Site-directed Mutagenesis Reveals Regions Implicated in the Stability and Fiber Formation of Human λ3r Light Chains*

Background: λ6a and λ3r are the most implicated germ lines in light chain amyloidosis. Results: Mutagenesis at N-terminal, loop 40–60, and CDR3 regions affected λ3r fibrillogenesis. Conclusion: Changes at residues 7, 48, and 91 increased fibril formation while changes at residues 8 and 40 reverted fibrillogenesis. Significance: Characterization of light chain germ lines helps to identify key regions implicated in amyloidosis. Light chain amyloidosis (AL) is a disease that affects vital organs by the fibrillar aggregation of monoclonal light chains. λ3r germ line is significantly implicated in this disease. In this work, we contrasted the thermodynamic stability and aggregation propensity of 3mJL2 (nonamyloidogenic) and 3rJL2 (amyloidogenic) λ3 germ lines. Because of an inherent limitation (extremely low expression), Cys at position 34 of the 3r germ line was replaced by Tyr reaching a good expression yield. A second substitution (W91A) was introduced in 3r to obtain a better template to incorporate additional mutations. Although the single mutant (C34Y) was not fibrillogenic, the second mutation located at CDR3 (W91A) induced fibrillogenesis. We propose, for the first time, that CDR3 (position 91) affects the stability and fiber formation of human λ3r light chains. Using the double mutant (3rJL2/YA) as template, other variants were constructed to evaluate the importance of those substitutions into the stability and aggregation propensity of λ3 light chains. A change in position 7 (P7D) boosted 3rJL2/YA fibrillogenic properties. Modification of position 48 (I48M) partially reverted 3rJL2/YA fibril aggregation. Finally, changes at positions 8 (P8S) or 40 (P40S) completely reverted fibril formation. These results confirm the influential roles of N-terminal region (positions 7 and 8) and the loop 40–60 (positions 40 and 48) on AL. X-ray crystallography revealed that the three-dimensional topology of the single and double λ3r mutants was not significantly altered. This mutagenic approach helped to identify key regions implicated in λ3 AL.

ing V H , allowing its free secretion (9,10). The loss of the Ig heterotetrameric structure may contribute to the amyloidogenicity of light chain variable domains because the fibrils of most AL patients predominantly comprise a single variable domain (11).
The comparison of amyloidogenic and nonamyloidogenic V L sequences has allowed other researchers to identify certain amino acid changes that destabilize the V L domain (12)(13)(14)(15)(16). The introduction of some of these mutations into nonamyloidogenic V L domains decreases their stability and increases their tendency to aggregate and form amyloid fibers in vivo (17). Several studies have suggested that a less stable V L domain is more likely to aggregate into amyloid fibers (17,18).
The light chains are responsible for the majority of AL cases, with a 3:1 ratio over the isotype (19,20). Several gene segments belonging to the 3 and 6 subgroups, particularly the 3r and 6a germ lines (6,21), are significantly associated with AL. Despite the predominance of the isotype in the disease, only one germ line has been characterized for its propensity to form amyloids. Our group previously reported the characterization of the recombinant germ line 6aJL2, a protein encoded by the 6a and the jl2 gene segments (22). 6aJL2 is thermodynamically more stable than other 6 light chains derived from patients with multiple myeloma, although it is capable of forming fibers in vitro after long periods of incubation (22).
The 3 light chain family comprises 21 genes, nine of which are polyclonal; only six of these nine genes have been associated with AL (23). Although several 3 amyloidogenic light chains isolated from patients have been analyzed (6, 24 -26), the role of germ line-encoded features (protein regions) in the amyloidogenic capability of the protein has not been determined.
Because the 3r subfamily might be intrinsically amyloidogenic, the aim of this work was to characterize the structural and biophysical properties of two germ lines of the 3 subgroup. We are interested in the 3r subgroup because of its high prevalence in AL, whereas the 3m germ line, which has not been associated with AL, was used as a control (6). Each 3m and 3r gene segment was joined to the jl2 segment to obtain the whole variable domains 3mJL2 and 3rJL2. Because of an inherent limitation (extremely low expression), Cys at position 34 of the 3r germ line was replaced by Tyr reaching a good expression yield. A second substitution at CDR3 (W91A) was introduced in 3r to obtain a better template to incorporate additional mutations. Taking the double mutant (3rJL2/YA) as template, other variants were constructed to evaluate the importance of those substitutions on the stability and aggregation propensity of 3 light chains. Mutations were introduced into two regions thought to protect against light chain fibril formation: the sheet switch (positions 7 and 8) (13) and the loop 40 -60 region (positions 40 and 48) (14). X-ray crystallography revealed that the three-dimensional topology of the single and double 3r mutants was not significantly different. The majority of 3 variants evaluated in this study showed higher stability compared with other germ lines associated with AL, such as 6a and IO18/O8 (22,27,28). This mutagenic approach helped to confirm or identify other key regions implicated in 3 AL.

