Crystal Structures of Diaryl Hydrazone and Sulfone Stabilizers in Complex with an Amyloidogenic Light Chain Reveal an Alternate Ligand-Binding Cavity

Abstract

Stabilization of amyloidogenic immunoglobulin light chains (LCs) by binding of small molecule “kinetic stabilizers” is under development as a novel treatment for light chain amyloidosis. From a high-throughput screen, we previously identified 16 full-length (FL) LC stabilizers from five distinct chemotypes. We then obtained structural biological information on two classes of hits, coumarins and hydantoins, revealing that both chemotypes bind to a pocket at the V_L—V_L interface of the FL LC dimer. Here, we report crystal structures of three screening hits from two other chemotypes, diaryl hydrazones and sulfones, in complex with an amyloidogenic FL LC. While two of these hits bind to the previously identified pocket, one diaryl hydrazone binds to a different pocket bisected by the C₂ symmetry axis of the dimer. These data further expand on the FL LC stabilizer-binding surface that could be used in design of more potent FL LC aggregation inhibitors.

Introduction

In the disease light chain amyloidosis (AL), the misfolding and aggregation of immunoglobulin light chains (LCs), with or without aberrant proteolysis, seem to drive organ degeneration pathology.^[1,2] The clonal expansion of plasma cells, which secrete the misfolding-prone LCs, is a critical “cancer” component of AL pathology.^[1] Effective diagnosis and treatment of AL are formidable, and the precise non-native LC conformers or aggregate structures that are responsible for organ deterioration are unknown.^[3] Moreover, each AL patient has an essentially unique LC sequence, rendering targeting the LC protein itself as a therapeutic strategy challenging.^[4,5] The most common treatment for AL is to use chemotherapy regimens^[6] or other cytotoxic therapies (e. g., the monoclonal antibody daratumumab^[7]) that eradicate the plasma cells secreting the misfolding-prone LC. However, chemotherapy cannot be tolerated by particularly sick patients, such as those with severe cardiac dysfunction, and does not fully eradicate the plasma cells in all patients, often leading to recurrence after an initial remission period.^[8] We propose the use of a full-length (FL) LC small molecule “kinetic stabilizer” as a mechanistically complementary treatment strategy for the treatment of AL. The kinetic stabilizer molecule binds the natively folded FL LC dimer structure, slowing the rate of FL LC dimer unfolding, and thereby inhibiting the downstream processes of misfolding, aberrant proteolysis, and aggregation that are thought to cause organ toxicity.[3] The small molecule kinetic stabilizer drug tafamidis is a widely used treatment for transthyretin amyloidosis, indicating that kinetic stabilization of misfolding-prone proteins can be effective in slowing disease progression in systemic amyloidosis patients.^[9]

In a previous publication, we reported a high-throughput screen (HTS) of 635,085 small molecules to identify kinetic stabilizers of FL LCs, using WIL-FL as a representative FL LC sequence.^[10] We identified 16 validated screening hits comprising five distinct chemotypes: coumarins, hydantoins, an aryl cyanoacrylamide, diaryl hydrazones, and sulfones (Figure 1). Much of the initial follow-up characterization of the screening hits focused on coumarin 1, for which we obtained a crystal structure in complex with the non-amyloidogenic FL LC JTO-FL.^[10] Later, we used computational docking to predict the binding modes of hydantoin and the aryl cyanoacrylamide chemotypes.^[11] The predicted binding mode of hydantoin 8 was confirmed through x-ray crystallography.^[11] We then utilized the crystallographic binding modes of coumarin 1 and hydantoin 8 to hypothesize a four-component design strategy for creating more potent FL LC stabilizers, which resulted in the discovery of kinetic stabilizers exhibiting nanomolar potency, representing a 3,000-fold potency improvement relative to coumarin 1.^[12,13]

Figure 1.

Chemotypes of FL LC kinetic stabilizers identified through high-throughput screening. The stabilizers highlighted in red are the focus of this study. The EC50 of each stabilizer, as measured by the proteolysis-coupled fluorescence polarization assay used in the high-throughput screen, is denoted. The structures shown in green have been previously crystallized in complex with an FL LC.

