Novel Transthyretin Amyloid Fibril Formation Inhibitors: Synthesis, Biological Evaluation, and X-Ray Structural Analysis

Transthyretin (TTR) is one of thirty non-homologous proteins whose misfolding, dissociation, aggregation, and deposition is linked to human amyloid diseases. Previous studies have identified that TTR amyloidogenesis can be inhibited through stabilization of the native tetramer state by small molecule binding to the thyroid hormone sites of TTR. We have evaluated a new series of β-aminoxypropionic acids (compounds 5–21), with a single aromatic moiety (aryl or fluorenyl) linked through a flexible oxime tether to a carboxylic acid. These compounds are structurally distinct from the native ligand thyroxine and typical halogenated biaryl NSAID-like inhibitors to avoid off-target hormonal or anti-inflammatory activity. Based on an in vitro fibril formation assay, five of these compounds showed significant inhibition of TTR amyloidogenesis, with two fluorenyl compounds displaying inhibitor efficacy comparable to the well-known TTR inhibitor diflunisal. Fluorenyl 15 is the most potent compound in this series and importantly does not show off-target anti-inflammatory activity. Crystal structures of the TTR∶inhibitor complexes, in agreement with molecular docking studies, revealed that the aromatic moiety, linked to the sp2-hybridized oxime carbon, specifically directed the ligand in either a forward or reverse binding mode. Compared to the aryl family members, the bulkier fluorenyl analogs achieved more extensive interactions with the binding pockets of TTR and demonstrated better inhibitory activity in the fibril formation assay. Preliminary optimization efforts are described that focused on replacement of the C-terminal acid in both the aryl and fluorenyl series (compounds 22–32). The compounds presented here constitute a new class of TTR inhibitors that may hold promise in treating amyloid diseases associated with TTR misfolding.