MATERIALS AND METHODS
Cloning, Expression, Extraction, and Purification-Germ line sequences were obtained from the VBASE database. 3rJL2 and 3mJL2 contained the sequences of the corresponding human V 3r and 3m germ lines. Both proteins contained the jl2 segment. The jl2 segment was used because it is frequently present in clonal plasma cells (6) 3r and 3m were synthesized by recursive PCR as described by Prodromou and Pearl (29). The DNA was cloned into the pET22b vector (Novagen, Darmstadt, Germany). Site-specific mutations in 3rJL2 were generated using a mutagenic mega-primer (30). All constructs were verified by nucleotide sequencing.
Vectors containing the light chain recombinant proteins were transformed into the Escherichia coli strain BL21(DE3). The cells were grown in 2XYT medium containing 100 g/ml ampicillin at 37°C and 150 rpm. When the culture reached an A 600 of 1-1.4, protein expression was induced by the addition of isopropyl-␤-D-thiogalactopyranoside at a final concentration of 0.2 mM. The cultures were grown for 5 h under low agitation (115 rpm) at 22°C.
The periplasmic proteins were extracted as described previously (22). The periplasmic extracts were precipitated twice with 30 -90% saturated ammonium sulfate and stored at 4°C. The precipitated proteins were removed by centrifugation (6000 rpm for 30 min), and the pellet was resuspended and dialyzed overnight in 50 mM NaCl and 40 mM Tris, pH 8.2 (buffer A). The proteins were separated by size exclusion chromatography using a Sephacryl S-100 HR column (Amersham Biosciences) connected to an Ä kta FPLC system (GE Healthcare). The column was previously equilibrated with buffer A at a flow rate of 2 ml/min. The peak corresponding to the light chains was collected and precipitated with 70% saturated ammonium sulfate.
The precipitated protein was harvested by centrifugation (6,000 rpm for 30 min), and the pellet was resuspended in buffer A. The soluble protein was extensively dialyzed at 4°C in buffer A while gradually decreasing the NaCl concentration. The final dialysis was performed in 40 mM Tris, pH 8.2. The dialyzed protein was applied onto an anion exchange chromatography column (Bioscale mini macro prep High Q; Bio-Rad) that was prewashed with 40 mM Tris, pH 8.2. The protein was eluted with a linear gradient of 0 -30% 500 mM NaCl and 40 mM Tris, pH 8.2, over 60 min. The fractions containing the light chain proteins were pooled and stored at 4°C until needed. The purity of the protein was verified by SDS-PAGE. The protein concentration was determined by UV absorption using the extinction coefficient value calculated with the ProtParam Tool in the ExPASy Web site.
Analytical Size Exclusion Chromatography-The oligomeric state of the light chain was determined using a Superdex 75 (10/30) prepacked column (Amersham Biosciences). The column was previously equilibrated with buffer A. The molecular masses were calculated using linear interpolation from a calibration curve that included serum albumin (66.3 kDa), ovalbumin (43.5 kDa), carbonic anhydrase (28.8 kDa), myoglobin (17 kDa), and cytochrome c (11.7 kDa) as molecular mass markers.
Unfolding Experiments: Data Analysis-The changes in tryptophan fluorescence were analyzed after normalization of the transition curves to the apparent fraction of unfolded molecules, F app , where Y obs is the observed fluorescence intensity at a given temperature or guanidine HCl concentration, and Y f and Y u are the fluorescence signals for the native and unfolded forms, respectively. A linear dependence of Y obs was observed in both the native and unfolded baseline regions; therefore, linear extrapolations from these baselines were made to obtain estimates of Y f and Y u in the transition region.
Guanidine HCl Denaturation Experiments-The intrinsic fluorescence emission measurements were recorded on an LS50B spectrofluorimeter (PerkinElmer Life Sciences) using a 1-cm-path length quartz cuvette. Protein samples were prepared at concentrations of 20, 50, and 200 g/ml in 20 mM sodium phosphate buffer, pH 7.5. Different concentrations of guanidine HCl from an 8 M stock solution were used to unfold the protein; the samples were then incubated for 12 h at 25°C (equilibrium conditions). Samples incubated for 6 or 12 h showed similar results. The fluorescence emission was collected at 355 nm using an excitation wavelength of 295 nm. Thermodynamic parameters were calculated assuming the unfolding reaction followed a two-state model in which the native monomers (N) are in equilibrium with the unfolded monomers (U), where K app ϭ F app /(1 Ϫ F app ). A linear dependence of ⌬G°on the concentration of guanidine HCl (GndHCl) was assumed (30), where ⌬G°represents the Gibbs free energy at a given concentration of guanidine HCl, ⌬G H 2 O represents the Gibbs free energy in the absence of guanidine HCl, m is the gradient of the linear transition region and reflects the cooperativity of the process, and C m is the denaturant concentration at which ⌬G ϭ 0 and was calculated as C m ϭ Ϫ/m. Thermal Unfolding-The samples were solubilized at 50 g/ml in 3 ml of PBS solution and then placed into quartz cuvettes. A micro-stir bar was used at a low speed to maintain a uniform solution temperature. The fluorescence intensity was measured as described for the chemical denaturing experiments. The protein solution was heated using a water bath with a recirculation system connected to the cuvette holder of the spectrofluorimeter (L550B; PerkinElmer Life Sciences). The temperature of the solution was measured using a thermistor thermometer. The thermal unfolding data were analyzed as described by Pace et al. (31) assuming a two-state process and the van't Hoff equation, where T is the temperature, ⌬H is the enthalpy, and R is the gas constant. The value of ⌬G at room temperature was determined using the following equation, where T is 298 K, and C p is the change in heat capacity associated with unfolding, calculated theoretically according to the method described by Milardi et al. (32).