The computational docking studies described in our earlier work were unable to produce reasonable docked poses of diaryl hydrazones 5 and 6 to FL LCs.^[11] Moreover, little is known about the binding mode of the sulfone hits (e. g., 17), which were initially de-prioritized from follow-up characterization due to their lower potency (EC₅₀ ~ 18 μM) compared to the other hits (EC₅₀ ~ 1 μM to 5 μM).^[10] Initial crystallization attempts of stabilizers 5 and 6 in complex with JTO-FL were unsuccessful, presumably due to the low solubility of the stabilizers. Structural biological information on diverse FL LC stabilizer chemotypes has proven to be invaluable for the design of LC stabilizers with improved potency and selectivity, as the different kinetic stabilizer chemotypes engage in different interactions with different FL LC binding site residues.^[11–13] We were interested in solving structures of all of the different kinetic stabilizer chemotypes bound to FL LCs to gain a comprehensive understanding of the binding mode(s) of each LC stabilizer chemotype from the initial HTS. This effort can therefore provide a detailed map of the FL LC binding surface(s) that could potentially be targeted by an optimized stabilizer.^[12,13] Based on the principles of structure-based drug design, this structural information can then be used to design and synthesize more potent and selective stabilizers that engage with FL LC binding pocket residues.^[14]

Toward this goal, we describe an improved FL LC · stabilizer crystallization protocol to more reliably generate diffraction-quality crystals while allowing crystallization of poorly soluble, small molecule stabilizers. This improved procedure was able to generate crystals and crystal structures of three more screening hits: diaryl hydrazones 5 and 6 and sulfone 17 in complex with an amyloidogenic FL LC, H9-FL. Herein, we report the first FL LC · stabilizer crystal structures utilizing an amyloidogenic FL LC in contrast to previously reported structures that were based on a more stable, non-amyloidogenic FL LC.^[10–13] From the perspective of mapping out new sites or subsites for stabilizing FL LCs by binding of small molecules, the crystal structure of stabilizer 5 in complex with H9-FL reveals an alternative stabilizer binding pocket formed by the V_L-V_L interface. This pocket could potentially be leveraged for the design of more potent FL LC kinetic stabilizers.

Results

Crystallization of H9-FL and Introduction of Small Molecule Stabilizers

Our existing crystallographic data on coumarin- and hydantoin-based FL LC small molecule kinetic stabilizers employed JTO-FL as the FL LC sequence.^[10,15] JTO-FL is non-amyloidogenic; however, it is structurally very similar to amyloidogenic non-liganded FL LCs such as WIL-FL^[16,17], which was employed in initial stabilizer · LC crystallization attempts, although these proved unsuccessful. In our existing JTO-FL · kinetic stabilizer crystallization protocol, we first grow crystals of the FL LC at room temperature and then introduce the small molecule stabilizer to the pre-formed crystals. Notably, JTO-FL crystals can exhibit two different morphologies: the first being diamond-shaped and the second consisting of plate-shaped crystals. Both crystal forms may appear in the same crystallization drop. However, only plate-shaped crystals provide useful diffraction data, as introducing small molecules to diamond-shaped crystals damages them. Our initial crystallization system employing JTO-FL had two disadvantages that limited its utility. First, the plate-shaped crystals that produced useful diffraction data were difficult to grow reproducibly, as most JTO-FL crystals formed in the diamond morphology. Second, JTO-FL is a non-amyloidogenic FL LC, and despite its “native state” being structurally similar to the native state structure of amyloidogenic FL LCs, structural data of kinetic stabilizers in complex with amyloidogenic FL LCs would have higher relevancy for AL. We therefore sought an alternative amyloidogenic FL LC sequence for crystallization to study kinetic stabilizer binding.

In 2017, Oberti and coworkers reported the crystal structures of several non-amyloidogenic and amyloidogenic FL LCs, which revealed that both types of FL LCs adopted very similar crystal structures of the presumed native state, despite displaying a wide range of stabilities in solution.^[17] Of the crystallized sequences, H9-FL is an amyloidogenic FL LC that diffracted to the highest resolution (1.7 Å). We therefore expressed and purified H9-FL from E. coli and found that this FL LC can be reliably crystallized under a variety of conditions at 4 °C. These crystals can be soaked with small molecule stabilizers at a high concentration (> 1 mM) affording H9-FL · kinetic stabilizer diffraction data at resolutions ranging from 1.8 to 2.3 Å. H9-FL was therefore chosen as the FL LC sequence to be used in crystallographic studies going forward. We then successfully obtained H9-FL dimer · kinetic stabilizer crystal structures of three screening hits (kinetic stabilizers 5, 6, and 17; Figure 1) representing two chemotypes (diaryl hydrazones and sulfones) that previously had no structural biological information available.

