A Residue-specific Shift in Stability and Amyloidogenicity of Antibody Variable Domains*

Background: Amyloid fibrils of variable (VL) domains are the main cause of death in light chain amyloidosis. Results: Conserved VL residue 2 is crucial for structural integrity and amyloidogenicity of VLκ domains. Conclusion: Our data reveal novel insights into the architecture of variable domains in general and the prerequisite for fibrillation. Significance: This is important for understanding the principles of antibody structure and amyloidogenicity. Variable (V) domains of antibodies are essential for antigen recognition by our adaptive immune system. However, some variants of the light chain V domains (VL) form pathogenic amyloid fibrils in patients. It is so far unclear which residues play a key role in governing these processes. Here, we show that the conserved residue 2 of VL domains is crucial for controlling its thermodynamic stability and fibril formation. Hydrophobic side chains at position 2 stabilize the domain, whereas charged residues destabilize and lead to amyloid fibril formation. NMR experiments identified several segments within the core of the VL domain to be affected by changes in residue 2. Furthermore, molecular dynamic simulations showed that hydrophobic side chains at position 2 remain buried in a hydrophobic pocket, and charged side chains show a high flexibility. This results in a predicted difference in the dissociation free energy of ∼10 kJ mol−1, which is in excellent agreement with our experimental values. Interestingly, this switch point is found only in VL domains of the κ family and not in VLλ or in VH domains, despite a highly similar domain architecture. Our results reveal novel insight into the architecture of variable domains and the prerequisites for formation of amyloid fibrils. This might also contribute to the rational design of stable variable antibody domains.

Variable (V) domains of antibodies are essential for antigen recognition by our adaptive immune system. However, some variants of the light chain V domains (V L ) form pathogenic amyloid fibrils in patients. It is so far unclear which residues play a key role in governing these processes. Here, we show that the conserved residue 2 of V L domains is crucial for controlling its thermodynamic stability and fibril formation. Hydrophobic side chains at position 2 stabilize the domain, whereas charged residues destabilize and lead to amyloid fibril formation. NMR experiments identified several segments within the core of the V L domain to be affected by changes in residue 2. Furthermore, molecular dynamic simulations showed that hydrophobic side chains at position 2 remain buried in a hydrophobic pocket, and charged side chains show a high flexibility. This results in a predicted difference in the dissociation free energy of ϳ10 kJ mol ؊1 , which is in excellent agreement with our experimental values. Interestingly, this switch point is found only in V L domains of the family and not in V L or in V H domains, despite a highly similar domain architecture. Our results reveal novel insight into the architecture of variable domains and the prerequisites for formation of amyloid fibrils. This might also contribute to the rational design of stable variable antibody domains.
Antibodies consist of structurally highly homologous domains all sharing the immunoglobulin fold. This tertiary structure is characterized by two ␤-sheets that are connected by a conserved disulfide bridge and form a ␤-barrel structure. In the variable domains, the anti-parallel ␤-pleated sheets include nine ␤-strands in a 4 ϩ 5 orientation (ABE(D/C)CЈCЉFG) (1), whereas those of the constant domains are made up of seven ␤-strands in a 4 ϩ 3 orientation (ABE(D/C)FG) (2). In the variable domains, this framework structure is expanded by three variable loops (CDR 2 1-3) that mediate antigen recognition. The principles that govern the stability of the variable domains are of special importance, as loops of different lengths and composition have to be tolerated. To achieve this, the amino acid composition of the framework is highly conserved at most positions. Single substitutions in variable domains can have deleterious effects, seen in some patients with monoclonal gammopathy or clonal B cell dyscrasias. In these patients, antibody variants are pathologically deposited in a fatal illness called light chain deposition disease (3,4) and systemic amyloid light chain (AL) amyloidosis (5,6). Analysis of proteins isolated from AL patient biopsies showed that the main constituent of amyloid deposits is the variable domain of the light chain (V L ) (7,8). The basis of the disease is the overproduction of unpaired antibody light chains (LCs), which are secreted into the bloodstream and taken up by multiple organs and tissues where they aggregate, severely compromising the functions of the affected organs and ultimately leading to death (9,10). AL amyloidosis remains the most common and most fatal systemic amyloidosis in western countries (7,11).
Despite the high similarity between the V L and V H domains, V H is less susceptible to amyloid fibril formation (12). Furthermore, why some V L s form amyloids and others do not still remains largely enigmatic. In general, it is therefore not clear which residues are of special importance for the structural integrity of the V L domain. As the C L domain is not involved in the deposits, it seems that structural elements exist that protect it from misfolding (13), whereas some factors within the V L domain might predispose it to the disease state. Highly conserved consensus sequences within the framework region are potentially important elements in this respect. A striking feature of variable domains is the conserved N terminus of ␤-strand A (FWR1). However, the role of the N-terminal residues of the LC on domain structure, stability, and amyloidogenic potential is still unresolved.
In this study, we performed a detailed analysis of the importance of the conserved N-terminal residues for the structural integrity, the conformational stability, and amyloidogenicity of variable domains. With biophysical approaches including CD, fluorescence, NMR, and molecular dynamics (MD) simulations, we demonstrate that specifically residue 2 affects the conformational integrity and is a key element controlling its amyloidogenic properties. This reveals novel insights into the architecture of antibody variable domains.

EXPERIMENTAL PROCEDURES
Oligonucleotides (primers) were purchased from MWG Biotec. Ultrapure GdmCl was obtained from Sigma. All other chemicals were purchased from Merck. The concentrations of GdmCl solutions were determined from refractive indices (14). Unless stated otherwise, all measurements were carried out at 25°C (pH 7.4) in PBS buffer.