In Vitro Fibril Formation-Protein samples at a concentration of 100 g/ml were filtered through a 0.22-m pore-size Millex GV membrane filter (Millipore) and poured into a polystyrene cuvette in PBS. The cuvette was incubated at 37°C under agitation with a magnetic stir bar. Fiber formation was monitored at different time points by following the incorporation of thioflavin T (ThT) (Sigma Aldrich) (33). The fluorescence emission was recorded at 482 nm using an excitation wavelength of 450 nm. The fibril formation kinetics was analyzed by fitting the time-dependent changes in the ThT fluorescence intensity to the following equation, where F ThT is the ThT fluorescence intensity, A is the ThT fluorescence intensity in the post-transition plateau, t i is the midpoint of the transition region, B is the fibril growth rate constant, and t is time. An absolute value of the nucleation lag time, t lag , was calculated by extrapolating the linear region of the hyperbolic phase back to the abscissa (34).
Circular Dichroism Experiments-CD spectra and thermal denaturation experiments were recorded on a JASCO J-715 spectropolarimeter (JASCO Inc., Easton, MD) equipped with a water-cooled peltier. Far-UV CD spectra were recorded using a 0.2-cm-path length closed quartz cell at a concentration of 200 g/ml protein in 20 mM Na 2 HPO 4 , pH 7.5.
The raw data were converted to molar ellipticity using the formula, where C is molar concentration, and l is the cell path length in cm.
Transmission Electron Microscopy-A Formvar/carboncoated copper G200 grid (Electron Microscopy Sciences, Hatfield, PA) was floated onto a 20-l drop of protein sample for 2 min. The excess liquid was drained off with filter paper, and the specimen was negatively stained by floating onto a 40-l drop of 4% (w/v) uranyl acetate for 2 min and blotted dry. The specimens were analyzed at 80 kV on a Zeiss EM900 Transmission electron microscope. The images were recorded with a CCD DualVision 300W camera (Gatan, Pleasanton, CA) at a resolution of 1030 ϫ 1300 pixels. The image processing was performed using Adobe Photoshop version 7.0.
Crystallization-Prior to the crystallization trials, the 3mJL2, 3rJL2/Y, and 3rJL2/YA proteins were maintained in 40 mM Tris, pH 8.2, containing 20 mM NaCl. Crystal screens 1 and 2 from Hampton Research (Laguna Niguel, CA) were chosen as the starting point in the search for crystallization conditions. The 3mJL2, 3rJL2/Y, and 3rJL2/YA proteins were concentrated to 3.5, 4.5, and 3.5 mg/ml, respectively. Drops containing 1 l of Hampton solution and 1 l of the respective protein were set up using the micro-batch method under paraffin oil. Micro-crystals were obtained under several conditions, as follows: 3mJL2 crystals with the best shape appeared using condition 5 from Crystal Screen 1 (Hampton Research, Aliso Viejo, CA) containing 0.2 M sodium citrate, 0.1 M HEPES, pH 7.5, and 30% MPD, and crystals of 3rJL2/Y and 3rJL2/YA appeared using condition 32 from Crystal Screen 1 (Hampton Research) containing 2 M ammonium sulfate. Upon optimization, 0.30 ϫ 0.30 ϫ 0.30-mm crystals were generated with the hanging drop method at 291 K after 15 days for each of the three proteins under the following conditions: 3mJL2, 0.2 M sodium citrate, 0.1 M HEPES, pH 7.5, and 40% MPD; 3rJL2/Y, 2.3 M ammonium sulfate and 40% trehalose; and 3rJL2/YA, 1.6 M ammonium sulfate and 40% trehalose. Because all of these conditions are suitable for cryocooling, the crystals were mounted in rayon cryoloops and flash-cooled in a 100 K nitrogen stream.
X-ray Data Collection-The data collection was performed at National Synchrotron Light Source Beamline X6A at two x-ray wavelengths: 0.9720 Å (for both 3r mutant crystals) and 0.9330 Å (for 3mJL2), using an Oxford Cryosystems 700 series Cryostream and an ADSC Q270 CCD detector (ADSC, Poway, CA).
Data Processing and Model Refinement-Indexing and integration were performed using MOSFLM for 3rJL2/Y (35) and XDS (36) for 3mJL2 and 3rJL2/YA. The integrated reflections were sorted, scaled, and truncated with SORTMTZ, SCALA, and TRUNCATE (37) from the CCP4 suite, respectively. Molecular replacement was carried out in PHASER (38) using the edited Protein Data Bank code 1LIL atomic coordinates, corresponding to the 3 immunoglobulin Cle, as the starting model (24). The resulting model for each protein was subjected to rigid body refinement followed by restrained refinement in REFMAC5 (39). Once convergence was reached, the refinement was continued in PHENIX 1.5 (40). The refinement was alternated with manual building in COOT (41). The refinement cycles ended when R work and R free values were lower than 0.18 and 0.22, respectively. Validation of the final models was performed using PROCHECK (42). The coordinates and structural factors of the final models were deposited in the Protein Data Bank (43) with the following codes: 4AIZ (3mJL2), 4AIX (3r JL2/ Y), and 4AJ0 (3rJL2/YA). The crystallographic data and refinement statistics for the three structures are shown in Table 1.