Introduction
Transthyretin (TTR) is a homotetrameric protein, consisting of four 127-amino acid b-sheet-rich subunits [1], and is present in mammals, birds, and reptiles [2]. Human TTR is involved in the transport of thyroxine (T4) in the cerebrospinal fluid and is a secondary carrier of T4 in plasma; approximately half of the TTR tetramer population in plasma is bound to retinol binding protein (RBP) [1,3,4,5,6,7,8]. TTR normally circulates as an innocuous soluble protein, but in some individuals it polymerizes to form amyloid fibrils. The fibrils are formed through a mechanism which most likely consists of a preliminary misfolding of the TTR tetramer [9,10,11], followed by self-assembly into amyloid fibrils [12,13]. The result is the formation of insoluble toxic fibrillar deposits associated with many diseases. Four types of amyloidosis have been observed: senile systemic amyloidogenesis (SSA) [14,15], familial amyloid cardiomyopathy (FAC) [15], familial amyloid polyneuropathy (FAP) [16], and central nervous systemselective amyloidosis (CNSA) [17,18]. SSA results from the fibrillization of wild-type TTR fibril in elderly individuals [14,15], whereas the origins of the familial diseases (FAC, FAP, and CNSA) are thought to be rooted in the fibrillogenesis of TTR mutants found in diverse populations all over the world [19]. In familial diseases, amyloid fibril aggregation may principally determine serious pathologies, including systemic and central neuropathies and cardiomyopathies leading to severe, lifethreatening conditions [20].
X-ray crystallographic studies have provided a rationale for the stabilization of the native state of TTR by T4 hormone, while offering insights into novel inhibitor designs [4,8,22,27]. Previous reports of the TTR tetramer structure depicted two funnel-shaped binding sites in the T4 hormone, each defined by its dimer-dimer interface [4,22]. Figure 2a depicts the tetrameric TTR and Figure 2b shows the close-up view of the hormone binding pocket. A junction of four Ser117 side chains may be observed, situated at the interface between the two identical T4 binding sites. Each hormone binding site can be divided into an inner and outer  Table 1 and Table 2 with general formula A and classical NSAIDs with aryl-propionic structure, respectively. doi:10.1371/journal.pone.0006290.g001 Figure 2. The T4 hormone binding channel of TTR tetramer. (a) Ribbon diagram of tetrameric structure of TTR with bound compound 15 (see results and discussion section). Each subunit (labeled A, B, C and D) of the tetramer is shown with its secondary structural elements and colored differently. The binding of 15 in both T4 binding pockets of TTR is shown as a stick model inside a transparent surface. The crystallographic asymmetric unit contains subunits A and B while subunits C and D were formed through crystallographic symmetry. Because of the two-fold axis along the binding channel (indicated by the arrows) a second symmetry-related binding conformation is present for the inhibitor molecule in both hormone binding sites. (b) Definition of the halogen binding pockets (HBPs) of the T4 hormone binding site of TTR based on the previously published crystal structure of T4 bound to the protein (PDB 1ROX) [4]. T4 is shown in both of its symmetry-related binding modes (shown in magenta and green inside the molecular surface mesh) that are related by a two-fold rotation axis. The carboxyl tail of T4 is positioned at the entry port of the binding site (outer binding pocket) and the iodines occupy the HBP1, HBP2, and HBP3 pockets. The inner binding pocket, HBP3, is located between the side chains of Ser117, Thr119, Ala108, and Leu110, the central HBP2 pocket is formed by the side chains of Leu17, Ala108, Ala109 and Leu110 and the outer pocket HBP1 is located between the side chains of Lys15, Leu17, Thr106 and Val121. doi:10.1371/journal.pone.0006290.g002 binding cavity. Six halogen binding pockets (HBP1, HBP1', HBP2, HBP2', HBP3 and HBP3') were also defined within each hormone binding pocket based on the positions of the halogen atoms of T4 in the TTR:T4 crystal structure, Figure 2b [4,22]. The inner binding cavity comprises HBP3 and HBP3', formed by the side chains of Ser117, Leu110, Thr119 and Ala108 of both subunits. The Ser117 hydroxyl groups mediate hydrogen bond interactions with bound inhibitors, as detailed in previously reported TTR:inhibitor complex structures [8,22,28]. The outer binding site is composed of HBP1 and HBP1', formed primarily by residues Lys15, Leu17, Thr106, and Val121 of both subunits. HBP2 and HBP2' are positioned at the interface of the inner and outer binding cavities, comprising residues Leu17, Ala108, Ala109 and Leu110 of both subunits. The associated binding pocket is highly lipophilic allowing the HBP2 and HBP2' residues to interact favorably with the hydrophobic portions of inhibitors. Typically, TTR inhibitors and T4 bind in what is referred to as the forward binding mode, where anionic substituents like carboxylate are positioned in the outer binding pocket engaging in electrostatic interaction with the Lys15 e-ammonium groups. However, the reverse binding mode, with the carboxylate oriented towards the inner binding pocket to hydrogen bond with Ser117 and Ser117', is not unusual and has also been observed previously in TTR complexed with diclofenac (a biarylamine), several diclofenac analogues, and some diflunisal analogs [24,25,29].
A common pharmacophore among small molecule stabilizers of the T4 hormone binding pocket of TTR tetramer is a carboxylic acid connected through a rigid spacer to an aromatic moiety ( Figure 1). Based on the co-crystal structure, Zanotti et al. have shown that all-trans-retinoic acid, a cyclohexene-linker-acid compound, binds to TTR in a forward binding mode similar to thyroxine [30]. However, to the best of our knowledge, the aryl-linker-acid architecture has not previously been utilized for TTR tetramer stabilization. Here in this study, we have evaluated a series of previously synthesized b-aminoxypropionic acids [31,32,33] (compounds 5-21 in Figure 3 and Table 1), as TTR amyloid inhibitors. These compounds contain a flexible oxime-based tether between the aromatic (aryl or fluorenyl) and acidic moieties. At the outset of this investigation, we hypothesized that the greater flexibility of the oxime ether tether could allow these small molecule ligands to more fully occupy the volume of the TTR hormone binding pocket. The effect of the bulkiness of the aromatic moiety was also probed with respect to inhibitor efficacy and ligand binding mode. The diversification from the typical biaryl system was designed to yield inhibitors devoid of off-target anti-inflammatory activity, as previous TTR inhibitor designs are roughly equipotent inhibitors of both TTR [22,34] and cyclooxygenase (prostaglandin endoperoxide synthase or COX 1-3) enzymes [35,36,37]. For example, FLP binds to TTR, COX-1 and COX-2 with similar sets of interactions and inhibits both TTR amyloidosis (70% inhibition at 7.2 mM), COX-1 (IC 50 = 0.41 mM) and COX-2 (IC 50 = 4.2 mM) enzyme activity [22,36]. Based on the TTR:FLP complex crystal structure, we have hypothesized that the bulkier compounds may bind specifically to TTR without inhibiting COX-1, characterized by comparatively smaller hydrophobic binding pockets [22]. Specifically, we suggested that the bulkier substituents projecting into the HBP1 and HBP2 regions of the TTR:FLP pocket might significantly improve TTR-inhibitor interactions and reduce its binding affinity for COX-1 and COX-2 due to steric hindrance. In fact, replacement of the phenyl ring in the NSAIDs biaryl system with a bulkier carborane moiety greatly decreased their COX activity with the significant inhibitor efficacy of TTR dissociation [34,38].
An in vitro TTR amyloid fibril formation assay was utilized to evaluate compound efficacy and the three-dimensional crystal structures of TTR in complex with four inhibitors were solved. Molecular docking was harnessed in concert with X-ray structural studies to examine the relationship between inhibitor chemical structure, efficacy, and binding mode. In an attempt to initiate the further optimization of these TTR inhibitors, novel compounds with longer chain substitutions and enhanced flexibility were prepared ( Figure 4 and Table 2): b-aminoxymethylsulfonylpropionamides (compounds [22][23][24][25][26][27], (b-aminoxypropanamido) acetic acids (compounds [28][29][30], and (b-aminoxypropanamido) propanoic acids (compounds 31 and 32). Based on the in vitro assay, three of these compounds exhibited significant inhibition against TTR amyloid fibrils providing a basis for the further exploration of SAR.

Design and Synthesis of Aryl and Fluorenyl Families
A common pharmacophore among small molecule stabilizers (including 1-4) of the TTR amyloidosis is a carboxylic acid connected through a rigid spacer to an aromatic moiety ( Figure 1). Based on these potent TTR inhibitors, the new series of compounds (5-21 in Figure 3 and 22-32 in Figure 4) were chosen for the current investigation. The rationale for these compounds involves two novel variations on the structures of previously known inhibitors: 1) the aryl moiety, displaying the carboxylate group, of the typical biaryl inhibitors is replaced by the C = N-O-C type linker and 2) the other ring structure in the biaryl scaffold is examined as either an aryl or fluorenyl moiety. It is noteworthy that the bulkier fluorenyl group has not been explored previously in this context.
Time course of TTR amyloid fibril inhibition for fluorenyl compounds 15, 16, and 20 Although the inhibitor efficacy of 15 and 16 at 3.6 mM was comparable to that of diflunisal, previous studies have shown that at 7.2 mM concentration the potency of diflunisal is increased from 63% to 96% [25]. Therefore, we evaluated 15, 16 and 20 further at 7.2 mM concentration, monitoring the fibril formation at the 0, 24, 48, 72 and 96 hours time points ( Figure 5) in three independent runs for each compound. All three inhibitors exhibited more than 90% inhibition and significantly suppressed fibril formation. Of the three tested compounds, 15 displayed the best inhibition (94%).