Cloning, Mutagenesis, Expression, and Purification of the Variable Domain Variants-The MAK33 V L wild type encoding plasmid was obtained as published previously (15). 1OPG V L , 1AQK V L , and 1VGE V H wild type sequences cloned into a pET28b vector between the NcoI and HindIII restriction sites were purchased from Geneart AG (Regensburg, Germany). Single, double, and triple point mutants were generated with QuikChange mutagenesis PCR with forward and reverse primers carrying the target mutation. Inserts of the wild type, single point, and double point mutants in the pET28b vector were used as PCR templates for the generation of single, double, and triple point mutants, respectively. The PCR was followed by digestion of template DNA with DpnI enzyme (New England Biolabs) for 1 h at 37°C. The PCR product was then transformed through a heat shock-dependent transformation into competent Escherichia coli Mach1 cells. Positive clones were selected on kanamycin LB plates overnight at 37°C. Plasmid DNA from a single colony was isolated with the Wizard Plus SV Miniprep kit (Promega) and sequenced using the T7 forward or pET-RP sequencing primer by Eurofins MWG Operon to verify the desired mutation. All MAK33 V L , 1OPG V L , 1AQK V L , and 1VGE V H variants were expressed and purified as described previously (15,16). In brief, the plasmid was transformed into E. coli BL21(DE3)-star cells (for V L and variants) or in E. coli JM109 cells (for V H variants) for expression at 37°C. At an A 600 of 0.6 -0.8, expression was induced using 1 mM isopropyl ␤-Dthiogalactopyranoside. Cells were harvested after overnight growth, and inclusion bodies were prepared as described previously (17,18). The pellet was solubilized and unfolded in 25 mM Tris-HCl (pH 8), 5 mM EDTA, 8 M urea, and 2 mM ␤-mercaptoethanol at room temperature for at least 2 h. The soluble fraction was then injected onto a Q-Sepharose column equilibrated in 25 mM Tris-HCl (pH 8), 5 mM EDTA, and 5 M urea. The proteins were eluted in the flow-through and diluted two times before being refolded by dialysis into 250 mM Tris-HCl (pH 8.0), 100 mM L-Arg (for V L and variants) or 400 mM L-Arg (for V H variants), 5 mM EDTA, 1 mM oxidized glutathione, and 0.5 mM reduced glutathione at 4°C overnight. To remove misfolded aggregates and remaining impurities, the protein was cleaned using a Superdex 75 gel filtration column (GE Healthcare) equilibrated in PBS buffer. The recovery and purity of intact protein were verified by SDS-PAGE and matrix-assisted laser desorption/ionization time-offlight mass spectrometry.
CD Measurements-CD measurements were performed using a Jasco J-720 spectropolarimeter (Jasco, Grossumstadt, Germany) equipped with a Peltier element. Far-UV CD spectra were measured using 10 M protein in a 1-mm path length cuvette between 260 and 200 nm. Near-UV CD was measured between 320 and 250 nm using 50 M protein in a 1-mm cuvette. All spectra were accumulated 16 times, and buffer was corrected. Thermal transitions were recorded using 10 M protein at 212 nm for V L and 205 nm for V L and V H variants with a heating and cooling rate of 20°C/h.
Tryptophan Fluorescence Measurements-Unless otherwise stated, all fluorescence measurements were carried out using a FluoroMax-4 spectrofluorimeter (Horiba Jobin Yvon, Bensheim, Germany). Tryptophan fluorescence measurements were performed with excitation and emission slit widths of 2 and 3 nm, respectively. The protein concentration in a 1-cm quartz cuvette was 1 M. The sample was excited at 280 nm, and spectra were recorded between 300 and 450 nm. Equilibrium unfolding and refolding transitions were carried out by denaturing 1 M protein overnight at different concentrations of GdmCl (between 0 and 3 M GdmCl). Fluorescence intensity at 358 nm was measured for 50 s, and the average was taken. Analysis of data was carried out using the linear extrapolation method described previously (14,19).
1-Anilino-8-naphthalene Sulfonate (ANS) Binding Assay-ANS binding was measured by fluorescence emission recorded from 400 to 650 nm, with excitation at 380 nm with excitation and emission slit widths of 2 and 3 nm, respectively. Experiments were performed at a protein concentration of 10 M and an ANS concentration of 100 M in 1-cm quartz cuvette. All spectra were accumulated three times and averaged.
Fibril Formation with Ultrasonication-For the standard reaction, the V L variants were diluted at a concentration of 30 M in PBS solution and 5 M thioflavin T (ThT). Then 0.2 ml of the V L variant solution was applied to each of the 96 wells of a microplate (Greiner Bio-One on a 96-well microplate 675-074 made of polystyrene with a size of 128 ϫ 85 mm). The microplate was set on the HANABI (HANdai Amyloid Burst Inducer), which combines the use of a water bath-type ultrasonicator (ELESTEIN, Elekon, Chiba, Japan) and a microplate reader (SH-9000, Corona Electric Co., Ibaraki, Japan). The water bath with 400 ϫ 250 mm ϫ 230 nm (height) contains about 12 liters of water, and three ultrasonic transducers are set, two on the sides and one at the bottom. Pulses were applied to the microplate from three directions, focusing on the center position, for cycles of 1 min followed by a quiescent period of 9 min. The measurement temperature was maintained at 37°C. The formation of fibrils was monitored by ThT fluorescence with excitation and emission wavelengths of 450 and 490 nm, respectively. Microplates were incubated at 37°C and shaken before the measurements. Transmission electron microscopy (TEM) images were taken using a HITACHI H-7650 transmission microscope, (Hitachi, Tokyo, Japan), with an acceleration voltage of 80 kV. Aliquots (5 l) of the samples were placed on a 400-mesh copper grid covered by a carboncoated colloidal film for 60 s. Two percent (w/v) uranyl acetate solution (5 l) was then placed on the grid to stain the samples for 60 s. Excess sample solutions on the grid were removed with a filter paper. The magnification on the representative pictures was 30,000-fold.
