A conservative point mutation in a dynamic antigen-binding loop of human immunoglobulin λ6 light chain promotes pathologic amyloid formation

Abstract

Immunoglobulin light chain (LC) amyloidosis (AL) is a life-threatening human disease wherein free monoclonal LCs deposit in vital organs. To determine what makes some LCs amyloidogenic, we explored patient-based amyloidogenic and non-amyloidogenic recombinant LCs from the λ6 subtype prevalent in AL. Hydrogen-deuterium exchange mass spectrometry, structural stability, proteolysis, and amyloid growth studies revealed that the antigen-binding CDR1 loop is the least protected part in the variable domain of λ6 LC, particularly in the AL variant. N32T substitution in CRD1 is identified as a driver of amyloid formation. Substitution N32T increased the amyloidogenic propensity of CDR1 loop, decreased its protection in the native structure, and accelerated amyloid growth in the context of other AL substitutions. The destabilizing effects of N32T propagated across the molecule increasing its dynamics in regions ~30Å away from the substitution site. Such striking long-range effects of a conservative point substitution in a dynamic surface loop may be relevant to Ig function. Comparison of patient-derived and engineered proteins showed that N32T interactions with other substitution sites must contribute to amyloidosis. The results suggest that CDR1 is critical in amyloid formation by other λ6 LCs.

Graphical Abstract

AL amyloidosis is the most common form of human systemic amyloid disease. This rapidly progressing life-threatening illness results from excessive circulating free monoclonal immunoglobulin (Ig) light chains (LCs) which are protein precursors of amyloid [1,2]. Extracellular deposition of monoclonal AL LCs as amyloid in kidney, heart and other organs causes organ damage and, if untreated, death [1–3]. In multiple myeloma (MM), another monoclonal Ig-associated disease, some (10–15%) patients develop AL while others do not [4]. The underlying basis is unclear and probably reflects conformational and aggregation properties of individual LCs. Current treatments for AL and MM target the source of the monoclonal LCs, the underlying plasma cell clones, and include chemotherapy, autologous stem cell transplantation, proteasome inhibitors, and immunomodulatory drugs [1,2]. These treatments have serious side effects and are not suited for all patients, necessitating the search for additional therapeutic approaches, including those targeting protein misfolding and amyloid deposition [1,5]. To this end it is essential to identify critical early steps and factors that influence amyloid deposition, by using the atomic structures of LCs as a guide.

Ig LCs have a two-domain architecture comprised of the variable-joined (V_L) and constant (C_L) regions, each containing ~100 amino acids in a β-sandwich fold. Circulating LCs readily form parallel head-to-head homodimers linked via the Cys215 disulfide near the C-terminus [6]. Unlike the non-polymorphic C_L region, the antigen-binding V_L region has a highly variable sequence resulting from somatic hypermutations accumulated during antibody maturation. This sequence variability may lead to variations in the local protein conformation, dynamics and physicochemical properties, which can critically influence amyloid formation. LCs accounting for most AL cases have V_L encoded by λ germline (GL) genes; of those, the λ6 family is prevalent [7]. Since each AL patient presents a unique LC sequence, the misfolding pathway and the amyloid structure are expected to differ for LCs from different patients. The immense variety of LC sequences makes it extremely difficult to identify specific protein properties that enhance amyloid formation and to determine how to prevent or correct LC misfolding [3, 8]. Another challenge is identifying which of the ~10 amino acids, which typically distinguish an AL or MM V_L from the corresponding GL sequence, are disease-causing [9, 10].

High-resolution structural studies of LCs in their native and fibrillar states indicate that the native structure must completely unfold to form amyloid [11–16]. This is consistent with the generally lower thermodynamic and kinetic stabilities of AL LCs compared to non-amyloidogenic counterparts [3, 15–23]. However, global destabilization is neither necessary nor sufficient to make the protein amyloidogenic, suggesting that local conformation in sensitive regions is paramount [9,10,20,21,24].

Several LC regions have been implicated in amyloid formation, through either stabilization of amyloid misfolding intermediates [11] or destabilization of the native structure. The misfolded states can be stabilized by buried salt bridges and other intramolecular electrostatic interactions that stabilize AL amyloid [12,16,22] and other pathologic amyloids [25]. Destabilization of the native structure in AL was proposed to involve several LC regions that modulate protein stability and proteolytic susceptibility. These include the domain interfaces V_L-V_L and C_L-C_L in the LC dimer. Since dimer dissociation into monomers precedes protein misfolding, a weaker monomer-monomer interface can promote amyloid formation [8,22,27,28]. Conversely, stabilization of this interface was proposed to interfere with disease [5]. Disruption of the V_L-C_L interface in LC monomer by mutations can also contribute to the AL development, as reported for a λ1 LC [9, 29]. Additionally, increased proteolytic susceptibility in specific regions may contribute to AL by generating amyloid-forming fragments [30].

Proteolytic cleavage downstream of the joined (J) region releases the V_L domain, which has a less stable structure than C_L domain and forms amyloid more readily than the corresponding full-length LC [17,31]. V_L and its fragments form major constituents of patient-derived AL deposits; however, the sites of proteolytic cleavage are heterogeneous, and the mechanism of cleavage, its timing, and actual role in disease are unclear [9, 30]. Moreover, full-length LCs, C_L domains, and their fragments have also been found in AL deposits, suggesting that C_L contributes to AL amyloidosis [20, 32–34].