Crystal Structure of Diaryl Hydrazone 6 in Complex with H9-FL

The crystal structure of stabilizer 6 in complex with H9-FL at 2.0 Å resolution reveals that stabilizer 6 binds in the same general hydrophobic region, termed the “core hydrophobic pocket” comprising the “anchor cavity” and the “aromatic slit” (Figure 2A–B, Table 1), that is occupied by coumarin- and hydantoin-based kinetic stabilizers.^[12] Stabilizer 6 contains a 2-phenyl-2-methylethanal-derived substructure that forms a hydrazone with the hydrazine substructure of the 1,3,4-trisubstituted nitrophenylsulfonamide, resulting in a 4-atom spacer between both aromatic rings (Figure 1). Only the (S) enantiomer of stabilizer 6 can be observed in the electron density map (Figure S1). The hydrazone and 1,3,4-trisubstituted phenyl ring occupy the slit-like cavity (the “aromatic slit”) formed by F98 of one LC protomer and P44’ of the other protomer, with the phenyl ring serving as the “aromatic core” substructure of stabilizer 6 (Figure 2A–B; light chain residues are numbered according to the Kabat system, and the prime symbol designates residues of the second protomer of the LC dimer). The hydrazone moiety in 6, which is buried from solvent, appears to form an intramolecular hydrogen bond with the nitro substituent of the phenyl ring. The nitro group of 6 also is in hydrogen bonding distance to the backbone amide of residue L46’. Formation of this hydrogen bond appears to be a common feature employed by other stabilizers that occupy this pocket. Because nitro groups are generally considered to be poor hydrogen bond acceptors,^[18] it is unclear to what extent these potential hydrogen bonding interactions contribute to binding energy. The sulfonamide substituent of the phenyl ring extends towards the exit of the binding pocket into the solvent-exposed region. This sulfonamide is in hydrogen bonding distance with the backbone carbonyl oxygen of residue G99, although a sizable contribution of this solvent-exposed H-bond to binding energy is unlikely.

Figure 2.

Crystal structure of diaryl hydrazone 6 (magenta, PDB: 8FO3) in complex with H9-FL (gray). (A) Overall structure of the H9-FL LC dimer bound with one molecule of 6 (magenta) in the “core hydrophobic pocket” formed at the VL-VL interface. (B) Close-up of stabilizer 6 (magenta) in the LC dimer binding site, referred to as the “core hydrophobic pocket”, which can be subdivided into the “anchor cavity” and the “aromatic slit”. Residues that comprise the binding site are labeled. Potential hydrogen bonds and their estimated distances are indicated with dashed lines. (C) Left: chemical structures of stabilizers 62 and 63. Right: alignment of the binding of 6 to H9-FL with the binding of 62 (pink) to JTO-FL (dark gray, PDB: 7LMQ) and 63 (blue) to JTO-FL (PDB: 7LMR). For clarity, atoms corresponding to the solvent-exposed Nmethylmorpholinyl substructures of 62 and 63 have been removed.

Table 1.

Data collection and refinement statistics for the H9-FL · 6, H9-FL · 5, H9-FL · 17 crystal structures.

Structure PDB code	H9-FL · 6 8FO3	H9-FL · 5 8FO4	H9-FL · 17 8FO5
Data collection
Beamline	APS 23-ID-B	APS 23-ID-B	ALS 5.0.1
Space group	P1	P1	P1
(a, b, c) (Å)	63.14, 95.65, 125.92	45.49, 63.19, 86.98	46.67, 63.57, 87.24
(α, ß, γ) (°)	106.31, 92.93, 90.16	84.76, 87.70, 81.18	84.39, 85.51, 80.09
Resolution range (Å)	47.58–2.00 (2.11–2.00)	48.58–2.00 (2.11–2.00)	48.69–1.89 (2.00–1.89)
Unique reflections	182,621 (26,347)	61,668 (8,686)	74,530 (9,836)
Completeness (%)	96.4 (95.0)	95.9 (92.4)	95.6 (86.6)
Rsym	0.086 (0.381)	0.083 (0.537)	0.073 (0.437)
Rpim	0.054 (0.241)	0.055 (0.341)	0.047 (0.275)
CC(1/2)	1.00 (0.87)	1.00 (0.79)	1.00 (0.85)
I/σ (I)	8.7 (3.4)	8.8 (2.4)	10.9 (2.9)
Redundancy	3.5 (3.4)	3.4 (3.4)	3.5 (3.5)
Wilson B factor (Å2)	22	21	23
No. complexes/asu	6	2	2
Refinement
Resolution range (Å)	47.62–2.00	48.62–2.00	48.73–1.89
No. reflections – work	173,255	58,568	70,774
No. reflections – free	9,361	3,085	3,736
Rwork	0.229	0.186	0.201
Rfree	0.280	0.238	0.250
RMS bond length (Å)	0.010	0.010	0.009
RMS bond angle (°)	1.55	1.58	1.58
Mean B value (Å2)
overall	35	37	30
protein	36	36	29
ligand	34	33	26
buffer	43	50	48
water	38	44	37
Ramachandran favored (%)	95.2	96.6	96.1
Ramachandran allowed (%)	99.6	100.0	100.0
Clashscore	4.26	3.46	5.12
No. atoms
total	20,516	7,079	7,477
protein	18,952	6,274	6,356
ligand	144	88	34
buffer	30	10	20
water	1390	707	1067

