Identification of two principal amyloid-driving segments in variable domains of Ig light chains in systemic light-chain amyloidosis

Systemic light-chain amyloidosis (AL) is a human disease caused by overexpression of monoclonal immunoglobulin light chains that form pathogenic amyloid fibrils. These amyloid fibrils deposit in tissues and cause organ failure. Proteins form amyloid fibrils when they partly or fully unfold and expose segments capable of stacking into β-sheets that pair and thereby form a tight, dehydrated interface. These structures, termed steric zippers, constitute the spines of amyloid fibrils. Here, using a combination of computational (with ZipperDB and Boston University ALBase), mutational, biochemical, and protein structural analyses, we identified segments within the variable domains of Ig light chains that drive the assembly of amyloid fibrils in AL. We demonstrate that there are at least two such segments and that each one can drive amyloid fibril assembly independently of the other. Our analysis revealed that peptides derived from these segments form steric zippers featuring a typical dry interface with high-surface complementarity and occupy the same spatial location of the Greek-key immunoglobulin fold in both λ and κ variable domains. Of note, some predicted steric-zipper segments did not form amyloid fibrils or assembled into fibrils only when removed from the whole protein. We conclude that steric-zipper propensity must be experimentally validated and that the two segments identified here may represent therapeutic targets. In addition to elucidating the molecular pathogenesis of AL, these findings also provide an experimental approach for identifying segments that drive fibril formation in other amyloid diseases.

ments predicted to form steric zippers are often buried in the interior and protected from forming fibrils. However, upon full or partial unfolding of the protein, these segments become exposed to solvent and can stack to form the amyloid fibril spine (22). Proteins known to assemble into amyloid fibrils may contain multiple segments capable of forming steric zippers. In the absence of experiments, it is unclear whether every, some, or only one of these segments is responsible for driving fibril formation. If multiple such segments can drive fibril formation independently, the protein could form amyloid fibrils having different spines and thus different structures (Fig. 1). The different fibril structures formed by the same protein are known as amyloid polymorphs (23)(24)(25)(26).
Here, we seek the molecular basis of the formation of amyloid fibrils from Ig light chains of both and families. The Human Gene Nomenclature Committee lists 33 functional genes for and 38 for variable domains (27)(28)(29). The repertoire of VLs is greatly expanded by the linkage of and domains to a variety of J-segments and constant domains. Consequently the Amyloid Light Chain Database lists 616 and 192 VLs (17). Sequence alignments of these VLs involved in AL do not reveal a single residue or segment that exclusively accounts for their amyloidogenic property, and differences in amino acids between VLs within the same family, or , affect the propensity to form amyloid fibrils (30 -32). Hence, the identification of the amyloidogenic segments of VLs is challenging.
We define an amyloidogenic segment by two criteria. 1) Introduction of amyloid-inhibiting residues, such as proline, into the segment by site-directed mutagenesis halts fibril formation of the parent VL. 2) The segment in isolation from the remainder of the VL forms a steric zipper.
Our search for amyloidogenic segments focuses on a human genomic reference VL sequence and two patient-derived VL sequences, one and one . We found two amyloidogenic segments in VLs and verified that in combination they fulfill criterion 1 by site-directed mutagenesis. Each of the two amyloidogenic segments independently causes amyloid fibril formation. Study of the isolated peptides shows that each segment also fulfills criterion 2 because each can form a steric zipper. By sequence comparison, we find that homologs of these segments are amyloidogenic in both and families of LCs and are the key propagators of fibril formation, thereby generalizing our model of amyloid formation to other LCs and suggesting these two LC segments as therapeutic targets.