Other LC regions that can trigger protein misfolding involve short adhesive segments with high sequence propensity to initiate aggregation in a cross-β-sheet conformation. Such amyloid hotspots have been predicted in various LCs; some predictions have been verified experimentally. Two hotspots in β-strands E_V and G_V have been proposed to independently initiate amyloid formation in λ-type LCs [11, 35]. The hotspot in βE_V was also inferred to drive the misfolding in a λ3 LC [16] and is prominent in this protein isotype. Other hotspots have been reported in βB_V [22], BC_V linker, βC_V [36,37], βB_C, and βC_C [33,38].

Whether and how individual hotspots contribute to amyloid formation is unclear. Most of these adhesive segments are located in the native β-strands and are thereby protected from aggregation. Individual hotspots may contribute differently to the misfolding of different LCs. Moreover, the amyloid fibril structures, which have been determined to atomic resolution for several LCs by cryo-EM [12,13,14] or solid state NMR ([11,16] and references therein) differ from each other, suggesting that the misfolding pathways also differ.

The current study probes the structural alterations in the critical early steps of LC misfolding. A λ6 LC has been cloned from a patient with AL amyloidosis, AL01–095, featuring multi-organ involvement [20]. To identify AL-associated changes in the global stability and local dynamics of the native structure, we used recombinant proteins representing this full-length AL LC and its isolated V_L and C_L domains, along with non-amyloidogenic MM and GL counterparts and engineered variants. (We refer to V_L and C_L as “regions” in the context of the full-length LC, and as “isolated domains” in the form of stand-alone constructs). Our approach integrated biochemical, spectroscopic and electron microscopic techniques with hydrogen deuterium exchange (HDX) mass spectrometry (MS). HDX monitored by MS or NMR is well suited to explore local conformation and backbone dynamics of a polypeptide chain in solution and in aggregates [38]. HDX MS has been used by us and others to study conformational dynamics of full-size antibodies and their isolated domains [39,40], to identify fibril core in LCs [15], and to unravel amyloidogenic pathways in other proteins [41]. HDX MS analysis of natively folded AL LCs, which was pioneered by Buchner’s team [9,10], is raised to a new level in the current study. The results help better understand what makes an LC amyloidogenic.

RESULTS

Protein Selection and Characterization

The total of 19 recombinant λ6 proteins were generated as described in Methods. This includes AL, MM, GL (Figure 1A shows primary structures) and two engineered proteins, one containing N32T substitution in GL (termed GL N32T) and another containing all other AL substitutions except for N32T (termed AL T32N). Each of these five proteins was generated in several forms: full-length LC; C215S LC; isolated V_L domain; the common isolated C_L domain; and C215S C_L. Two additional proteins, AL Y75S and AL T32N/Y75S, were generated as full-length LCs. The X-ray crystal structure of MM LC, determined to 1.75Å resolution [5], was used to interpret the results.

Fig. 1.

LC protein structures. (A) Top to bottom: Primary structures of k6 LCs used in this study show VL domains of AL, MM and GL LCs and their common CL domain. Amino acid substitutions in VL of AL and MM versus GL are in blue italics. Dots indicate cysteines C22 - C91 and C138 - C197 forming intradomain disulfides, and C215, which can form an intermolecular disulfide. Amyloid hotspots are colored according to the AmylPred2 consensus number, 2–8 (weakly to strongly amyloidogenic). Secondary structures in linear representation are based upon the crystal structure of the native k6 MM LC (“Native k6”, PDB: 6MG4), and the amyloid fibril structures for fragments of k6 LC VL (“Fibril k6”, PDB: 6HUD) and k1 LC VL (“Fibril k1” PDB: 6Z1I). In VL, framework regions (FR1 to FR3), complementarity-determining regions (CDR1 to CDR3), and the joined (J) region are indicated according to the international ImMunoGeneTics system (IMGT). (B-D) Ribbon diagrams showing the structure of k6 MM LC (PDB: 6MG4). (B) Red balls in VL indicate a-carbons of AL substitutions from the current study. The dotted line indicates the VL - CL boundary. (C) LC monomer, with b-strands indicated and the major amyloid hotspots color-coded. (D) LC dimer. Monomer 1 (gray) with CDRs in blue and J-region in teal; monomer 2 (orange).

In vivo, free LCs circulate as monomers or covalent homodimers linked via Cys215 [6]. Intact mass spectrometry showed that all full-length LCs were covalent dimers and indicated the number of disulfide bonds in each protein (Table S1). As fully monomeric controls, we prepared full-length LCs containing a C215S substitution. All LCs migrated as monomers in reducing SDS-PAGE, as a monomer/dimer mixture in non-reducing SDS-PAGE, and primarily as dimers on size exclusion chromatography (SEC) (Figure S1). This is consistent with the high dimerization propensity of free LC monomers, driven by both non-covalent association and disulfide bonding.

AL V_L Has Increased Amyloidogenic Sequence Propensity

Amino acid sequence differences between V_L of AL, MM and GL influence their predicted amyloidogenic propensity. Sequence analysis using a consensus prediction method, AmylPred2 [42], revealed that several point substitutions in V_L of AL and MM proteins significantly altered the amyloid hotspots (Figure 1A,C). AL V_L, which contained nine point substitutions versus GL, showed two major differences. First, an N32T replacement enhanced the amyloidogenic potential of the βC_v hotspot and extended it into the BC_V linker; second, an S75Y substitution increased the amyloidogenic potential of the βE_V hotspot. In MM V_L, with 11 substitutions versus GL, the βC_V hotspot diminished upon A30D replacement, while the βC’_V hotspot increased with T46I. The remaining substitutions in AL and MM did not significantly alter their amyloidogenic propensity. In summary, the predicted amyloidogenic sequence propensity of AL V_L increases upon N32T and S75Y substitutions, while in MM the net effect is less clear. Sequence analysis using LICTOR, a new machine learning approach designed to predict LC toxicity based on the somatic mutations [43], identified AL and MM LCs as “toxic” and GL as “non-toxic”.