We compared the binding of 6 to H9-FL to structures of the previously reported coumarin-based stabilizers 62 and 63 bound to JTO-FL (Figure 2C).^[12] All three phenyl ring-comprising anchor substructures bind in a similar orientation and at the same site in FL LCs, while the aromatic cores of the stabilizers are oriented more divergently in the aromatic slit and form a π-π stacking aromatic interaction with F98. The displacement of the aromatic core relative to F98 is dependent on the number of atoms in the spacer between the two aromatic rings. This displacement is greatest for the four-atom spacer of stabilizer 6, less for the three-atom of stabilizer 63, and least for the two-atom spacer associated with stabilizer 62.

Crystal Structure of Diaryl Hydrazone 5 in Complex with H9-FL

The crystal structure of diaryl hydrazone kinetic stabilizer 5 in complex with H9-FL at 2.0 Å resolution reveals that kinetic stabilizer 5 binds in a different small molecule binding pocket in the FL LC dimer (Figure 3A) than in all other stabilizer · FL LC complexes that have been structurally characterized to date.^[10–13] Importantly, the FL LC dimer conformation required to bind 5 is different than that required to bind all previously published FL LC kinetic stabilizers. The FL LC dimer binding pocket occupied by 5 is bisected by the C₂ symmetry axis of the LC dimer and comprises a different region of the V_L-V_L interface compared to that utilized to bind to other kinetic stabilizers. We therefore refer to this FL LC stabilizer binding pocket as the “C₂ symmetry pocket.” Since a symmetry axis bisects the C₂ symmetry pocket, two symmetry-related molecules of stabilizer 5 can be modeled into the electron density, although only one stabilizer molecule can bind one LC dimer at a time (Figure S2). Despite the stark differences in the binding pocket utilized, kinetic stabilizer 5 has comparable activity to the other top kinetic stabilizer hits reported in our previous publication.^[10]

Figure 3.

Crystal structure of diaryl hydrazone 5 (orange, PDB: 8FO4) in complex with H9-FL (gray). (A) Overview of the binding site of 5 (orange), which occupies a deep pocket bisected by the C2 symmetry axis of the VL-VL interface in the H9-FL LC dimer. (B) Subdivision of 5 into three distinct substructures that are used to describe its binding. (C) Close-up of the binding of 5 to H9-FL. Residues comprising the binding site are labeled. Potential hydrogen bonds and their estimated distances are indicated by dashed lines. Ordered water molecules are depicted as red spheres. (D) Conservation of the “C2 symmetry pocket” binding site residues in H9-FL. “Conservation” refers to the proportion of 347 AL-associated LC sequences in AL-Base that include the given residue.

To describe the binding of stabilizer 5 to H9-FL, we subdivided the structure of 5 into three distinct substructures: the “base”, the “linker”, and the “headpiece” (Figure 3B). The deepest part of the C₂ symmetry pocket overlaps with the previously described “anchor cavity” that binds the “anchor substructure” component of other LC kinetic stabilizers reported (Figure 2B).^[12] The nitrophenyl “base substructure” of stabilizer 5 occupies the “hydrophobic base cavity,” formed by the Q38, P44, and Y87 residues of both LC protomers and overlapping with the previously described “anchor cavity.” The two Q38 residues comprise the floor of the hydrophobic base cavity and hydrogen bond with each other through their side chains (Figure 3C). The conjugated alkene-hydrazone linker occupies a narrow “tunnel” that is formed by the Y36 and F98 residues from both protomers (Figure 3C). Notably, this tunnel is not occupied by any previously known structurally characterized kinetic stabilizers. The hydrazone N–H is within hydrogen bonding distance of the hydroxyl group of the Y36 side chain. The pyrimidine “headpiece substructure” of stabilizer 5 occupies a cavity that is closest to the surface of the LC dimer. The ceiling of the “headpiece binding cavity” is formed by the Y91 residues of each protomer (Figure 3C). Serine residues (S34, S89) from both protomers and two ordered water molecules form the sides of the “headpiece binding cavity”. The pyrimidine substructure hydrogen bonds with the serine side chains via ordered water molecules.