Crystal structures of inhibitors 11, 13, 15 and 16 bound to TTR
High-resolution X-ray crystal structures of 11, 13, 15 and 16 bound to wt-TTR were obtained by soaking TTR crystals with a fivefold molar excess of inhibitor for 3-4 weeks. Compounds 11 and 13 were selected for crystal structure investigation as they showed significant inhibitory activity and represent the aryl family, while the two most potent inhibitors in this report (compounds 15 and 16) are members of the fluorenyl class. In each case, the crystals diffracted to approximately 1.8 Å resolution and the crystal data and refinement statistics for each complex structure are reported in Table 3. The overall molecular structure of the TTR:inhibitor complexes are very similar to that of native TTR. Figure 2a shows the binding of 15 to both hormone binding pockets of the TTR tetramer. There are two hormone binding  Table 1. In vitro acid-mediated wt-TTR (3.6 mM) amyloidogenesis inhibition activity of compounds 5-21 (see Figure 3). sites per tetramer (AC between the subunits A and C, and BD between the subunits B and D), each of which has an intrinsic twofold symmetry as these are located on the crystallographic two-fold axis (indicated by an arrow in Figure 2a). All TTR:inhibitor complex structures showed significant additional electron density in both hormone binding pockets, confirming binding of the inhibitor. The electron density corresponding to the inhibitor observed in the inner and outer hormone binding pockets of the BD dimer is well ordered in the crystal structures with compounds 11, 13, 15, and 16. (Figures 6 and 7). In contrast, the electron density of the carboxyl substituted region of these inhibitors is only partially ordered in the AC dimer (Figures 6a, 6b, 7a and 7b). The observed electron density for the substituted region and the clear electron density for the side chain atoms of the protomer allowed us to finalize the binding modes in both the AC and BD dimers without ambiguity. Moreover, similar electron density differences between the two hormone binding pockets have been observed in many of the previously reported structures including the TTR:oxime-ether crystal structure [39].

Binding of 11 to wt-TTR
Compound 11 binds to TTR in the forward binding mode, with the carboxylate-substituted hydrophilic chain oriented in the outer binding pocket to interact with Lys15 and Lys15' (Figures 6a and  6c). In each hormone binding site, two different binding conformations of the ligand were identified with approximately equal occupancy (shown in magenta and green in Figures 6a and  6c). In the inner binding pocket of the BD dimer, the aryl ring is stacked between the hydrophobic side chains of Ala108, Ala108', Leu110, Leu110', Thr119 (through C c ), and Thr119' (also through C c ). The trifluoromethyl substituent occupies halogen binding pocket 3 (HBP3) in the inner cavity. One of the fluorine atoms of the trifluoromethyl substituent makes what may be termed as a favorable through space electrostatic interaction [40,41] with the backbone amide N-H of Leu110 and Leu110' at a distance of 2.9 Å . In both the AC and BD dimers, Ser117 and Ser117' orient their hydroxyl groups away from the inhibitor, contributing to additional hydrophobic interactions with the aryl moiety of the inhibitor via their b-CH 2 groups (the C6 carbon atom of the aryl moiety of 11 is at 3.3 Å distance from C b carbon atom of Ser117). An ordered water molecule (labeled W1 in Figure 6a) is located between the adjacent Ser117 residues of the AC dimer facilitating hydrogen bonding interactions between the serine O c atoms (Ser117 O c -W1-Ser117' O c at a distance of 2.8 Å ) and is not involved in any direct interactions with the inhibitor. However, the side chain of Thr119 and Thr119' of the AC dimer makes adjustments by adopting two conformations and contributes to additional hydrophobic and hydrogen bond interactions (O c atom is at 2.9 Å and C c atom is at 3.3 Å from the closest fluorine atom of 11) with the inhibitors. The multiple conformations of Thr119 are not observed in the BD dimer, but all other interactions of the inner binding pocket are very similar in both the AC and BD dimers. The electron density of the linker atoms is well ordered in the BD dimer and only partially ordered  in the AC dimer. The electron density of the carboxylate group of the inhibitor is clearly visible. In both dimers, the linkers are stabilized by the side chains of residues Leu17, Leu17' (the closest distance from the inhibitor is 3.3 Å ), Ala108, Ala108' (at 3.9 Å distance). However, in both inhibitor binding pockets, the linker oxygen and nitrogen atoms do not appear to make hydrogen bond   interactions with TTR. The electron density map of all binding pocket residues was clear and well ordered. However, Lys15 and Lys15' of the AC and BD dimers showed weak electron density, indicating their flexibility, with their best fit conformations placing the N e atom at hydrogen bonding distance (3.0 Å ) from the carboxyl substituent of the inhibitor.