RESULTS
3mJL2 and 3rJL2/YA as Nonamyloidogenic and Amyloidogenic 3 Variable Domains-3mJL2 was expressed and purified efficiently. In contrast, 3rJL2 was scarcely expressed. 3 germ line sequences were analyzed to identify single substitutions that would allow improve protein expression (Fig. 1). We found that 3r is the only germ line with a cysteine residue at position 34 (according to Kabat numbering), a solvent-exposed position at the N terminus of strand C. Although both amyloidogenic and nonamyloidogenic 3r-derived sequences bear a Cys in this position, other germ lines (such as 3m) contain Tyr or another polar residue (Fig. 1). These data suggested that Cys-34 should not substantially modify the fibrillogenic properties of the proteins derived from the 3r sub-family. Because the presence of a solvent-exposed cysteine residue may affect 3rJL2 expression, we constructed 3rJL2/Y, a mutant in which Cys-34 was mutated to tyrosine (3rJL2/C34Y), the equivalent residue found in the 3m germ line (Fig. 1). The expression, purification, and yield of 3rJL2/Y were comparable with those of 3mJL2. We assessed the thermodynamic stability of 3mJL2 and 3rJL2/Y by thermal and chemical unfolding experiments. Based on the tryptophan fluorescence measurements, monophasic and reversible transitions were observed; therefore, as a first approximation, we analyzed the data assuming a two-state process (Fig. 2). The unfolding kinetics of 3rJL2/Y and 3mJL2 were very similar, as deduced from the comparison of the midpoints of chemical or thermal unfolding (C m and T m , respectively; Table 2) and the values for the cooperativity of the unfolding process (m and ⌬H VH , respectively; Table 2). The ⌬G values showed that 3rJL2/Y and 3mJL2 were more stable than 6aJL2 and IO18/08, the only amyloid germ lines previously evaluated (22,27,28). We found that 3rJL2/Y had refolding transitions that were not superimposable with the unfolding transitions (Fig. 2B). These results may reflect the presence of an irreversible refolding process and/or photo-physical damage of a Trp residue. Fluorescence spectra of samples kept in the dark or illuminated during the thermal melt were very similar, indicating that photo-oxidation is not responsible for these differences. Size exclusion chroma-tography and electron microscopy analyses were performed to know whether some type of aggregation would explain irreversibility. No detectable aggregates were found using both alternatives (data not shown). The far UV CD spectra of 3rJL2/Y before The amino acid residues are colored by relative conservation using the ALIGN program (54). The dots indicate gaps inserted to maximize the alignment. The top group shows the 3germ line sequences, including the 6aJL2 germ line sequence. Green, slate, and purple triangles above the alignment mark the sites that were mutated in 3rJL2, 3rJL2/Y, and 3rJL2/YA, respectively. Black rectangles above the alignment mark the potential aggregation regions predicted by TANGO, PASTA, and Aggrescan servers. Framework (FR), complementarity determining regions (CDR), and joining (JL2) segments are indicated below the alignment with pink, indigo, and gray stripes, respectively. The secondary structure elements of 3rJL2/Y are marked below the alignment. The middle and lower groups comprise the 3 sequences derived from patients with AL amyloidosis and healthy individuals, respectively. The germ line sequences were obtained from the VBASE database, and the 3 sequences derived from patients and healthy individuals were obtained from the National Center for Biotechnology Information server.
and after the thermal melt are not the same (Fig. 3), suggesting that some sort of irreversibility not linked to aggregation is present in this protein.
All of the recombinant chain proteins previously analyzed have shown reversible temperature-induced transitions. In addition to the buried Trp-35 in the hydrophobic core of all light chains, 3rJL2 contains a second tryptophan residue at position 91, which is also present in most 3 light chain sequences (Fig. 1). To simplify the interpretation of unfolding transition reversibility followed by fluorescence, Trp-91 was  Table 2.

TABLE 2 Thermodynamic and fibrillogenesis parameters of 3 proteins
The error shown is the standard deviations from three independent experiments. NA, not applicable, i.e. fibril formation was not observed after 70 h.  changed to alanine (the equivalent residue present in the sequence of 3m). This double mutant named 3rJL2/YA (3rJL2/ C34Y/W91A) showed superimposable unfolding and refolding transitions. Far UV CD results indicate that the double mutant recovered its native structure more significantly as compared with the single mutant. The better recovering of native properties after thermal unfolding encouraged us to use this double mutant as a template for the evaluation of the fibril formation propensity of additional mutants generated by site-directed mutagenesis.
It should be noted, however, that the thermodynamic analysis of the double mutant showed that W91A change in 3rJL2/YA caused a significant destabilization, as demonstrated by the ⌬G decrement compared with that of 3rJL2/Y ( Table 2). The C m and T m values decreased by 0.4 M and 7°C, respectively ( Table 2). The unfolding of 6aJL2 was evaluated to assess the thermodynamic stability difference between the 3 and 6 families. The results for 6aJL2 were similar to those previously reported (22). Far UV CD data showed that 6aJL2 germ line is completely reversible (Fig. 3). 3mJL2 and 3rJL2/Y had higher T m values than 6aJL2, indicating that the 3 variable domains are more stable ( Table 2).
