Small molecule stabilization of diverse amyloidogenic immunoglobulin light chains revealed by hydrogen-deuterium exchange mass spectrometry

Abstract

Immunoglobulin light chains, a component of antibodies, can misfold and aggregate to cause systemic AL amyloidosis. Aggregation, including amyloid fibril formation, requires unfolding of the full-length light chain from its native state, and in most cases aberrant proteolysis. Small molecules that bind to the native state of light chains to stabilize them against conformational excursions and proteolysis are under development as drug candidates for AL amyloidosis. Since each patient has a unique light chain sequence, a challenge for candidate stabilizer drugs is to bind multiple light chains and suppress their dynamics. Here, we used hydrogen-deuterium exchange coupled to mass spectrometry to characterize the binding of small molecule stabilizers to λ light chain proteins. We measured the effects of six small molecules on a previously studied light chain, then characterized a further six amyloidogenic light chains in the presence and absence of the most efficacious stabilizer. Despite structural and dynamic differences among the light chains, stabilizer binding led to decreased hydrogen exchange rates, consistent with reduced local and global unfolding. Protection upon binding was most prominent in residues within complementarity determining region 3 and framework region 4 of the light chain variable domains, which undergo major conformational changes enabling amyloid formation. Stabilizer binding reduced the rate at which all light chains were cleaved by protease, and delayed amyloid formation in vitro. These data show that these stabilizers suppress the range of conformational dynamics associated with light chain aggregation, supporting their therapeutic potential.

Introduction

Unfolding and subsequent aberrant proteolysis of full-length (FL) immunoglobulin light chain (LC) proteins are generally required for the formation of amyloid light chain (AL) deposits in the disease AL amyloidosis (Figure S1), one of a spectrum of systemic amyloidoses [1]. Accumulation of amyloid fibrils derived from normally soluble LCs in multiple tissues leads to progressive and eventually fatal organ dysfunction [2,3]. Soluble, misfolded LCs can also contribute to cardiotoxicity that is the major cause of mortality [3]. FL LCs are subunits of antibodies that normally function in the immune system to recognize antigens [4]. Antibodies are covalent heterotetramers of two FL LCs and two heavy chains, which circulate in blood alongside “free” LC monomers and homodimers—FL LCs are produced in excess to enable efficient antibody secretion [5]. Overproduction of a single monoclonal free LC by an aberrant clonal cell population can cause LC fragment accumulation as amyloid fibrils within tissues. Each AL patient has an essentially unique LC sequence that forms amyloid [6]. The amino acid sequence of each LC appears to define its propensity to aggregate [4,7]. Amyloid formation is also driven, in part, by increased concentration of the monoclonal free FL LC or fragments thereof, so existing therapies kill the clonal cells to reduce the total amyloidogenic LC concentration [2,3]. An alternative and potentially less toxic approach is to prevent secreted LCs from misfolding and aggregating. Protein conformational control by small molecules that modulate protein assembly, folding and stability have become an important class of drugs, targeting proteins associated with multiple disease categories [8]. Kinetic stabilizers are a subset of these molecules, which selectively bind to the native state and slow the rate of protein unfolding. Kinetic stabilization of circulating LCs should protect residual free LCs from misfolding, aberrant proteolysis and aggregation, thus complementing cytotoxic chemotherapies [9], and this concept is currently being tested in clinical trials.

There are two types of LC, κ and λ, of which λ LCs are more frequently involved in AL amyloidosis [3,6,10]. FL LCs are secreted as 210–220-residue proteins that fold into two independent structural domains (Figure 1A), a variable (V_L) domain and a constant (C_L) domain [4,5]. Both domains have a ß-sandwich structure, termed an immunoglobulin (Ig) fold, stabilized by an intramolecular disulfide bond. The N-terminal variable domain (residues 1–113 in the alignment shown in Figures 1B and 1C) is involved in antigen binding when incorporated into antibodies via three highly polymorphic loops known as complementarity determining regions (CDR, Figure 1C). Other residue segments, defined as framework regions (FR), are more conserved. The C-terminal constant LC domain (residues 114–219) has a structural role within antibodies, and its sequence is highly conserved between LCs. Free FL LCs can form homodimers in circulation. Such free LC dimers have a parallel orientation where the V_L and C_L domains from one protomer interact with their counterpart in the other protomer (Figure 1A). Residues from ß-strands C_V, C’_V, F_V and G_V form a V_L-V_L interface, while residues from ß-strands D_C, E_C, B_C and A_C form the C_L-C_L interface. A cysteine residue near the C-terminus of C_L (referred to here as C218, relative to the alignment) can make an intermolecular disulfide bond within the homodimer. In patients, amyloidogenic FL LCs can form both monomers and homodimers [11–13].

Figure 1: Kinetic stabilizers to inhibit aggregation of diverse antibody light chains.

(A) Structures of MM LC, an example full-length LC, in complex with kinetic stabilizer M63 (PDB entry 7LMR). Left, a single LC molecule comprises N-terminal variable (V_L, yellow) and C-terminal constant (C_L, grey) domains. Each domain harbors an intramolecular disulfide bond, shown in stick representation. The ß-strands that form the homodimeric interface are indicated. Middle, overall structure of a stabilizer-bound LC homodimer. Right, detail of stabilizer M63 bound at the interface between the V_L domains M63 is shown in stick representation. (B) Sequence logo plot of 554 V_L domains of λ LCs taken from AL-Base. The height of each letter is proportional to its frequency in the alignment and the overall height of the column is proportional to the information content at that position. Amino acids interacting with kinetic stabilizers are indicated according to the numbering used in the original crystal structure (PDB code 6MG5) and color-coded according to the substructure of the M83 stabilizer to which they bind (see the structure of M83 in panel (D): red, anchor substructure; blue, aromatic core; green, linker module; purple, distal substructure). (C) Multiple sequence alignment of the 9 selected LCs explored in this study. The shading indicates the degree of conservation of amino acid residues. V_L domain framework regions (FR1–4) and complementarity determining regions (CDR1–3) are shown in green and yellow, respectively. Blue and red boxes show the locations of the native secondary structure elements. Individual ß-strands are labeled. Intramolecular disulfide bonds are shown by orange lines. Note that the IMGT regions are defined to allow comparison between different immunoglobulin domains and do not correspond precisely to LC secondary structure elements. (D) Structures of the kinetic stabilizers used in this study. M83, the stabilizer molecule that showed the highest potency in LC stabilization, has its substructures highlighted.

The structured cores of patient-derived amyloid fibrils primarily comprise residues from the V_L domain in non-native, cross-ß-sheet conformations [14,15]. Isolated V_L domains can aggregate under mild conditions in vitro, whereas FL LCs remain soluble for extended periods under the same conditions and are much more resistant to aggregation [16–18]. Amyloid-associated V_L domains tend to unfold and aggregate more readily than non-AL V_L domains, although substantial variation in aggregation rates has been observed [7,19–22]. Furthermore, amyloid-associated FL LCs are generally less conformationally stable than their non-AL counterparts and are more prone to local or global unfolding that enables aberrant endoproteolysis, releasing aggregation-prone fragments, including the V_L domain [16,23–26]. In some cases, proteolysis can occur after aggregation [27]. Amyloid fibrils formed by V_L domains can accelerate, or seed, the aggregation of full-length LCs [18].

Collectively, these data suggest a role for the local conformational dynamics in sensitive protein regions. These regions have been proposed to involve: i) domain-domain interfaces in free LCs, such as the V_L-V_L and V_L-C_L interfaces [28–30]; ii) regions involving the internal disulfide bond in V_L, a region that unfolds during amyloid formation [14,31]; iii) amyloid-promoting regions (APRs), which are sequence regions with high intrinsic aggregation propensity that are proposed to initiate protein misfolding [31–33]; iv) mutation, misfolding or proteolysis of the C_L domains [23,26,34–36]. The roles of these sensitive regions in amyloid formation may differ for different LC sequences, complicating their therapeutic targeting.