Binding of 13 to wt-TTR
Substitution of a methoxy group in place of the trifluoromethyl in compound 11 does not significantly affect the binding mode of 13 to TTR, while affording a slightly more potent inhibitor. Compound 13 binds to TTR in the forward binding mode (Figure 6b) with the carboxyl-substituted hydrophilic chain oriented in the outer binding pocket towards residues Lys15 and Lys15'. The electron densities of the aryl ring and linker atoms of the inhibitor are well ordered in the BD dimer and only partially ordered in the AC dimer. Thus, the binding interactions are described for the BD dimer bound inhibitor. In both T4 hormone binding pockets, the inhibitor adopts two conformations with equal occupancy (shown in magenta and green in Figure 6b). In both binding conformations, the methoxy substituent at the orthoposition of the aryl ring of 13 protrudes into halogen binding pockets 3 and 39. The oxygen atom of the ortho-O-methyl group of 13 forms a hydrogen bond with the O c atom of Ser117 at a distance of 3.0 Å (Ser117' for the second inhibitor conformation) and the carbon atom is oriented away from Ser117 (Ser117') into HBP3 or HBP3'. In HBP3 and 39, the methoxy group of the inhibitor makes additional hydrophobic interactions with subunit residues: Ala108 (C b to OCH 3 is 3.7 Å ) and Thr119 (C c to OCH 3 is 3.9 Å ). As with the structure of bound compound 11, the hydrophobic side chains of the inner binding pocket residues of the BD dimer (Ala108, Ala108', Leu110, Leu110', Thr119 and Thr119') stack with the aryl ring of the inhibitor. Interestingly, in contrast to 11, water molecules (labeled W1 and W2 in Figure 6b) are located between the Ser117 and Ser117' residues of the AC and BD dimers facilitating a hydrogen bonding network through the O c atom of Ser117 residues of all subunits. Ser117 and Ser117' are also hydrogen bonded to the inhibitor. It should be noted that the oxygen atom of the methoxy group of 13 does not appear to be close enough to form a hydrogen bond (3.6 Å distance) with either water molecule. In both dimers, the linkers are stabilized by the side chains of residues Leu17, Leu17' (the closest distance from the b carbon atom of the inhibitor is 3.5 Å ), Ala108, and Ala108' (3.8 Å distance from the nitrogen atom of the inhibitor). The electron density map of all binding pocket residues was clear and well ordered with the exception of the Lys15 and Lys15' in both the AC and BD dimers. With the best fit conformation, the Lys15 and Lys15' placed their N e atom at hydrogen bonding distance (2.9 Å ) from the carboxyl substituent of the inhibitor. Like the TTR:11 structure, Thr119 of the AC dimer maintains two conformations in the inner binding pocket and interacts with the aryl ring of the inhibitor.

Binding of 15 to wt-TTR
Inhibitor 15 binds in the reverse binding mode by orienting its carboxyl substituent towards the inner binding pocket (Figures 7a and 7c) to interact with HBP3 and HBP3' near Ser117-Ser117'. The linker region of the inhibitor is only partially ordered in AC  [50] omit maps are contoured at the 1 s level. The blob feature in XtalView [49] was applied to limit the electron density display to within 1.5 Å of the inhibitor and the final figure was rendered with Raster3D [51]. The inhibitor molecules were omitted from the model before the map calculation. (c) The binding interactions of 11 with the hormone binding pocket of TTR, the better ordered BD binding pocket is shown here. Like most of the TTR bound ligands, 11 also binds in two symmetry-related binding modes (shown in magenta and green). The key interacting residues are labeled, primed and unprimed residues refer to two neighboring symmetry related monomers comprising the T4 site. Compound 11 binds in the forward binding mode by orienting its carboxylate substituent to the outer binding pocket residue Lys15. The aryl moiety of 11 is anchored by its trifluro group to HBP3 and HBP3' of the inner most binding pocket. Like compound 11, compound 13 also binds in the forward binding mode with similar interactions (not shown here). doi:10.1371/journal.pone.0006290.g006 dimer while the electron density corresponding to the BD dimer inhibitor is well ordered as shown in Figure 7a. However, in both hormone binding pockets the carboxyl group and fluorenyl moiety occupy similar positions in both inner and outer binding pockets to make comparable interactions with the residues near Ser117 and Lys15 (Figures 7a and 7c). The fluorenyl ring moiety of the inhibitor is located at the outer binding pocket formed by residues Ala108, Lys15, Leu17, Thr106, and Val121 of both TTR subunits. In contrast to the TTR:11 structure, all protomer residues of the binding pocket, including Lys15, were well ordered in the electron density map. Compared to the aryl compounds, the bulky fluorenyl ring occupies the outer binding pocket more extensively and forms several non-bonding interactions with the protomer residues. The fluorenyl moiety of the inhibitor is comfortably sandwiched in between the hydrophobic groups of residues Lys15 (at the closest distance of 3.8 Å from the fluorenyl ring), Lys15', Leu17 (C d1 at 3.3 Å ), Leu17', Thr106 (C c at 3.7 Å ), Thr106', Ala108 (C b at 3.4 Å ), Ala108', Val121 (C c2 at 3.3 Å ) and Val121'. In the dimer binding pocket, the fluorenyl ring is nearly centered on the two-fold symmetry axis, giving the appearance of a single binding conformation. However, the spacer maintains two conformations (shown in magenta and green in Figures 7a and 7c) and simultaneously interacts with Ser117 and Thr119 residues of both HBP3 and HBP3' of the inner binding pocket as detailed below. In the inner binding pocket the carboxyl group forms four hydrogen bonds per subunit. Both oxygen atoms of the carboxyl group of 15 form hydrogen bonds with the side chain O c of Ser117 at distances of 2.7 and 2.8 Å . Thr119 residue orients itself towards the inhibitor and hydrogen bonds (O c of Thr119 is at 2.8 Å distance from one of the carboxylate oxygen atoms) with one of the oxygen atoms of the inhibitor carboxylate, also positioning its C b atom close to the inhibitor (the carbon alpha atom to the carboxylate group of the inhibitor is at 3.6 Å distance). Interestingly, a water molecule is located between the adjacent Ser117 and Ser117' residues at the two-fold axis and forms a hydrogen bond with the carboxylate oxygen atom (3.1 Å distance) and the O c atom of Ser117 and Ser117' (3.0 Å distance). Thus, the reverse binding mode of 15 facilitates a network of hydrogen bonds in the inner binding pocket connecting the Ser117 residues of all subunits, the nearby water molecules, and the carboxylate substituents of the inhibitor (Figures 7a and 7c). The alpha-and beta-carbon atoms of the linker are also sandwiched between the side chain atoms of Leu110 (the closest distance from the alpha carbon atom of the linker is 3.8 Å ) and Leu110' to further augment the binding.   [50] omit maps were contoured at the 1 s level. The blob feature in XtalView has been applied to limit the electron density display to within 1.5 Å of the inhibitor and the final figure was rendered with Raster3D [51]. The inhibitor molecules were omitted from the model before the map calculation. (c) The binding interactions of 15 with the hormone binding pocket of TTR, the better ordered BD binding pocket is shown here. Inhibitor 15 also binds in two symmetry-related binding modes (shown in magenta and green). The key interacting residues are labeled, primed and unprimed residues refer to two neighboring symmetry related monomers comprising the T4 site. Compound 15 binds in the reverse binding mode by orienting its carboxylate substituent to the inner binding pocket and its fluroneyl ring to the outer binding pocket. The fluroneyl ring is optimally sandwiched between the side chain atoms of the outer binding pocket, while the corboxylate group makes a network of direct and water mediated hydrogen bonds. Leu110 and Leu110' stack on top of the linker region but are not shown for clarity. Compound 16 also binds in similar fashion (not shown here). doi:10.1371/journal.pone.0006290.g007 mode by pointing its linker toward the inner binding pocket and placing its fluorenyl ring into the hydrophobic pocket formed by residues Ala108, Lys15, Leu17, Thr106, and Val121. As in the TTR:15 structure, the aliphatic linker of 16 takes on two conformations with equal occupancy to simultaneously interact with both HBP3 and HBP3' residues ( Figure 7b). Although the outer binding pocket interactions are very similar for both 15 and 16, compound 16 shows notable differences in the binding interactions at the inner binding pocket of TTR. Substitution of an (S)-methyl group in the alpha position to the carboxylic group kinks the carboxylic acid to the side of the inhibitor, into HBP3 or HBP3'. As a result, in both conformations of the inhibitor, the methyl substituent and the alpha-and beta-carbon atoms of the linker are sandwiched between the side chain atoms of Leu110 and Leu110'; C d2 of Leu110 is at 3.3 Å from the carbon alpha atom to the carboxylic group of 16, the (S)-methyl carbon of 16 is at 3.6 Å from C d2 of Leu110, and C b of Leu110 is at 4.0 Å from the beta-carbon of 16. One of the oxygens of the carboxyl group hydrogen bonds with the hydroxyl group of Ser117 (at a distance of 2.6 Å ) and O c atom of Thr119 (at a distance of 2.5 Å ). The second conformation of the inhibitor is generated through crystallographic two-fold symmetry and thus makes identical interactions with the hormone binding pockets of TTR. Together, the methyl substitution contributes to additional hydrophobic interactions with the inner binding pocket of TTR. The W1 and W2 water molecules observed between the Ser117 and Ser117' residues of the TTR:15 structure are also conserved in TTR:16 (in almost identical positions). However, the carboxylate group of the inhibitors does not make any hydrogen bonds with the water. Similar to the structures with bound 11, 13 and 15, the electron density of the linker region is only partially visible in the AC binding pocket.

