Inhibition of the mechano-enzymatic amyloidogenesis of transthyretin: role of ligand affinity, binding cooperativity and occupancy of the inner channel

Dissociation of the native transthyretin (TTR) tetramer is widely accepted as the critical step in TTR amyloid fibrillogenesis. It is modelled by exposure of the protein to non-physiological low pH in vitro and is inhibited by small molecule compounds, such as the drug tafamidis. We have recently identified a new mechano-enzymatic pathway of TTR fibrillogenesis in vitro, catalysed by selective proteolytic cleavage, which produces a high yield of genuine amyloid fibrils. This pathway is efficiently inhibited only by ligands that occupy both binding sites in TTR. Tolcapone, which is bound with similar high affinity in both TTR binding sites without the usual negative cooperativity, is therefore of interest. Here we show that TTR fibrillogenesis by the mechano-enzymatic pathway is indeed more potently inhibited by tolcapone than by tafamidis but neither, even in large molar excess, completely prevents amyloid fibril formation. In contrast, mds84, the prototype of our previously reported bivalent ligand TTR ‘superstabiliser’ family, is notably more potent than the monovalent ligands and we show here that this apparently reflects the critical additional interactions of its linker within the TTR central channel. Our findings have major implications for therapeutic approaches in TTR amyloidosis.

The seminal observation that the native non-covalent TTR homotetramer dissociates at low pH into dimers and monomers that self-assemble into amyloid fibrils 1 is the basis for the current, widely accepted model for TTR amyloid formation. However low pH treatment of both wild type and amyloidogenic TTR variants produces mostly heterogeneous amorphous aggregates with a very low yield of authentic amyloid fibrils showing pathognomonic green birefringence in polarized light after Congo red staining, characteristic fibrillar electron microscopic appearance and the cross-β X-ray fibre diffraction signature. Ex vivo TTR amyloid deposits, especially in the heart, usually contain a substantial proportion of the C-terminal TTR fragment generated by proteolytic cleavage at Lys48-Thr49 2 . Following this observation we have identified and characterized a novel mechanism of TTR amyloid fibrillogenesis mediated by selective tryptic cleavage at residue 48. The highly amyloidogenic TTR C-terminal residue 49-127 polypeptide is released, catalysing amyloid fibril formation in vitro, and the whole process is strongly enhanced by biomechanical forces 3,4 . Abundant authentic amyloid fibrils are produced with pathognomonic features indistinguishable from natural ex vivo fibrils.
Compounds able to stabilize TTR, intended for use as drugs to treat and prevent systemic ATTR amyloidosis, have hitherto been identified exclusively by their capacity to inhibit TTR dissociation and aggregation induced by low pH in vitro. However, we have found that inhibition of the mechano-enzymatic pathway of TTR amyloid fibrillogenesis at physiological pH, ionic strength and temperature requires occupation of both thyroxine binding sites in each native TTR tetramer. This is most efficiently achieved by bivalent ligands, exemplified by the palindromic compound mds84 5 , that is spontaneously bound simultaneously and pseudo-irreversibly in both sites. Sant' Anna and colleagues 6 have recently reported that tolcapone is a potent inhibitor of TTR dissociation and aggregation under denaturing conditions. They attribute its efficacy to occupation of both binding pockets with similar high affinity in contrast to the notable negative cooperativity observed with thyroxine itself and other monovalent ligands, including the TTR stabiliser drugs, tafamidis 7, 8 and diflunisal 9 , which are now in clinical use. This important observation prompted us to test the capacity of tolcapone to inhibit mechano-enzymatic amyloidogenesis of TTR using the V122I variant which is the most prevalent cause of hereditary cardiac amyloidosis 10 .