Structural Characterization-The asymmetric units of the 3mJL2, 3rJL2/Y, and 3rJL2/YA crystal structures each contain four monomers, and each monomer comprises residues 2-107 of the light chain variable domain. The crystallographic data and refinement statistics are presented in Table 1. The superposition of monomers from each of the structures showed a backbone root mean square deviation between 0.4 and 0.9 Å, indicating that the global topology of the three models is similar (Fig. 4). However, there are subtle backbone deviations between the 3mJL2, 3rJL2/Y, and 3rJL2/YA structures at CDR1, CDR3, the region around residue Pro-40, and the loop connecting strands CЉ and D (Fig. 4C).
The conformation of the N-terminal region in 3mJL2, 3rJL2/Y, and 3rJL2/YA was similar to that of 6aJL2 (13), including the proline ␤-bulge between the two segments of strand A. However, strand A in the 3 structures was shorter and only comprised the segment after the ␤-bulge (Fig. 4B). The same hydrogen bonds are formed between the carbonyls of Pro-7 and Pro-8 in strand A and the carbonyls of Thr-103 and Lys-104 in strand G in the 3 and 6aJL2 structures (Fig. 4B). 3mJL2 has an insertion (Y95b) at CDR3 compared with the other 3 germ lines (Fig. 1). The OH of residue Y95b formed a hydrogen bond with the OE2 of residue E3, tightening the N terminus to the core of the domain (Fig. 4B and Table 3).
We compared the 3 structures and 6aJL2 and found deviations of ϳ2 Å in the C␣ position in the loop connecting ␤-strands C and CЈ (K39 to Q42) (Fig. 4A). This difference has been previously reported in AL proteins and may reflect the high mobility of this region (16).
In the 3rJL2/Y mutant, Tyr-34 (at ␤-strand C) forms part of a hydrophobic cluster with residues Tyr-32, Tyr-49, Leu-46, and Trp-91 (Fig. 4D). Trp-91 forms an additional hydrophobic interaction with Val-96 (Fig. 4D). The 3rJL2/YA mutant maintains the same hydrophobic interactions, with the exception of position 91. The change from cysteine to tyrosine at position 34 increases this hydrophobic patch in 3r structures. Because the backbone conformations of 3rJL2/Y and 3rJL2/YA are similar, we hypothesize that the W91A mutation (␤-strand G) loosens the contacts at ␤-strands C, F, and G, explaining the slightly higher stability of 3rJL2/Y (Table 2) and the fibril formation capacity of 3rJL2/YA (see below and Table 3). Our structural analysis suggests that each germ line has a particular set of contacts that stabilizes its structure.
Many variable domains and complete light chains form dimers at physiological conditions (44,45). Because the dimers do not seem to be involved in fiber elongation, they could modulate the aggregation of pathological light chain monomers (45). Therefore, variable domain dimerization could be a security lock to prevent fiber formation. We analyzed the V L -V L interactions in the crystal lattices of the 3 variants. In the asymmetric unit of 3mJL2, each monomer interacted with another monomer to form the equivalent of a Bence-Jones dimer (Fig. 4E), whereas the 3rJL2/Y and 3rJL2/YA monomers formed crystallographic dimers in which the second monomer was rotated 180°with respect to the corresponding 3mJL2 monomer (Fig. 4F). The 3r monomers would only form the equivalent of a Bence-Jones dimer with their symmetric partners. At the center of the 3mJL2 B strand, the Arg-20 side chain pointed toward the solvent, preventing the formation of the alternate interface found in the 3rJL2/Y and 3rJL2/YA mutants, whereas the Ser-20 side chain is buried in the crystal dimer interface (Fig. 4F). It was previously proposed that the presence of a charged amino acid at the center of an edge strand may protect against the aggregation of proteins that contain a ␤-sandwich structure (8). The majority of 3 germ lines and 3 variants isolated from patients presented an arginine residue at position 20 (Fig. 1). The germ line 3r contains a Ser at this position, and this is an important difference between 3m and 3r. Presumably, Arg-20 may be one of the security locks that prevent fibril formation in the 3m germ line.
PISA (Protein Interfaces, Surfaces, and Assemblies) (46) analysis indicated that the 3mJL2 dimer interface contains 1,820 Å 2 of buried accessible interface area. This value is significantly higher than the canonical crystallographic dimer interfaces formed by 3rJL2/Y, 3rJL2/YA, and 6aJL2 (1,546, 1,361, and 1,227 Å 2 , respectively) and other dimer interfaces in antibody fragments (V H -V L ) and light chains (V L -V L ) ( Table 4). The size of the canonical interface of the 3mJL2 dimer would block conformations prone to aggregation.