AL Proteins Have Less Stable Structure and Form Amyloid Faster

Secondary structure and stability were assessed by circular dichroism (CD). All proteins showed far-UV CD spectra typical of an Ig fold (Figures 2A, S2A, S3A). Protein structural stability was assessed using thermal denaturation monitored by CD at 206 nm for β-sheet unfolding. Compared to GL, AL LCs showed a small decrease in thermal stability indicated by a decrease in the apparent T_m by ~2°C, while MM LCs showed no significant changes (Figure 2B). The decreased stability was amplified in C215S LCs and, particularly, in the isolated V_L that showed a low-temperature shift by ~6°C in AL versus GL or MM (Figures S2, S3). This agrees with our previous studies of these LC and V_L, which showed lower thermodynamic and kinetic stabilities for AL compared to GL [20]. Protein stability was further probed using limited proteolysis by either trypsin or proteinase K. The time course of proteolysis was monitored by SDS PAGE for 4 hours. All protein forms including LC, LC C215S, and isolated V_L showed similar trends: AL variants were digested by both proteases slightly faster than MM and much faster than GL (Figures 2C, S2C, S4). In summary, thermal denaturation and limited proteolysis consistently showed a rank order of stability, GL≥MM>AL, for all protein forms.

Fig. 2.

LC secondary structure, stability and proteolytic susceptibility. Color-coding throughout the paper: AL (red), MM (blue), GL (green). Details are in SI Methods. (A) Far-UV CD spectra. (B) Thermal unfolding monitored by CD at 206 nm during protein heating shows data (dotted lines) and data fitting by a sigmoidal function (solid lines). (C) SDS PAGE shows the time course of proteolysis by trypsin. Incubation time is indicated; 0 represents no trypsin. Positions corresponding to the covalent LC dimer (LC2), the monomer (LC1), and the isolated CL are indicated. (D) Transmission electron micrographs of LC samples after 140 hours incubation with trypsin. Similar data for LC C215S and for VL are shown in Figures S2 and S3. (E) Time course of amyloid formation. The proteins were incubated at 37 C with shaking, with or without SDS micelles. ThT binding to amyloid-like structure was monitored by fluorescence as described in SI Methods.

Prolonged protein incubation with trypsin at 37°C for up to 140 hours resulted in amyloid formation. After incubation, all samples became positive for thioflavin T (ThT) – a fluorescent dye whose emission increases upon binding to amyloid-like structures – and the products were visualized by transmission electron microscopy (Figure 2D). All proteins formed amorphous aggregates along with fibrils ~10 nm in width, which is typical of amyloid. For C215S LCs, all variant proteins formed fibrils (Figure S2D). Therefore, all LCs formed amyloid upon prolonged incubation with trypsin, and the amyloid morphology was protein-dependent. To determine the protein composition of the amyloid, AL, MM and GL C215S LCs were incubated with trypsin for 10 days under fibril-promoting conditions and the aggregated material was subjected to non-reducing SDS PAGE. All proteins showed a major band ~10 kDa (Figure S2E). MS analysis of this band (described in SI Methods) showed peptide fragments spanning the entire LC length. Therefore, both V_L and C_L formed amyloid in vitro. This is consistent with the LC protein composition of ex vivo amyloid deposits determined for this AL case [20].

Amyloid formation in the presence of SDS micelles was monitored in real time by ThT fluorescence. Hyperbolic amyloid growth kinetics were observed (Figures 2E, S2F, S3C), which is consistent with other LC studies [9] reflecting the ability of periodic anionic arrays to catalyze amyloid nucleation [45]. Notably, AL proteins showed a much larger increase in ThT emission compared to MM and GL, suggesting faster amyloid growth. Collectively, these results show that AL proteins have a slightly lower thermal stability, are significantly less resilient to broad-specificity proteases, and form amyloid faster compared to GL or MM counterparts.

General Trends Revealed by HDX MS Analysis of AL, MM and GL λ6 LCs

To explore local backbone conformation and dynamics, we performed HDX MS of native AL, MM and GL proteins in different forms (LC, LC C215S, and isolated V_L). Protein deuteration was performed for multiple time points, from 10 seconds to 4 hours, followed by peptic digestion, LC separation, and MS analysis (see SI Methods). Figure 3 shows a summary of deuterium uptake as a function of time for representative protein regions, comparing LC, LC C215S, and isolated V_L (violet, magenta, and gray, respectively) with each other in AL, MM, or GL; SI Excel contains the full dataset with all overlapping peptides. Figure 4 compares AL, MM and GL full-length LCs, LC C215S forms, or isolated V_L and C_L forms. Figure 3 compares LC, LC C215S and V_L over the entire labeling time course, while Figure 4 compares HDX for the full-length AL, MM and GL using a single time point. Minor EX1 signatures at longer times suggest variable domain packing in some proteins (SI Results).

Fig. 3.

Deuterium uptake plots for representative parts of VL. Top to bottom: FRs and CDRs, the native secondary structure derived from k6 MM (as in Figure 1A), the major amyloid hotspots (purple bars), and the amino acid sequence of GL. Rainbow-colored bars show representative non-overlapping peptide fragments explored by HDX MS. Each row corresponds to a different LC variant, AL, MM, or GL. Each column represents a common peptide (shown by residue numbers); asterisks indicate peptides that slightly differ among the three LCs. Each uptake plot shows the measured deuterium level versus labeling time for LC (violet), LC C215S (magenta), and isolated VL (gray). The maximum y-axis value is the maximal theoretically allowed deuteration assuming 100% labeling and loss of the N-terminal amide. To the right on each X-axis is the measured incorporation for the maximally deuterated (maxD) control sample; the dotted horizontal line indicates the measured deuteration level of maxD. Replicate data and their points spread calculated by DynamX (which considers multiple charge states for each peptide) are shown for each time point. Raw values and graphs for all peptide fragments are in the SI Excel.