Because LCs have diverse sequences and no natural small-molecule ligands, historically they epitomized a challenging drug target [37]. Nonetheless, we carried out a high-throughput screen that identified small molecules that protect folded FL LCs from undergoing conformational excursions enabling proteolysis, compounds that we call LC kinetic stabilizers; these molecules bind highly conserved residues at the V_L-V_L interface of the LC homodimer [38] (Figure 1B, colored residues). Binding of stabilizer molecules enhances the dimerization affinity of the two VL domains. For V_L domains and FL LCs without an intermolecular disulfide bond, which equilibrate between monomeric and dimeric states, stabilizer binding increases the proportion of dimer in solution. Within a disulfide-linked dimer, stabilizer binding is hypothesized to suppress domain movements and transient unfolding. Subsequent structure-based medicinal chemistry efforts led to more potent kinetic stabilizers (Figure 1D) [39–43]. Screening and optimization of hit molecules was mainly carried out using a single full-length LC protein, termed WIL-FL. A key challenge for ongoing drug development is to identify kinetic stabilizers that bind tightly and specifically to multiple LCs with diverse sequences. These kinetic stabilizers slow LC unfolding by strengthening the V_L-V_L domain interface within the LC homodimer, which is expected to decelerate proteolysis and aggregation. However, the effects of small-molecule kinetic stabilizers on localized unfolding and proteolysis of a broader collection of FL LCs associated with amyloidosis have not been scrutinized experimentally.

To explore the conformational dynamics of multiple FL LCs we used hydrogen-deuterium exchange (HDX) mass spectrometry (MS). Previously, we used HDX-MS to investigate the native structure and dynamics of isolated V_L domains, FL LCs, and IgG proteins to identify specific regions that potentially contribute to protein misfolding enabling aberrant proteolysis [31,32,44]. In the current study we investigated localized unfolding in nine different FL LCs and the influence that the binding of small-molecule kinetic stabilizers has on these dynamics. The stabilizers explored were either screening hits (P1 and P5, Figure 1D) or structure-based designed stabilizers (B20, M63, M53 and M83). We provide molecular details about stabilizers’ mechanism of action and gain insights to guide the development of novel therapies for the treatment of AL amyloidosis.

Materials and Methods

Protein selection, expression and purification

Nine LC proteins selected for analysis are listed in Table 1. Amino acid sequences of these LCs are listed in the Supplementary Information. All residues are numbered according to the sequence alignment shown in Figure 1C. Domain boundaries and CDR residues are defined according to the IMunoGeneTics (IMGT) system [45]. Cysteine 218 in the alignment, which forms an intermolecular disulphide bond, corresponds to residue 214 in the Kabat numbering system [46] and residue 127 of the C-region in the IMGT numbering system. Variable and constant germline genes were assigned using IMGT’s tools for peptide and nucleotide analysis [47], as appropriate.

Table 1:

LC proteins used in the current study. Genes are defined according to the IMGT system. PDB identifiers are provided where experimental structures are available. For AL LC and H16 LC, structural models were generated using AlphaFold3.

Name (alternative names)	Germline variable gene	Germline constant gene	Sequence NCBI accession	PDB code	Reference	Disease
AL (AL–150L; 01–095)	IGLV6–57	IGLC2	EF589390	N/A	[70]	AL
MCG	IGLV2-8	IGLC1	3MCG_1	3MCG	[71]	AL
H3 (AL30)	IGLV1-44	IGLC3	KC433670	5MTL	[24]	AL
H6 (AL50)	IGLV1-51	IGLC2	KY471433	5MUD	[24]	AL
H7 (AL48)	IGLV1-51	IGLC3	KC433671	5MUH	[24]	AL
H9 (AL52)	IGLV2-8	IGLC2	KY471435	5M6A	[24]	AL
H16 (AL54)	IGLV2-14	IGLC3	KY471437	N/A	[19]	AL
MM (JTO)	IGLV6-57	IGLC3	6MG4_A	6MG4; 1CD0	[32]	MM
GL (6aJL2)	IGLV6-57	IGLC2	2W0K_A (VJ only)	2W0K	[32]	Germline sequence

Each LC listed in Table 1 was expressed as a full-length protein containing Cys218, which predominantly forms disulfide-linked dimers when expressed in E. coli [32]. We refer to these tabulated LCs as “AL LC”, “MCG LC”, etc. We also explored three additional proteins: the stand-alone V_L domains of the AL and MM sequences (AL V_L and MM V_L) and the AL C218S variant (Kabat: C214), where replacement of the cysteine prevents formation of the intermolecular disulfide bond. In total, 12 proteins were studied. Unless otherwise specified, experiments were carried out in phosphate buffered saline (PBS), defined here as 20 mM sodium phosphate buffer, pH 7.4, containing 150 mM NaCl. For the AL and H16 proteins where no crystal structure was available, unliganded models were generated using the AlphaFold 3 algorithm, implemented via www.alphafoldserver.com [48]. For each model, two copies of the full-length sequence were used as the input with the default parameters. The relative orientations of the V_L domains within each model were similar but the “elbow” angle between the V_L and C_L domains varied; for clarity, only the V_L domains are shown when multiple structures are compared.

Variants of AL, MM and JTO LCs were expressed in E. coli with N-terminal polyhistidine tags and purified by affinity chromatography as previously described [32]. Untagged MCG, H3, H6, H7, H9 and H16 were expressed as inclusion bodies and purified by ion exchange and size exclusion chromatography, as previously described [24,42]. Protein purity was 95%+ verified by SDS-PAGE (Figure S2). Oligomeric states were determined by size exclusion chromatography (Figure S3). Protein molar concentration is reported as monomer equivalents. For LCs characterized to date, stabilizers bind with a stoichiometry of one small molecule to two LC monomers.

Small-molecule stabilizers

We tested six small-molecule kinetic stabilizers shown in Figure 1D and described in our previous work [37]: P1, P5, B20, M53, M63 and M83. Numbering corresponds to that from the original paper [40]. Letters indicate the origin of the molecules: P1 and P5 were originally identified as hits in a high-throughput screen [38]; B20, M53, M63 and M83 were designed and synthesized during medicinal chemistry efforts to improve stabilization efficacy [40,41]. These molecules are elaborations on the coumarin scaffold of P1. B20 and M53 retain the original diethylamino substitution in position 7 of the coumarin heterocycle. M63 and M83 incorporate phenylethoxy modifications that were optimized for binding within the core of the LC dimer. Synthesis and purification of these molecules was carried out as previously described [40]. Additional M83 was synthesized by BioDuro Ltd, Irvine, CA, USA. Small molecules were solubilized in dimethyl sulfoxide (DMSO) vehicle and diluted to the desired concentration. DMSO concentrations were 1.5% during equilibration and 0.15% during deuterium labeling unless otherwise specified.

Sequence analysis

All AL-associated λ LCs with complete V_L domain sequences were taken from AL-Base, available via https://wwwapp.bumc.bu.edu/BEDAC_ALBase/, and aligned according to the IMGT numbering system. Fully conserved gaps, which are used to facilitate structural comparisons between different types of immunoglobulin domains, were removed from the alignment, leaving 113 occupied positions. This alignment was visualized using the ggseqlogo [49] package in R v 4.2.2 [50], and the binding site residues were highlighted manually.

The sequence-based consensus method AmylPred2 [51] was used to predict the amyloidogenic sequence propensities of the nine LCs under study. APRs are defined as segments predicted by 6 or more of the 10 algorithms available via the AmlyPred2 website.