Finding amyloid-driving segments
To identify the amyloidogenic steric zippers within VLs, we employed two complementary approaches: a computational assessment using ZipperDB to identify candidate segments with a high propensity to form steric zippers, and experimental mutagenesis to inhibit the ability of candidates to form amyloid fibrils, as detected by ThT assays (33). We identified a genomic variant of a VL2-8 with J1-connecting segment (VL2-8 -J1) as a useful reference model for our experiments; upon exposure to destabilizing conditions, VL2-8 -J1 forms amyloid fibrils (Fig. 2).
Within VL2-8 -J1, ZipperDB (https://services.mbi.ucla.edu/ zipperdb/) 4 (58) identified five high-propensity steric zipperforming regions, with each region containing conserved residues (Fig. 3, regions A-E). ZipperDB ranks the relative stabilities of the putative zipper structures; however, the potential of a segment to drive amyloid formation depends on additional factors such as environmental conditions and the shielding of the segment by neighboring parts of the sequence. Hence, we performed site-directed mutagenesis to experimentally determine the degree of involvement of these segments in formation Amyloid fibrils form through adhesive segments that stack to form steric zippers. Ig VLs contain two segments capable of forming steric-zipper fibril spines, shown here schematically in blue and yellow. Each one of the steric zipper-forming segments can induce assembly of a particular amyloid polymorph with its distinctive spine.

Amyloid-driving segments of VL domains in AL amyloidosis
of amyloid fibrils. We replaced conserved amino acids with proline residues. Prolines introduce geometric strain into ␤-sheets and eliminate a main-chain hydrogen bond, thus destabilizing the amyloid spine and impeding formation of amyloid fibrils (34). We sequentially replaced every amino acid with proline within each of the five predicted steric zipper-rich regions A-E. A single mutation, L75P in segment C, stopped formation of amyloid fibrils by the reference sequence, VL2-8 -J1, whereas prolines in other positions did not.
Upon identification of the amyloid-inhibiting proline mutation VL2-8 -J1-L75P, we examined whether residue Leu-75 is required for formation of amyloid fibrils by other pathologic VLs: a -type VL (Mcg) and a -type VL (AL09) (35,36). Both Mcg and AL09 are specific VL variants isolated from patients afflicted with AL. The same site-directed mutation, L75P, did not stop amyloid fibril formation by Mcg (Fig. 2). The opposite effects of the L75P mutation in two different VL types provide an important insight: a single steric zipper is not responsible for formation of amyloid fibrils by all VL types. This insight led us to hypothesize that VLs may form different polymorphs, and there may be more than a single segment that can independently induce formation of amyloid fibrils.

Finding a second amyloid-driving segment
The inability of the L75P to stop amyloid fibril formation by Mcg led us to perform a "proline-scan" experiment. With the aim of identifying segments required for amyloid fibril formation, we introduced proline mutations in four consecutive residues in each of 28 separate constructs that span the entire polypeptide sequence of Mcg, except that we did not replace either of the two structural cysteines with proline (Fig. 4). Table  1 summarizes the results of the proline-scan experiment: all 28 mutated constructs consistently formed amyloid fibrils, except for construct-25, which contains mutations in region E (residues F99P/V100P/F101P/G102P), in which several segments with a ZipperDB-predicted high propensity for amyloid formation are located. For this construct, several independent batches formed amyloid fibrils, whereas others did not. These data suggest that at least for Mcg, no single steric zipper individually accounts for the amyloid-forming property of VLs, and several different polymorphs may exist. When combined, L75P and F99P/V100P/F101P/G102P mutations abolished the ability of Mcg to form amyloid fibrils (Figs. 5 and 6). To summarize, every Mcg construct with tetra-proline mutations covering the entire VL sequence forms amyloid fibrils. However, the Mcg with L75P and F99P/V100P/F101P/G102P together does not form fibrils. These observations support our conjecture that no single steric zipper fully accounts for formation of amyloid fibrils by VLs; rather, at least two segments drive formation of steric zippers.
To verify whether these two segments are essential for formation of amyloid fibrils by another VL type, we introduced the corresponding mutations into a well-studied, patient-derived pathologic VL, AL09 (37). Sequence alignment of AL09 and Mcg identified the corresponding mutation sites: Mcg-L75P corresponds to AL09-F73P, and Mcg-F99P/V100P/F101P/ G102P corresponds to AL09-Y96P/T97P/F98P/G99P. As with Mcg, only the combined mutant, AL09-F73P/Y96P/T97P/ F98P/G99P, stopped amyloid fibril formation; the individual mutations did not affect the amyloidogenic potential of AL09 (Fig. 5). The reproducibility of results in both types of VLs, and , corroborates the involvement of two segments in formation of VL amyloid fibrils.