Results and Discussion
We compared tolcapone with tafamidis 7, 8 and the experimental bifunctional compound, mds84 5 (Fig. 1A). When stirred with V122I TTR at 37 °C in PBS in the presence and absence of trypsin for 96 h, all three ligands inhibited proteolytic cleavage of the protein in a dose-dependent manner ( Fig. 1B and Supplementary Fig. 1). At molar  Fig. S1). Intensities of the electrophoretic bands corresponding to the intact protomer in the whole mixture were normalized to 100% for the same band of the protein before addition of trypsin. The solid lines represent the nonlinear fit to the experimental mean (SD) of three replicates using GraphPad Prism v5. Two way ANOVA gave a P value < 0.001 for tafamidis vs mds84 at 9 and 18 µM; for tolcapone vs mds84 at 9 µM. (C) Aggregation of 18 µM V122I TTR was quantified as spectrophotometric turbidity at 400 nm normalized to 100% for aggregation of the protein in the absence of ligands. We know from previous work that TTR aggregation in this system is in the form of authentic amyloid fibrils 3,4 . All data shown represent mean (SD) of three independent experiments, and *** represents P < 0.001. ratios of ligand:TTR tetramer higher than 2:1, the three compounds had similar efficacy with apparent almost complete inhibition of TTR cleavage. However at lower molar ratios, mds84 was the most potent inhibitor, followed by tolcapone and tafamidis.
Even when we compared the effect of the three compounds on the inhibition of TTR acidic-mediated aggregation 11 , mds84 at equimolar concentration with TTR was the only ligand able to completely inhibit the process (Supplementary Information and Supplementary Fig. S2).
The inhibition of the mechano-enzymatic pathway of TTR amyloidogenesis, which we believe to be the most likely pathophysiological mechanism in vivo, depends on the occupation of both thyroxine binding sites. The superior potency of tolcapone among monovalent TTR ligands is consistent with its unique property of not inducing negative cooperativity 6 .
The much more effective inhibition of proteolysis-mediated TTR fibrillogenesis by mds84 results from the simultaneous occupation of both thyroxine binding sites, and the internal channel between them, by this palindromic molecule 5 . It occurs rapidly and completely at equimolar concentrations of TTR and ligand. Binding of mds84 by native TTR is pseudo-irreversible under physiological conditions 5 and generates a stable complex. This in contrast to the reversible monovalent ligands for which a higher molar excess is required to saturate both binding sites in vivo. These limitations on efficacy against mechano-enzymatic mediated amyloidogenesis may explain the modest therapeutic benefit of monovalent ligands in the clinical studies reported so far 12 .
To understand the superior inhibitory effect of bivalent ligands, we analysed the deposited X-ray structures of TTR complexed with monovalent ligands, tolcapone 6 and tafamidis 7 , (Fig. 2) in comparison with bivalent ligands, mds84 5 and compounds 20 and 22 of Green et al. 13 (Supplementary Fig. S3). For a general description of the crystal structures we refer to a very exhaustive review by Palaninathan 14 . There is no PDB structure of the mds84-V122I TTR complex so the analysis is restricted to wild type TTR. This choice is justified by the almost perfect superposition of the wild type and V122I TTR structures in the absence of ligands ( Supplementary  Fig. S4A,B) with minor deviations when bound to tolcapone ( Supplementary Fig. S4C,D). The average root mean square deviations (rmsd) for the TTR backbone residues are 0.45 and 0.9 Å for the two proteins in their free and bound form, respectively. Moreover, Val122 and Ile122 side chains point towards the external part of the pocket cavity and residue 122 is not present in the halogen binding pockets (HBPs; see Methods). Importantly, neither mono-nor bivalent molecules induce major rearrangements of the HBP residues with rmsd values calculated for the backbone atoms of HBP1, HBP2 and HBP3 ranging from 0.2 to 0.3 Å ( Table 1). The superposition of the binding pockets of TTR with and without ligands is shown in Supplementary Fig. S5. Also there is no correlation between the number of ligand-protein hydrogen bonds, (Table 1) and inhibitory efficacy. A full list of hydrogen bonds between the ligand and the HBPs is reported in the Supplementary Information. However, in contrast with the monovalent ligands which interact only with the ligand pocket itself, the methylene linker of the bivalent compounds occupies the narrow TTR inner channel with an interaction surface increased by 48 Å 2 , for mds84 compared with tolcapone. Binding of mds84 5 , compound 20 and compound 22 13 by TTR creates, respectively, up to 9, 5 and 4 hydrophobic interactions, mainly with S117 Cβ and Leu110 Cδ2. These interactions make an estimated favourable energetic contribution of −4.6, −2.5 and −1.0 kcal/mol for the three ligands respectively. Occupancy of the central channel of TTR, estimated for mds84 to be around 80% of the void volume, may thus have a crucial role in preventing subunit sliding and also provide long distance stabilisation affecting the dynamics of CD loop, the selective cleavage of which potently primes amyloid fibrillogenesis. The pioneering study of Green et al. 13 identified the enhanced stabilisation of TTR by bivalent ligands with a central linker but their compounds did not interact with native TTR. They were bound only by renaturation of dissociated denatured TTR around the ligand. 13 In sharp contrast, both the mds84 prototype and its analogues are rapidly bound by the native TTR tetramer, traversing the inner channel to occupy it with the linker. 5 The unique total binding site in TTR, comprising the two halogen binding pockets and the inner channel, is entered by these ligands with a decreased entropic cost compared to all monovalent ligands 15 . The key factors responsible for the 'superstabiliser' property of mds84 and its analogues are thus likely to be the combination of entropic gain, increased hydrophobic contributions of the linker and a reduced void volume in the interface cavity.
