Identification of Small Molecule Ligand Binding Sites On and In the ARNT PAS-B Domain

Transcription factors are generally challenging to target with small molecule inhibitors due to their structural plasticity and lack of catalytic sites. Notable exceptions to this include a number of transcription factors which are naturally ligand-regulated, a strategy we have successfully exploited with the heterodimeric HIF-2 transcription factor, showing that a ligand-binding internal pocket in the HIF-2α PAS-B domain could be utilized to disrupt its dimerization with its partner, ARNT. Here, we explore the feasibility of directly targeting small molecules to the structurally similar ARNT PAS-B domain, potentially opening a promising route to simultaneously modulate several ARNT-mediated signaling pathways. Using solution NMR screening of an in-house fragment library, we previously identified several compounds that bind ARNT PAS-B and, in certain cases, antagonize ARNT association with the TACC3 transcriptional coactivator. However, these ligands only have mid-micromolar binding affinities, complicating characterization of their binding sites. Here we combine NMR, MD simulations, and ensemble docking to identify ligand-binding ‘hotspots’ on and within the ARNT PAS-B domain. Our data indicate that the two ARNT/TACC3 inhibitors, KG-548 and KG-655, bind to a β-sheet surface implicated in both HIF-2 dimerization and coactivator recruitment. Furthermore, KG-548 binds exclusively to the β-sheet surface, while KG-655 binds to the same site but can also enter a water-accessible internal cavity in ARNT PAS-B. Finally, KG-279, while not a coactivator inhibitor, exemplifies ligands that preferentially bind only to the internal cavity. Taken together, our findings provide a comprehensive overview of ARNT PAS-B ligand-binding sites and may guide the development of more potent coactivator inhibitors for cellular and functional studies.

Table S3 GAFF2 atomic charges derived for KG-655 drug ligand.Atom names in the structure of the ligand correspond to those used for ligand parameterization.Note that these atom names differ from labels by IPUAC locants in Tables S1 and S2.S1 and S2.
Figure S1.Evaluation of KG-548 and KG-655 ligand interactions with ARNT PAS-B using 1D 19 F NMR at 278K.1 and 298.1K.The surface-bound spectra of both ligands showed an additional peak between -61 and -61.5 ppm.The free ligand peak of KG-655 was also broadened in the presence of ARNT PAS-B, indicating a possible second binding mode.

Figure S2 .
Figure S2.Comparisons of the four Ile to Val mutants to the WT ARNT PAS-B protein. 15N/ 1 H-HSQC spectra of the four Ile to Val ARNT PAS-B mutants (250 µM) indicated these mutants were folded and adopted similar structures to the WT protein, as evident by good chemical shift dispersion and overlap with the spectrum of the WT ARNT PAS-B.

Figure S3 .
Figure S3. 13C/ 1 H-HSQC spectra of the ARNT PAS-B mutants compared with the wildtype (WT) protein.13C/ 1 H-HSQC spectra of the four Ile to Val ARNT PAS-B mutants (250 µM) were used to confirm the Ile methyl assignments.The methyl cross-peak locations of the mutated residues are marked with arrows.Peaks other than the mutated Ile residues showed good overlap with the spectrum of the WT protein.

Figure S4 .
Figure S4.I364V/I458V double mutation further disrupts surface binding. 19F-spectra of 1 mM KG-548 mixed with 250 µM ARNT PAS-B WT, I458V, or I364V/I458V double mutant, zoomed in on the bound-fraction of the ligand.Mutation I458V substantially decreased binding affinity, with the I364V/I458V double mutation resulting in the lowest amount of bound ligand.

Figure S5 .
Figure S5.Stereospecific assignments of Leu and Val methyl resonances.Pro-R methyl groups (g1 and d1 for Val and Leu, respectively) yield negative signals in this constant time (CT) -13 C/ 1 H HSQC experiment.Pro-S methyl groups (g2 and d2 for Val and Leu, respectively) yield positive signals.Pro-S methyl groups are the migrating methyl groups.When they are 13 C labeled, the adjoining carbons are highly unlikely also to be labeled.Signs of the signals are determined by checking the signs of the Met methyl peaks, which are positives here.

Figure S6 .
Figure S6.The surface and internal binding modes of KG-655 are independent of each other.A double-filtered HSQC-NOESY experiment was performed using KG-655 and ARNT PAS-B I364V/I458V double mutant.Heteronuclear NOE correlations between KG-655 and methyl groups of all of the internal-facing protein residues identified previously using the WT ARNT PAS-B remain unaffected.

Figure S7 .
Figure S7.Comparison between ARNT PAS-B apo and KG-548 bound states.Superimposition of the crystal structure of the ARNT PAS-B apo (PDB: 4EQ1, lime green) and KG-548 bound state (PDB: 8G4A, magenta), with residues within 5Å of the ligand shown in sticks.Residue Y456 moves away from KG-548 in the ligand-bound state to prevent clashes.R366 also reorients and forms polar contacts with the tetrazole group of KG-548 (yellow dashed lines).

Figure S8 .
Figure S8.Mutation Y456T disrupts the surface binding of KG-655. 19F spectra of ligands KG-548 or KG-655 (1 mM) mixed with ARNT PAS-B WT and Y456T (250 µM), collected at 278.1K.Mutation Y456T affected the surface binding of the two ligands differently, shifting the ligand-bound peak of KG-548 upfield by 0.65 ppm while completely abolishing the surface binding of KG-655.The internal binding mode of KG-655 also appeared to be affected (less broadening of the peak at -63.1 ppm) by the mutation.

Figure S9 .
Figure S9.The ARNT PAS-B/KG-548 interface is an important surface binding 'hotspot'.The same b-sheet surface has also been shown to interact with HIF-2a PAS-B and TACC3 coactivator.

Table S4 GAFF2 atomic charges derived for KG-279 drug ligand
. Atom names in the structure of the ligand correspond to those used for ligand parameterization.Note that these atom names differ from labels by IPUAC locants in Tables