Atomic structures of amyloid-driving segments
Crystal structures of individual peptides derived from amyloidogenic segments, two from the VL Mcg (ASLTVS and NFVFGT) and one from the AL09 (YTFGQ), reveal that these peptides form steric zippers (Fig. 6). Each of these steric zippers features the typical dry interface with high-surface complementarity. The crystals of the fourth segment, EFTFTIS from AL09, were of insufficient resolution to resolve its atomic structure. However, cylindrical averaging of the single-crystal diffraction data shows the characteristic reflections of steric zippers (Fig. 7), as do the fibril diffraction patterns calculated from the atomic coordinates of the other three.
Our three crystal structures of VL amyloidogenic segments reveal steric zippers belonging to different symmetry classes (20). Symmetry classes define the spatial relationship between identical strands in the zipper. The steric zipper of segment

Amyloid-driving segments of VL domains in AL amyloidosis
YTFGQ from AL09 is composed of pairs of parallel sheets mated face-to-face, identifying it as class 1 (Fig. 6C). The same crystal structure also reveals a different pair of sheets mated back-to-back, corresponding to a second polymorph of symmetry class 1. ASLTVS from Mcg crystallizes as anti-parallel, antifacial ␤-sheets packed face-to-face (Fig. 6). This symmetry pat-tern is class 5. Furthermore, its ␤-sheets are out-of-register ( Fig.  8A) because its strands are not perpendicular to the fibril axis, but instead are inclined. The third steric zipper, NFVFGT from Mcg, displays a pattern of symmetry more complex than any previously observed steric zipper. The sheets that compose the 10 previously described steric-zipper symmetry classes can be Four-residue segments of the proline-scanning, site-directed mutagenesis experiment are shown in black bars below the energy histogram. Peptide-scanning was performed by mutating four consecutive residues in each segment to proline. The mutated residues are identified at the bottom of the histogram and represent individual constructs each containing a tetra-proline replacement. No single tetra-proline mutation inhibited amyloid fibril formation of Mcg, but constructs L75P and 25 together did. Red arrows identify amino acids that were identified as involved in formation of amyloid fibrils.

Table 1 Effects of site-directed mutations on amyloid fibril formation of three different VLs
The tripartite experiment comprises the following: 1) an amyloid-inhibiting mutation in a reference variant of a VL; 2) a variant from a patient; and 3) a variant from another patient. The main conclusion is that there are two amyloidogenic segments in VLs capable of driving formation of amyloid fibrils.

Detection of fibrils in electron micrographs
Ability of the site-directed mutation to stop amyloid fibrils

3) variable domain from an AL patient
Ϫ Ϫ Positive a ϩ/Ϫ indicates that some batches formed amyloid fibrils, and others did not. Three batches formed fibrils, and two did not. All other mutants were tested in at least three batch-independent repetitions.

Amyloid-driving segments of VL domains in AL amyloidosis
built from translational repeats of one strand along the fibril axis (classes 1-4) or two strands (classes 5-10). Sheets of NFVFGT contain four orientations of the strand, and so they require a minimum of four strands per translational repeat. The symmetry pattern of this sheet is one of six conceivable unique arrangements possible for a sheet with a four-strand translational repeat (Fig. 8B).