To determine whether the “C₂ symmetry pocket” could exist in other FL LC sequences adopting a dimeric structure, we examined the conservation of the binding site residues among 347 AL sequences from AL-Base, a database for LC sequences (Figure 3D).^[4] The residues that comprise the “hydrophobic base cavity” (Q38, P44, Y87), which overlaps with the “anchor cavity” used for binding by most stabilizers (albeit in a different protein conformation), are known to be highly conserved, appearing in > 95 % of LC sequences. The Y36 residues, which form a narrow tunnel occupied by the hydrazone linker substructure of 5, are also highly conserved. However, the residues that comprise the “headpiece binding cavity”, including S34, S89, and Y91, are poorly to moderately conserved, appearing in 50 % or fewer of AL FL LC sequences.

We compared the crystal structures of unliganded H9-FL, H9-FL · 5, and H9-FL · 6 to assess conformational changes that occur upon binding the small molecule kinetic stabilizers. Unbound H9-FL adopts a dimeric structure containing the “core hydrophobic pocket” that was characterized in our previous work (Figure 4A).^[10] To our knowledge, H9-FL is the only full-length light chain sequence with a known crystal structure in which the pocket is present in the unbound FL LC dimer. The H9-FL · 6 structure does not exhibit significant conformational differences compared to the unliganded H9-FL structure,^[17] as the “anchor cavity” and the “aromatic slit” comprising the “core hydrophobic pocket” and required for binding of 6 are already present in the apo FL LC dimer structure. Other FL LC dimers with available crystal structures, such as JTO-FL, do not contain this binding pocket in the unbound structure, but reveal it upon binding to a variety of small molecule kinetic stabilizers.^[10] The stabilizer binding-competent conformation in JTO-FL is formed by a rotation of the two V_L-domains relative to each other.^[10]

Figure 4.

Comparison of crystal structures of unbound H9-FL (gray, PDB: 5 M6 A), H9-FL · 6 (pink, PDB: 8FO3) and H9-FL · 5 (yellow, PDB: 8FO4). (A) Van der Waals surface representation of the VL-domains of each FL LC conformation. The core hydrophobic pocket is labeled and is only present in unbound H9-FL and H9-FL · 6. (B) Small molecule stabilizer binding pocket of diaryl hydrazones 6 and 5 are shown in two different views (“front” and “side”). The binding pockets are labeled in the “front view” images (left two panels). The binding pocket subregions are labeled in the “side view” images (right two panels). (C) Cartoon overlay of H9-FL · 6 and H9-FL · 5 structures. (D) Binding site residue overlay to illustrate residue-specific conformational changes between unbound H9-FL, H9-FL · 6, and H9-FL · 5, shown in two different views (front and top). For clarity, the bound stabilizers have been removed.

However, the FL LC structure harboring the “C₂ symmetry pocket” required for binding of 5 by H9-FL contains notable conformational differences compared to H9-FL · 6 or apo H9-FL (Figure 4B). Although the sizes of the hydrophobic base cavity in H9-FL · 5 and the overlapping anchor cavity in H9-FL · 6 are similar, the major conformational differences relate to the accessibility of neighboring binding subsites. In particular, the “aromatic slit,” which binds the “aromatic core” substructure of most FL LC stabilizers, is absent in the H9-FL · 5 structure (compare Figure 4B panel labeled H9-FL · 6, “side,” to H9-FL · 5, “side.”) The narrow tunnel formed by the Y36 and F98 residues is enlarged and complete compared to the conformations of apo H9-FL and H9-FL · 6, wherein the tunnel is incompletely formed (Figure 4B). Finally, the cavity that binds the pyrimidine “headpiece” of stabilizer 5 is enlarged compared to the cavity in the absence of stabilizer 5. The primary conformational change differentiating the FL LC conformations of unbound H9-FL/H9-FL · 6 and H9-FL · 5 is a rotation of one of the V_L-domains relative to the other (Figure 4C). The distance between the Cα residues of F98 and P44’ decreases from 10.3 Å in unbound H9-FL to 7.9 Å in H9-FL · 5, which closes the aromatic slit when stabilizer 5 is bound. The enlargement and complete formation of the tunnel and the headpiece cavity when 5 is bound is partially enabled by displacement of the Y36’, S34’, and S89’ residues, with root-mean-square deviation (RMSD) values of 1.92, 1.54, and 1.54 Å respectively, when comparing unbound H9-FL and H9-FL · 5 (Figure 4D). The most notable conformational change is a ~ 60° rotation of the Y91 residue on each protomer. These residues form a “lid” at the top of this “C₂ symmetry pocket,” and the side-chain rotation enlarges the headpiece cavity. Comparing unbound H9-FL and H9-FL · 5, the RMSD values of Y91 and Y91’ are 3.17 and 3.24 Å, respectively. Thus, the “C₂ symmetry pocket” and the “core hydrophobic pocket,” have different backbone and side-chain conformational requirements and cannot exist simultaneously.