Fig. 4.

Deuterium uptake by GL, AL and MM proteins compared. HDX MS skyline plots for AL (red), MM (blue), and GL (green) after 1 min labeling for (A) LC, (B) LC C215S, and (C) isolated VL and CL. The plots were created using common peptides for the three LCs, except for fragments 66–76 (66–75 in AL, red asterisk) and 23–32 (23–33 in MM, blue asterisk). Percent deuteration (Y-axis) was calculated after back-exchange correction. CDRs are shaded; the dotted line defines the VL-CL domain boundary. Underneath are the native secondary structure derived from k6 MM, the major amyloid hotspots (purple bars), and the proteolysis-prone sites (yellow bars) identified for patient-derived deposits of a k6LC [30]. HDX MS skyline plots for all time points are in Figure S5, and all uptake values for all peptides are in SI Excel. (D, E) Deuteration after 1 min labeling of VL in (D) GL and AL LCs, and (E) the difference in deuteration between GL and AL, mapped onto the VL of the native structure for MM LC (PDB: 6MG4). Peptides are colored in panel D from low to high deuteration (blue to brown); selected peptides (numbered 1–6) are color-coded from negative to positive difference between AL and GL LCs (blue to red, E). Figure S5D shows deuteration levels and differences between GL and AL for the entire LC.

Several major conclusions follow from these HDX data. First, the results are generally consistent with the native structure of LC determined by X-ray crystallography or NMR: peptide fragments overlapping exposed loops in V_L and C_L showed higher deuteration (lower protection from exchange) than the β-strands (Figure 4D). The only exception was the relatively low protection in βA_C and, particularly, βG_C, the first and last β-strands of C_L, which are not involved in interdomain interactions. Importantly, in the V_L of all protein forms explored, CDR1 showed the highest deuteration after 1 minute of labeling (Figures 3, 4A–4D). The exact rank order of protection in CDRs and other regions of the V_L was different for GL, MM and AL proteins; however, CDR1 remained the least protected region in the V_L of all proteins. Therefore, CDR1 loop is the most dynamic/exposed region in V_L. Since LCs of λ6 subtype originate from a common GL sequence, the low protection of CDR1 observed in GL must be characteristic of all λ6 LCs.

Second, in various protein forms, most regions of V_L, including all CDRs, showed more deuterium uptake in AL compared to GL or MM (Figures 4, S5). One exception includes residues 23–32 where the deuteration of V_L, LC, and LC C215S was similar in AL, MM or GL proteins. Another exception was residues 78–88 whose deuteration was lower (in LC and LC C215S) or similar (in isolated V_L) in AL compared to MM or GL proteins. Variations in interdomain interactions probably contribute to this effect, as segment 78–88 overlaps the EF_V linker involved in V_L-C_L interactions in full-length LC. The difference in the deuteration of AL versus GL LC after 1 minute of exchange was mapped on the 3D structure of V_L to show where and by how much AL differs from GL LC (Figure 4E). AL consistently showed decreased protection at the “top” of V_L encompassing CDRs and adjacent β-strands. Conversely, residues 78–88 at the “bottom” of V_L showed increased protection in AL compared to GL (Figures 4, S5). Unlike V_L, C_L showed similar deuterium uptake in all proteins (Figure 4A–C, S5D). These findings are consistent with the variable nature of V_L, the constant features of C_L, and the greater structural stability of isolated C_L versus V_L [20].

Third, V_L showed significantly higher deuteration as a separate domain versus part of the full-length protein (gray, violet and magenta lines in Figure 4; Figure 5A–C; Figure S5). The overall deuteration in each variant protein (AL, MM or GL) decreased in order V_L>LC C215S>LC; reduced deuterium levels are consistent with V_L interactions with C_L increasing the protection of V_L in full-length LC monomer and, especially, in the covalent dimer.

Fig. 5.

Effects of the amyloidogenic substitution N32T on the structural stability, proteolytic susceptibility and amyloid formation by LC. The data for AL (red), AL T32N (orange), GL (green) and GL N32T (dark green) LCs are shown. Figures S6, S7 show similar data for C215S LCs and the isolated VL domains. (A) Far-UV CD spectra, (B) the CD melting data at 206 nm, (C) SDS PAGE showing the time course of limited proteolysis by trypsin, and (D) ThT emission showing the time course of amyloid formation at 37 C with or without SDS micelles. Positions corresponding to LC dimer (LC2), monomer (LC1), and the individual VL and CL domains are indicated. For details see Figures 2, 3 and the Methods.

Another finding stems from the comparison of our HDX results with the MS analysis of an AL λ6 amyloid tissue sample [30]. This analysis identified three proteolysis-prone LC sites; the V_L site overlapped CDR2 and showed low protection in AL, but two other sites in residues 115–139 and 170–195 mapped on the well-protected parts of C_L (Figure 4). This raises the possibility that proteolytic cleavage of C_L in the natively folded λ6 LCs does not occur in vivo, supporting the concept that the cleavage follows amyloid formation rather than initiates it [30].

ln summary, most V_L regions in the various forms of AL proteins showed less protection than MM and GL proteins (Figures 3, 4, S5), while the protection of C_L did not change. This is consistent with the rank order of the global protein stability indicated by thermal unfolding and limited proteolysis, GL≥MM>AL, and with the more rapid formation of amyloid observed in AL compared to MM and GL proteins (Figures 2, S2–S4). Importantly, CDR1 was the least protected region in the V_L domains of all proteins. The protection of CDR1 was lower in AL compared to GL and MM proteins, and in isolated V_L domains compared to V_L regions in full-length LCs. These results are in general agreement with HDX NMR study of the same MM protein (called JTO), which also showed lowest stability near residue 30 in CDR1 [22].

