with other BF-3 mutants that reduce AR LBD activity.
Calculation of correlation coefficients (r 2 ) along the MD trajectories between the residues forming BF-3, AF-2, and LBP (where DHT is bound) sites underscores impressions from the mobility correlation matrices (Table 1).Activating mutants (with the exception of N833R and R840A) have BF-3/AF-2 correlation r 2 more than 0.7 (significantly higher than WT), whereas inhibiting mutants have r 2 around 0.5-0.6,smaller than that of the activating mutants, but still higher than WT.Thus, there is increased allosteric coupling between BF-3 and AF-2 when the transcriptional output is enhanced activation.
We do not observe any correlation between AF-2 or BF-3 and residues that line the LBP for WT AR or any of the analyzed AR mutants.Curiously, if we exclude the most activating mutants (N833R, R840A, I672R, and F826L), there is some correlation between the average DHT ASA relative to WT and activity (Supplemental Fig. 4).This raises the possibility that enhanced plasticity of LBP observed during MD simulations may favor AR activity.

BF-3 mutations change dynamics of AF-2
Because the results obtained from our MD simulations suggest close structural communication between BF-3 and AF-2 surface pockets, we analyzed effects of BF-3 mutations on AF-2 conformation in more detail.To do this, we computed all possible contact pairs of AF-2 residues that exhibited at least a 15% reduction or increase in their pair-wise interaction frequency relative to WT AR during simulations (see Materials and Methods for details).Figure .3 and Supplemental Figs. 5 and 6 show evolution of contacts between selected residue pairs along the simulations.
BF-3 mutations result in significant changes in pairwise interactions within AF-2 residues over time.For activating mutants I672R and F826L, the proximity between the AF-2 key residues K720-M734 is greatly reduced relative to WT AR or R840E (Fig. 3).In contrast, K720 (one of the charge clamp AF-2 residues) forms more extensive contacts with H729, N727, L728, and V730.Among other AF-2 residues, M734 establishes frequent contacts with L712, M894, and I898, whereas V730 forms more frequent contacts with V716 in the context of both activating mutants.Effects of the AR R840E (inhibitor) mutation are opposite to that of the activating mutants.Specifically, pair-wise interactions between K720 and M734 are enhanced, whereas interactions between R726, K720, Q733, and V730 appear less frequently, Mol Endocrinol, July 2012, 26 (7):0000 -0000 mend.endojournals.org7

F3
V730 remains further away from K720 and V716, and contacts between K720 and Q733 are less prevalent (whereas contacts between K720 and H729, N727, or L728 are never formed, as in WT).
The net effect of these changes in pair-wise interactions is that there are alterations in the frequency of formation of AF-2 subpockets that accommodate bulky side chains of F 1 xxLF 5 peptides present in AR-specific coactivators (Supplemental Fig. 7).In the x-ray WT AR structure (PDB 1T7T) (19), K720 and M734 are in contact, closing the entrance for the second F side chain (at position 5) of the FxxLF motif (Fig. 4A).During the WT MD simulation, these residues open frequently to create an organization that resembles that of AR in complex with the ARA70 FxxLF motif (PDB 1T5Z) (20) (Fig. 4B).In the I672R mutant simulations, K720 and M734 separate even more widely and frequently, opening a larger hydrophobic pocket (Fig. 4E), an effect that can also be seen in the other activating mutant F826L (data not shown).By contrast, in the AR R840E (inhibitor) mutant, the arrangement of K720 and M734 along the simulation resembles that of the original AR x-ray structure PDB 1T7T (19), and the deep AF-2 subpocket fails to open (Fig. 4F), unlike the dynamic opening and closing of the subpocket seen in simulations with WT AR (Fig. 4D).Thus, MD simulations seem to suggest that BF-3 mutations alter the propensity of AF-2 to form subpockets that accommodate FxxLF peptides; activating mutations enhance sub-pocket formation, whereas the inhibitory mutation reduces subpocket formation.