Genomic analysis of amyloidogenic light chains
Analysis of light chain sequences of AL patients suggests that the segments identified in Mcg and AL09 are the principal amy-loid-driving segments in some other individuals with AL. We analyzed protein sequences from individuals with AL, AL/myeloma (597 , 186 ), and a normal repertoire (NR) (158 , 401 ), all obtained from the Boston University ALBase. The protein sequences were aligned with ClustalW. Segments corresponding to segments 1 and 2 of Mcg were isolated from the sequences, and segments corresponding to segment 2 of AL09 were isolated from the sequences. ZipperDB (https://services. mbi.ucla.edu/zipperdb/) 4 (58) was then used to predict the propensity of the segments to form steric zippers. We found distributions of Rosetta energy scores for both segments 1 and 2 are significantly different between AL patients and NR individuals (Fig. 9, A and B). Specifically, scores for these segments tend to be lower (i.e. greater propensity to form steric zippers) in AL patients than in NR individuals. This result is based on Wilcoxon rank sum tests (segment 1: U ϭ 26,240, p Ͻ 2.2e-16; segment 2: U ϭ 24,574, p ϭ 5.197e-7). We note that the scores are not normally distributed, and the minimum scores are similar for each segment. However, there is a clear shift from a relatively higher frequency of higher scores in NR individuals to a higher frequency of lower scores in AL patients (Fig. 10, A and  B). ZipperDB scores for segment 2 also rarely fall below the threshold that predicts a hexa-peptide would form a fibril, suggesting segment 1 may be a more dominant factor in driving the formation of amyloid fibrils of the full-length protein.
We found no significant difference in the distribution of sequences between AL and NR individuals; however, 227 of the 587 sequences were not possible to evaluate with ZipperDB due to the presence of proline residues. In contrast, only 3 of the 753 segment 1 sequences and 137 of the 753 segment 2 sequences were discarded due to proline residues. Because of the insufficiency of data, analysis of the sequences was abandoned. Nevertheless, the sequences offer compelling evidence that segments 1 and 2 play an important role in amyloid formation in many AL patients.

Discussion
Our results confirm that computational prediction of steric-zipper propensity must be validated by experiments to correctly identify amyloidogenic segments. Computational prediction suggested several steric zippers within VLs, but sitedirected mutagenesis verified only two such segments. We found that both are amyloidogenic independently of each other. Some of the other predicted high-propensity steric-zipper segments do indeed form amyloid fibrils, but only in isolation from the rest of the protein. Specifically, in the context of a full-length protein, not all predicted steric zippers become exposed and available to drive formation of amyloid fibrils. Through a combination of computational, biochemical, and structural work, we have demonstrated the involvement of two distinct amyloidogenic segments in the formation of VL amyloid fibrils.
Although we performed ThT assays under acidic conditions that facilitate a feasible time scale to assess the formation of amyloid fibrils, we show that at least one of the variants, VL2-8 -J1, also forms amyloid fibrils at nearly neutral pH (Fig. S1). Other variants, such as AL09, may require a longer time frame, perhaps months or years, to demonstrate formation of amyloid Figure 6. Crystal structures of VL segments that drive the assembly of amyloid fibrils. Each of the three segments adopted a steric-zipper motif, characterized by pairs of sheets, tightly mated by the interdigitation of side chains across the interface and exclusion of water molecules. For each segment, two views are given: fibril axis vertical (left column) and fibril axis normal to the page (right column). Sheet architecture and symmetry differ among the three segments as follows. A, segment 1 of Mcg, ASLTVS. Sheet architecture is anti-parallel, antifacial, and out-of-register (class 5). The translational repeat distance along the fibril axis is 11.7 Å. B, segment 2 of Mcg, NFVFGT. Each sheet contains four orientations of strands in the asymmetric unit, creating a level of complexity beyond any of the 10 previously described steric-zipper symmetry classes. The sheet is out-of-register, with an unusually long translation repeat distance of 19.8 Å. C, segment 2 of AL09, YTFGQ. Sheet architecture is parallel, in-register, with a conventional translational repeat distance of 4.8 Å. Two types of dry interfaces are evident in the crystal packing: faceto-face and back-to-back (both class 1). Distances between sheets are given in the right column.