Fibril Formation without Ultrasonication-Samples of V L at a concentration of 30 M in PBS buffer (pH 7.4), containing 0.05% sodium azide (or acetate buffer (pH 2)) were incubated with shaking (gentle agitation) in a roller shaker at 37°C for at least 1 week. At various intervals, a 20-l aliquot was withdrawn for ThT binding and TEM. For the ThT binding, 10 l of protein aliquot was added to 480 l of a PBS buffer (pH 7.4) followed by the addition of 10 l of 500 M ThT prepared in PBS buffer (pH 7.4) (ThT final concentration 10 M). This was left to incubate for 90 s before acquiring the emitted signal in a 1-cm quartz cuvette from 430 to 570 nm with excitation at 440 nm. The spectra were recorded using an FP-6500 spectrofluorimeter (Jasco) equipped with a Jasco ADP-303T Peltier as temperature controller. The excitation and emission slit widths were both 3 nm. A scan speed of 100 nm/min, an integration time of 1 s, and a 0.5-nm recording interval were used. Three scans were averaged for each sample. Rayleigh (elastic) light scattering at an angle of 90°was also measured at the excitation wavelength peak of 440 nm (20,21). Therefore, for the same sample, information on its light scattering ability (presence of macromolecular assemblies or aggregation) and ThT binding (fibril formation) was assessed. For comparison, the Rayleigh scattering intensity at 440 nm and the ThT fluorescence intensity at 480 nm were used.
For TEM measurements, a 10-l protein aliquot was fixed on a 200-mesh activated copper grid and incubated for 60 s. After a washing step with 10 l of H 2 O, the samples were stained with 10 l of a 1.5% uranyl acetate solution. Samples were recorded with an acceleration voltage of 100 kV at a magnification of ϫ33,000 on a transmission electron microscope JEM100CX (Jeol).
Thermally Induced Aggregation-Native MAK33 or 1OPG V L at a concentration of 10 M in PBS buffer (pH 7.4) with 10 M ThT was subjected to increasing temperatures from 20 to 80°C in steps of 5°C in a 1-cm path cuvette. After thermal equilibration for 3 min, emission spectra were recorded for Rayleigh scattering and ThT fluorescence (21) as stated under "Fibril Formation without Ultrasonication." NMR Spectroscopy-All NMR spectra were recorded using uniformly 15 N, 13 C-labeled proteins in 20 mM phosphate, 50 mM NaCl (pH 6.5) at 298 K. Backbone resonance assignments of MAK33 V L -WT and MAK33 V L -I2E were transferred from previous assignments of MAK33 V L -S20N (data not shown). The assignment was accomplished based on threedimensional HNCA, HNCACB, CBCACONH, HNCO, and HNCACO experiments (22). Experiments were carried out at 1 mM protein concentration. Those residues, for which the assignment could not be transferred from MAK33 V L -S20N, were assigned based on an HNCA experiment using a 100 M MAK33 V L -I2E sample. All spectra were recorded on a Bruker AVANCE 900MHz spectrometer (Bruker Biospin) equipped with a triple-resonance cryoprobe. Spectra were processed using TOPSPIN 3.2 (Bruker BioSpin) and analyzed with CcpNmr 2.2.2 (23). For determination of residues affected by the I2E substitution, 15 N HSQC spectra of MAK33 V L -WT and MAK33 V L -I2E, both at 50 M, were acquired on a Bruker Avance 600MHz spectrometer equipped with a cryoprobe. For all residues, chemical shift perturbations were determined using Equation 1, with ⌬␦ res being the weighted chemical shift difference, ⌬␦ 1H as the chemical shift of the amide proton in ppm, and ⌬␦ 15N as the chemical shift of the amide nitrogen in ppm. For comparison of signal intensities, all signals, which displayed significant overlap in one of the proteins, were ignored. Intensities were normalized for both proteins to the same average intensity.
MD and Umbrella Sampling (US) Simulations-Start structures of MAK33 V L domains and V L and V H domains were obtained by extracting the corresponding coordinates from the crystal structures 1FH5 (MAK33 V L ), 1OPG, 1AQK, and 1VGE. Residue 2 of MAK33 V L , 1AQK V L , 1VGE V L , and 1VGE V H was substituted in silico with Glu although residue 2 of 1OPG, and V L was replaced with Ile. In silico substitutions were performed using the SPDBV package (24), while selecting the best fitting side chain rotamer. All MD simulations and the analysis of root mean square deviations (r.m.s.d.) and fluctuations were performed using the Amber12 package (25). Proteins were solvated in octahedral boxes, including explicit ions and explicit (TIP3P) water molecules (26). The simulation systems were first energy-minimized (5000 steps) followed by heating up to 300 K in steps of 100 K with position restraints on all heavy atoms of the proteins. Subsequently, positional restraints were gradually removed from an initial 25 kcal mol Ϫ1 Å Ϫ2 to 0.5 kcal mol Ϫ1 Å Ϫ2 within 0.5 ns followed by a 1-ns unrestrained equilibration at 300 K. All production simulations were performed at a temperature of 300 K and a pressure of 1 bar. US simulations were performed using the distance between the C␣ atom of residue 2 and the C␣ of residue 32 at the floor of the binding region for residue 2 in the V L domains as a reaction coordinate. A quadratic penalty potential (k (d C␣-C␣ Ϫ d ref ) 2 , force constant k ϭ 2.0 kcal mol Ϫ1 Å Ϫ2 ) for the C␣-C␣ distance was used with reference distances varying from 11.5 to 16 Å in 0.5-Å steps and from 16 to 20 Å in 1-Å steps. At ϳ12-13 Å, residue 2 stays bound to the protein in the cavity as observed in the experimental x-ray structure, whereas it adopts a fully exposed state at distances of Ͼ16 Å. The associated potential of mean force was calculated using the weighted histogram analysis method (27).