The mechanism of binding of TTR by bivalent compounds cannot be fully clarified on the basis of the available crystallographic structures. These data reveal that the TTR inner cavity is very narrow and therefore apparently inaccessible to the bulky groups of the ligands. We hypothesize that the tetramer dynamics may be larger than expected and current studies in our laboratory using NMR might shed light on this important issue.
Finally, even at large molar excess of tafamidis or tolcapone that inhibited obvious in vitro aggregation, ultracentrifugation of the reaction mixture yielded a small amount of insoluble material which stained with Congo red to give typical green birefringence ( Fig. 3A and Supplementary Fig. S6) and was fibrillar in the electron microscope (Fig. 3B).
In the presence of mds84 such insoluble material was minimal but extensive searching by electron microscopy revealed some fibrils (Fig. 3B). Evidently, the mechano-enzymatic mechanism enables slow formation of some fibrils even when both binding sites are fully occupied in most TTR molecules. Effective targeting of all TTR molecules with stabilizer drugs is much more challenging in vivo than in vitro and it is therefore crucial to consider the significant pathogenic impact of even minimal amyloid fibril formation in vivo. There are potentially major implications, that should not be ignored, both for interpretation of current clinical studies in TTR amyloidosis and for the design of future therapeutic approaches such as the exploitation of non-natural peptides inhibitors of TTR aggregation 16 or antibodies recognizing cryptic epitope exposed only on the surface of aggregated TTR and not in the native tetrameric state 17 . At the end of the aggregation, susceptibility to trypsin was monitored using SDS-homogeneous 15% PAGE (GE Healthcare) under reducing conditions. Two-way Anova was performed using GraphPad Prism 5 for pairwise multiple comparison among tafamidis, tolcapone and mds84. The pellet was harvested from each protein sample in the absence and in the presence of ligands by ultracentrifugation in a Beckman Optima TL ultracentrifuge at 135,000 g, 20 min. After resuspension of the pellet with a minimal volume of PBS, samples were stained with alkaline alcoholic Congo red and examined by high intensity cross polarized light microscopy 19 . Amyloid load was scored in the Congo red stained pellet of TTR in the presence of fourfold molar excess of each ligand (Supplementary Fig. S6). A blind quantification was carried out by an expert operator on six slides per each group of treatment using the following grading score: 0 (no spot detected), 1 (occasional spots), 2 (green birefringent spots clearly visible and corresponding to the stained material in the bright field), 3 (surface homogeneously covered by green birefringent material). The non-parametric Kruskall-Wallis test for mds84 vs tafamidis (or tolcapone) was applied using GraphPad Prism 5.
Samples were also examined by negative staining transmission electron microscopy. Briefly, a drop of each sample was allowed to dry on formvar coated copper EM grids for 2 min before blotting with filter paper to remove excess solvent and staining with 2% w/v uranyl acetate for 1 min. After further blotting and drying in air, transmission electron microscope (Jeol1200EX) images were obtained at 80 kV.