To rule out those possibilities, we assessed the oligomerization state of the variable domains of 3mJL2, 3rJL2/Y, 3rJL2/YA, and 6aJL2, previously analyzed by SDS-PAGE (Fig. 5). Different protein concentrations (20, 50, and 200 g/ml) were examined through analytical size exclusion chromatography (Fig. 5). The purified variable domains eluted as a single peak with an elution volume corresponding to the molecular mass of a monomer (Fig. 5). We did not observe any oligomeric species in the elution profile. We next performed guanidine HCl and temperature-induced unfolding experiments using the same protein concentrations (20, 50, and 200 g/ml) (Fig. 6). These results, in addition to the hydrodynamic and thermodynamic data (Table  5), allowed the conclusion that the variable domains were mainly monomeric at the protein concentrations evaluated.

Site-directed Mutagenesis of the Protective Edges Identifies a Putative Fibrillogenic Region in 3r-derived Proteins-Proteins
rich in ␤-structures have structural features that prevent their aggregation under harsh conditions or as a consequence of mutations. Light chains have two such anti-aggregation motifs that have been proposed to prevent intermolecular associations of edge strands that could lead to aggregation and fibril formation (8,47). The N-terminal region of 3 (residues 1-14) contains a "sheet switch" motif, a structural feature in light chains that is proposed to act as an anti-aggregation domain for ␤-strands B and G. Prolines at positions 7 and 8 are highly conserved in light chains (Fig. 1); these residues have been proposed to stabilize the sheet switch (13) (Fig. 4B). The functions of these residues were assessed using the 3rJL2/YA mutant. Pro-7 was mutated to aspartic acid, which is normally present in the 3l germ line. We named the resulting protein (3rJL2/ C34Y/W91A/P7D) 3rJL2/YA/P7D. We also mutated Pro-8 to serine, the residue present at the equivalent position in the 2-19 germ line. The resulting protein was named 3rJL2/YA/P8S (3rJL2/C34Y/W91A/P8S). The thermodynamic data for these mutants are shown in Table 2 and Fig. 2. The thermodynamic stability of 3rJL2/YA/P7D was significantly affected, whereas the stability of 3rJL2/YA/P8S was unaffected as compared with the double mutant (Table 2 and Fig. 2). Far UV CD data showed that mutation at position 7 exerted a subtle influence on the reversibility to the native structure. In the case of position 8, the effect on the reversibility of thermal unfolding was minimal (Fig. 3). Residues 40 -60 formed the other protective edge of light chains and shield strands C and CЉ (Fig. 1). Residue Pro-40 was of particular interest because it is highly conserved in both and isotypes. Furthermore, residue Pro-40 of the multiple myeloma LEN protein has been shown to confer stability onto several amyloid-related light chains (14). We evaluated the role of Pro-40 using the 3rJL2/YA/P40S mutant (3rJL2/C34Y/ W91A/P40S). We introduced this mutation based on the 3p sequence, the only 3 germ line that does not contain a proline residue at this position (Fig. 1). The thermodynamic parameters of 3rJL2/YA/P40S indicated that this change did not affect the stability of the mutant protein (Table 2 and Fig. 2). Far UV CD data showed that mutations at position 8 and 40 exerted a minimal influence on the reversibility to the native structure. (Fig. 3).
We examined protein aggregation because it promotes amyloid fibril formation. The TANGO, PASTA, and Aggrescan servers are useful bioinformatics tools to evaluate the propensity of residues within a ␤-sheet to be facing one another on neighboring strands (48 -50). Following these criteria, we identified four regions in 3 sequences (designated a-d) with different aggregation propensities (Fig. 7). Region b (residues 44 -50) obtained the highest scores. 3rJL2/YA was subjected to in silico amino acid scanning around this region to find key residues that could modify its aggregation profile. When we mutated the conserved residue Ile-48 to methionine or proline, the servers predicted a decrease in the aggregation profile of this region in 3rJL2/YA (Fig. 7). Several light chain sequences contain a methionine residue at this position (Fig. 1). These results motivated us to generate the corresponding mutants; however, we were only successful on expressing the 3rJL2/YA/I48M (3rJL2/ C34Y/W91A/I48M) mutant. Because Ile-48 is buried in the hydrophobic core of the protein (Fig. 4C), the mutation to proline presumably destabilized the mutant so severely that it was not expressed. The thermodynamic results of 3rJL2/YA/I48M indicated that this mutant showed a decrease in its thermodynamic stability ( Table 2 and Fig. 2). Far UV CD data showed that mutation at position 48 exerted a moderate influence on the reversibility to the native structure (Fig. 3).
According to their thermodynamic parameters, the 3rJL2/ YA/P8S and 3rJL2/YA/P40S mutants were considered stable. In contrast, 3rJL2/YA/I48M and 3rJL2/YA/P7D were the least stable. In fact, the C m and T m values of 3rJL2/YA/P7D and 3rJL2/YA/I48M are characteristic of relatively unstable proteins (22,51). Based on the C m and T m values, the stability of the

TABLE 4 Interface areas of different antibody formats
To calculate the interface area of the dimer, only the variable domain was taken into account. In the Fab structures, the constant domains were not included in the surface calculations. The interface areas were calculated using PISA analysis (46). 3r mutants can be ordered as follows: 3rJL2/Y Ͼ 3rJL2/YA Ͼ 3rJL2/YA/P40S ϭ 3rJL2/YA/P8S Ͼ 3rJL2/YA/I48M Ͼ 3rJL2/ YA/P7D. The mutants had different transition slopes (m and ⌬H values) and different chemical and thermal unfolding curves ( Table 2 and Fig. 2, C and D). In other words, although the global stability was not significantly altered, the cooperativity of the unfolding process was modified in each mutant. The thermodynamic parameters as a whole showed that the more unstable mutants were 3rJL2/YA/P7D and 3rL2/YA/I48M. 3rJL2/YA/P7D and 6aJL2 Have Similar Kinetics during in Vitro Fibrillogenesis-One of the characteristics of amyloidogenic proteins is their capacity to form fibers in vitro (18,52).   Table 5.