Protection in Hotspots, CDRs and the J-region Suggests Possible Drivers of Amyloid

Decreased protection of amyloid hotspots in a globular protein is expected to augment amyloid formation ([41,42] and references therein). To investigate the roles of the predicted hotspots in LC amyloid formation, we compared deuterium uptake for selected peptides overlapping these hotspots. All proteins showed the lowest protection from exchange in the CDR1 residue segment 23–32, which overlaps the amyloid hotspot from the BC_V linker and βC_V (Figures 3, 4, S5). This segment showed especially low protection (including early time points) for AL versus MM or GL in all protein forms (Figure 3). This low protection combined with the high amyloid-forming sequence propensity of CDR1, which increases and extends into the BC_V surface loop upon AL substitution N32T (Figure 1), suggests that CDR1 is key to amyloid formation in this AL LC.

Another amyloid hotspot, which showed increased amyloid-forming sequence propensity due to the AL substitution S75Y (Figure 1), overlaps βE_V (residues 73–78). Unlike βC_V, βE_V strand showed substantial protection from exchange with minimal/slow dynamics in the LCs in all studied proteins (Figures 3, S5), consistent with the central location of this strand in the Ig fold, where βE_V is hydrogen-bonded to its flanking strands, βB_V and βD_V (Figure 1C). Therefore, the βE_V hotspot is unlikely to initiate amyloid formation by this LC.

Similarly, other amyloid hotspots showed general protection from exchange (Figure 4). These include: i) segment 17–22 overlapping βB_V and stabilized by the internal disulfide C22-C91; ii) segment 46–50 overlapping βC_V, which shows increased amyloidogenic sequence propensity in MM due to the T46I substitution; iii) a weakly amyloidogenic segment 99–110; iv) the sole C_L hotspot in residues 134–154, which overlaps βB_C and βC_C (Figure 1C). General protection of these adhesive segments (which if unprotected are thought to trigger protein aggregation and amyloid formation [41, 42]) suggests that they are unlikely to initiate amyloid formation by this LC.

Next, we compared deuterium uptake in other dynamic protein regions, including CDR2 and CDR3. For some peptides explored by HDX MS, the deuterium incorporation was similar for AL, MM and GL proteins; however, segment 49–63 which overlaps CDR2 showed much more rapid deuteration in AL compared to GL or MM proteins (Figures 3, 4, S5). This suggests that the CDR2 loop conformation in AL differs from that in MM and GL. This difference likely stems from a cluster of AL mutations in or near CDR2 (T47N, Y50F, Q54E, R55E; Figure 1B). CDR3 (residues 93–102 overlapping 4 βF_V and βG_V), typified by peptide fragment 90–98, also showed more deuteration in AL versus MM or GL (Figure 4). This trend was observed in all protein forms at early exchange times, as expected for solvent-exposed dynamic regions. The substantial amyloidogenic sequence propensity of βG_V, especially in AL and MM proteins (Figure 1A), combined with the decreased protection of CDR3, potentially contribute to amyloid formation.

Finally, we explored deuterium uptake by the J-region (residues 101–111), which defines the domain orientation in full-length LC and its proteolysis to release V_L and C_L fragments. The J-region conformation reportedly modulates amyloid formation by a λ1 AL LC [29]. Our HDX data at longer exchange times show that the peptide overlapping the J-region (residues 101–107) is less protected in AL and MM compared to GL (Figures 3, S5). This is consistent with faster proteolytic degradation of AL and MM compared to GL (Figures 2C, S2C, S4) and suggests that conformational or dynamic differences in the J-region may contribute to the differences in amyloid formation between λ6 AL and GL proteins.

In summary, compared to GL, AL proteins consistently showed lower protection from exchange in flexible regions of V_L, most notably in the CDRs (Figures 4, S5). This is consistent with the overall decrease in stability of AL proteins indicated by melting data and limited proteolysis (Figures 2, S2–S4). Our results suggest that some of these flexible regions may contribute to amyloid formation. In particular, CDR1 shows the highest sequence propensity to form amyloid (Figure 1) along with the lowest protection from exchange in V_L for all proteins explored, especially in AL (Figures 3, 4). Therefore, we posited that CDR1 can trigger amyloid formation by λ6 AL01–095 LC, and tested it as follows.

Probing the Role of N32T and Other AL Substitutions

Out of nine amino acid differences between AL and GL, N32T is the sole AL substitution in CDR1, which accounts for its increased amyloidogenic sequence propensity (Figure 1). In the fibril core structure of a homologous λ6 LC fragment [14], N32 is located next to I29 and S26 in a linker between two β-strands, one of which overlaps the amyloid hotspot (Figure 1); N32T is not expected to disrupt this structure. To explore the roles of this and other AL substitutions in amyloid formation, we engineered two additional proteins. In the first protein termed GL N32T, a single amino acid (N32) in GL is replaced to T. In the second protein termed AL T32N, the native residue (T32) in AL is replaced to N. The proteins were prepared in three different forms: LC, LC C215S, and isolated V_L.