Amyloid-driving segments of VL domains in AL amyloidosis
fibril formation at neutral pH. The investigation of variable domain variants, Mcg and AL09, with sequences previously isolated from patients with systemic light-chain amyloidosis, supports the clinical relevance of our findings.
Our conclusions are consistent with previous solid-state nuclear magnetic resonance (ssNMR) experiments. ssNMR experiments identified several segments having immobile conformations in VL amyloid fibrils, suggesting that some of these segments are involved in the formation of amyloid fibril spines. Piehl et al. (38,39) and Hora et al. (40) each assigned amino acids to these segments in the ssNMR spectra of amyloid fibrils of two VLs, human AL09 and murine MAK33. Two of the ssNMR-assigned segments correlate with our findings: they overlap with the amyloidogenic segments identified by our computational and site-directed mutagenesis experiments (segment 1 in strand E of the Greek-key fold of immunoglobulin; and segment 2 in strands F-G (Fig. 11)). Although AL09 and MAK33 are different in their amino acid sequences and hence their amyloid fibrils are different, both include the amyloidogenic segments 1 and 2. In our work, we have experimentally validated the involvement of segments 1 and 2 in forming the steric zipper spine of the amyloid fibril. Thus, three different experiments, two ssNMR assignments of peptides in amyloid fibrils performed by independent laboratories and our site-directed mutagenesis, identify common regions involved in formation of VL amyloid fibrils (Fig. 11A).
In support of our findings, we succeeded in constructing geometrically-reasonable models of full-length VL amyloid fibrils with minimal deviation from crystallographic coordinates of the steric zippers (Fig. 12). The geometric plausibility of these models in which little unfolding occurs outside of segments 1 and 2 supports our hypothesis that these segments are geometrically capable of driving amyloid formation, even at early time points when unfolding may be incomplete. We envisage that with time, additional segments may unfold and contribute to the cross-␤-structure and further stabilize the fibril. As some measure of validation, we find that the diameters of the fibril models are consistent with the measured values for the diameters of amyloid fibrils in negative-stain electron micrographs, ϳ10 -11 nm.
Identification of amyloidogenic segments has translational implications. Our conclusions have direct bearing on strategies for developing peptide-based inhibitors of VL amyloid fibrils. Current approaches to create peptide-based inhibitors generally aim to impede elongation of the amyloid fibril spine; modified peptides are designed to incorporate into the amyloid fibril spine with one face while arresting the propagation of the spine with the opposite face (41). However, in light of our results, it is clear that inhibitor design must simultaneously block at least two independent spines of VL amyloid fibrils. Future therapies may require a mixture of two or more inhibitors of amyloidogenic segments to halt fibril formation.
The structure of NFVFGT from the VL Mcg reveals unprecedented complexity in its ␤-sheet geometry. It contains four distinct orientations of the ␤-strand within a single out-ofregister ␤-sheet. These features give the ␤-sheet a translational periodicity of 19.8 Å, which exceeds the more frequently observed in-register parallel sheets (4.8 Å), in-register antiparallel sheets (9.6 Å), and out-of-register antiparallel sheets (11.7 Å). Such lengthy periods, if truly present in biological fibrils,  Perutz banknote represents a protein segment within a steric zipper; it has N and C termini, two distinct faces, and up-and down-direction of hydrogen bonds within a sheet. Arrows represent 2 1 symmetry axes, meaning that the peptides are related by a 180°rotation about the arrow, and translated onehalf of the distance between Max Perutz banknotes along the arrow. The YTFGQ steric zipper belongs to class 1. The ASLTVS steric zipper belongs to class 5. The packing of ␤-strands is out-of-register with its strands inclined relative to the fibril axis. B, NFVFGT peptide crystallizes in a sheet containing four strands in the asymmetric unit. This sheet belongs to the one of six conceivable arrangements of a sheet with a modulus of four strands and internal symmetry. Arrows represent 2-fold symmetry operators; an ellipse represents a 2-fold symmetry axis perpendicular to the page, and T represents translation.