Protein Interaction Format Area Protein Data Bank code
3mJL2 and the proteins derived from the 3r germ line were continually stirred to induce fibril formation. The fibril formation kinetics is illustrated in Fig. 8A. 3mJL2 and 3rJL2/Y samples did not show any change in ThT fluorescence after 70 h (Fig. 8A). These protein samples only became slightly turbid as the experiment progressed. The transmission electron micrographs showed the presence of amorphous deposits but no amyloid fibrils (data not shown). This is in agreement with the thermodynamic data, which indicate that both 3mJL2 and 3rJL2/Y are stable variable domains. Although the 3rJL2/YA and 3rJL2/YA/I48M mutants are borderline stable, they formed fibrils, as indicated by the significant increase in ThT fluorescence after 10 h of stirring (Fig. 8A). The 3rJL2/YA and 3rJL2/ YA/I48M mutant fibril formation kinetics showed slightly different nucleation times (t lag ) and growth rates. 3rJL2/YA displayed a t lag of 11.25 h and required 40 h for fiber extension. 3rJL2/YA/I48M showed a t lag of 11.74 h and required a longer time for fiber extension (56 h). 3rJL2/YA and 3rJL2/ YA/I48M showed similar growth rates, with a tendency of 3rJL2/YA/I48M to form fewer fibers with slower kinetics ( Table 2).
3rJL2/YA/P8S and 3rJL2/YA/P40S did not form fibrils despite being marginally less stable than their parental protein (Fig. 8A). The transmission electron microscopy of 3rJL2/YA samples after fiber formation confirmed the presence of thin, unbranched, individual fibrils, and amorphous aggregates (Fig.  8B). 3rJL2/YA/P7D fibers appeared to be long and slightly curved, which is an unusual morphological feature. Smooth  bundles containing two to four well defined fibrils were easily observed in this preparation (Fig. 8B). This mutant was the least stable with a t lag of 9.2 h and a fibril formation plateau at ϳ20 h ( Table 2 and Fig. 8A). In the 3rJL2/YA/I48M mutant, there were fewer fibrils compared with the amorphous deposits, but the fibrils were organized as wide, rod-like bundles containing several linear fibrils (Fig. 8B). 6aJL2 also formed fibrils under the same conditions (Fig. 8B). According to the kinetics of fibrillogenesis, 3rJL2/YA/P7D and 6aJL2 have similar extension times of ϳ10 and 8 h, respectively, but the nucleation time of 6aJL2 is faster than that of 3rJL2/YA/P7D (5.17 and 9.24 h, respectively) ( Fig. 8A and Table 2). In addition, 6aJL2 had a faster growth rate ( Table 2).

DISCUSSION
To the best of our knowledge, this work provides the first biophysical characterization of a 3 germ line and 3 germ line-derived mutant proteins. Our initial hypothesis postulated that the high association of the 3r germ line with AL is due to the intrinsic thermodynamic instability of the corresponding light chain. Thermodynamically, 3rJL2/Y and 3rJL2/YA should be as stable as 3mJL2, a model protein that corresponds to a nonamyloid 3 germ line (Table 2). We expected that 3rJL2/Y would form fibrils in vitro because 3r is among the germ lines most frequently associated with AL (6). However, we did not find any fibrils in the recombinant protein sample (data not shown). The relatively high stability of 3rJL2/Y may be attributed to the new interactions formed by residue Tyr-34 (Table 2 and Fig. 4D), which may have also increased the expression level of the protein. However, these results did not directly answer whether the 3r germ line is intrinsically amyloidogenic.
Although 3mJL2, 3rJL2/Y, and 3rJL2/YA share similar thermodynamic properties, 3rJL2/YA was the only mutant that formed fibrils in vitro. Structural comparisons indicate that the W91A mutation in 3rJL2/YA decreased its hydrophobic interactions compared with the other 3 structures (Table 3). The results indicate that mutations at positions Cys-34 and Trp-91 had opposite effects in the 3rJL2 derivatives. The mutation in position 34 may have reinforced the interactions among the ␤-strands C, F, and G, resulting in a more stable domain. The second mutation, in position 91, may have diminished these interactions rendering a less stable variant and more prone to fibril aggregation (Fig. 4D).
