Throughout this manuscript, we use the IMGT nomenclature and numbering system for LC genes, proteins and residues (47, 57). For each antibody, single germline precursor variable and joining genes (IGV_L and IGJ_L , encompassing both κ and λ gene fragments from the IGK and IGL loci, respectively) are recombined to yield a variable region (IGKV-IGKJ for κ or IGLV-IGLJ for λ), encoding the ~110-residue LC V_L-domain protein. The IMGT numbering system has 127 possible residue positions in LCs, including conserved gaps to accommodate insertions. This numbering system allows direct comparison between the different types of immunoglobulin domains, so several positions such as 58–64 are very rarely occupied by residues within human LCs. “Residue position” always refers to IMGT numbering.

This study focuses on the frequency of individual residue changes within LC V_L-domains associated with AL amyloidosis, MM, or the polyclonal immune repertoire represented by OAS (56). We refer to these sequences as AL, MM and OAS LCs, respectively. The overall workflow is shown in Supplementary Figure 1 . A “set” of sequences is defined as all the LCs from a specific germline gene and disease or repertoire of origin. Only complete IGV_L -IGJ_L sequences with no ambiguous or missing residues were used for analysis. Sequences with CDR3 insertions longer than the available space in IMGT numbering were also excluded. We did not consider nucleotide sequences or C_L domains, since these are not available for all LCs. We refer to “residue changes” rather than “mutations” since only protein sequences were analyzed and to avoid ambiguity between germline and somatic mutations.

In total, 746 AL, 969 MM and 8,047,747 OAS LC sequences were analyzed. AL LCs included monoclonal LC sequences from the AL-Base “AL-PCD” category, comprising all sequences from individuals with a diagnosis of AL amyloidosis, regardless of the underlying hematological malignancy. Note that this selection differed from that used in our previous studies (11, 50), which excluded LC sequences from the AL/MM subcategory. MM LCs included sequences from the “MM” and “SMM” (smoldering MM) subcategories, but excluded those with a diagnosis of light chain deposition disease. Unique OAS LCs were identified from 66 OAS samples derived from healthy adults as previously described (11).

Twenty IGV_L genes (8 IGKV and 12 IGLV), for which at least five AL LCs were available in the AL-Base data, were analyzed: IGKV1-12, IGKV1-16, IGKV1-33, IGKV1-39, IGKV1-5, IGKV3-15, IGKV3-20, IGKV4-1; and IGLV1-36, IGLV1-40, IGLV1-44, IGLV1-47, IGLV1-51, IGLV2-8, IGLV2-14, IGLV2-23, IGLV3-1, IGLV3-19, IGLV3–21 and IGLV6-57. Results in the main body of the manuscript focus on LCs derived from IGKV1–33 and IGLV2-14, which are frequently observed in AL, MM and OAS LCs. Data for other germline genes is shown in Supplementary Data Sheet 1 .

LC nucleotide and protein sequences were assigned to germline precursor genes using the IMGT High-VQuest or DomainGapAlign tools, respectively (58). Paralogous IGKV genes from the proximal and distal loci (59) were counted as being from the proximal locus (e.g., LCs assigned to IGKV1-33 and IGKV1D-33 were both counted as being derived from IGKV1-33). For nucleotide sequences, the translation of the sequence provided by High-VQuest was used. OAS LCs are assigned to a precursor gene as part of the deposition process (56) so these assignments were not reanalyzed. Residues were aligned to the IMGT numbering system using the ANARCI tool (60), which more consistently assigned sequence gaps than other tools for these sequences. The analyses described below are sensitive to misaligned gaps in the IMGT numbering system, which may be misinterpreted as insertions or deletions (indels). There were 71 AL-Base sequences where alignment by ANARCI led to indels, relative to the assigned germline sequence. These were checked manually and 43 alignments were modified to favor residue replacements over indels, so as to maintain the positions of gaps that are present in germline sequences and minimize alignment biases.

For each IGV_L gene, we separately aligned the AL, MM or OAS V_L domain sequences using ANARCI (60). Following the approach of Sheng and coworkers (13), we constructed consensus matrices to define the distribution of residues at every position. An example matrix is shown in Supplementary Figure 2 . The identity of the assigned IGV_L allele and IGJ_L gene were not considered. The number and fraction of sequences with each residue at all 127 IMGT positions was calculated. For each group of sequences, this procedure yielded a 127 × 21 matrix, corresponding to the fraction of each residue, including gaps, at each position. We created a total of 59 matrices, corresponding to AL, MM and OAS LCs from each gene except IGLV1-36, for which no MM LCs were present in AL-Base. This approach avoided multiple sequence alignments, which are computationally expensive for large numbers of sequences and are prone to errors around the indels which are frequent in CDR3. The OAS matrices are provided in Supplementary Table 1 .

