Binding of a Pyrene-Based Fluorescent Amyloid Ligand to Transthyretin: A Combined Crystallographic and Molecular Dynamics Study

Misfolding and aggregation of transthyretin (TTR) cause several amyloid diseases. Besides being an amyloidogenic protein, TTR has an affinity for bicyclic small-molecule ligands in its thyroxine (T4) binding site. One class of TTR ligands are trans-stilbenes. The trans-stilbene scaffold is also widely applied for amyloid fibril-specific ligands used as fluorescence probes and as positron emission tomography tracers for amyloid detection and diagnosis of amyloidosis. We have shown that native tetrameric TTR binds to amyloid ligands based on the trans-stilbene scaffold providing a platform for the determination of high-resolution structures of these important molecules bound to protein. In this study, we provide spectroscopic evidence of binding and X-ray crystallographic structure data on tetrameric TTR complex with the fluorescent salicylic acid-based pyrene amyloid ligand (Py1SA), an analogue of the Congo red analogue X-34. The ambiguous electron density from the X-ray diffraction, however, did not permit Py1SA placement with enough confidence likely due to partial ligand occupancy. Instead, the preferred orientation of the Py1SA ligand in the binding pocket was determined by molecular dynamics and umbrella sampling approaches. We find a distinct preference for the binding modes with the salicylic acid group pointing into the pocket and the pyrene moiety outward to the opening of the T4 binding site. Our work provides insight into TTR binding mode preference for trans-stilbene salicylic acid derivatives as well as a framework for determining structures of TTR–ligand complexes.

and was resuspended in 20 mM Tris-HCl, pH 8.0, 100 mM NaCl, (Buffer A) and lysed by sonication. The lysate was cleared by centrifugation at 25,000 × g for 30 min at 4 • C. The supernatant was heated to 60 • C for 30 min. After heating, the precipitated material was removed by centrifugation at 14,000 × g for 30 min at 4 • C followed by filtration through 0.45 µm cellulose acetate membrane (Millipore) and applied to a Source-15Q 10/10 ion exchange chromatography column. The column was washed extensively with buffer A and elution was carried out with a linear gradient of 1M NaCl in buffer A. Sample containing TTR was further purified by a subsequent size-exclusion chromatography on a HiPrep 16/60 superdex 75 column (Cytiva) equilibrated with 10 mM Na-phosphate buffer, 100 mM KCl pH 7.6 at 20 • C. Fractions containing pure TTR were collected, pooled, and concentrated using an Amicon Ultra centrifugal filter device (Millipore, 3 kDa molecular-weight cutoff). Protein concentration was determined by using the absorption extinction coefficient 73,156 M −1 cm −1 at 280 nm applied for tetrameric TTR. Protein quality and purity were accessed by SDS-PAGE prior to experiments. Aliquots of purified TTR were flash-cooled in liquid nitrogen and stored at -80 • C until use.

S-1.3 Crystallization of the Py1SA-TTR complex
The protein was crystallized as described previously. 3 The purified TTR was dialyzed against 10 mM Na-phosphate buffer with 100 mM KCl (pH 7.6) and concentrated to 5.2 mg·mL −1 using an Amicon Ultra centrifugal filter device (Millipore, 3 kDa molecular-weight cutoff) and co-crystallized at room temperature with 500 µM concentration of Py1SA added from DMSO stock solutions at 10 mM, using the vapor-diffusion hanging drop method. A drop containing 3 µL protein solution was mixed with 3 µL precipitant and equilibrated against 1 mL reservoir solution containing 1.3-1.6 M sodium citrate and 3.5 % v/v glycerol at pH 5.5 in 24-well Linbro-plates. Crystals grew to dimensions of 0.1 × 0.1 × 0.4 mm 3 after 5-7 days. Once fully grown, the crystals were further transferred into a new equilibrated drop containing the same amount of ligand and were incubated for three days. The crystals were S3 cryo-protected with 12.5 % v/v glycerol and to avoid the possibility of the ligand washing out of the crystals during the brief cryo-protection step, the final cryo-solution always contained the same amount of ligand.
S-1.4 X-ray data collection, integration, and processing The X-ray diffraction data of Py1SA-TTR were collected under cryogenic conditions at the MAX IV facility (MAXIV), Sweden, using PILATUS detectors at a wavelength of 0.97993Å. These data were processed to a resolution of 1.4Å using XDS 4 and AIMLESS from the CCP4 software suite. 5 Data collection statistics are summarized in Table S-I. Phasing was done by molecular replacement using Phaser 6 with a search model derived from the published coordinates 1F41. In short, residues 11-98 and 104-122 were included in the initial model omitting a known flexible region. The model was refined against all the diffraction data using REFMAC. 7 Manual map inspections were performed with COOT. 8 Ligands and solvent were placed in density after 1 to 2 rounds of rebuilding the protein model with COOT and refinement using REFMAC.