Analytical size exclusion chromatography

Protein (AL V_L, 0.24 mg/mL, 20 μM) was incubated for 30 minutes at 37°C in the presence or absence of M83 (100 μM) in PBS, pH 7.4, containing 1 % (v/v) DMSO. Following incubation, the protein mixture was loaded onto a Superdex^® 75 10/300 column (Cytiva) equilibrated with PBS, pH 7.4, and eluted using an AKTA Pure FPLC system (Cytiva) at a flow rate of 0.5 mL/min, while monitoring the absorbance at 280 nm. M83 was added at a 1:5 (mol:mol) ratio. The column was calibrated using protein standards (Sigma): blue dextran (2000 kDa), bovine serum albumin (66 kDa), carbonic anhydrase (29 kDa), cytochrome C (12.4 kDa), and aprotinin (6.5 kDa).

Hydrogen deuterium exchange mass spectrometry (HDX-MS)

At least two independent HDX-MS replicates were measured for each LC in accordance with current recommendations [52,53]. Experimental parameters, full details of replicate acquisitions for each protein state, and all processed values are found in the Supplemental Data File. HDX-MS data have been deposited to the ProteomeXchange Consortium via the PRIDE repository [54] with the dataset identifier PXD068745.

Deuterium labeling

LCs were freshly diluted to 30 μM in biological replicates from stock concentrations in PBS, pH 7.4, prepared in H₂O. Before labeling, the protein solutions were mixed with an equal volume of 300 μM solution of stabilizer molecule in PBS containing 3% (v/v) DMSO. Deuterium labeling commenced with a 10-fold dilution (18 μL) of the D₂O labeling buffer (10 mM sodium phosphate, 150 mM NaCl, pD 7.4, 99.9% D₂O) added to 2 μL of each 15 μM protein solution. After each labeling time (10 seconds, 1 minute, 10 minutes, 1 hour, 4 hours, 16 hours) at 20 °C, the labeling reaction was quenched with the addition of 20 μL of ice-cold quenching buffer [200 mM sodium phosphate, 4 M guanidine hydrochloride (GdnHCl), 0.72 M TCEP, pH 2.37, H₂O] and analyzed immediately by liquid chromatography coupled mass spectrometry (LC-MS). The maximally deuterated samples of each protein that were used for back-exchange correction were prepared as described [55]. Protein solutions (15 μL at 15 μM) were lyophilized, resuspended in 7 M GdnHCl containing 50 mM DTT (15 μL), and heated at 90 °C for 5 min. After cooling to 20 °C, 18 μL of labeling buffer (as above) was added to 2 μL of cooled, denatured protein solution, and the exchange reaction was allowed to proceed at 50 °C for 10 min. These maximally deuterated protein samples were then cooled to 0 °C, exchange quenched with the addition of 20 μL of ice-cold quenching buffer (as above) and analyzed immediately by LC-MS.

Mass spectrometry

LC-MS was performed with a Waters HDX system containing an HDX unit and two Acquity I-class UPLC pumps [56]. Deuterated and control samples were digested online in the HDX cooling unit, where the digestion chamber was held at 15 °C, using an AffiPro Pepsin column (AffiPro, AP-PC-001, 2.1 mm × 20 mm). Peptides were trapped and desalted on a VanGuard PreColumn trap [2.1 mm × 5 mm, ACQUITY UPLC BEH C18, 1.7 μm, (Waters, 186002346)] for 3 minutes at 100 μL/min. Peptides were then eluted from the trap using a 5%–35% gradient of acetonitrile over 6 minutes at a flow rate of 100 μL/min, and separated using an ACQUITY UPLC HSS T3, 1.8 μm, 1.0 mm × 50 mm column (Waters, 186003535). The main cooling chamber of the Waters HDX unit, which housed all the chromatographic elements, was held at 0.0 ± 0.1 °C for the entire time of the measurements. The back pressure averaged ~9000 psi at 0 °C and 5% acetonitrile 95% water, 0.1% formic acid. To eliminate peptide carryover, a wash solution (1.5 M GdnHCl, 0.8% formic acid and 4% acetonitrile) was injected over the pepsin column during each analytical run. Mass spectra were acquired using a Waters Synapt XS HDMSE mass spectrometer operated in ion mobility mode. The mass spectrometer was calibrated with direct infusion of a solution of Glu-fibrinopeptide (Sigma, F3261) at 200 femtomole/μL at a flow rate of 5 μL/min prior to data collection. A conventional electrospray source was used, and the instrument was scanned over the range 50 to 2000 m/z. The instrument configuration was: capillary voltage 2.5 kV, trap collision energy at 4 V, sampling cone at 35 V, source temperature of 80 °C, and desolvation temperature of 175 °C. All experiments were done under identical experimental conditions. In this experimental setup, the error in determining peptide deuterium levels was ±0.20 Da, as established using deuterated peptide standards.

Data processing

Peptides were identified using PLGS 3.0.1 software (Waters, 720001408EN) with replicates of undeuterated samples used as controls. Raw MS data were imported into DynamX 3.0 (Waters, 720005145EN) and filtered with minimum consecutive products of 2 and minimum number of products per amino acid of 0.25. The peptides meeting the filtering criteria were further processed automatically by DynamX followed by manual inspection of all spectra and data processing. The amount of deuterium in each peptide was determined by subtracting the centroid mass of the undeuterated form of each peptide from that of the deuterated form, at each time point, for each condition. Correction for back exchange to report percent deuteration (%D) was done as described [57] according to the equation:

$$\%D_{\mathit{\text{level}}}=\frac{(m_t-m_0)}{(m_{\mathit{\text{maxD}}}-m_0)}$$

where m_t is the observed peptide centroid mass at a given labeling time-point t, m₀ is the undeuterated peptide centroid mass, and m_maxD is the maximally deuterated peptide centroid mass.

Peptide-level “chiclet” plot heatmaps were used to visualize the HDX-MS data [58]. These visualizations directly show the acquired data, but because the analyzed peptides may not be equivalent between different proteins it can be difficult to make direct comparisons. To more easily visualize differences between LCs, we used the PyHDX web server to reduce the peptide deuteration level information to the single amino acid level [59]. We assumed that both subunits of the LC dimers were equivalent. All measured and processed values can be found in the Excel supplemental data file. Residue-level percent deuteration values, corrected for back-exchange, were used to generate skyline plots, heat maps, difference heat maps and structural images. Data were plotted onto structural models of the appropriate LC using PyMOL (Schrödinger LLC). For a simple comparison between LCs, we calculated the area under the exchange curve (AUC) over the residues with measurable HDX data in all experiments, using a trapezoidal method.

PLIMSTEX analysis of binding affinity

Binding affinity between small-molecule stabilizers and proteins was measured by HDX-MS using the PLIMSTEX approach [60]. Binding isotherms were generated by incubating 5 μM of protein (AL LC, C218S, or V_L) with 2-fold serial dilutions of stabilizer molecules (500 to 1.9 μM) at 20°C in 10 mM sodium phosphate buffer, pH 7.4, containing 150 mM NaCl and 5% DMSO (v/v). After 30 minutes of equilibration at room temperature, 3 μL of the protein-ligand mixture was diluted with 57 μL of deuterium labeling buffer (10 mM sodium phosphate, 150 mM NaCl, pD 7.4, 99.9% D₂O, final DMSO 0.25% v/v) and incubated for 1 hour and 30 minutes at 20°C. After dilution, protein concentration was 0.25 μM and small-molecule stabilizers ranged in concentration from 0.095 to 25 μM. The reaction was acid-quenched and immediately injected into the mass spectrometer for analysis. All experiments were performed in duplicate.