Allosteric paths on AF-2
We examined snapshots of the MD simulations to search for dynamic structural rearrangements that could be responsible for the allosteric transmission of information from BF-3 to AF-2 (Fig. 5).Analysis of the simulations of activating AR mutants I672R and F826L revealed a large conformational change in the H3-H4/5 loop (residues 723-734) toward the end of the simulation (see a conformational snapshot of I672R in Fig. 5B).Movements of this loop relative to the initial position (Supplemental Fig. 8) were more extensive than the ones seen in WT AR or any other mutant simulation [interestingly, the mutants with higher fold increase in activity in the presence of GRIP1 (V757A, R726L and, to a lower extent, F826R), have these loops more extensively moved than WT along the dynamics].Within the H3-H4/5 loop, there are striking changes in the spatial location of residues H729 and V730 (Fig. 3).A possible mechanism for this effect is that the I672R side chain interacts with E837 (located in H9 and forming a salt bridge with R840 in WT AR).This effect would alter H9 position and free the H3-H4/5 loop so that K720 (H3) can interact with H729 (H3-H4/5 loop).H729 moves from being in contact with a lysine residue (K822, loop S3-H9) to contacting K720 backbone as a result of the new loop conformation.
Similar analysis for F826L (Fig. 5E) also reveals opening of the H3-H4/5 loop.Early in the MD simulation, L826-N823 interaction breaks leading to distortion of the S3-H9 loop and breakage of K822-H729 contact.This effect, which is not seen in WT AR, leaves H729 free to interact with K720 and results in a H729-K720 conformation that is exactly the same as the one seen in the I672R simulation and is not found in any other mutant.By contrast, the K822-H729 contact is more stable in the simulations with the R840E inhibitory mutant relative to WT-AR and certainly much more stable than in activating mutants (Fig. 3), underscoring the potential importance of this interaction in BF-3/AF-2 communication.
These new conformations of the H3-H4/5 and S3-H9 loops (especially in the activating mutants) open a small hydrophobic cavity around Y739 (H4) that is adjacent to
Our studies confirm and extend our previous data, which show that BF-3 mutants alter overall AR activity in different, and hard to predict, ways.We show that mutants in BF-3 and residues that lie between BF-3 and AF-2 produce a range of effects from superactivators (I672R, N833R, R840A), moderate enhancers (F826L, N727K), weak inhibitors (F826R, R726L), and very potent inhib-itors (R840E).We have also shown a wide range of effects upon coregulator binding; BF-3 mutations do not completely inhibit functional interactions with GRIP1 or NTD but do affect coactivation differentially by both proteins.This stands in marked contrast to effects of mutations in the AF-2 surface that consistently block AR interactions with GRIP1, the AR NTD, and other coregulators.Finally, BF-3 mutations moderately inhibit corepressor interactions, but, here again, there are diverse effects with some mutations strongly inhibiting N-CoR and SMRT binding, others exhibiting milder effects or no effects, and some distinguishing between N-CoR and SMRT.
Our MD simulations do not systematically explain effects of all AR BF-3 mutations but do suggest a plausible general explanation for their diverse effects on AR activity; BF-3 is coupled to AF-2, and BF-3 mutations alter the propensity of the AF-2 surface to form deep subpockets that accommodate bulky side chains of coregulator motifs.MD simulation on AR LBD shows a dynamic link between BF-3 and AF-2, and two regions that form the boundary between AF-2 and BF-3 pockets play a key role in allosteric communication: the H3-H4/5 loop, where R726 and N727 are located, and the S3-H9 loop, where F826 resides.Interestingly, consideration of mobility of the entire LBD reveals that three mutations that enhance AR LBD in vitro activity (I672R, F826L, N727K), exhibit a greater mobility than WT.On the contrary, AR inhibitory mutations generally feature smaller mobility with respect to WT. Analysis of the flexibility per pocket indicates that BF-3 flexibility degree is inversely correlated to the AR LBD in vitro function, but there is no correlation with the AF-2 mobility.
BF-3 mutations induce conformational changes in several side chains of the adjacent AF-2 pocket.For activating mutants I672R and F826L, residues K720 and M734 stand out as already observed in the x-ray crystal structures of coactivator peptides bound to AF-2 (20).Pairwise residue contacts with several key AF-2 residues are manifested, specially implicating L712, H729, N727, L728, V730, and H12 M894 and I898.Effects of the AR R840E (inhibitor) mutation on the observed pair-wise contact formation are opposite to that of the activating mutants.The net effect of these changes in pair-wise in-  teractions is alterations in the frequency of formation of AF-2 subpockets that accommodate NR boxes present in coactivator peptides and the NTD domain.There is thus a shift in the conformational ensemble of the AF-2 groove.Residues K720 and M734, which are in a closed conformation in the x-ray WT AR structure without coactivator peptide (PDB 1T7T) (19), remain more frequently along the MD trajectory in the AF-2 open conformation, similar to the one observed in the crystal structure of AR bound to the ARA70 FxxLF motif (PDB 1T5Z) (20).In the I672R and F826L mutant MD simulation, K720 and M734 separate even more widely, frequently opening a larger hydrophobic pocket.However, K720 and M734 are found in the closed conformation along the MD trajectory in the inhibitory AR mutant R840E.Thus, MD simulation suggests that BF-3 mutations alter the propensity of AF-2 to reorganize its subpockets to accept the entering coactivator peptides.Whereas several activating mutations enhance the formation of subpockets, inhibitory ones reduce them.
Other paths may also be involved in the allosteric transmission of information from BF-3 to AF-2, and from/to other areas of the receptor.Our studies have evidenced possible roles for identified residues located in the H3-H4/5 and S3-H9 loops that are responsible for conformational changes and allosteric cross talk among AR LBD surfaces.These new conformations of the H3-H4/5 and S3-H9 loops (especially in the activating mutants) open a small hydrophobic cavity around Y739 (H4) that is adjacent to AF-2 and close to BF-3.Comparison of the organizations of this cavity reveals that it is closed throughout the simulations in the inhibitor mutation R840E.Interestingly, we observed weak binding of an apolar small molecule (salycylaldehyde) at this cavity adjacent to AF-2 in our initial screening of surface inhibitor compounds (PDB 2PIR) (30).Y739 is one of the residues that interacted with this surface binder drug, and, most importantly, K822 (loop S3-H9) was at a distance that allowed electrostatic stabilization of the drug (30).We call this groove "binding function 4" (BF-4).We did not observe correlation between AF-2 or BF-3 and the residues that line the LBP for WT AR or any of the AR analyzed mutants.However, it is interesting to point out that if the most activating mutants are excluded (N833R, R840A, I672R, and F826L), there seems to be a correlation between the average ASA of the LBP, where DHT is cocooned, indicating that an increase in LBP adaptability during the MD simulation may influence AR activity.
Our data indicate that two superactivating BF-3 mutations may work by a distinct mechanism.Our MD simulations evidence that N833R and R840A do not exhibit larger overall flexibility than WT AR, unlike I672R and other activating mutations, suggesting that they enhance AR activity via a mechanism that differs from allosteric communication.Because BF-3 exhibits characteristics of a protein interaction surface, it is attractive to speculate that both mutations could alter direct contacts with an unknown protein(s) that could either potentiate or silence AR function.Several lines of evidence suggest that BF-3  could be involved in protein contact.Equivalent regions of the thyroid hormone receptor (70) and the nuclear receptor LRH-1 bind to corepressor and the wnt-signaling dependent coactivator ␤-catenin (71).Furthermore, functional evidence links this region of AR to contacts with the chaperone FKBP52 (72).It is even possible that BF-3 might contact GRIP1 coactivator, the NTD, and corepressors and that these proteins have an interaction surface on AR LBD that extends beyond the AF-2 pocket toward BF-3 as we have previously suggested (30,33).