The analysis of the triple mutants revealed the critical role of Pro-7 and Ile-48 on the stability and fibril formation of the 3r domain. Our results suggest a relationship between cooperativity and fiber formation, with m and ⌬H being key parameters that inhibit fibrillogenesis. A previous report stated that the sheet switch at the N terminus of the light chains may impede and edge strand-mediated aggregation (8). The 3rJL2/YA/P7D mutant was less thermodynamically stable than 3rJL2/YA. The decrease in the stability of mutant 3rJL2/YA/P7D is comparable with the decrease in that of mutant 6aJL2/P7S (13). The P8S mutation, despite causing a much lower effect on the stability of 3rJL2/YA/P8S, abolished fibril formation (Table 2). These results suggest that Pro-8 is important for fibril formation. The analysis of the 6aJL2-derived mutants P7S and H8P and the double mutant P7S/H8P revealed that only the mutation in position 7 decreased the stability of the protein (13). Our results confirm that position 7 contributes to the protective role of the N terminus of the germ lines 3r and 6a. The protective effect of the sheet switch against fiber formation may depend on the combinatorial amino acid sequence of residues 7 and 8. Furthermore, the similar features observed in the mutants studied in the context of the 6a and 3r scaffolds suggest that the protective role of the ␤-bulge of ␤-strand A (sheet switch) against fibril formation may be conserved in the family.
The substitution of residue Pro-40, located in the loop that connects ␤-strands C and CЈ, has been associated with amyloidosis in several light chains, including BRE, ARN, WR, and MCG. In these proteins, the Pro-40 mutation enhanced their fibril formation capacity (53). The 3rJL2/YA/P40S mutant did not exhibit a significant change in stability compared with the double mutant. A plausible explanation can be that residue Pro-40 only establishes stabilizing interactions through the peptide backbone (Table 3). Unexpectedly, this mutant did not form fibrils like the 3rJL2/YA/P8S mutant.
As previously mentioned, the Ile-48 residue is conserved in families 3 and 6. The decreased thermodynamic stability of the 3rJL2/YA/I48M mutant is likely related to the presence of the bulky methionine side chain, which may decrease the rate of the fiber formation process without completely inhibiting it. It is likely that the methionine residue induced a structural rearrangement between strands C and CЈ, decreasing the self-complementation between the ␤-sheets and restricting the correct coupling of the monomers during fiber extension. In summary, position 48 seems to be important for maintaining the stability of the light chain and, when disrupted, favors fiber formation.
Our analysis of the 3rJL2/YA/P40S and 3rJL2/YA/I48M mutants indicates that loop 40 -60, which connects the two ␤-sheets, contributes to fibril formation when the protein is destabilized. The protection of loop 40 -60 is associated with the canonical V H -V L interaction, which limits its exposure (Fig. 4E).
Importantly, although region 40 -60 in germ lines 3m and 3r presented a high ␤-aggregation profile score, the same region in germ line 6a had a low score (Fig. 7). 6a differs from germ lines 3r and 3m in only three residues at this region (Fig. 1). Consistent with these observations, minimal changes in the loop 40 -60 are likely to modify its aggregation profile.
To the best of our knowledge, light chain dimerization has been reported only in proteins isolated from patients, such as the Bence-Jones proteins. Baden et al. (28) reported that the monomeric form of germ line IO18/O8 is more stable than its dimeric form, and the opposite occurs for its amyloidogenic counterpart AL-09, i.e. the dimer of this protein is more stable than the monomer (28). The calculated interface areas of the native dimers in the 3 crystallographic structures are very similar to the V H -V L and V L -V L interfaces of several other antibodies (Table 4). These observations led us to assess the presence of dimers in the 3mJL2, 3rJL2/C34Y, 3rJL2/YA, and 6aJL2 proteins through various methods. Our results indicated that the variable domains are mainly monomeric at the protein concen-trations tested. It is possible that the 3 germ line proteins exist mainly as monomers, similarly to IO18/O8. This is in contrast to the variable domains isolated from patients with AL, in which both monomers and dimers are present (28). Those variable domains contain several mutations at different regions of the interface, favoring the V L -V L interaction. Additionally, the complete light chain or the variable domain plus a fragment of the constant region may also promote V L -V L interactions leading to dimerization.
In summary, our data indicate that the 3m germ line is relatively stable and did not form fibrils in vitro. This is in agreement with the very low percentage (2%) of 3m-derived proteins in reported AL cases (6). Before a light chain acquires its fibrillogenic potential, it has to accumulate destabilizing mutations to generate new molecular interactions favoring fibril aggregation under micro-environmental conditions, such as pH, temperature, and high local urea concentrations. In the mutants derived from 3r germ line, we identified residues that affect the stability of the protein, influence fiber formation, and alter their anti-aggregation properties, similarly to previously reported mutants (13,28,51).
The 6a and 3r germ lines show low polyclonal expression in B lymphocytes (2 and 8%, respectively). The percentage of AL cases in which 6a is the germ line donor gene precursor can reach up to 40%. Although 3r is expressed at a 4-fold higher level than 6a, the percentage of cases with 3r as the donor gene precursor hardly reaches 20% (6). The differences in polyclonal expression levels, stability, and fibril nucleation times and the differential gain or loss of particular interactions between 6 and 3 light chains mutants may explain the different "aggressiveness" of these two light chain families, of which 6a is the most pathogenic.
The characterization of other germ lines will lead to a more complete understanding of how changes in the original sequence of a light chain domain can modify its protective effect against aggregation. Determining how a germ line domain gives rise to amyloidogenic variants will allow us to understand why certain light chain subfamilies are highly implicated in AL and to develop new therapeutic strategies.