Far-UV CD spectra showed little change upon T32N substitution in AL LC and no significant change upon N32T substitution in GL LC (Figure 5A). However, T32N substitution in AL LC significantly increased the protein thermal stability and delayed the tryptic digestion (Figure 5B–C, orange and red). N32T substitution in GL LC showed only marginal effects on the thermal stability and proteolytic degradation (Figure 5B–C, green). LC C215S and isolated V_L showed similar but clearer trends: the protein stability to thermal denaturation and proteolytic degradation was lowest in AL, highest in GL, and intermediate in AL T32N and GL N32T (Figures S6, S7). These effects were unexpected, as N32T is a conservative substitution in a dynamic surface loop not involved in interdomain interactions (Figure 1B, D).

Moreover, substitution at residue 32 influenced amyloid formation in AL proteins. Of all variant proteins explored, AL LC showed the largest amplitude of the hyperbolic increase in ThT emission during incubation at 37°C with SDS; this increase was greatly diminished in AL T32N LC (Figure 5D, red and orange). LC C215S and isolated V_L showed similar trends (Figures S6, S7). In contrast, the difference in the extent of amyloid growth between GL N32T and GL proteins was only marginal for LC, LC C215S and isolated V_L (Figures 5D, S6, S7, green lines). In summary, in all protein forms, the N32T replacement significantly enhanced amyloid formation in AL versus AL T32N; however, GL N32T versus GL showed little difference. Therefore, T in position 32 is amyloidogenic in the context of other AL substitutions but note alone; conversely, other AL substitutions without N32T are not amyloidogenic. This suggests coupling between this and other sites of AL substitutions, which again was unexpected given the surface location of N32 and the absence of extensive interactions formed by this residue in the native structure.

HDX MS provided sharper insights into the structural and dynamic effects of this substitution. The N32T substitution in GL increased the deuteration not only locally, but also across the V_L. This increased deuteration was observed in all protein forms and was largest in isolated V_L. Specifically, compared to GL, GL N32T proteins showed more deuterium at short exchange times in the dynamic CDR1 region near the mutation site (Figure 6A, bottom). At longer exchange times, the destabilizing effect of the N32T mutation propagated across V_L to well-ordered regions, particularly to segments containing βC_V-C_V’, βE_V, and the J-region located on the opposite side of V_L (Figure 6A, bottom). The only protein region that showed no significant substitution-induced changes encompassed CDR2, despite enhanced dynamics in CDR2 and its proximity to the mutation site in CDR1 (Figure 1B).

Fig. 6.

Effects of the amyloidogenic substitution N32T on the local conformation in VL domain. (A, B) Top: the percent deuterium uptake by selected peptides from (A) GL, GL N32T, (B) AL, and AL T32N is shown at various labeling times and color-coded as indicated (low to high %D, black to brown). Bottom: differential uptake (A) between GL N32T and GL or (B) between AL T32N and AL, color-coded negative to positive (blue to red). X-axis - amino acid numbers; Y-axis - labeling time. SI Excel lists all uptake values for all peptides. (C) Structural model of GL created from the MM crystal structure (PDB: 6MG4) using Swiss-Model. N32 side chain (in magenta) can form H-bonds with R24 Ng and Y94 CO (dotted lines). These and other nearby residues, which form an H-bonded network in the MM structure and in the GL model, are shown in sticks (O – red, N – blue).

Conversely, the T32N substitution in AL partially restored the protection from exchange across the V_L (Figure 6B). Again, CDR1 showed large mutation-induced changes at short exchange times, while at longer times these changes propagated to other parts of V_L, except for CDR2 that showed no change (Figure 6B, bottom). The latter suggests that decreased protection of CDR2 observed in AL proteins (Figure 3) stems from replacements other than N32T; perhaps these include AL replacements located in or near CDR2 (Figure 1). Similar to other LCs studied here, C_L deuteration showed little change in all full-length proteins (Figure S8). Comparison of the deuteration levels of GL, GL N32T, AL, and AL T32N showed that N32T alone does not fully account for the observed differences between AL and GL proteins. Consequently, other AL mutations must contribute to these differences.

The HDX MS results in Figure 6 complement our proteolysis experiments, which revealed significantly faster protein degradation in AL versus AL T32N, but marginally slower degradation in GL versus GL N32T (Figure 5C). In comparison, HDX results showed significant destabilization in several regions of AL versus AL T32N and a large stabilization in GL versus GL N32T (Figure 6). Together, these results suggest that increased structural stability of AL upon T32N substitution decelerates proteolysis; conversely, N32T substitution in GL has an opposite effect (Figures 5B, 5C, 6). Importantly, the proteolysis and HDX MS data are consistent with the amyloid growth kinetics which were much faster in AL compared to AL T32N (Figure 5D, S6D, S7C). Collectively, the results in Figures 5, 6 and S6–S8 show that the N32T substitution plays a major role in destabilizing LC and augmenting its proteolytic cleavage and amyloid formation, particularly in the context of other AL mutations.

Additional support for this conclusion comes from the comparison of two restorative substitutions, AL T32N in hotspot βC_v and AL Y75S in hotspot βE_v (Figure S9). Each of these substitutions replaces an amino acid found in AL with its GL counterpart and decreases the amyloidogenic sequence propensity of its respective hotspot (Figure 1), yet the effects on the protein properties are dramatically different. AL T32N substitution stabilizes the AL LC against thermal denaturation and proteolytic degradation and protects it from forming amyloid. In stark contrast, AL Y75S substitution slightly destabilizes the AL LC, accelerates its tryptic digestion, and has no significant effect on amyloid growth; the latter was observed both for the single and the double substitution, AL Y75S and AL T32N/Y75S (Figure S9). These results indicate that, unlike N32T, S75Y substitution is not a driver of amyloid formation, and support our conclusion that hotspot in βC_v, but not in βE_v, initiates amyloid formation in this AL LC.