Amyloid-driving segments of VL domains in AL amyloidosis
may produce more complex surface patches with affinity for more complex ligands than thioflavin T or Congo red, typically known to bind amyloid fibrils. It remains to be seen whether such complexity will be observed in other amyloid structures.
The VL segments found to form steric zippers through crystal structures were analyzed in 597 and 186 VLs of patients with AL as well as 158 and 401 VLs of patients without AL. In the sequences, Rosetta energy scores of each of these segments calculated by ZipperDB tend to be lower for segments from patients with AL, indicating a higher propensity to form fibrils. The sequences could not be reliably analyzed the same way due to the presence of proline residues in an exceedingly large proportion of the gathered sequences. These results provide further evidence for the segments identified here to be amyloid-driving segments in light-chain amyloidosis. The dashed red line represents the energetic threshold of Ϫ23 kcal/mol. ZipperDB predicts hexa-peptides at or below this threshold to form fibrils. Three data points were omitted from segment 1 AL sequences, zero from segment 1 NR sequences, 121 from segment 2 AL sequences, and 16 from segment 2 NR sequences due to the presence of proline residues or missing sequence data making ZipperDB analysis impossible.

Amyloid-driving segments of VL domains in AL amyloidosis
In summary, we experimentally identified two amyloidogenic segments within VLs. Each of the segments is a key propagator of amyloid fibrils, giving rise to distinct polymorphs. Both amyloidogenic segments occupy the same spatial location of the Greek-key immunoglobulin fold in both and VLs (segment 1 in strand E of the Greek-key fold; and segment 2 in strands F-G) (42). We found that this pattern of dual steric zippers is found in VLs belonging to both and classes, suggesting that homologous segments may be responsible for fibril formation in other VLs. Disrupting a single steric zipper is insufficient for blocking amyloid formation because an alternative zipper-forming region remains intact. These results explain previous difficulties in identifying a single residue or segment within VLs responsible for amyloid formation; there can be more than one. Our work offers an experimental framework to identify multiple amyloidogenic segments in VLs and other amyloid-forming proteins.

Preparation of recombinant proteins
Protein samples were prepared as described previously. Site-directed mutagenesis was performed with PCR (44). Although not a formal test, if the notches of two boxes do not overlap, it is evidence that their medians differ significantly. A Wilcoxon-Mann-Whitney test was performed on each pair, and the median Rosetta energies of the AL group was significantly lower than that of the NR group in both segments (A, U ϭ 26240, p Ͻ 2.2e-16; B, U ϭ 24574, p ϭ 5.179e-7). Three data points were omitted from segment 1 AL sequences, zero from segment 1 NR sequences, 121 from segment 2 AL sequences, and 16 from segment 2 NR sequences due to the presence of proline residues or missing sequence data making ZipperDB analysis impossible.

Amyloid-driving segments of VL domains in AL amyloidosis
DNA of every construct was verified twice by an external DNA-sequencing service (Laragen, Inc.). In summary, VLs were expressed in Escherichia coli, purified in denaturing conditions, refolded, and concentrated (35). In the prolinescanning experiment, we found that four is the largest number of sequential prolines that permits formation of the intended constructs. Longer proline segments resulted in degradation of several VL constructs and hampered the systematic experiments.

Amyloid fibril formation assays
ThT assays were performed as described previously (45). Assays were conducted in acidic conditions (50 mM NaAc/HAc (pH 4), and 150 mM NaCl), 0.5 mg/ml VL (40 M), at 37°C with constant shaking at 300 rpm with Teflon balls with a 0.125-inch radius as stirrers (46,47). ThT fluorescence spectra were recorded at 440/ 480 nm excitation/emission wavelengths. Each experiment was performed with freshly prepared samples in at least three independent repeats on different dates. We refer to an independent repeat as a new preparation of analytes, including production in E. coli, purification, ThT assays, and analysis. Each experiment was performed at least three times on the same batch of analytes. Analysis of the genomic variant and its amyloid-stopping mutation (VL2-8 -J1 and VL2-8 -J1-L75P) was performed in at least 14 different independent repeats.
The amyloid fibril formation assays were performed in acidic conditions. Although the Mcg and AL09 are pathologic VLs, derived from patients, their behavior in vitro somewhat differs. Although Mcg and VL2-8 -J1 readily form amyloid fibrils in both acidic conditions (pH 4) and conditions close to physiologic (pH 7.4) as monitored by ThT fluorescence (Fig. S1), AL09 only forms amyloid fibrils in acidic conditions (pH 4). The application of acidic condi-