Docking of b-aminoxypropionic acids to TTR
In an effort to complement our X-ray crystallographic studies with molecular modeling exercises, we performed docking studies of compounds 5-21 with the Gold v3.1 program (Cambridge Crystallographic Data Centre), starting from the crystallographic structure of the TTR-FLU complex (PDB ID: 1BM7 Klabunde, 2000 #16). For the arylideneaminoxy-derivative compounds 5-14, the Gold program predicted the typical thyroxine orientation (forward binding mode) using the Goldscore functions; The aromatic ring is inserted into HBP3, and the b-aminoxyethyl chain directs the carboxylic group towards Lys15, (see Figure 8a for compound 11). In the docked structure, the ortho-CF 3 substitution on the phenyl ring is engaged in significant non-bonding interactions with HBP3 residues Ala108, Leu110, Ser117 and Thr119. However, substituents in the metaand parapositions are not predicted to interact with HBP3 to the same extent, since they occupy the side part of this pocket. For 9-fluorenyliden-aminoxy derivatives 15-17 and 20, the Gold program suggested the reverse binding modes (see Figures 8b, 8c and 8d) using Goldscore functions. The TTR:15 and TTR:11 crystal structure reveals that 15 binds to the hormone binding pocket in the reverse binding mode and 11 bind in the forward binding mode with an orientation similar to the one obtained through the Goldscore algorithm. Therefore, this docking protocol was subsequently used to predict the binding of all other b-aminoxypropionic derivatives. The 9-fluorenyliden-aminoxy derivatives 16 and 17 may be placed in a reverse binding mode (Figure 8c). Only compound 16 reveals an effective interaction with HBP3. The corresponding (R)enantiomer 17 inserts the methyl substituent into HBP3, and directs the carboxylic group away from the Ser117 or Ser117' residue, without any productive interactions. Interestingly, the hypothesized binding mode for compound 20 (Figure 8d) is different from the other inhibitors; the molecule interacts with HBP3 and HBP3' simultaneously, through the carboxylic moiety and the aminoxymethyl chain, respectively. Based on the docking analysis of compounds 5-21, we have also evaluated the docking of two potent b-aminoxy-N-(methylsulfonyl) propionamides (compounds 24 and 26) to TTR hormone binding pocket using Goldscore algorithm. As shown in Figure 8e the aryl compound 24 predicted to bind in a forward binding mode while the fluorenyl analog 26 binds in the reverse binding mode (Figure 8f) with extended interactions in HBP3s and HBP1s respectively.