PLIMSTEX experiments were conducted on a Waters SELECT Series CyclicIMS system coupled to two I-Class UPLC pumps (Waters). The separation gradient was 5–35% acetonitrile over 3 minutes at a 100 μL/min flow rate. The mass spectrometer was calibrated using the Major Mix (Waters) and operated with the following parameters: capillary voltage 2.8 kV, trap collision energy 4 V, sampling cone 30 V, source temperature 80°C, and desolvation temperature 175°C.

For the PLIMSTEX data analysis of AL LC titration with M83, the peptide fragment ¹⁰¹NWVFGGGTKL¹¹⁰ (numbered according to the sequence alignment, equivalent to residues 98–107 of the AL LC sequence) was selected. This peptide was chosen because it contains residues that form the stabilizer binding site in WIL-FL, notably F104 [38]. Accordingly, it exhibited one of the strongest responses to kinetic stabilizer binding.

Limited proteolysis

LCs (0.2 mg/mL, ~8.3 μM) were incubated in PBS, pH 7.4, containing 1% DMSO at 37°C with agitation at 450 rpm in the presence or absence of 100 μM of M83. Proteolysis was initiated by adding trypsin (Promega, 0.5 mg/mL stock concentration, ~20.8 uM). At designated time intervals, aliquots (10 μg) of each proteolysis mixture were quenched with 0.2 % v/v formic acid and separated by denaturing electrophoresis on 10–20% acrylamide gradient Tris-Glycine gels under reducing conditions, followed by Coomassie staining. Intact LCs was quantified by densitometric analysis of the gel bands (Gel Analyzer v23.1, Istvan Lazar Jr. and Istvan Lazar Sr., available at www.gelanalyzer.com). Residual LC fractions were calculated for each time point. Proteolysis kinetics were measured over 24 h. Proteolysis rates were calculated by fitting the data to a single exponential decay, and areas under the curve (AUC) calculated using a trapezoidal method. These data were used to optimize a single timepoint experiment to determine the maximum protection afforded by kinetic stabilizer binding. Enzyme to substrate (E:S) ratios ranged from 1:16 to 1:64 (w:w) and incubation times from 5 hours to 24 hours, depending on the inherent proteolytic susceptibility of each LC variant.

Aggregation kinetics

LCs were induced to aggregate in PBS, pH 7.4, in 96 well, clear bottomed, black walled plates (Corning #3631), as previously published [30–32]. Plates were sealed black vinyl film (Nunc, #236703) and covered with a lid to minimize evaporation. Samples were centrifuged at 20,000 g for 10 minutes to remove pre-formed aggregates after thawing and filtered before use. Protocols were different for FL LCs, which aggregate only after proteolysis with trypsin, and isolated V_L domains, which aggregate spontaneously upon shaking, for consistency with our prior studies. Data were visualized without further processing, using R v. 4.2.2 [ref 50].

For FL LCs (n=3 technical replicates per condition), proteins were incubated at 0.3 mg/mL (12.5 μM) in buffer containing 10 μM thioflavin T, 0.01 mg/ml trypsin and either 60 μM M83 in 1% DMSO or 1% DMSO vehicle control. Samples were shaken continuously at 37 °C in an Infinite PRO 200 microplate reader (Tecan Trading AG, Switzerland) and fluorescence readings (λex = 450 nm, λem = 482 nm) were taken every 10 minutes.

For V_L domains (n=6 technical replicates per condition), proteins were incubated at 10 μM LC in buffer containing containing 2 μM thioflavin T and 10 μM M83 in 1% DMSO or 1% DMSO vehicle control. Samples were incubated at 37 °C, 500 rpm on an orbital shaker and fluorescence readings (λex = 440 nm, λem = 480 nm) were taken at regular time intervals in a SpectraMax M5 platereader (Molecular Devices, San Jose, CA, USA).

Results

Structural dynamics vary among different light chains

We investigated seven FL λ LC proteins whose sequences were originally identified in patients with AL amyloidosis, along with two FL λ LC control sequences (Figure 1, Table 1). We chose amyloidogenic sequences based on the availability of structural and dynamic data on these sequences. As in our prior study [32], our controls were two non-amyloidogenic λ LCs: MM, originally identified in a patient with multiple myeloma but without amyloid deposition; and GL, the protein with the precursor germline sequence from which the AL and MM sequences are derived, IGLV6–57. These sequences are derived from germline genes which are over-represented in AL amyloidosis compared to the normal repertoire [6,10]. All proteins were expressed in E. coli and purified using column chromatography (Figure S2). All FL LCs eluted from a size exclusion column at a volume consistent with their predicted size as obligate, disulfide-linked dimers, although MM FL exhibited a broad peak that may be due to interactions with the column matrix (Figure S3).

To identify APRs within the FL LC sequences we used the AmylPred2 server [51], which combines the results of multiple algorithms, producing a consensus score from 0 to 10. Similar regions of sequences were identified in each FL LC, although the strength of the predictions varied between sequences (Figure 2A). V_L domains harbored two (H6) to six (GL, MM, H16 and MCG) APRs (AmylPred2 score > 5). Two segments in the C_L domain, common to all nine FL LCs, were also predicted as APRs. These segments predominantly form ß-strands within the native LC structure (Figure 2A) but may drive self-association upon local or global unfolding.

Figure 2: Hydrogen-deuterium exchange profiles of LCs.

(A) Sequence alignment of the nine LCs under study, highlighting in purple the residues with amyloidogenic propensity as predicted by the AmylPred2 webserver. The shading represents the number of algorithms within AmylPred2 that identified each residue as amyloidogenic. (B) Skyline plot showing the percent of deuterium incorporation (%D) across the aligned LC amino acid sequences at 1-minute labeling time point. Gray shading represents CDRs 1, 2 and 3 within V_L and residues 133–139 within C_L. Top, middle, and bottom panels correspond to LCs derived from variable gene families IGLV6, IGLV1 and IGLV2, respectively. Dashed lines show the boundary between the variable and constant domains. The secondary structure scheme is shown at the top of the panels, with blue boxes representing ß-sheets and red indicating α-helices. Intramolecular disulfide bonds are shown by orange lines. (C) Deuterium uptake plots (%D) are shown for specific peptides covering regions of interest, including CDR1, CDR2, CDR3, and the constant region (C_L). Each line corresponds to a specific LC variant, with peptide sequence numbers indicated. Note that these plots are not directly comparable due to sequence differences between the peptides. Complete HDX-MS results for all proteins and labeling times are reported in Figures S5, S8 and S9.

We first employed HDX-MS to investigate the local backbone conformational dynamics of FL LCs under identical conditions in the absence of kinetic stabilizer small molecules. HDX-MS provides detailed insights into solvent exposure and hydrogen bond stability by monitoring the exchange of backbone amide hydrogens with deuterons in the solvent [61]. The exchange was initiated under native-like conditions where the protein was folded. Greater deuterium incorporation reflected greater conformational dynamics and/or solvent exposure. Exchange was quenched at specific timepoints, from 10 sec to 16 h, by denaturation of the protein at acidic pH, where the average intrinsic rate of backbone amide HDX is minimal, followed by peptic digestion to generate peptides for LC-MS analysis. The extent of exchange at different time points was determined by measuring deuterium incorporation relative to that of the undeuterated protein, and was normalized relative to the experimental maximal deuterium uptake (maxD) observed for each peptide, as previously described [55]. For each LC, sequence coverage exceeded 95% (Figure S4).