Crystal Structure of Sulfone 17 in Complex with H9-FL

The crystal structure of kinetic stabilizer 17 in complex with H9-FL at 1.9 Å resolution reveals that stabilizer 17 binds in the “core hydrophobic pocket” comprising the “anchor cavity” and the “aromatic slit”.^[12] The core hydrophobic pocket of the FL LC dimer binding site is occupied by stabilizer 6, as well as by coumarin 1 and its analogs and by hydantoin 8, as described in earlier publications (Figures 5A–B and S3, Table 1).^[10–12] The phenyl and sulfone substructures of 17 serve as the “anchor substructure” and occupy the hydrophobic “anchor cavity” (Figure 5C). The acetylated pyrrole serves as the “aromatic core” and binds the “aromatic slit”-binding subsite comprising F98 within the FL LC dimer (Figure 5C). The ketone substituent is positioned to hydrogen bond with the backbone amide of L46’. However, the distance between the ketone oxygen and the backbone amide proton is 3.6 Å, so it is likely that any potential interaction would be relatively weak in terms of contributing to binding energy.

Figure 5.

Crystal structure of sulfone 17 (green, PDB: 8FO5) in complex with H9-FL (gray). (A) Overview of the binding site of 17 (green), which binds the “core hydrophobic pocket” comprising the “anchor cavity” and “aromatic slit.” The “core hydrophobic pocket” is the binding site occupied by most other FL LC stabilizer chemotypes. (B) Close-up of the “core hydrophobic pocket” binding site. 17 is overlaid with 6 (purple) for comparison. (C) Labeling of binding site residues. The distance between the ketone of 17 and the backbone amide proton of L46’ is indicated with a gray dashed line.

Discussion

The crystal structures of three chemically distinct small molecule kinetic stabilizers in complex with the amyloidogenic FL LC dimer, H9-FL, reaffirm the importance of previously characterized interactions and binding sites for the molecular recognition of 6 and 17. However, a different “C₂ symmetry pocket” is revealed within the FL LC dimer that binds kinetic stabilizer 5, which could be further leveraged for the design of potent kinetic stabilizers. Diaryl hydrazone 6 and sulfone 17 bind in a similar mode to all previously characterized coumarin and hydantoin-based FL LC stabilizers.^[10–12] These binding modes share common features: (1) the deepest part of the “core hydrophobic pocket” is occupied by a hydrophobic substructure termed the “anchor substructure,” and (2) the narrow “aromatic slit” serves as the cavity opening and is occupied by an aromatic substructure of the stabilizer called the “aromatic core” that makes a π stacking aromatic interaction with the side chain of F98. These features are energetically favorable for high affinity binding and FL LC dimer kinetic stabilization. Design of new LC stabilizer chemotypes with improved properties should satisfy these interactions. The crystal structures of H9-FL · 6 and H9-FL · 17 reveal unoccupied space in the core hydrophobic pocket that could serve as initial avenues for optimizing binding affinity and selectivity. Notably, stabilizer 17 leaves significant unoccupied space in the “aromatic slit” cavity of the binding pocket, and the potential hydrogen bond that 17 forms with L46’ may be too long to be optimal. These factors could help explain its lower potency compared to other stabilizer screening hits. Because coumarin-based stabilizers have liabilities owing to their metabolic instability,^[12] sulfone 17 could serve as a useful starting point for alternative structure-based drug design.

Comparison of the crystal structures of H9-FL · 6 to JTO-FL · 62, JTO-FL · 63 and other previously reported coumarin-based stabilizers^[12] reveals similarities and a relationship between the binding of the anchor substructure and the binding of the aromatic core – one that could be utilized in structure-based optimization. The longer the anchor substructure, the greater the shift in the position of the aromatic core towards the exit of the “aromatic slit” binding pocket. A carefully chosen anchor substructure enables monocyclic aromatic cores to be optimally placed in the aromatic slit, i. e. enabling an optimal π-π stacking aromatic interaction with F98.