Amyloid-driving segments of VL domains in AL amyloidosis
tions is common; its purpose is to facilitate and expedite the formation of amyloid fibrils in vitro (48,49). In conditions other than acidic, amyloid fibril formation by pathologic VLs shows some increased variability; in some experiments, the VLs form amyloid fibrils, although in others they do not. However, in destabilizing conditions the results are consistent, which enables experimental analysis of the capability to form amyloid fibrils.

Electron microscopy (EM)
Samples from the ThT fibril formation assays were diluted with water to 10% v/v and applied onto copper grids with Formvar-carbon coating (Ted Pella, Inc., catalog no. 01810). Negative staining was performed with 2% w/v uranyl acetate, and images were collected by means of a Tecnai T12 electron microscope at 120 kV with a Gatan CCD camera.

Crystal structure determination
Peptides were purchased from Genscript, with TFA salt substitution for HCl. The purity of peptides was greater than 95%. Peptides were dissolved in water to a concentration of 20 mg/ml, and crystallization trials were set up using a Mosquito robot, with a 1:1 ratio of the peptide solution and crystallization screen. Crystals grew in hanging drops and were of needle-like morphology. Solutions that yielded the bestdiffracting crystals are listed in Table 2. Crystals were drymounted onto pulled glass capillaries, and diffraction data were collected at Advanced Photon Source beamlines 24-ID-C and 24-ID-E (ASLTVS, 24-ID-E; NFVFGT, 24-ID-C; and YTFGQ, 24-ID-C). Phases were determined by molecular replacement with Phaser (ASLTVS) or by direct methods with Shelx (50, 51) (NFVFGT and YTFGQ). Molecular structures were refined with Refmac5 and deposited into the PDB with codes 6DJ0 for ASLTVS, 6DIX for NFVFGT, and 6DIY for YTFGQ (52).

Construction of amyloid fibril models
In constructing our models of full-length VL amyloid fibrils, we strived to make as few assumptions as possible. Therefore, atoms in the VL zipper spine were strongly restrained to maintain the arrangements observed in the crystal structures of their amyloidogenic segments. Because the exact unfolding mechanism of the variable domains is unknown, the remaining atoms of the VL were restrained to maintain the native globular folds (PDB codes 4UNU (35) and 3CDY (53)).
The largest challenge we faced in constructing the models was avoidance of steric clashes among globular domains, because the VL chains are very closely spaced along the fibril axis (4.8 Å for YTFGQ; 5.8 Å for ASLTVS; and 5.0 Å for NFVFGT). We performed a systematic search of orientation space, counting the number of steric clashes incurred for each possible orientation of the globular domains (in 5°increments). To apply the incremental rotation operations and to evaluate steric clashes, we used the programs PDBSET and CONTACT, respectively, from the CCP4 suite (54). We did not impose any twist in our fibril model, because there was no characteristic twist pitch length evident in electron micrographs of the fibrils. As a matter of convenience, we chose to model either 8 or 16 orientations of the globular domains around the fibril axis. We call this collection of unique orientations the asymmetric unit of the fibril. The fibril is then composed of translational repeats of this asymmetric unit (38.4 Å for YTFGQ, 19.8 Å for NFVFGT, and 46.3 Å for ASLTVS) along the fibril axis. For simplicity, the molecules within the asymmetric unit were restrained to have either cyclic or dihedral symmetry. Those models that produced the least number of clashes were manu-

Amyloid-driving segments of VL domains in AL amyloidosis
ally edited with the program Coot (55) to build connections between the zipper spine and globular domain. The models were geometrically and energetically minimized with the programs CNS (56) and Rosetta (57).