To facilitate comparisons among different isoforms, HDX-MS results were visualized as skyline plots, which show percent deuteration at each residue interpolated from peptide-level exchange data. Figure 2B shows the data for 1 minute of exchange and Figure S5 shows other timepoints. Peptide-level data are included in the Supplemental data file. No systemic differences between FL LCs with a histidine tag (AL, MM and GL) compared to other (untagged) FL LCs were observed by HDX (Figure 2B and S5). Figure 2C compares deuterium uptake kinetics for representative peptides in specific regions of interest, including CDR1, CDR2, CDR3, and the residue segment 133–139 from C_L. The latter segment was selected because it partially overlaps with the amyloidogenic region 136–145 and encompasses a known proteolytic hotspot whose cleavage can generate a V_L-containing LC fragment found in human amyloid deposits [26,27,62]. We observed large variations in HDX within and between the FL LCs, mainly in the V_L domains, reflecting their variable sequences. As expected, the solvent-exposed CDR loops exhibit more deuteration, corresponding to the low structural protection of amides from HDX, as compared to the FRs, which contain more hydrogen-bonded β-sheets. The deuteration level in each region varied among different LCs. The least protected regions were adjacent to the APRs identified by AmylPred2. However, portions of these APRs showed low exchange indicating high protection by the native LC structure. This included high protection in C21 and C140 that are located in amyloidogenic segments and form internal disulfides in the V_L and C_L domains, respectively. The counterpart cysteine residues in the intermolecular disulfide bonds, C92 and C200, respectively, are located in regions of variable solvent exposure: C92 resides in a relatively exposed area near CDR3, whereas C200 is positioned within a more protected region.

As expected, the extent of C_L deuterium incorporation is similar for the nine LCs studied, reflecting the sequence and structural conservation of C_L. Most differences in HDX were observed in the first 20 residues of C_L which are adjacent to the V_L domain. Within these residues, segments from AL, MM, GL, H3 and H16 showed lower structural protection (greater exchange) after 1 minute than those of H6, H7, H9 and MCG. These differences may be associated with differential interactions between the two domains.

Overall, no clear and strict correlation was identified between the observed deuteration levels in specific regions and the amyloid-forming properties of the LCs under study. However, compared to the non-amyloidogenic isoforms GL and MM, which share a very similar HDX profile, all amyloidogenic counterparts showed lower protection in the V_L domain, but not in the C_L domain. AL, H3, H9, and H16 exhibited generally rapid exchange across the entire V_L, particularly in the CDRs. MCG showed greater deprotection in CDR1 and CDR3 at shorter time points, while FR3 exchanged to a greater extent at longer time points. H6 and H7 were more difficult to compare due to differences in peptide lengths and positions, but they exhibited increased uptake in FR4, FR1, and FR3 while remaining more protected in CDR1 and CDR2.

Stabilizer binding reduces FL AL LC deuterium incorporation

We next investigated the effects of stabilizer molecules on deuterium uptake of AL LC. HDX-MS was carried out using 1.5 μM AL LC in the presence of 15 μM stabilizer. These concentrations are comparable with the circulating concentrations of free LCs [63] and an approved kinetic stabilizer drug for transthyretin amyloidosis [64]. Figure 3A shows differences in normalized deuterium uptake of AL LC peptic peptides in the presence of P1, P5, B20, M53, M63 and M83, compared with the vehicle control (0.15% v/v DMSO), presented as peptide-level heatmaps (“chiclet plots”) across all labeling times. For clarity, only peptides from the V_L domain are shown; complete data are shown in Figure S6. Example data for the peptide covering residues 101–110 (¹⁰¹NWVFGGGTKL¹¹⁰, numbered according to the sequence alignment in Figure 1, corresponding to sequential positions 98–107 of AL LC) is shown in Figure 3B. This peptide comprises residues in CDR3 and FR4 and also forms part of the V_L-V_L interface. No differences were detected at short labeling times; at longer times, increased stabilizer-induced protection from HDX was detected for four out of six stabilizer molecules (M83, B20, M53, P1). The region with the greatest change in protection from HDX was residues 92–110 (equivalent to sequential positions 89–107), which extends from C92 of the intra-domain disulfide bond through CDR3 to FR4. An additional site exhibiting less increased protection from HDX was observed between residues 34 and 48 (Figure 3A). Only minor stabilizer-induced changes were detected in C_L domain, which were most evident for residues 165–180 in the presence of M83 (Figure S6). Stabilizer molecules P5 and M63 did not alter deuterium uptake. The other four stabilizers resulted in increased protection from HDX in the order M83>B20>M53>P1. This is the same order as the small-molecules’ efficacy determined in the assay that was originally used for drug optimization, which measured protection of the WIL-FL LC from endoproteolysis [38,40]. Similar residues were protected from exchange by all four kinetic stabilizers.

Figure 3: Differential stabilization of AL LC by six small molecule stabilizers.

(A) Peptide-level chiclet plot of deuterium uptake of FL AL LC in the presence or absence of 15 μM kinetic stabilizers. Only peptides originating from the V_L domain are shown; complete data are provided in Figure S6. “N” and “C” indicate the N- and C-termini, respectively, of the peptic peptides, numbered according to the sequence of AL LC. The AL LC peptide ¹⁰¹NWVFGGGTKL¹¹⁰ (corresponding to sequential positions 98–107) is highlighted with a red arrow. Vehicle (Veh) represents DMSO at 0.15% v/v during exchange. (B) Deuterium uptake curves for peptide ¹⁰¹NWVFGGGTKL¹¹⁰. (C) PLIMSTEX curves for peptide ¹⁰¹NWVFGGGTKL¹¹⁰ titrated with P1 (blue), B20 (green), M53 (magenta), and M83 (red) are shown as a function of the stabilizer molecule concentration.

To determine the relative efficacy of the small-molecule stabilizers, we measured deuterium uptake by AL LC (0.25 μM) as a function of stabilizer concentration. M83, B20, M53, and P1 were titrated from 0.2 to 25 μM. We analyzed the data using the PLIMSTEX approach (Protein Ligand Interaction by Mass Spectrometry, Titration, and EXchange) [60,65]. In a PLIMSTEX experiment the labeling time is fixed, and protein-ligand interaction is monitored through changes in deuterium incorporation across a titration series of the ligand. AL LC peptide ¹⁰¹NWVFGGGTKL¹¹⁰, described above, was selected for PLIMSTEX analysis. This peptide exhibits a large change in deuterium uptake upon stabilizer binding. The analysis was conducted at a labeling time of 1 hour and 30 minutes, chosen to capture the maximum observable protection from exchange while allowing the system to approach the near steady-state equilibrium of the exchangeable hydrogens [65]. PLIMSTEX analysis of the HDX data indicated that M83 decreases deuterium uptake at concentrations 100-fold lower than that of the other stabilizer molecules (Figure 3C).

We chose to focus the remainder of our studies on M83, the most efficacious stabilizer, in order to investigate the mechanism by which M83 protects AL LC from exchange. We used two additional forms of the AL protein: AL LC C218S, which cannot form the intermolecular disulfide linking the C_L domains; and AL V_L, the stand-alone V_L domain. Both these proteins equilibrate between monomer and non-covalent dimer in solution, with monomer being the major species under the conditions used.

Figures 4A–4C show HDX-MS data from AL LC, AL C218S and AL V_L, respectively, in the presence or absence of M83. Consistent with our previous studies, unliganded AL LC exhibited greater protection from HDX than AL C218S or AL V_L (difference heatmaps in the right-most columns of Figures 4B and 4C). These data are consistent with progressive stabilization associated with the presence of the C_L domain and intramolecular disulfide bond that has been observed in other FL LCs [28,31,32,66]. Figure S7 shows the same data as peptide-level chiclet plots. Binding of the stabilizer molecule led to a large decrease in deuteration (i.e., an increase in HDX protection) observed from middle to long exchange times. For V_L, this change in deuteration was no longer apparent at 16 hours, likely because both bound and unbound states of the protein had reached equilibrium with the solvent. Notably, this increased protection from HDX was observed in similar V_L segments of all three protein constructs. These segments were also more protected in AL LC than in AL C218S or AL V_L, indicating that stabilizer binding mimics the effect of increased dimer formation that was previously observed for these proteins [32]. Unlike V_L, the C_L domain showed only a small increase in protection upon stabilizer binding to FL LCs, mainly in residues 165–180, which are also more protected in AL LC than in AL C218S. The observed deuterium incorporation was only slightly reduced in M83-bound AL LC compared to M83-bound AL C218S. Accordingly, the change in deuterium incorporation upon M83 binding observed for AL C218S was greater than that observed for AL LC.