FIG. 1 .
FIG. 1.Mutations in AR LBD, AF-2 transcriptional activation, and mammalian two-hybrid assays.A, Simplified model representation of AR LBD structure.Helices H1, H3, H4 -5, H9, and H12 are depicted as gray cylinders.The AF-2 coactivator binding pocket is lined by H3, H5, and H12, whereas the BF-3 pocket is formed by H1, H9, and the loop linking H3 with H4 -5 (L3/4), which is shown as a thin gray wire.BF-3 residues studied herein are shown as green sticks.Key AF-2 residues implicated in engaging in important contacts with coactivator peptides as shown in previous crystal structures are highlighted in raspberry, and the charge clamp residues K720 and E897 are shown as rasperry sticks.The residues R726 and N727 belonging to the boundary between AF-2 and BF-3 pockets are shown as dark blue and green sticks, respectively.The residue V757 the mutation of which, V757A, has been associated with PCa is depicted as a blue stick located at the end of H5 and is surface exposed.B, Solidsurface representation of AR LBD in gray showing the residues lining BF-3 in green, the residues lining AF-2 in raspberry and residue R726 in blue.C, Western blot showing the protein expression level of all AR LBDs.D and E, AR AF-2 activity in the absence (D) or presence (E) of GRIP1 coactivator.WT AR LBD AF-2 activity is 1%.F-H, Mammalian two-hybrid assays with (F) AR NTD domain, and corepressors (G) N-CoR or (H) SMRT.HeLa cells were transfected with 100 ng DNA and treated with (D-F) 100 nM or 10 nM (G-H) DHT.Results are the mean of at least five independent experiments performed in triplicate.