Biophysical Characteristics, Stability, and Amyloidogenic Propensity of Two Highly Homologous V L Domains-MAK33
is a well studied IgG antibody with respect to folding and association. In this context, the folding pathway of the V L domain has been analyzed in detail (15); its amino acid composition is typical for a murine /IgG1 light chain variable domain (PDB code 1FH5 (28)). Interestingly, another antibody V L domain exists (PDB code 1OPG (29)), which has identical CDRs but five differences in the framework region (Fig. 1a). We were wondering whether these differences affect the structural properties of the domain. Both isolated domains were expressed, purified, and characterized. The far-UV circular dichroism (CD) spectra of the V L domains were similar with a minimum at 218 nm ( Fig. 1b) characteristic of the ␤-sheet conformation. However, the shapes of both native spectra suggest a large contribution of random coils, which might be due to long flexible CDR loops. Their near-UV CD spectra (Fig. 1c) with a minimum at 275 nm confirmed that the tertiary structure of both proteins is correctly formed (30,31). The presence of a buried tryptophan in close proximity to the disulfide in antibody domains makes it a sensitive probe for studying conformational changes by fluorescence. Intrinsic tryptophan fluorescence spectra of both V L domains showed MAK33 V L to have a large (ϳ2.5-fold) increase in fluorescence emission intensity (maximum at ϳ358 nm) in the presence of 3 M GdmCl (Fig. 1d), as opposed to a small increase observed in 1OPG V L (Fig. 1e). This suggests already a partial exposure of tryptophan in the native 1OPG V L . To verify whether the V L domains exposed hydrophobic patches on their surfaces, ANS binding assay was performed. As depicted in Fig. 1f, 1OPG V L binds ANS with an increase and a blue-shifted fluorescence emission maximum (shift from 525 DVLMTQTPLS LPVSLGDQAS ISCKSSQSIV HSSGNTYFEW YLQKPGQSPK LLIYKVSNRF : ::**:* : *. : *: .. ::* :**.: : * * **. ::*: *** .*. to 515 nm), suggesting the presence of exposed hydrophobic patches. MAK33 V L shows no ANS binding (Fig. 1f). The propensity of both MAK33 and 1OPG V L domains to aggregate with increasing temperatures was monitored by recording Rayleigh (elastic) scattering of ThT fluorescence excitation (440 nm) light (Fig. 1f) (20,21). Surprisingly, 1OPG V L already starts aggregating at ϳ35°C, although MAK33 V L only starts transforming into aggregates at ϳ50°C (a difference of ϳ15°C). Maximum aggregation for both proteins occurred at ϳ65°C. No ThT binding was observed for any of the proteins at the experimental temperature range of 20 -80°C (data not shown), which implies that the aggregates formed by thermal induction did not contain amyloid fibrils. When we determined their stabilities against temperature monitored by far UV CD, we found, interestingly, that the 1OPG V L is 16°C less stable than MAK33 V L as judged from their transition midpoints (Fig. 2b), which correlates very well with thermal-induced aggregation monitored by light scattering at 440 nm. Thermally induced unfolding transitions were not reversible for any of the proteins. For equilibrium unfolding and refolding in the presence of GdmCl monitored by Trp fluorescence, the MAK33 V L with a cooperative sigmoidal transition had identical midpoints in both the unfolding and refolding directions (Fig. 2c). This sigmoidal transition was not observed for 1OPG V L , for which a native baseline could not be obtained (Fig. 2c). The GdmCl-induced unfolding transitions were cooperative and reversible. Assuming two-state transitions, the data were fitted using linear extrapolation (14,19) that yielded an intrinsic stability of Ϫ17.8 kJ mol Ϫ1 for MAK33 V L and Ϫ3.0 kJ mol Ϫ1 for 1OPG V L and m values of 18.0 Ϯ 2.4 and 12.9 Ϯ 5.3, respec- tively, which reflects the different cooperativities observed. Although the stability of MAK33 V L falls within the range of those reported for amyloidogenic V L domains (ϳ15-20 kJ mol Ϫ 1 ) (32,33), it is known to be nonamyloidogenic in vitro at physiological pH (15). Accordingly, after 1 week of incubation in PBS buffer at pH 7.4 and 37°C with gentle agitation, MAK33 V L did not form any amyloid fibrils (Fig. 2c). However, under these conditions, 1OPG V L readily turned into well defined fibrillar structures (Fig. 2c). When the proteins were incubated in an acidic buffer (pH 2) under similar conditions, fibrils were detected for both MAK33 V L and 1OPG V L samples (data not shown), confirming our previous results on MAK33 V L (15). This supports the notion that under appropriate destabilizing conditions, even proteins not known to form amyloid fibrils under normal physiological conditions can indeed fibrillize (34 -36).
To assess how conserved the frameworks of both V L domains are in comparison with other V L sequences, a comprehensive sequence analysis was performed using the abYsis database, which integrates sequences from the Kabat (37) and IMGT (38) databases. Fig. 1a depicts five representative V L sequences from this alignment. Considering amino acid side chain chemistry, all framework residues are very similar, except for residue 2. In 1OPG V L , this residue is a Glu, although in most other V L frameworks an aliphatic residue is found at the same position (Fig. 1a). A residue frequency distribution analysis using the abYsis database with all available (ϳ20,000 nonidentical) LC sequences revealed that Ile occurs with the highest frequency (46%) at position 2 (Fig. 3). Other aliphatic residues like Val, Leu, and Ala represent 7-9% at this position in V L domains and charged residues (Glu, Asp, Lys, and Arg) only Ͻ1% (Fig. 3a). In addition, looking at the residue frequency distribution at position 102 of these V L domains, we observed a highly conserved Thr with a frequency of 98%, although all other residues had a Ͻ1% frequency of occurrence (Fig. 3d). In the more conformationally stable MAK33 V L , position 102 is occupied by Thr, whereas in the less stable 1OPG V L , the less frequent Ser is found. With positions 2 and 102 showing huge frequency distribution discrepancies between the two V L domains, we wondered whether these positions were responsible for the different behaviors of the two proteins. In consequence, we exchanged these residues between the two V L domains and monitored their effects on stability and amyloid fibril formation.