Figure 4: Effects of the kinetic stabilizer M83 binding to AL constructs monitored by HDX-MS.

Residue-level heatmaps display percent deuterium incorporation (%D) for the proteins in the presence or absence of M83, as well as the corresponding differences in deuterium uptake (Δ%D). “Veh” represents DMSO vehicle at 0.15% v/v during exchange. Horizontal lines indicate the boundaries between the variable (V_L) and constant (C_L) domains. (A) AL LC; (B) AL C218S; (C) AL V_L. Peptide-level data are provided in Figure S7. Differences between constructs in the absence of M83 (C218S – LC and V_L – LC) are also shown. The protection induced by the kinetic stabilizer is greatest at peptides corresponding to the dimeric interface regions, but is not restricted to these residues. (D) Change in deuteration (Δ%D [(AL LC + M83) - AL LC]) after 1 hour of labeling mapped onto an AlphaFold3 model of the AL LC dimeric structure. Regions of increased protection are shown in shades of blue. The names of the β-strands within the V_L domain are shown for one monomer.

We mapped the differences in AL LC deuteration in the presence and absence of M83 onto an AlphaFold3 model of AL LC, since no experimental structure of this FL LC is available (Figure 4D). Multiple backbone regions exhibited increased protection from HDX in the presence of M83. The greatest increase in protection was observed in peptides that form the V_L-V_L interface. Altered exchange in this region is consistent with M83 binding in a similar site to that observed for other stabilizer molecules in co-crystal structures with LCs [38–40,42,43]. In addition, residues from the solvent-exposed ß-strands C_V, C’_V, F_V and H_V, of the V_L domain also show increased protection upon M83 binding. C_L residues 165–180, which show a small increase in protection upon stabilizer binding, form the C_L-C_L interface and are located close to the C_L-V_L domain interface.

Kinetic stabilizer binding enhances self-association of V_L domains

To further investigate the interaction between stabilizer binding, dimerization and HDX, we applied the PLIMSTEX approach, as above, to measure the apparent affinity of M83 for the three AL constructs. AL C218S and AL V_L can equilibrate between monomeric and dimeric forms in solution [32], but are expected to be primarily monomeric at the concentrations used for HDX-MS (1.5 μM) in the absence of M83. We asked whether binding of M83 could yield a form with protection from HDX equivalent to that of the unbound or bound disulfide-linked dimer. M83 was titrated against each AL construct and the deuterium incorporation measured after 90 minutes. Raw mass spectra of peptide ¹⁰¹NWVFGGGTKL¹¹⁰ at varying concentrations of M83 are presented in Figure 5A, and the averaged centroid values of the spectra plotted against concentration of M83 are shown in Figure 5B. Without M83, deuterium incorporation for peptide ¹⁰¹NWVFGGGTKL¹¹⁰ differed significantly among the three AL constructs, reflecting differences in their intrinsic stabilities shown, consistent with previous experiments [32]. AL V_L exhibited the highest deuterium uptake (D₀ = 6.34 Da, Figure 5A), AL C218S showed intermediate uptake (D₀ = 5.20 Da), and AL LC showed the lowest uptake (D₀ = 3.73 Da). M83 titration resulted in a concentration-dependent reduction in deuterium uptake across all three protein constructs, confirming specific binding. The maximum reduction in deuterium incorporation (ΔD, measured between the centroids of the m/z distributions) was observed at 25 μM M83, the highest stabilizer concentration investigated and 100-fold in excess of protein. The ΔD values for AL LC, AL C218S and AL V_L were −1.37 Da, −2.85 Da and 2.61, respectively. The change in deuteration was greater for the non-covalent constructs than for AL LC upon M83 binding. However, this quantitative analysis showed that neither AL C218S nor AL V_L reached the same protection from exchange as M83-bound AL LC. M83 binding was sufficient to reduce uptake of deuterium by AL C218S, but not AL V_L, to below the level measured for unliganded AL LC. Fitting the observed ΔD to a unimolecular binding model (1:1 LC dimer:ligand) yielded protection midpoints (EC₅₀, the concentration of 50% maximal effect) of 0.65 μM for AL LC, 1.8 μM for AL C214S and 2.2 μM for AL V_L (Figure 5B; reported values are the mean of two independent measurements). These values are not directly comparable because the dimerization equilibria and exchange endpoints are different for the three AL constructs and hence, these values should not be treated as dissociation constants.

Figure 5: Stabilizer binding promotes LC dimerization.

(A) Raw mass spectra of peptide ¹⁰¹NWVFGGGTKL¹¹⁰ from AL LC, AL C218S, and AL V_L in the presence of varying concentrations of M83 at a single time point of 90 minutes. Proteins concentration was 0.25 μM monomer equivalent. The blue peaks represent the isotopic distribution of the peptide, with centroid shifts from the undeuterated state indicating the average extent of deuterium incorporation (ΔD) (B) PLIMSTEX titration data for peptide ¹⁰¹NWVFGGGTKL¹¹⁰ from AL LC (red), AL C218S (black), and AL V_L (teal), showing measured deuterium uptake values (average of duplicates, where error bars show the range of the measurements) with corresponding fitted curves. (C) Size exclusion chromatography (SEC) of AL LC, AL C218S, AL V_L and MM V_L in the absence (black solid lines) and presence (red dashed lines) of 100 μM M83, illustrating its effect on dimer formation of otherwise monomeric LCs. Peaks corresponding to molecular weight standards are shown above.

The distributions of discrete m/z states observed in each experiment for peptide ¹⁰¹NWVFGGGTKL¹¹⁰ report on the ensembles of species within the exchange reaction, averaged over the 90-minute timecourse. AL LC spectra had similar distributions of peaks at all concentrations of M83, consistent with homogeneous exchange. The spectra of AL C218S and AL V_L were more complex, indicating the presence of multiple populations of exchanged species. Unliganded AL C218S had a broad distribution of states that becomes narrower as the concentration of M83 increases. AL V_L exhibited different behavior, where increasing M83 led to the appearance of a population with less incorporation of deuterium. We hypothesize that this behavior reflects ligand-induced dimerization of both non-covalent species. Under the maximum concentration of M83 tested, only a sub-population of AL V_L is stabilized.

We next asked whether kinetic stabilizer binding could enhance LC dimerization. M83 increased the fraction of dimeric AL C218S and AL V_L, as determined by analytical size exclusion chromatography at an initial concentration of 20 μM (Figure 5C). At the concentrations used for HDX-MS (1.5 μM) both constructs are expected to be predominantly monomeric in the absence of M83. Nevertheless, the greater protection of the V_L domain in AL C218S compared to the isolated V_L construct suggests that AL C218S retains a higher dimer population under these conditions, likely due to the contribution of the C_L domain [28]. However, the dimerization affinities and associated rates have not been determined. A small population of a larger AL V_L species was also formed, which we were not able to identify. This observation is consistent with drug binding in the V_L-V_L interfacial cavity, and with the previous observation that P1 binding favors dimerization of WIL-FL C218S [38]. M83 binding did not alter the elution volume of AL LC, which forms a disulfide-stabilized dimer, or MM V_L, which is primarily dimeric in solution (Figure 5C).