Discussion
The majority of previously reported TTR amyloid inhibitors, including diflunisal, are halogenated compounds designed to occupy one or more TTR hormone binding pockets [8]. With several potent biaryl families of TTR inhibitors available, the current drug discovery initiatives have targeted diverse chemotypes, focusing on various amyloidogenic conditions, including TTR-mutant amyloidosis, sparing off-target anti-inflammatory activity, and reducing the toxic effects of long-term NSAID administration. Recently published studies have focused on optimizing the aryl halogen substituent(s) [42], the linker between the two aryl rings [43] and minimization of cyclooxygenase (COX) inhibition (a common unwanted side effect of TTR inhibitors) [34,38]. However, these studies were solely focused on halogenated biaryl compounds. It is interesting to note that among all compounds tested here, the most potent inhibitors (compounds 15, 16, 20 and 27) are from the non-halogenated fluorenyl family, while the phenyl-substituted relatives (2 halogenated and 2 nonhalogenated) showed comparatively lower inhibition. We seek to further discuss these findings, their correlation with inhibitor binding mode, and reflect on their significance with regard to known inhibitors. In particular, it is noteworthy that the fluorenyl family constitutes a novel class of TTR fibrillization inhibitors.

Insight from aryl analogs
The chemical structures of 11, 13 and all other aryl derivatives discussed in Table 1 may be compared with the previously reported bisaryloxime ethers, potent inhibitors of TTR [39]. For example, 4carboxybenzaldehyde-O-(2-trifluoromethylphenyl)-oxime (OE1) showed more than 90% inhibition at a 7.2 mM concentration against acid mediated fibrils of 3.6 mM TTR. The TTR:OE1 crystal structure is shown in Figure 9a [39]. This inhibitor contains an ortho-trifluoromethylphenyl O-linked through an oxime with a para-substituted benzoic acid and binds to TTR in the forward binding mode. Similar to the TTR:OE1 crystal structure, inhibitors 11 and 13 bind in the typical forward binding mode by pointing their carboxylate functionality towards the outer binding pocket residues Lys15 and Lys15' (Figure 6). In contrast to the bisaryloxime ethers, the inhibitors reported here only contain a single aryl group and the carboxyl group is directly tethered through a flexible linker. Compared to OE1, inhibitors 11 and 13 bind more deeply in the inner binding pocket of TTR (Figure 9a). In the TTR:OE1 structure, the halogenated phenyl ring moves away from the inner binding pocket and does not make significant interactions with the Ser117 and Ser117'. For inhibitors 11 and 13, the aryl ring makes both hydrophobic and hydrogen bond interactions with Ser117, Ser117', Thr119, and Thr119' residues and points its orthosubstituent into the center of the HBP3 and HBP3' pockets. Taken together, the X-ray crystal structural data demonstrate that, like the OE1 bis(aryl) family, compounds 11 and 13 can bind to TTR in the forward binding mode. However, the aryl family reported herein has the advantage of more fully occupying the inner hormone binding pockets, further stabilizing the native tetrameric structure and, in turn, inhibiting amyloid formation.
In the in vitro fibril formation assay, inhibitors 11 and 13 showed 43% and 51% inhibition, respectively, at a 3.6 mM concentration against an equimolar concentration of TTR. Analysis of the fibril inhibitory activity for our somewhat small set of aryl analogs suggests that TTR inhibition scales with the size of the orthosubstituent, with 11 (CF 3 ) and 13 (OMe) having the largest groups at this position. Based on X-ray structural data and docking, the o-CF 3 or o-OMe group is able to interact with HBP3 and HBP3' residues Ala108, Leu110, Ser117 and Thr119. When this moiety is in the metaor paraposition, these interactions with HBP3 appear to be lost, resulting in a loss of activity (cf. compounds 10 and 14 of Table 1). BBased on the crystal structures, docking studies and SAR for the aryl class, we propose that further optimization of the ortho-substituted aryl ring may be achieved through the direct substitution of a halogen atom at the 5-position of the aryl ring which will protrude into the HBP3' while the CF 3 is still anchoring at HBP3 position. In addition, placement of a hydrogen bond donor and/or acceptor, i.e. hydroxyl, at the metaor para-position may lead to enhanced interactions with Ser117/Ser117' and, thus, enhanced inhibitory activity.
Based on the inhibitor efficacy data of aryl compounds 22, 23, 24, 28, 29, and 31 it is clear that further lengthening of the linker affects TTR fibril formation. Compounds 22,23,28,29, and 31 with an aryl moiety substituted with either meta-Me, meta-F, 3,4-OCH 2 O, or the acid isostere CONHSO 2 CH 3, did not exhibit significant activity. However, compound 24, featuring an orthotrifluoromethyl group and the CONHSO 2 CH 3 modification of the linker, showed a 13% improvement in fibril inhibition activity over 11. Based on the docking analysis (Figure 8e), the ortho-CF 3 group was predicted to be critical for anchoring the aryl moiety of 24 to the inner binding pocket, favoring the forward binding mode. Interestingly in the outer binding pocket, the acylsulfonamide moiety of 24 may mimic the negative charge of the carboxylate in 11, while having the potential to make additional hydrogen bonds with Lys15 and Lys15' residues. The sulfonamide pendant methyl group may also interact with the HBP1 and HBP1' pockets of the TTR outer binding pockets. Clearly, further optimization of the terminal carboxylic acid/acid isostere, in conjunction with continued examination of the role of the orthoaryl substituent, will be necessary to clarify the precise structural requirements for enhanced inhibition of fibrillization.