The H9-FL · 5 crystal structure reveals a small molecule kinetic stabilizer binding site that we name the “C₂ symmetry pocket.” The hydrophobic base cavity (Q38, P44, Y87) overlaps with the anchor cavity of the “core hydrophobic pocket.” However, it is unlikely that the “C₂ symmetry pocket” and the “core hydrophobic pocket” are concurrently accessible, as the crystallographic data indicate that the solvent-accessible aromatic slit is closed when the C₂ symmetry pocket is formed and occupied by a stabilizer such as 5. The mutual exclusivity of these two binding sites is supported by our previously reported competition binding experiments using the fluorogenic coumarin stabilizer 1 (binds to the same site as 6). In this competition binding experiment, we titrated an FL LC (20 nM to 150 μM) into 1 (1 μM) in the presence of the non-fluorescent stabilizer 5 (10 μM).^[10] Stabilizer 5 reduced the occupancy of stabilizer 1 because the stabilizers either occupy the “core hydrophobic pocket” (1, 6) or the overlapping “C₂ symmetry pocket” (5) lacking the “aromatic slit” binding subsite. Simultaneous binding of 5 and either 1 or 6 seems to be precluded. However, the H9-FL · 5 structure reveals that we may be able to design kinetic stabilizers targeting the “core hydrophobic pocket” and that occupy the anchor cavity, the aromatic slit, and the Y36/F98 tunnel, which appears to be at least partially accessible when the anchor cavity and the aromatic slit are occupied by stabilizers like 1 or 6, compared to the expanded tunnel found in the H9-FL · 5 structure. This design strategy is supported by the conservation of the amino acid residues making up the binding sites, as residues forming the anchor cavity, aromatic slit, and tunnel are highly conserved,^[10] while residues that form the “headpiece cavity” of the “C₂ symmetry pocket” are less conserved. Targeting the most conserved residues in designing potent FL LC dimer stabilizers is most likely to result in lead stabilizers with high affinity for most AL LC sequences.

The existence of the “C₂ symmetry pocket” was broadly established by early work done by Edmundson and coworkers and later studied by Brumshtein and coworkers.^[19–22] The original reports, which used a different amyloidogenic LC known as “MCG,” described a long, cylindrical pocket that can accommodate a variety of small aromatic molecules and synthetic peptides. Edmundson and Brumshtein subdivided the cavity into three subsites: A, B, and C, with the A subsite located closest to the surface of the protein and the C subsite being the deepest. The C subsite equivalent in H9-FL comprises residues Y36, L46, Q38, and Y87 and therefore includes the “hydrophobic base cavity” and the “tunnel” as we defined these regions of the FL LC dimer.^[21,22] The B subsite includes residues F98, which also line the tunnel, and S34 and S89, which comprise the headpiece binding cavity. The A subsite includes Y91, which forms a “lid” over the headpiece binding cavity. Aromatic ligands studied by Edmundson non-selectively bound any of the three sites in MCG-FL when soaked into pre-formed crystals, while others preferentially bound one or two of the sites.^[19,22] However, no ligand appeared to bind all three sites at once. A later study from Brumshtein identified one small molecule, sulfasalazine, that occupied all three sites.^[21] However, the isolated V_L-domains of MCG were used as a construct for those studies instead of MCG-FL. MCG-V_L can dimerize in a conformation that is not observed in FL LCs, so the binding pocket of sulfasalazine is not present in FL LCs. While sulfasalazine can stabilize MCG-V_L, in our hands, it cannot stabilize FL LCs,^[10] which may be due to the inaccessible binding pocket conformation in FL LCs.

Our work describes the first high-resolution crystallographic data from small molecule 5 bound in the “C₂ symmetry pocket” as it exists in a conformation accessible in FL LC dimers. This molecule binds all three subsites that were described in earlier studies, which may be a critical feature for stabilization of FL LC dimers. Occupancy of the “hydrophobic base cavity” (C subsite) appears to be essential for small molecule stabilization of FL LCs, regardless of whether the “C₂ symmetry pocket” or the “core hydrophobic pocket” are occupied. In the case of MCG-FL, the C₂ symmetry pocket is described as highly flexible and able to accommodate a variety of ligands through induced fit. However, much of this flexibility appears to be derived from residues forming the A subsite, which are poorly conserved among LC sequences due to being part of hypervariable complementarity-determining region (CDR) loops. It is unclear to what extent this flexibility can be observed in other FL LC sequences, which may an important consideration for designing stabilizers that bind the headpiece cavity (A and B subsites).