Stabilizer molecules reduce hydrogen exchange and endoproteolysis of diverse LCs

We next asked whether binding of M83 alters the stability and dynamics of other FL LCs. We measured HDX kinetics for each FL LC (1.5 μM) in the presence and absence of 15 μM M83. Figure 6A shows the difference in deuteration between FL LCs with and without M83 as residue-level heatmaps, while Figure 6B shows the data mapped onto the dimeric V_L domains of structural models. All FL LCs exhibited reduced deuteration in the presence of M83, particularly between 1 and 4 hours of exchange (further data are provided in Figures S8 and S9, and the supplemental data file). The protein segments showing the most significant stabilizer-induced protection from HDX were consistent across eight of the nine proteins, with the strongest effects observed in FR4 and, in some cases, CDR3. An exception was H6, which exhibited increased protection primarily within FR2. Only AL LC and GL LC exhibited increased protection within the C_L domain.

Figure 6: Effect of M83 binding on multiple LC proteins monitored by HDX-MS.

(A) Single-residue aligned heatmaps of percent deuterium difference (Δ%D) across the nine LCs. The solid horizontal line indicates the boundary between the V_L and C_L. Protection from deuterium exchange is observed for all LCs upon incubation with M83. Complete data are provided in Figure S8 and S9. (B) Structural mapping of Δ%D onto V_L dimer structures for each isoform highlights regions of stabilization upon M83 binding after 1 hour of exchange. Crystal structures are shown where available, other models were created using AlphaFold3. The relative orientation of the V_L and C_L domains differs between the models, so for clarity only V_L is shown.

Analytical size exclusion chromatography showed that stabilizer binding did not alter the apparent oligomeric state of any FL LC (Figure S10). However, H3, H6, H7 and MCG had slightly longer elution times in the presence of M83, consistent with structural compaction or increased interaction with the column matrix. The elution time of H7 was slightly reduced, indicating expansion or reduced interactions.

To further assess the structural impact of M83, we conducted limited proteolysis experiments using trypsin at neutral pH on FL LCs in the presence or absence of M83 (Figure 7). We initially measured timecourses for each FL LC under identical conditions, with timepoints ranging from 1 minute to 24 hours (Figure 7A and Figure S11). Although we were able to fit data for some FL LCs to exponential decay models (Figure 7A–B and Figure S11), this was not successful for all FL LCs. Therefore we calculated the area under the curve (AUC) for these data to define a simple metric for protease sensitivity (Figure 7A and Figure S11). For comparison, we calculated an equivalent AUC for HDX data (Figure 7B and Figure S11). Since the intrinsic proteolysis rates and the magnitude of M83-induced protection varied across the FL LCs, we tailored single timepoint proteolysis experiments by selecting optimized conditions for each FL LC (Figure 7D). The protease sensitivity of the FL LCs and the effect of M83 varied substantially. There was only a weak correlation between the intrinsic protease resistance of each FL LC and its protection from HDX (Pearson’s correlation coefficient R = 0.14, Figure 7E). Furthermore, the extent of protection from protease and HDX were weakly correlated, both when comparing the AUC for the proteolysis timecourses (R = −0.3, Figure 7F) and the optimized, single timepoint data (R = 0 069, Figure 7G). These observations indicate that limited trypsin proteolysis and HDX measure different aspects of stability, likely due to differential cleavage of unstructured peptide bonds within the folded states. However, both methods demonstrated stabilization of all FL LCs upon binding to M83.

Figure 7: Effect of M83 binding on multiple LC proteins monitored by limited proteolysis.

(A) Kinetics of digestion of H16 LC by trypsin in the presence (red) or absence (black) of M83, analyzed by SDS-PAGE. Data were analyzed by fitting to a single-exponential decay and by calculating the area under the curve (AUC). Data for other LCs are provided in Figure S8. (C) Proteolysis rates from fits of timecourse data to single exponential models. Hollow symbols and bars represent rates and fitting errors for successful fits (see Figure S11), solid symbols show data for which no rate could be determined. (B) AUC analysis of H16 HDX in the presence and absence of M83 for comparison with (A). (D) Single point proteolysis using optimized conditions to determine the maximal protection from protease afforded by M83 binding. Quantitation of the data is shown by the bars. (E–G) Correlations between proteolysis and HDX measurements of the intrinsic LC stability or stabilization imparted by M83. The blue dashed lines indicate linear regression lines with associated Pearson correlation coefficients (R) and significance values (p).

Stabilizer molecules inhibit LC aggregation

Finally, we asked whether stabilizer binding would inhibit aggregation of LCs as amyloid fibrils. Full-length LCs do not readily aggregate under native-like conditions [16] but cleavage with trypsin releases aggregation-prone fragments [32]. Our stabilizers were envisaged to inhibit such proteolysis-coupled aggregation by binding to FL LCs [38]. Accordingly, M83 prevents proteolysis and subsequent aggregation of AL LC, as measured by thioflavin T spectroscopy (Figure 8A). Inhibition of V_L domain aggregation is more challenging, because V_L domains are much more prone to aggregate without proteolysis than their parent FL LCs [16] and because stabilizer binding is weaker (Figure 5). We measured the effect of 10 μM M83 binding on the aggregation of 10 μM AL V_L and MM V_L, which corresponds to a 2-fold molar excess of M83 over V_L dimer. In the absence of stabilizer, both variable domains aggregate after a similar delay (Figures 8 B and C). Stabilizer binding delayed MM V_L aggregation beyond the 14-hour experiment time course, but did not consistently alter aggregation of AL V_L.

Figure 8: Inhibition of LC aggregation by M83 binding.

Aggregation of LCs measured by thioflavin T fluorescence. LCs were incubated at 37 °C with shaking in PBS, pH 7.4, in the presence (red lines) or absence (black lines) of M83. (A) Proteolysis-coupled aggregation of 12 μM FL AL LC incubated with trypsin, with or without 60 μM M83. (B) Aggregation of 10 μM AL V_L with or without 10 μM M83. (C) Aggregation of 10 μM MM V_L with or without 10 μM M83.

Discussion

This work demonstrates the ability of small-molecule kinetic stabilizers to bind to multiple LCs and thereby suppress their dynamics and proteolytic susceptibility. LC dynamics, including local and global unfolding, enable self-association of non-native LCs to form amyloid, as well as proteolytic cleavage that yields more amyloidogenic V_L-containing LC fragments. Binding of the most potent small molecule, M83, leads to a reduction in exchange of amide hydrogens with the solvent and decreased susceptibility to proteolytic cleavage for all nine LCs studied (Figures 6 and 7). The degree of stabilization varies between LCs, consistent with their distinct sequences and dynamics (Figure 2). These data extend the previous observation that small molecule binding stabilizes WIL-FL, the IGLV6–57-derived LC that was used for the high throughput screen and hit optimization to generate M83. Accordingly, the greatest degree of stabilization was observed for AL FL, which is also derived from the IGLV6–57 gene. The precursor germline genes represented by the LCs studied here, IGLV1–44, IGLV1–51, IGLV2–8, IGLV2–14 and IGLV6–57, together account for 386/847 (46%) of all amyloidogenic LCs and 386/621 (62%) of amyloidogenic λ LCs in the AL-Base repository [6]. By demonstrating that M83 stabilizes several LCs derived from these λ LC genes, our results support the hypothesis that pharmacologic stabilization of LCs is a promising therapeutic strategy that could complement existing and emerging approaches.