Insight from the fluorenyl compounds
Compared to the aryl family, the fluorenyl derivatives presented herein do not have close analogs that have previously been well-characterized as TTR inhibitors. Dibenzofuran-4,6-dicarboxylic acid (DDBF) [22] and m-trifluoromethyl-substituted N-aryl phenoxazine (PHENOX) have been explored with the hypothesis that the bulkier tricyclic moieties should bind with higher affinity to the TTR hormone binding pocket [22]. The TTR:DDBF cocrystal structure showed that the tricyclic ring of DDBF effectively utilizes the hydrophobicity of the outer binding pocket and binds in a forward binding mode, orienting its directly linked carboxyl moiety towards Lys15. However, the inner binding pocket is completely unoccupied (Figure 9b), and as a result DDBF exhibited only moderate inhibition (46%) at a 3.6 mM concentration [22]. Substituting a halogenated aryl ring for one of the benzene rings of DDBF increased its inhibitory activity to ,70% [22], but the binding mode has yet to be explored [22]. In the case of the TTR:PHENOX structure [22], the inhibitor binds in the outer binding pocket between Glu54 and Thr119 (Figure 9b). The biaryl system facilitated additional hydrophobic interactions with the HBP1 residues ( Figure 9b) and positioned its carboxyl group close to Glu54 to form additional hydrogen bond interactions, resulting in an improved fibrillization inhibition [22] of 73% inhibition at 3.6 mM. The fluorenyl compounds presented in Table 1 constitute a new class of inhibitors as they can span the entire binding pocket and simultaneously interact with both the inner and outer binding pockets. In the TTR:15 and TTR:16 structures, the tricyclic moiety of the inhibitor completely fills the outer binding pocket and the carboxylated chain adopts two conformations to interact with both the HBP3 and the HBP3' residues of the inner binding pocket. Consequently, a watermediated hydrogen bonding network is enabled, engaging neighboring TTR subunits AC and BD (Figure 7). The fluorenyl (outer pocket)-tether (linker space)-carboxyl (inner pocket) architecture of 15 and 16 fully occupies the space between Lys15 and Ser117 and stabilizes the tetrameric structure of TTR by tethering each of its subunits through numerous interactions. The fluorenyl moiety is sandwiched between the hydrophobic groups of residues Lys15, Lys15', Leu17, Leu17', Thr106, Thr106', Ala108, Ala108', Val121 and Val121' in the outer binding pockets of both AC and BD dimer. The carboxyl group simultaneously interacts with Ser117 and Thr119 residues of both HBP3 and HBP3' and facilitates a network of hydrogen bonds in the inner binding pocket connecting the Ser117 residues of all subunits, the nearby water molecules, and the carboxylate substituents of all the inhibitors. In addition, the linker carbon atoms are also sandwiched between the side chain atoms of Leu110 and Leu110' to further augment the inhibitor protomer interactions. Thus, 15 and 16 exhibit approximately 70% inhibitory activity at 3.6 mM concentration and more than 90% inhibitory activity at 7.2 mM against 3.6 mM TTR fibrils. It is interesting to note that the R-enantiomer (17) of the most potent inhibitor (16) of this series is completely inactive, suggesting that the conformation of the linker is critical for the activity of 16. This is, however, not the case with the modified linker compounds 26 and 27, in which the activity remained similar for both Sand R-enantiomers.
The X-ray structural data demonstrate that the aromatic group (phenyl or fluorenyl) is what determines the binding mode of the TTR inhibitor in this series. Particularly, the complementarity of the fluorenyl moiety to the outer binding pocket supports this conclusion. The two-carbon linker of 15 compliments the reverse binding of fluorenyl derivatives and yielded the most potent compound of this series. It is interesting to note that alterations to the linker, including stereodefined a-methylation (to afford 16 and 17) and introduction of an a-methoxyimino moiety (to afford 20), did not improve inhibition of fibrillization. Clearly, the stereochemistry of the appended a-methyl group is important and this information should be of utility in future designs. The limited number of replacements for the terminal carboxylic acid in the fluorenyl series produced mixed results. In the case of substitution of the acid with an acylsulfonamide, we observed a complete loss of activity (15 versus 25), maintenance of inhibition (16 versus 26), and significant gain of efficacy (17 versus 27). Clearly, a larger subset of compounds will be required to clarify this SAR along with the critical structural insights that X-ray crystallography can provide. Similar efforts may also allow us to realize benefit from the use of amide-acid termini, such as in 30 and 31, which currently do not exhibit a benefit compared to the carboxylic acid terminated 15. Furthermore, we will seek to examine the role of substitutions on both the fluorenyl ring and two-carbon linker of 15 to further optimize this potent class of fibrillization inhibitors.

Concluding remarks
Previous studies have shown that diverse families of biaryl compounds possess potent inhibitory activity against TTR amyloidosis. We have evaluated a novel family of inhibitors, featuring an aryl or fluorenyl substituent tethered through a flexible oxime to a carboxylic acid, or methylsulfonamide group. Of the 28 compounds tested, based on the in vitro fibril formation assay, eight compounds showed significant inhibition of TTR amyloidogenesis at a concentration of 3.6 mM, equal to the concentration of tetrameric wt-TTR. The x-ray crystal structures of the TTR-inhibitor complexes presented here illustrates the key molecular features for inhibitor binding and provide the structural basis for the stabilization of the native tetrameric TTR. The crystallographic and docking studies also suggest that the aryl compounds prefer the forward binding mode and the flurorenyl analogs orient in a reverse binding mode. Analysis of the TTR:inhibitor complexes reveals the advantage of the fluorenyl group over the aryl moiety, in combination with the flexible oxime based tether, in tetramer stabilization. With regard to further diversification from T4, none of these fluorenyl compounds, including four potent TTR amyloid inhibitors, contain halogen substitutions, in striking contrast to the majority of previously reported biaryl inhibitors. Fluorenyl compound 15 exhibited the best activity, reducing fibril formation by nearly 70% at 3.6 mM concentration and 95% at 7.2 mM concentration. Another important advantage of 15 is that it is devoid of anti-inflammatory activity in an in vivo carrageenan-induced paw edema assay in rats [31,33]. This selectivity, in contrast to NSAID-based TTR inhibitors such as Flu, is hypothesized to derive from the bulky nature of the fluorenyl moiety. The bulkier fluorenyl system may render 15 incompatible with the comparatively smaller COX enzyme binding pockets (Supplementary Figure S2). Thus, 15 may represent a good starting point for a new family of compounds that stabilize the TTR tetramer, while sparing off-target antiinflammatory activity and reducing the toxic effects of long-term NSAID administration.