Residue 2 Is Crucial for the Stability of V L Domains-To identify residue(s) that predispose 1OPG V L for lower stability and fibril formation, single point variants were generated by replacing residues in 1OPG V L with corresponding residues of MAK33 V L at any of the five different positions (OP-E2I, OP-L3V, OP-G100A, OP-S102T, and OP-I106L). All five substitutions were correctly folded as revealed by UV CD spectra. The thermal unfolding of these variants showed a striking increase in thermal stability for OP-E2I with a T melt value of 43.1°C, compared with 31.6°C for 1OPG V L -WT. Also, OP-E2I was significantly more stable toward GdmCl than 1OPG V L -WT (Fig. 4, a and c; Table 1). OP-S102T was only slightly more stable, and the other single point substitution variants were all similar to the less stable 1OPG V L -WT (Table  1), implying that the exchange of these residues is not important for the conformational stability of the domain. With the high increase in stability seen in OP-E2I and the small increase observed in OP-S102T, we next tested whether a combination of both substitutions (OP-E2I/S102T) and a triple substitution OP-E2I/G100A/S102T (OP-IAT) would further lead to increased stability. The double and triple substitution variants both unfolded with a T melt of about 45°C quite similar to that obtained for OP-E2I (Table 1), confirming only a small influence of the G100A and S102T substitutions on the thermal stability of the protein. However, in terms of stability to GdmCl, both variants were more stable than the respective single substitution variants ( Fig. 4c; Table 1). These stability values observed for the double and triple substitution variants of 1OPG V L are similar to that of MAK33 V L wild type (Table  1). This demonstrates an additive effect of exchanges at different positions.
Because Glu-2 and Ser-102 to a minor extent were determined to be the destabilizing residues in 1OPG V L , we were also interested in how they affect MAK33 V L . Interestingly, MK-I2E was destabilized, whereas MK-T102S had stability values similar to that of MAK33 V L -WT (Fig. 4, b and c; Table 1). Specifically, MK-I2E was ϳ10°C less stable with a free energy of ϳ12 kJ mol Ϫ1 less than that of MAK33 V L -WT. The combined substitution variant (MK-I2E/T102S) was less stable than both single substitution variants with a T melt of 33.6°C and a chemical stability of Ϫ8.0 kJ mol Ϫ (Fig. 4c; Table 1). This again emphasizes the additive effect of both positions. To demonstrate that our findings on the pivotal role of residue 2 are general to the V L family, we investigated a human /IgG1 antibody V L domain (1VGE V L ) (39), which differs greatly in amino acid sequence from 1OPG and MAK33 V L . The N-terminal motif is ELV as opposed to DIV in MAK33 and DEL in 1OPG V L (Fig. 1a). To assess the role of residue 2 in this human V L , Leu was replaced with Glu (1VGE V L -L2E). Notably, analysis of their thermal and thermodynamic stabilities also revealed a decrease in thermal stability of ϳ10°C and a 2-fold decrease in thermodynamic stability for 1VGE V L -L2E compared with the wild type (Fig. 4, d and e; Table 1). This correlates very well with results obtained for MAK33 and 1OPG V L , implying that the effects mediated by residue 2 are conserved within the V L family.
Influence of Sequence Variation on Amyloid Fibril Formation-To examine whether the increase in the stability of 1OPG V L variants also correlates with the resistance to amyloid formation, we set up ultrasonication amyloid induction assays at neutral pH at 37°C (40 -45). ThT fluorescence was used to monitor the fibrillation process (46,47). All variants that had a thermodynamic stability of Ͻ10.0 kJ mol Ϫ1 (Table 1) bound ThT as indicated by an increase in fluorescence after a lag time of several hours, whereas the more stable variants (OP-E2I/ S102T, OP-IAT, MAK33 V L -WT, and MK-T102S) showed no ThT fluorescence (Fig. 5, a and b). The large fluctuations in ThT fluorescence amplitude are likely due to differences in higher order structure of amyloid fibrils as ThT fluorescence can change depending on the morphology of amyloid fibrils (41)(42)(43)48). These variations in ThT fluorescence amplitudes might also depend on the mutants, and it is worth noting that ThT fluorescence is not always proportional to the amount of amyloid fibrils (48).
To confirm the presence of fibrils, transmission electron microscopy was performed. As expected, amyloid fibrils were detected in all variants for which ThT fluorescence was observed (Fig. 5c). Similar results were obtained with gentle agitation without ultrasonication after 4 or 10 days of incubation of the V L variants (data not shown). However, although the stable variants were protected from amyloid fibril formation, they did transform into amorphous aggregates upon prolonged incubation at 37°C, revealed by the increase in Rayleigh scattering intensity (data not shown) Thus, the changes in V L domain stability induced by modulating residue 2 inversely correlated with its ability to form amyloid fibrils.
Residue 2 Is a Decisive Factor for V L Domain Architecture-Alignment of V L sequences revealed a highly conserved N terminus for this family, with residue 1 always being an acidic residue (Asp or Glu) and residue 2 mostly an aliphatic residue (Ile, Leu, or Val) (Fig. 1a). We wondered whether the conserved residue 1 also affects the stability of V L . Thus, N-terminal truncation variants of MAK33 V L were generated in which either the first residue (Asp-1) was deleted (MK-V L ⌬D1) or one in which both the first (Asp-1) and the second (Ile-2) residues were truncated (MK-V L ⌬D1/I2). Deletion of these N-terminal residues did not influence the structure of the domain as determined by CD spectroscopy. Surprisingly, the deletion of Asp-1 (MK-V L ⌬D1) did not lead to changes in the thermodynamic stability when compared with the wild type domain. In contrast, the deletion of both Asp-1 and Ile-2 (MK-V L ⌬D1/I2) resulted in a decrease in T melt and thermodynamic stability of ϳ7°C and ϳ10 kJ mol Ϫ1 , respectively (Fig. 6, a and b; Table 1), which is similar to the stability obtained by substituting Ile-2 alone with Glu (MK-I2E). These results show that residue 2 and not residue 1 is important for the stability of V L domains.