DISCUSSION

This study integrated HDX MS with analyses of the amino acid sequence, structural stability, limited proteolysis, and amyloid formation to explore patient-derived and engineered recombinant Ig LCs of λ6 subtype, which is predominant in AL amyloidosis. Our results provide the first direct evidence that CDR1, which has substantial amyloidogenic sequence propensity (Figure 1A), is the most dynamic/exposed region in V_L domains of λ6 GL, MM and AL proteins studied (Figures 3, 4). Since all λ6 LCs have a common GL sequence, our results suggest strongly that CDR1 can contribute to the pathological misfolding of other AL LCs from λ6 subtype. This role of CDR1 is supported by computational and amyloid growth studies using tryptic digests of a model λ6 LC, 6aJL2 [37]. Additional support comes from HDX NMR analysis of JTO V_L (which has a similar sequence to MM V_L explored in the current study) showing that the central residue in CDR1 has the lowest stability in this V_L [22].

The current study not only reveals a key role for CDR1 in amyloid formation by λ6 AL01–095, but also implicates N32T, which is the sole AL substitution in CDR1 of this LC, as a major player in this process. We show that the N32T substitution increases the amyloid-forming sequence propensity of this dynamic surface loop (Figure 1A) and further decreases protection from exchange (Figures 6A, S8), a combination that is expected and observed to be pro-amyloidogenic. In fact, N32T combined with other AL mutations increases proteolytic susceptibility of this LC and augments amyloid growth in vitro (Figures 5, S6, S7). Unexpectedly, the destabilizing effects of N32T propagate across V_L changing protection in other regions up to 30Å away, e.g., in residues 78–88 at the opposite end of V_L (Figures 4E, 6). This destabilization includes the J-region (Figure 6A,B) whose proteolytic cleavage can release the V_L domain, a major protein precursor of AL amyloid. All these effects are expected to augment amyloid formation.

Comparative analysis of AL, AL T32N, GL, and GL N32T proteins showed that the N32T substitution cannot fully account for the observed differences in the secondary structure, proteolytic susceptibility, amyloid growth and local protection of AL versus GL proteins (Figures 3–6). Consequently, not one but several AL substitutions drive amyloid formation in this LC. Moreover, the observation that N32T promotes amyloid formation in the context of other AL substitutions, but not alone (Figures 5D, S6D, S7C), indicates coupling between this and other sites. Coupling between different sites of AL substitutions has been previously reported for one other LC of λ3 subtype [10]. Our finding of coupled AL substitution sites λ6 LC suggests that this may be a general phenomenon wherein several substitutions cooperate to make the protein amyloidogenic.

The long-range effects of N32T, a conservative substitution in a dynamic surface loop, are unusual. In other proteins and their complexes, effects such as these are thought to underlie the ensemble nature of allostery [45]. Our finding resembles the pro-amyloidogenic effect of D76N substitution in a surface loop of β2 microglobulin, which destabilized the edge strands in the β-sandwich [46]. Another example includes a λ3 LC V_L wherein combined effects of two glycine substitutions in the loops, G49R in CDR2 and G94A in CDR3, increased the dynamics in the protein core and promoted amyloidogenesis [10]. Unlike these substitutions, N32T is highly conservative as it changes neither charge nor the loop entropy. Therefore, the long-range effects of N32T on the V_L dynamics and amyloid formation are particularly striking. We hypothesize that in antibodies, which are highly dynamic molecules [32], such a long-range propagation of conformational changes from the antigen-binding CDRs across the Ig domains helps optimize antigen recognition and binding.

To envision a possible mechanism by which the destabilizing effects of N32T propagate across V_L, we examined the crystal structure of a MM λ6 LC (Figure 6C). In this structure, N32 is located at the end of a short helix (in residues 28–32, SIDSN, which correspond to SIASN in GL). The solvent-exposed N32 side chain in MM contributes to an extensive hydrogen-bonding network involving multiple main and side chains, wherein N32 Oδ is H-bonded to R24 Nη and Y94 CO. Since R24, N32, and Y94 are common to MM and GL, some of these H-bonds are probably retained in GL. Replacement of N32 with a smaller less hydrophilic threonine abrogates the H-bonds to R24 and/or Y94. Propagation of this perturbation across the H-bonding network throughout V_L helps explain the observed increase in the Ig core dynamics in upon N32T substitution (Figures 5, 6, S6–S9). Similarly, NMR and molecular dynamics studies of D76N β2 microglobulin implicated perturbed H-bonded network in β-sandwich destabilization and amyloid formation [46].

In a recent study of an AL LC from λ1 subtype, mutational frequency at specific positions in the LC repertoire was proposed to predict the clinical implications of individual AL mutations: only replacements in highly conserved positions were postulated to promote amyloidosis [9]. Does this conjecture extend to other LC families such as λ6? To assess residue replacement frequency at nine positions differing between λ6 AL01–095 V_L and its GL counterpart, we used ALBase [47] that currently lists 94 AL LCs and 49 non-amyloidogenic LCs. Numerous substitutions in eight out of nine positions were found. For example, N32 was replaced in 23 AL LCs (~25% frequency) and in 4 non-amyloidogenic LCs (~8% frequency). Interestingly, one of these non-amyloidogenic LCs (NCBI code KY471436) contained the N32T substitution. This finding agrees with our in vitro analysis of GL N32T showing that N32T alone is insufficient to make a LC amyloidogenic (Figures 5, S6–S8). Notably, only the S75Y substitution was unique to λ6 LC, AL01–095; ALBase sequences listed no other substitutions at this position. S75Y is located at a surface site in the relatively well-ordered βE_V strand and enhances its amyloid hotspot (Figure 1). Although it is possible that S75Y contributes to amyloidosis involving AL01–095, our results suggest that this contribution is much smaller than that of N32T and counteracts its destabilizing effect (Figure S9). Collectively, our findings indicate that amino acid replacement frequency does not necessarily predict the pathologic role of a specific residue, and emphasize the combined effects of multiple substitutions in amyloidosis. Furthermore, analysis using LICTOR [43] did not attribute changes in toxic / non-toxic status to any single or double substitution that differentiates AL from GL V_L, including N32T, S75Y or combination thereof.