Consistent with our previous studies [28,31,32,44] and those of other groups [66–69], the extent of amide exchange varies across the sequence of the LCs. The V_L domains of the seven amyloidogenic proteins have distinct patterns of deuterium uptake. There are no clear patterns that could represent a “signature” of amyloid propensity other than our previous observations that AL LC is less protected from exchange than the non-amyloidogenic IGLV6–57-derived MM LC or GL LC. None of the LCs investigated here exhibited increased exchange in the C_V and C_V’ strands (residues 38–49) that were identified as rapidly exchanging in the amyloidogenic IGLV6–57-derived LC AL55 [66]. Instead, we observed more rapid HDX in the region encompassing CDR2 and strand D_V (Figure 2). In contrast, similar patterns of deuterium uptake were observed in all nine C_L domains, consistent with their similar sequences and structures. C_L domain mutations have been associated with instability and amyloid formation [28,34,36], but our data suggest that in the absence of such mutations, the dynamics of the unmutated C_L sequences are indistinguishable within free LC dimers. Overall, these results and our previous study of κ LCs [31] are consistent with the hypothesis that multiple types of sequence and structural changes can lead to amyloidogenic LCs.

We asked whether stabilizer binding would lead to similar structural changes in LCs with different local dynamics. The stabilization imparted to all tested LCs is consistent with similar modes of binding between LCs, as observed in co-crystal structures of stabilizers with MM and H9 [40,42,43]. However, our analysis only measures the effect of stabilizer binding on HDX and cannot directly report on the conformation of the LC or small molecule. It is possible that M83 can bind in different conformations to that observed for other stabilizers and LCs, particularly in the case of H6, which has a different pattern of differentially-protected regions than the other LCs (Figure 6).

We hypothesize that stabilizer binding enhances structural coupling between the constituent domains of a full-length LC dimer [28], so that both global and local structural fluctuations are reduced. The V_L-V_L interface within FL LC dimers appears to be labile [28,38,66,69,70], leading to domain movements that are necessary for stabilizer binding [38]. An “open” structure where the V_L-V_L interface was not formed has been observed for the AL55 LC [66,69,70], and this conformation is likely accessible to other LCs in solution. Stabilizer binding enhances dimerization and increases protection in both the residues that define the binding pocket and residues forming the dimeric interface, such as strand A, FR2, strands E and F, CDR3 and J-region of the V_L domain as well as strand E of the C_L domain (Figures 4–6). On the other hand, relatively little change in protection was observed for the most labile protons in the loops of CDR1 and CDR2, indicating that these residues do not form new hydrogen bonds. However, the global protection from protease digestion afforded by stabilizer binding indicates that these mobile regions are not susceptible to trypsin proteolysis without local or global unfolding. Importantly, M83 binding increased local protection in CDR3 and FR4 regions in all proteins but H6 (Figure 6A). CDR3 is the most variable region in LCs, is invariably observed in the structured core of AL fibrils, and is involved in the V_L-V_L interface, so residue changes here would potentially alter multiple aspects of amyloid formation and stabilizer binding. Furthermore, CDR3 is adjacent to C92 that forms an internal disulfide (C21–C92) in V_L. The structure around this conserved disulfide must fully unfold to enable the reorientation of the peptide chain that is observed in AL fibrils, from parallel in native V_L to antiparallel in amyloid [31]. This backbone rotation likely provides a major contribution to the kinetic barrier for LC misfolding in amyloid; if so, suppression of local backbone dynamics in the regions adjacent to the C21–C92 disulfide, particularly the highly dynamic CDR1 and CDR3, is expected to decelerate amyloid formation [32].

The largest reduction in deuterium uptake induced by M83 was observed for AL LC (Figure 6A). Other amyloidogenic LCs showed a reduced level of stabilization, which may be due to the optimization of stabilizers for binding to IGLV6–57-derived LCs. The reduced protection afforded to GL LC and MM LC by stabilizer binding appears to be due to their increased stability in the absence of the small molecule. Binding of M83 is not sufficient to globally reduce the level of exchange of AL LC to that of the non-amyloidogenic LCs (Figure 6 and Figures S8–S9). However, the stabilization of AL LC’s structured regions by M83 reduced their exchange to a level similar to that observed for equivalent regions of unbound GL LC. The M83-bound states of AL LC and AL C218S exhibited similar exchange kinetics (Figures 4A and 5), where the M83-bound state of AL C218S was slightly less stable than M83-bound AL LC but more stable than unbound AL LC. Similarly, stabilizer binding increased the proportion of dimer in solution as measured by SEC (Figure 5C). We therefore suggest that stabilizer binding can partially compensate for the loss of stability associated with reduction of the intermolecular disulfide bond, in part due to enhancement of homodimer formation. Therefore, a sufficiently efficacious stabilizer may be able to suppress unfolding, proteolysis and aggregation of normally monomeric LCs in patients.

Under the conditions tested, M83 binding was able to suppress tryptic cleavage of AL FL that generated amyloidogenic fragments (Figure 8A) and delay aggregation of MM V_L (Figure 8C), but not that of AL V_L (Figure 8B). We attribute this to the different affinities of M83 for the three proteins, which correlates with their propensity to form dimers. Dimerization has been reported to be protective against unfolding and aggregation of several LCs [28,29,69]. Stabilizer binding promotes dimerization and thereby increases the free energy barrier to unfolding and subsequent amyloid formation, but the extent of this effect depends on the intrinsic properties of the target LC. AL V_L is less stable than MM V_L [38,40] and has a weaker propensity to form dimers and bind M83 (Figure 5), therefore M83 binding imparts less stabilization to AL V_L than to MM V_L. However, the two V_L domains aggregate at similar rates in the absence of stabilizer, which highlights the roles of other factors in amyloidogenesis, such as differing amyloid promoting regions (Figure 2).

The major limitation of our study is that these nine LCs represent only a fraction of the vast potential diversity of amyloidogenic LCs. Although we expect that the stabilizer can bind to most λ LCs, it is not clear how much stabilization will be imparted in each case and whether this will be sufficient for clinical efficacy. M83 is a tool compound that allows us to investigate LC stability and aggregation in vitro, but not suitable for use as a medicine. Coumarin-based stabilizers including M83 bind relatively weakly to κ LCs [6], which account for approximately 20% of AL amyloidosis clones [43]. They also bind tightly to other serum proteins including albumin, which makes them unsuitable as clinical candidates [40]. Furthermore, it is not clear how the dynamics that are measured by HDX relate to amyloid propensity, since there is no clear correlation between the patterns of exchange protection observed for each LC and its pathogenicity. Our data will support further work to identify such mechanistic details.

Stabilization of LCs is a novel and potentially valuable therapeutic approach that could complement existing therapies. This work demonstrates that kinetic stabilizers can bind to multiple LCs and reduce the conformational dynamics associated with proteolysis and aggregation, which is essential for their broader utility. In combination with our previous work developing stabilizers with increased affinity [43] and specificity for LCs in blood [32], the current study supports further development of these molecules as therapeutic candidates.

Highlights.

• Small-molecule kinetic stabilizers suppress conformational dynamics and aberrant proteolysis of diverse amyloidogenic full-length λ light chains.

• HDX-MS indicates conserved protection in regions forming the interface between the two variable domains of the light chain dimer, corresponding to residues that form the core of patient-derived amyloid fibrils.

• PLIMSTEX binding titrations show coupled binding, dimerization and stabilization at sub-micromolar kinetic stabilizer concentrations.

• Small molecule kinetic stabilization across different λ sequences supports kinetic stabilizers as a promising therapeutic strategy for AL amyloidosis.

Abbreviations:

AL

amyloid light chain

CDR

complementarity determining region

C_L

light chain constant domain

FR

framework region

HDX

hydrogen-deuterium exchange

GL

germline

LC

immunoglobulin light chain

MM

multiple myeloma

MS

mass spectrometry

SDS PAGE

sodium dodecyl sulfate polyacrylamide gel electrophoresis

ThT

thioflavin

V_L

light chain variable domain