With Glu at position 2 identified to be destabilizing and Ile or Leu at the same position are stabilizing, we were interested to determine how residues with other side chains might affect the stability of these V L domains. Considering the residue frequency distribution at position 2 of V L (Fig. 3a), residue 2 of

TABLE 1 Thermal and chemical stabilities of 1OPG V L and MAK33 V L and 1VGE V L variants
Stabilities against the thermal and chemical (GdmCl) denaturation of different variants are shown. Midpoints of thermal transitions are shown as T melt . Because the GdmCl-induced unfolding transitions were reversible, the data were fitted to a two-state equilibrium unfolding model to obtain the thermodynamic stability of unfolding (⌬G U ), as well as the cooperativity parameter (m value), for a qualitative comparison of the data. 1OPG V L -WT and MAK33 V L -WT were replaced with Ala, Leu, Asp, Lys, or Gln. Interestingly, the thermal unfolding transitions of these variants revealed striking differences. All vari-ants with an uncharged residue at position 2 had a high thermal stability ( Table 2). Although the respective 1OPG V L variants showed an increase in T melt of 4 -12°C compared with their  SEPTEMBER 26, 2014 • VOLUME 289 • NUMBER 39 wild type, MAK33 V L variants with the same uncharged residues at position 2 had T melt values similar to the more stable MAK33 V L -WT ( Fig. 6e; Table 2). However, all variants with a charged residue (whether positive or negative) at position 2 were strongly destabilized. OP-E2D and OP-E2K were as unstable as 1OPG V L -WT, whereas the same substitutions in MAK33 V L resulted in an ϳ10°C decrease in T melt (Fig. 6e; Table 2).   a conserved framework. We were interested to determine whether residue 2 also plays a similar role in the V L and in the V H domains. A multiple sequence alignment (Fig. 7, a and b) and a residue frequency distribution analysis at position 2 of V L and V H showed this position in V L to be mostly occupied by Ser or Ala (Fig. 3a), although at the same position in V H domains, Val occurs with a frequency of 83% (Fig. 3f). To ascer-tain the effects of residue 2 in these variable domains, a human /IgG V L (1AQK (49)) and a human IgG1 V H (1VGE (39)) were examined. In the 1AQK V L domain, residue 2 (Asn-2) was replaced with either a Glu (1AQK V L -N2E) or Ser (1AQK V L -N2S). Unexpectedly, stability studies did not reveal any differences between the wild type and the 1AQK V L -N2E mutant (Fig. 7, c and d; Table 3). This suggests that the stabiliz-  SEPTEMBER 26, 2014 • VOLUME 289 • NUMBER 39

Role of Residue 2 for Integrity of Antibody Variable Domains
ing effects of residue 2 seen in the V L family cannot be ascribed to the V L family. Moreover, the mutant that contained Ser at position 2 (1AQK V L -N2S) was slightly more stable than the wild type and 1AQK V L -N2E mutant. When Val-2 of 1VGE V H was substituted with a Glu (1VGE V H -V2E) or Ala (1VGE V H -V2A), again no pronounced thermal stability differences were observed ( Fig. 7e; Table 3). Both 1VGE V H -V2E and 1VGE V H -V2A mutants were ϳ3°C less stable than the wild type. Unlike in the V L family, the effects of residue 2 seen in V H are minor and not specific for any amino acid side chain tested.
Influence of Residue 2 on Domain Structure and Dynamics-To obtain a structural understanding of the influence of the residue at position 2 of V L , both MAK33 V L -WT and MAK33 V L -I2E were analyzed by solution-state NMR spectroscopy. 71% of the non-proline residues could be assigned (Fig. 8a). Backbone assignment was hampered by chemical exchange dynamics. The unassigned residues mainly include the regions from the N terminus to Pro-8, from His-41 to Tyr-50, and from Cys-88 to Gly-101. Fig. 8b depicts chemical shift changes caused by the I2E substitution. We find particularly large chemical shift differences for residues Cys-23 to Ser-40 FIGURE 8. Structural properties of V L variants monitored by NMR spectroscopy. a, 1 H-15 N-HSQC spectra of MAK33 V L -WT and MAK33 V L -I2E. Both spectra were acquired at a protein concentration of 50 M in 20 mM phosphate, 50 mM NaCl (pH 6.5) at 298 K on a 600 MHz spectrometer equipped with a cryoprobe. b, chemical shift changes of MAK33 V L caused by the I2E substitution. 1 H and 15 N backbone chemical shifts were determined at a protein concentration of 50 M at 298 K in 20 mM phosphate, 50 mM NaCl (pH 6.5). c, effects of I2E substitution in MAK33 V L . Strongly affected residues are marked on the MAK33 VL-WT crystal structure (PDB code 1FH5). Ile-2 is shown in red, and residues with chemical shift changes Ͼ0.05 ppm are shown in dark purple, and unassigned residues are shown in gray.

TABLE 3 Thermal and chemical stabilities of 1AQK V L and 1VGE V H variants
Stabilities against thermal and chemical (GdmCl) denaturation of the different 1AQK V L and 1VGE V H variants are shown. Midpoints of thermal transitions are shown as T melt . Because GdmCl-induced unfolding transitions acquired for 1AQK V L were reversible, the data were fitted to a two-state equilibrium unfolding model to obtain the thermodynamic stability of unfolding (⌬G U ), as well as the cooperativity parameter (m value), for a qualitative comparison of the data. When due to too less an amount of proteins, the chemical stability experiments could not be performed (NA). and Ser-67 to Leu-73. The first region contains the structurally important residues Cys-23, which forms a conserved disulfide bond with Cys-88. The adjacent Arg-24 is involved in a conserved salt bridge with Asp-70 and the conserved Trp-35 lies in the core of Ig domains (50). Furthermore, the I2E substitution causes structural changes also at distant residues like Tyr-86 and Phe-87 (Fig. 8c). In consequence, a rearrangement of the hydrophobic core involving residues Ala-25, Ile-29, Trp-35, Leu-73, and Tyr-86, which exhibit major chemical shifts changes, is induced. The CDRs also show shift perturbations, especially the loop including residues Arg-24 to Ile-29, which is close to the N terminus.