FIG. 2 .
FIG. 2. Simulation of AR LBD WT and mutants.A, Evolution of root mean square deviation (RMSD) with respect to the initial structure along the MD trajectory.B, Evolution of global energy along MD trajectory.C, Comparison between experimental (in red) and AMBER-based simulated B factors (in green).For consistency, experimental B factors from PDB 1T7T were transformed to be comparable to simulated values (see Materials and Methods for details).AMBER B-factors per residue were computed from an ensemble of 200 frames along the MD trajectory selected every 200 psec.D, Simulated B-factors mapping on AR.The AR receptor B-factors are shown as worms with variable thickness and color according to their corresponding value (B-factors Ͻ 46 in blue; 46 Ͻ B-factors Ͻ 77 in white; B-factors Ͼ 119 in red).ns, Nanoseconds.

FIG. 4 .
FIG. 4. Surface residues of the AF-2 pocket.A, WT AR x-ray structure, PDB 1T7T.B, AR bound to ARA70 peptide x-ray structure, PDB 1T5Z.C, AR bound to ARA70 peptide, with peptide removed for the sake of clarity.D, WT AR during MD simulation (snapshot taken at 10.4 nsec, representative of the first third of the trajectory).E, I672R mutant during MD simulation (snapshot taken at 29.6 nsec, representative of the last half of the trajectory).F, R840E mutant during MD simulation (snapshot taken at 4 nsec, representative of the first half of the trajectory).

FIG. 5 .
FIG. 5. Conformational changes in WT AR and mutants during MD simulation.A, Detail of the H3-H4/5 loop in WT AR during MD simulation (in gray, snapshot taken at 10.4 nsec, representative of the first third of the trajectory) with respect to the x-ray structure in yellow.B, I672R mutant after MD simulation (in gray, snapshot taken at 29.6 nsec, representative of the last half of the trajectory) compared with WT AR x-ray structure (in yellow).C, Same loop detail in R840E mutant after MD simulation (in gray, snapshot taken at 4 nsec, representative of the first half of the trajectory) compared with WT AR x-ray structure (in yellow).D, Residue pairs in which contact frequency during MD simulation in I672R mutant significantly change with respect to the AR WT simulation; residues are represented as spheres, and the link is colored in red or green depending whether the contact frequency in the corresponding AR mutant MD simulation is significantly smaller or larger, respectively, than in WT AR MD simulation.E and F, Similar analysis for F826L mutant (E), and for R840E mutant (F).