Other key differences between the current HDX MS study and that of the λ1 AL LC [9] as well as the HDX NMR study of λ6 MM LC [22] can be summarized as follows. First, relative protection in specific V_L segments is different; e.g., the least protected segment in λ6 LC studied here is CDR1 (Figure 4), while in the λ1 LC it is apparently CDR3 based on the limited HDX data reported in [9]. This difference highlights antibody-specific variations in the conformational dynamics of CDR loops and suggests that individual CDRs play different roles in amyloid formation by λ1 and λ6 LCs.

Second, no significant effects of AL replacements on the protection of the C_L domain were observed in λ6 LC (Figure 4A–C). In contrast, such effects were seen in the λ1 LC and were attributed to a highly destabilizing V81L substitution at the C_L-V_L interface [9]. This difference highlights LC-specific variations in domain-domain interactions. Notably, HDX NMR analysis of JTO LC (which is similar to our MM LC) and its variants yielded free energy profiles suggesting domain-domain interactions [22]. There are important similarities, as well as some differences, in the behavior of this protein observed by NMR and MS. The differences may result, at least in part, from very different experimental conditions used in MS (plasma pH, low protein concentrations approximating those in plasma) versus the NMR (acidic pH and ~10-fold higher protein concentrations). The pH of the analysis of JTO LC was more acidic (pH 6.4 in H₂O, pD 5.0 in D₂O) in NMR studies versus pH 7.4 of the current MS work. The protein concentrations for NMR were about an order of magnitude higher than those of the current MS study (~0.2 mg/ml), and protein lyophilization was employed for NMR versus freshly prepared proteins used in the current MS work. The time scales of HDX measurements in NMR and MS were different, and the data processing was also distinct: NMR data were presented as local stability ΔG=RT ln(k_on/k_off) shifted by a theoretical constant along the Y-axis, while MS data were shown as protection factors measured on the absolute scale (no Y-axis shift). Finally, unlike the NMR study [22], the results of the current study involved minimal assumptions in the data processing.

Third, the overall destabilization of the AL LC compared to its GL counterpart was modest for the λ6 protein, with a decrease in the apparent T_m ranging from ~2°C for the full-length LC (Figure 2B) to 4–6°C for the isolated V_L domain (Figure S7B) [20]. This contrasts with the λ1 LC showing a larger destabilization by 8°C [9]. Together, these distinctions reveal unique structural and dynamic features in the native states of individual LCs and suggest that early triggers of amyloidosis are LC-specific. This conjecture is consistent with a recent report that structural variations induced by AL substitutions in CDR2 and CDR3 drive amyloid formation by a λ3 LC [10] and with the observation of dissimilar structures of amyloid fibrils extracted from different AL patients, each containing a unique LC [11–14,16]. Together, these findings suggest distinct amyloid-forming pathways for different LCs.

Importantly, the native LC structure in the current study showed protection from exchange at the sites of proteolytic cleavage of C_L in vivo (Figure 4), which were reported for a homologous λ6 LC for patient-derived AL deposits [30]. In the amyloid fibril structure of this LC, these sites were largely disordered [14,30]. These results suggest that fragmentation of λ6 LC follows amyloid formation rather than triggers it. This finding is important for therapeutic targeting of AL amyloidosis.

In summary, studies integrating HDX MS with other techniques provide sharp insights into protein-specific drivers of amyloid formation. Expanding such studies to include a wider repertoire of LCs from λ and κ classes found in AL amyloidosis should help establish the repertoire of critical determinants for LC misfolding, and thereby help guide the therapeutic targeting of this incurable disease.

Materials and Methods

AL λ6 LC was cloned from bone marrow of an AL01–095 patient (NCBI code EF58390) as described [20]. MM λ6 LC was from an MMJTO patient (NCBI code 1CDO_A) with MM but no amyloidosis [31]. The closest germline sequence, with V_L and C_L encoded by the GL gene donors IGLV6–57*01 and LC3*04, respectively, was assigned using the IMGT/V-QUEST tool, http://www.imgt.org/ [48]. The proteins were cloned and expressed using an E. coli-based system and purified to 95%+ purity as described [20]. All other materials were of highest available purity. Methods of sample preparation and analysis are described in SI Materials and Methods. All HDX MS data have been deposited to the ProteomeXchange Consortium via the PRIDE [49] partner repository with the dataset identifier PXD026528.

Highlights.

• Drivers of Ig λ6 LC misfolding were identified by HDX MS and biochemical methods

• CDR1 loop is the least protected part of the variable domain in λ6 LC class

• N32T substitution in CDR1 increases protein dynamics and drives aggregation

• Substitutions at several interacting sites contribute to AL amyloidosis

• Substitution in a conserved position does not always predict AL

Abbreviations:

Ig

immunoglobulin

AL

amyloid light chain

MM

multiple myeloma

GL

germline

LC

light chain

C_L

light chain constant domain

V_L

light chain variable-joined domain

CDR

complementarity-determining region

FR

framework region

J-region

joined region

HDX

hydrogen-deuterium exchange

MS

mass spectrometry

CD

circular dichroism

ThT

thioflavin T

SEC

size exclusion chromatography