S-1.5 Forcefield parameters for Py1SA
The geometry optimization using the Gaussian (version 16.B.01) program 9 was initially performed at the B3LYP level of theory in combination with the 6-31G(d,p) basis set to identify the lowest energy conformers. Next, the initial forcefield was generated by deriving the RESP charges for the most stable conformer (Conf3 in Fig parametrization, refer to supporting information of our previous work. 13 To validate the forcefield, the ground state energies of conformers computed using MM and DFT methods were compared (see Table S-II). The highest error between the two meth-

S-1.6 Molecular dynamics simulations
All molecular dynamics (MD) simulations were performed using the Gromacs version 2019. 3 14 with the Amber ff14SB force field 15 for the TTR protein and re-parametrized General Amber Force Field (GAFF) 11,12 for Py1SA (see Section S-1.5).
The starting structures of the TTR protein with Py1SA ligand existing in two configurations with respect to protein were based on the X-ray crystallography data from the present study (see Figure 3 in the main text). As a result of electron density being averaged along the AA' BB' symmetry axis, crystallography could not differentiate between four potential binding models for the ligand, i.e., forward-B, forward-B', reverse-B, and reverse-B'. All models were presented as possible starting points for modeling. All initial structures from S7 the X-ray analysis were cleaned using the module pdbfixer, 16 Gromacs 14 utilities, and the Gauss view 17 program by removing water molecules, adding H atoms to the protein and the ligand, and adding three missing amino acid residues to the two chains of the TTR protein.
We used Gromacs tools to solvate the protein-ligand system in a TIP3P 18 water box of size 7 × 7 × 9 nm 3 with a total of approximately 12,000 water molecules. 21 Na + ions were then added to obtain system charge neutrality. The long-range electrostatic interactions were calculated using the particle mesh Ewald (PME) method 19 with a long-range cutoff of 1 nm which also is the cutoff of the short-range van der Waals interactions. Default settings were used for Fourier spacing (0.12 nm) and PME order (4)

S-1.7 Umbrella sampling simulations
We computed the potential of mean force (PMF) surface using umbrella sampling. 21-23 MD simulations were employed to extract the initial coordinates for binding free energy calculations. We applied the same computational settings as in the molecular dynamics simulations S8 but using Gromacs version 2021.3 14 and except the box size was set to 7 × 11 × 9 nm 3 .
We further only constrained the backbone throughout the protein in the US simulations.
Starting from the last snapshot of each initial structure from the equilibration phase, Py1SA was pulled away from its binding site about 2.5 nm; we used a force of 5,000 kJ mol

S-4 Umbrella sampling
We have obtained two PMF trajectories for every starting structure. The results are shown in Figures S-9 -S-12. In our analysis, we focus on the results from the forward-B' and reverse-B'. We disregard forward-B due to rather pronounced Coulombic and hydrogen-bond interaction in the pulling trajectories at distances up to more than 1.5 nm from the original binding pocket (see the first trajectory of forward-B in Figure S-10). The second trajectory of forward-B exhibits Lennard-Jones interaction at about the same distance. In both cases, the PMF shows significantly smaller energetic differences between the minimum and endpoint than for the trajectories for the forward-B' (Figure S-9). These observations suggest that the unbinding process in these two trajectories is not fully finished so we cannot conclude on a free binding energy in case of the pulling trajectories obtained from the forward-B initial structure. Similar Coulombic interaction is also obtained for the second pulling trajectory of the reverse-B' initial structure and first pulling trajectory of reverse-B initial structure (see Figure S -12). In contrast to the first case, the latter case is also accompanied by a smaller energetic difference of the minimum with the endpoint. This is why we disregarded this trajectory.
For the reverse-B' trajectories, we observe a shift in the ϕ 1 angle close to the minimum for the pulling trajectories which we do not observe for the remaining pulling trajectory of reverse-B (c.f. Figure S-12). However, due to the small number of pulling trajectories, there is no clear evidence for another minor binding mode driven by the ϕ 1 angle. If we include the PMF for reverse-B in the bootstrap for the reverse mode, we obtain a binding free energy of 80±4 kJ/mol and if we do treat speculative possible minor modes separately, we obtain binding free energies of 83±5 and 78±4 kJ/mol for the reverse modes. All reverse mode binding free energies are, hence, clearly larger than the 67±4 kJ/mol obtained for the forward modes. This leaves the conclusion unchanged that the reverse mode is dominant. In S20