The C₂ symmetry pocket could be potentially leveraged for the design of covalent FL LC kinetic stabilizers, which could be utilized as first-in-class small molecule probes of FL LCs or be further developed as drug candidates, provided they exhibit sufficient FL LC dimer conjugation selectivity. The Y36 and Y36’ residue side chains, in addition to being highly conserved, are oriented directly at the pyrimidine headpiece of 5. We envision analogues of 5 being synthesized in which a tyrosine-reactive covalent warhead, such as a sulfonyl fluoride or an arylfluorosulfate,^[23,24] is attached to the headpiece or linker of the stabilizer, allowing for formation of a covalent bond with the hydroxyl oxygen of Y36 or Y36’. One limitation to this design approach is that it is unclear if the tyrosine hydroxyl groups are sufficiently nucleophilic to undergo sulfur (VI) exchange chemistry, as there are no nearby basic residues (e. g., Lys, Arg) that are known to increase Tyr-OH reactivity through pKa perturbation and possibly by catalyzing fluoride displacement.^[24,25]

Materials and Methods

Expression and Purification of H9-FL

The coding sequence for H9-FL was ordered as a gBlock and cloned into the pET11a vector. Using this expression construct, H9-FL was expressed and purified from inclusion bodies in E. coli according to the previously described methods for purification of FL LCs.^[26]

Crystallization, Data Collection, and Refinement

Purified H9-FL was concentrated to 200 μM dimer in phosphate-buffered saline (PBS, pH 7.4). We were unable to grow crystals of H9-FL using reported literature conditions.^[17] Therefore, we submitted a sample for high-throughput crystallization trials on our robotic CrystalMation system (Rigaku) at the Scripps Research Institute. Crystals of H9-FL were grown using sitting-drop vapor diffusion by mixing 2 μL of H9-FL and 2 μL of a crystallization buffer consisting of 20 % PEG 3350 and 200 mM KH₂PO₄ (among multiple other salts in 20 % PEG 3350) at 4 °C. Crystals reached maximal size after 1–2 days and appeared as clusters of plates. Small molecule stabilizer solutions were prepared by diluting 100 mM stabilizer in DMSO stock solutions into 50 % PEG 3350 to achieve 10 mM stabilizer. A portion (0.2 μL) of the resulting 10 mM solutions were added to crystallization drops, which were incubated at 4 °C for at least 2–3 days. Crystals were cryo-protected by brief immersion in a 25 % glycerol and 1 mM small molecule stabilizer solution in crystallization buffer and flash cooled in liquid nitrogen. For H9-FL · 6 and H9-FL · 5, data were collected at beamline 23-ID–B at the Advanced Photon Source (Argonne, IL) at a wavelength of 1.0332 Å and a temperature of 100 K. For H9-FL · 17, data were collected at beamline 5.0.1 at the Advanced Light Source (Berkeley, CA) at a wavelength of 0.97741 Å and a temperature of 100 K. Frames were indexed and integrated using XDS,^[27] the space group was assigned as P1 using Pointless, and data were scaled using Scala.^[28] Five percent of reflections (randomly distributed) were flagged for model cross-validation using R_free.^[29] It is of note that H9-FL · 6 data were assigned markedly different unit cell parameters by XDS (indicative of a larger unit cell) compared to H9-FL · 5 or H9-FL · 17 (Table 1). We were unable to successfully index and integrate the H9-FL · 6 data by assigning the unit cell parameters of H9-FL · 5 or H9-FL · 17. The larger unit cell of H9-FL · 6 is likely due to an uncommon occurrence of growing two different crystal forms of H9-FL in the same drops, one of which differed from those observed in other H9-FL structures. It seems unlikely that introducing stabilizer 6 would have caused alteration of the original crystal form to this extent.

The H9-FL · 5, H9-FL · 17, and H9-FL · 6 structures were solved by molecular replacement (MR) with Phaser,^[30] using the unbound H9-FL dimer (PDB code: 5M6A^[17]) as a search model. Two copies of H9-FL · 5 and H9-FL · 17, and six copies of H9-FL · 6, were found in the crystal asymmetric units. Small molecule stabilizer CIF dictionaries were generated using the Grade Web Server. Each model was refined with iterative cycles of manual adjustment in Coot^[31] and refinement in Refmac5 using isotropic thermal parameters and hydrogen atoms at calculated positions.^[32] Final adjustments were made after analysis with MolProbity and the wwPDB Validation System.^[33] The crystal structures were deposited to the Protein Data Bank with accession codes of 8FO3 (H9-FL · 5), 8FO4 (H9-FL · 6), and 8FO5 (H9-FL · 17).

Supplementary Material

Supporting information for this article is available on the WWW under https://doi.org/10.1002/ijch.202300002

Data Availability Statement

The crystal structures associated with this study were deposited to the Protein Data Bank (https://www.rcsb.org/) with accession codes of 8FO3 (H9-FL · 5), 8FO4 (H9-FL · 6), and 8FO5 (H9-FL · 17).