Variable domain variant
To gain further insight in the dynamics of the V L domains, MD simulations in explicit solvent were performed on the MAK33 V L -WT and 1OPG V L -WT domains and on the MAK33 V L -I2E and 1OPG V L -E2I mutants. The substitutions were generated in silico based on the native crystal structures followed by extensive energy minimization before starting MD simulations (see "Experimental Procedures" for details). For all four cases, the r.m.s.d. of the backbone stayed close to the starting structure within Ͻ2.0 Å (Fig. 9). In the variants with an Ile at position 2 (MAK33 V L -WT and 1OPG V L -E2I), the Ile residue remained close to the starting conformation with the side chain located in a hydrophobic cavity (Fig. 10, a, b, and d).
During the entire simulation, the r.m.s.d. of the heavy atoms of Ile-2 stayed at a level similar or only slightly larger than that of the average backbone r.m.s.d. of the complete domains (black curves in Fig. 10, a and b). Interestingly, one can distinguish two rotameric sub-states of the Ile side chain during the simulation with slightly different r.m.s.d. with respect to the starting structure. Contrary to this, in the two variants with a Glu at position 2 (MAK33 V L -I2E and 1OPG V L -WT), significant deviations and fluctuations of the Glu-2 side chains with respect to the starting structure were observed (red curves in Fig. 10, a and b). Deviations of up to 6 -8 Å from the initial placement were observed that correspond to a partial or full dissociation of the Glu-2 side chain from the cavity, thereby adopting a fully solvent-exposed conformation (compare Fig. 10, d and e). The frequent disruption of contacts between Glu-2 and other side chains indicates that Glu does not contribute significantly to the stability.
The analysis of root mean square fluctuations of each residue with respect to the mean structure (Fig. 9, c and d) showed similar fluctuations for wild type and corresponding mutant chains. However, significantly larger fluctuations were observed for variants with a Glu at position 2 compared with those with Ile-2. Slightly increased fluctuations were also seen at loop segments that flank residue 2 (e.g. the loop around Ala-25), also identified in NMR experiments to have strong chemical shift perturbations. To quantify the relative stability of solvent-exposed and buried states of the side chain at residue 2, US simulations were performed to induce a dissociation of the side chain from the N-terminal cavity region. The distance between the C␣ atom of residue 2 and the C␣ atom of a residue at the floor of the N-terminal contact region served as reaction coordinate (see "Experimental Procedures" for details). The US simulations allowed for the calculation of a potential-of-mean force or associated free energy change for the dissociation of residue 2 (Fig. 10c). A distance of ϳ12 Å corresponded to a location of residue 2 in contact with the residues forming the N-terminal cavity region (smaller distances resulting in sterical repulsion). At distances of 16 -18 Å, the side chain of residue 2 had lost contacts with the N-terminal cavity region (larger distances resulting in disruption of additional contacts of subsequent residues along the chain). The US simulations indicate a significantly higher dissociation free energy barrier for the case of an Ile at position 2 compared with Glu-2 in both MAK33 and 1OPG V L domains. Because in the folded form the Ile-2 residue stays in an associated conformation, one can assign ϳ2.5 kcal/mol (ϳ10 kJ mol Ϫ1 ) as a free energy difference of dissociation compared with Glu-2 as a folding stabilization free energy contribution of the Ile-2 variants. This is in excellent agreement with our experimental values.
In addition to the V L domains, the V L and V H domains (taken from the x-ray structures of 1AQK (49) and 1VGE (39), respectively), were also analyzed. MD simulations were performed starting from the x-ray conformations of the wild type structures and from variants with a Glu at position 2. In all simulations, the domain conformations stayed close (within ϳ2 Å) to the start structure for a 50-ns sampling time (Fig. 11, a and  b). In the case of 1AQK V L , the Asn-2 wild type residue is partially solvent-exposed forming contacts with polar and charged residues near the N terminus (Fig. 10h). During the simulations, it transiently dissociates to more solvent-exposed states. Similarly, the Glu-2 variant (1AQK V L -N2E) also shows fluctuations to partially solvent-exposed states with limited contacts to the N-terminal region of the protein (Fig. 11c). In the case of 1VGE V H , residue 2 is in a fully solvent-exposed state FIGURE 10. Heavy atom r.m.s.d. of residue 2 versus simulation time after best superposition on the complete backbone of the start structure for 1OPG V L -WT (red, Glu-2) and 1OPG V L -E2I (black, Ile-2) (a) and MAK33 V L -WT (black, Ile-2) and MAK33 V L -I2E (red, Glu-2) (b). c, calculated potential-of-mean force for the dissociation of the Glu-2 in 1OPG V L -WT (black curve), Ile-2 in 1OPG V L -E2I (green), Ile-2 in MAK33 V L -WT (red curve), and Glu-2 in MAK33 V L -I2E (blue curve) from the hydrophobic cavity region. d, N-terminal cavity region of MAK33 V L with the Ile-2 side chain buried in the cavity (protein schematic with Ile-2 as sticks model and adjacent side chains as van der Waals spheres). e, example of a simulation snapshot with a fully solvent-exposed Glu-2 side chain and several water molecules at the rim of the N-terminal cavity region of 1OPG V L . f, solvent-accessible surface representation in yellow of the binding cavity region for residue 2 in the case of MAK33 V L -WT (stick model of Ile-2); g. same for 1VGE V H -WT with Val-2 as stick representation, and h, same for 1AQK V L -WT with Asn-2.
already in the x-ray starting structure (Fig. 10g). No cavity near the N terminus is present that could serve as a stable anchor region for residue 2 as found for the domains (Fig. 10f). Both the simulations of 1VGE V H -WT and the V2E mutant showed large fluctuations of residue 2 and no stable binding mode could be identified (Fig. 11d). Thus, in these cases neither the wild type nor the Glu-2 variants adopt a stable state with a buried side chain of residue 2 that could contribute to the stability of the protein, in contrast to what was observed for the Ile-2 variants of MAK33 and 1OPG V L . This is largely due to the absence of an appropriate cavity region near the N terminus and explains why mutation of residue 2 has little or no influence on protein stability in the case of the V L and V H domains.