Chemistry
Melting points were determined on a Kofler hot-stage apparatus and are uncorrected. 1 (C, H, N).
The E configuration around the C = N double bond of the baminoxy-N-(methylsulfonyl)propionamides 22-24, the (b-aminoxy-propanamido)-acetic acids 28,29 and the propanoic acid 31 was assigned on the basis of the knowledge of the configuration of starting acids and on the basis the observation that in their 1 H NMR spectra the proton linked to the oxymic carbon atom, resonates at chemical shift values identical or very close to those of the same proton of starting acids of E configuration

Molecular modeling
The crystal structure of the TTR:FLU dimeric complex (PDB ID: 1BM7 [27]) was used to generate the correspondent tetramer through the Unit Cell Tool of the Chimera program [44]. All of the inhibitors were submitted to a conformational search of 1000 steps with a 2 kcal/mol energy window and then minimized using the conjugated gradient method until a convergence value of 0.05 kcal/mol by means of the MACROMODEL (Macromodel, ver. 8.5; Schrödinger Inc.: Portland, OR, 1999) program. The algorithm used was the Monte Carlo method with MMFFs as the force field and a distance-dependent dielectric constant of 4.0. The docking studies were carried out through the standard mode of GOLD 3.1 [45], which uses a genetic algorithm search strategy. A binding site was defined as all atoms of TTR within 5 Å of the FLU, and the Cavity Detection algorithm [46] was enabled. The number of generated poses was set to 20, without the early termination option, and the default calculation mode was selected. By default, the GA (Genetic Algorithm) run comprised 100,000 genetic operations on an initial population of 100 members divided into five subpopulations, and the annealing parameters of the fitness function were set at 3.0 for van der Waals and 2.5 for hydrogen bonding. The two fitness functions implemented in Gold, GoldScore was used to identify the better binding mode.

Fibril formation assay
Wild-type TTR was purified from an E. coli expression system and the rate of acid-mediated fibril formation and inhibitor efficacy of each compound was determined by monitoring the turbidity of wt-TTR at pH 4.4, as described previously [22,25,27]. Each compound was dissolved in DMSO at a concentration of 720 mM starting from a primary stock solution of 7.2 mM. Five microliters of a solution of the compound being evaluated was added to 495 mL of a 7.2 mM TTR solution in 10 mM phosphate (pH 7.6), 100 mM KCl, 1 mM EDTA buffer, allowing the compound to incubate with TTR for 30 min at room temperature. Five hundred microliters of 200 mM acetate buffer (pH 4.2), 100 mM KCl, 1 mM EDTA was added to yield final TTR and inhibitor concentrations of 3.6 mM each and a final pH of 4.4. The final 1 mL mixture was vortexed, then incubated at 37uC for 72 hr, after which the tubes were vortexed, and the optical density was measured at 400 nm in a disposable UV cuvette. All assays were performed in triplicate. The percentage of fibril formation was determined by normalizing each optical density by that of TTR without inhibitor, defined to be 100% fibril formation. Control solutions of each compound in the absence of TTR were tested, all compounds were soluble and none absorbed appreciably at 400 nm, ensuring that turbidity was the result of TTR amyloid formation.

X-ray Data collection and structure determination
Crystals of wt-TTR were obtained from 5-7 mg/mL protein solutions (in 100 mM KCl, 1 mM EDTA, 10 mM sodium phosphate, pH 7.0, 0.3 M ammonium sulfate) equilibrated against 2 M ammonium sulfate in hanging drops. All TTR:ligand complexes were prepared from crystals soaked with a fivefold molar excess of ligand for 3-4 weeks to ensure saturation of both binding sites without affecting the diffraction quality of the crystals. An R-axis IV++ detector coupled to an Rigaku Micromax 007 rotating anode X-ray generator was used for data collection of all four complex structures. A single crystal was placed in paratone oil as a cryoprotectant and cooled to 120 K for diffraction experiments. Crystals of all TTR:ligand complex structures are isomorphous with the apo crystal form with the space group P2 1 2 1 2 with two subunits in the asymmetric unit (Table 3). All data sets were processed and scaled with the Crystal Clear suite (Rigaku Corporation).

Structure refinement
The protein atomic coordinates for wt-TTR from the Protein Data Bank (accession number 1BMZ [27]) were used as a starting model during the rigid body refinement in CCP4-Refmac [47,48]. For each binding pocket of the TTR tetramer, the resulting difference Fourier maps revealed two ligand-binding conformations. The electron density corresponding to the inhibitor observed in the inner and outer hormone binding pockets of the BD dimer was well ordered in all TTR:ligand crystal structures (Figures 6  and 7). In contrast, the electron density of the carboxyl substituted region of these inhibitors was only partially ordered in the AC dimer. However, the ligands could be unambiguously placed into the existing density of both hormone binding pockets and were included in the crystallographic refinement. The subsequent mapfitting was done in XtalView/Xfit [49] using the unbiased weighted electron density map calculated by the Shake&wARP bias removal protocol [50]. All binding conformations of the ligand were in good agreement with the unbiased annealed omit maps as well as the Shake&wARP unbiased weighted maps phased in the absence of the inhibitor. Final cycles of the refinement were carried out by the maximum likelihood restrained refinement protocol of CCP4-Refmac [47]. Due to the lack of interpretable electron densities in the final map, the nine N-terminal and three C-terminal residues were not included in the final model. A summary of the crystallographic analysis is presented in Table 3. Water molecules were identified using XtalView/Xfit. The figures were prepared using XtalView [49], Raster3D [51] and UCSF Chimera [44]. The atomic coordinates of the TTR inhibitor complexes have been deposited in the Protein Data Bank with the codes 3GLZ, 3GS7, 3GS4, and 3GS0.