DISCUSSION
We show here that residue 2 is crucial for the integrity and stability of V L domains of the family, and this residue is a key factor that controls its amyloidogenic properties. Specifically, uncharged residues at position 2 of V L domains are important for the maintenance of the structural integrity and a high stability. In agreement with their extremely low frequency in naturally occurring V L domains, charged residues do not support this.
During NMR backbone assignment, we encountered several regions that are exchanged-broadened and therefore could not be assigned completely. Interestingly, these regions match well with dynamic residues in the dimer interface of the human V L sequence LEN, as identified by Mukherjee et al. (51). In this V L domain (LEN), residues 1-9, 37-58, and 89 -100 were shown to exhibit pH-dependent millisecond dynamics, which seem to be related to the process of amyloid fibril formation. The presence of residue 2 within these dynamic regions is another indicator for its relevance for amyloidogenicity. This is supported by our MD simulations that show an increase in N-terminal dynamics caused by a charged residue at position 2. Overall, the I2E substitution affects large parts of the V L structure. Both the disruption of the conserved salt bridge and the changes in the hydrophobic core presumably contribute to the loss of thermodynamic stability. However, the short strand-connecting helix (residues 79 -83), which is highly conserved in antibody variable domains and known to be important for early folding events of the domain (13) was not affected by the I2E substitution, implying that this replacement may not be involved in the amyloidogenicity of the V L variants studied here.
Hydrophobic core packing is well known to be an important stabilizing factor in proteins in general (52)(53)(54) and antibody domains in particular (55). Therefore, the destabilizing effects of charged residues at position 2 could be explained by the fact that they are energetically unfavorable within the hydrophobic core. A charged residue at position 2 of a V L domain leads to a partial exposure of the hydrophobic interior and to the disruption of an optimal packing around residue 2. This in turn causes rearrangements of the whole hydrophobic core region. Consequently, this results in the destabilization of the fold and an increased aggregation propensity. For the V L and V H domains, the residue at position 2 is largely solvent-exposed or forms only transient polar contacts with the surrounding protein surface. In MD simulations, both the wild type residues at position 2 or substitutions with Glu resulted in fluctuating solvent-exposed conformational states due to the lack of a stable binding region. Hence, in the case of the V L and V H domains, residue 2 is not part of the hydrophobic core, and substitution of this residue also does not disrupt the hydrophobic packing of the FIGURE 11. a, comparison of backbone r.m.s.d. from the corresponding experimental start structure versus data gathering simulation time for 1AQK V L -WT (black curve) and 1AQK V L -N2E (red curve). b, same for 1VGE V H -WT (black curve) and 1VGE V H -V2E (red curve). Heavy atom r.m.s.d. of residue 2 versus simulation time after best superposition on the complete backbone of the start structure for 1AQK V L -WT (black, Asn-2) and 1AQK V L -N2E (red, Glu-2) (c) and 1VGE V H -WT (black, Val-2) and 1VGE V H -V2E (red, Glu-2) (d).
domains. Whereas the binding of the N-terminal segment to the hydrophobic cavity is crucial for stability in V L domains, it appears to be less important for V L and V H domains. Unlike in the V L domains where the hydrophobic cavity requires binding of the N-terminal residue 2 to prevent the unfavorable exposure of nonpolar residues, in the V L and V H domains, the respective hydrophobic cavities are less hydrophobic and do not require binding of residue 2 for protection.
It has been suggested that the formation of amyloid fibrils does not begin from the native state of a protein but more likely from a partially folded state or an intermediate state (32,56,57). In this context, the occurrence of particularly nonconservative mutations that involve a change in side chain chemistry in a structurally important region and the correlation with an increased propensity to form amyloid have been reported (32,58). Mutations and changes in conditions that lead to destabilization of the native state might therefore serve to increase the population of an intermediate state or partially folded species, thereby enabling more molecules to be diverted into an amyloid-forming pathway (59,60). In this study, the less stable MAK33 and 10PG V L variants readily formed fibrils, whereas their stable counterparts did not. Although stability is an important factor determining the amyloidogenic propensity of proteins, the amino acid sequence plays a major role (61) as this determines the amyloid fibril formation kinetics and the types of intermediates populated (13,32,56). Although some residues protect against amyloid formation, their chemistry and position within the protein appear to be more important. This protective role is largely attributed to edge strands of ␤-sheet proteins (62)(63)(64). Residue 2 in antibody variable domains appears in ␤-strand A, which is one of four edge strands (A, D and CЉ, G) of the ␤-sandwich topology. Uncharged residues at this position might have evolved to prevent undesirable ␤-sheet self-propagation (63). The frequent dissociation and solvent exposure of charged residues at position 2 of V L domains presumably lead to the disruption of native H-bonds and the subsequent exposure of H-bond donors and acceptors on the edge strand interface. This might result in V L edge-to-edge aggregation (65). The rare occurrence of charged residues at this position within edge strand A and the high amyloidogenic propensity of these V L variants are in line with this explanation. Although there exists a huge similarity between the three different families of variable domains, it has also been shown that they differ significantly in their biophysical properties and propensity to aggregate to amyloid fibrils (55,66). Our study identifies an important property of the V L family, not pertaining to the V L and V H domains, which controls its amyloidogenicity.
A compelling amount of evidence from studies on sequence and structural features that can predispose to amyloid formation has often been limited to given V L subfamilies (e.g. 1 or 6) with hardly one feature pertaining to all or an entire variable domain family (67)(68)(69). Our results identify a key feature that can destabilize and as a result predispose an entire variable domain family to fibrillar aggregation. The importance of residue 2 for the V L family determined here adds to the understanding of antibody LC amyloidosis linking the destabilization of native interactions with amyloid formation. In the context of AL amyloidosis, and also in the broader context of amyloid diseases, however, it remains to be seen whether point mutations promoting fibril formation generally affect stability. It seems reasonable to assume that other mechanisms may also apply.