To evaluate the residue-wise variability at each position within the matrices, we calculated the Gini coefficient (61), implemented in the R ineq package (62). This is a measure of the inequality of a distribution, which varies between 0 (corresponding to equal fractions of all possible residues) and 1 (a single residue observed in all sequences). To compare the residue-wise differences between sets of sequences, we calculated the Pearson correlation coefficient (ρ) between the columns of the relevant consensus matrices for each pair of residues (ρ = 1 indicates identical distributions of residues).

To compare pairs of consensus matrices, we calculated difference matrices by subtraction. For global comparison between groups of sequences, we calculated the pairwise Euclidian distance between each consensus matrix, which is the square root of the sum of squared differences between each of the positions in the matrices. These distances are equivalent to the total magnitude of the corresponding difference matrix. We subjected the results to hierarchical clustering analysis in R (63) to yield a phylogenetic tree (or dendrogram).

For all AL-Base sequences analyzed, the frequency of each residue within the appropriate OAS consensus matrix was calculated. Based on the distributions of these frequencies, we defined “common” residues as those appearing in ≥ 10% of OAS sequences derived from that germline gene and “uncommon” residues as those appearing in < 10% of OAS sequences. To visualize the frequency in OAS of each residue within a sequence of interest, we plotted the values from the corresponding consensus matrix as a “frequency profile”.

For each group of LCs, the number sequences harboring common and uncommon residues at each position was calculated. For prospective analysis and creation of frequency profile plots, all positions were considered. For previously identified positions, numbers of sequences were counted with both the specific residue change as published, and with any uncommon residue at that position. Residue changes described by Hurle and coworkers (41) were assigned to the germline gene of the LC in which they were originally described. Insertion of proline within CDR3 of an IGKV1-33-derived LC was identified as potentially amyloidogenic by Randles and coworkers (64). This is referred to as 95ProIns in the original publication, which uses the Kabat numbering system, and P115PP here using the IMGT numbering system. Identifying these insertions is difficult because of the variable length of CDR3, which can lead to ambiguous alignments. Sequences were identified where two consecutive proline residues occur around IMGT position 115 and at least one residue is inserted into the gap within CDR3, which corresponds to positions 110–113 of the germline IGKV1-33 sequence.

To compare the frequencies of residue changes within groups of LCs, we calculated the odds ratios (OR), 95% confidence intervals (CI) and associated p-values for each pairwise comparison. These parameters were calculated directly from counts, rather than estimated from a generalized linear model. To avoid division by zero, a correction factor of 0.1 was added to all counts before ORs were calculated. OR significance was calculated directly from the natural logarithms of ORs and their standard errors using the normal distribution, implemented in R (63). Positions with a positive OR for the AL vs. OAS comparison, but where only germline residues were observed in AL LCs, were excluded because these were determined to be artifacts caused by the large difference in sample sizes. P-values were corrected for multiple testing using the false discovery rate (FDR) method (65), where FDR < 0.05 was considered statistically significant.

To visualize residue frequency at each position across the LC sequence, we generated customized sequence logo plots (shown in Supplementary Data Sheet 1 ) using the R package ggseqlogo (66) that differed from the original sequence logo concept (67). Because residue changes were distributed throughout the LC sequence, the information content at each position was dominated by the germline sequence. To highlight non-germline residues, we excluded the most frequent residue from the consensus matrix and created a logo where the height of each letter represented only its frequency in the alignment.

Crystal structures of germline IGKV1-33, IGLV2-14 and IGLV6-57 V_L-domains were available for analysis (PDB entries 2Q20 (43) 6SM1 (44) and 2W0K (68), respectively). In addition, we generated V_L domain models for the 20 IGV_L genes studied using the Alphafold 3 server https://alphafoldserver.com/ (69). Protein sequences corresponding to the *01 allele of each IGKV or IGLV gene and appropriate IGKJ1*01 or IGLJ1*01 gene were submitted using the default parameters. The resulting models had good agreement with the crystal structures (Cα root mean square deviations of 0.476, 0.435 and 0.679 Å for IGKV1-33, IGLV2-14 and IGLV6-57, respectively). Secondary structures and solvent-accessible surface areas were calculated using DSSP (70, 71). Residues with < 10% surface exposure, relative to reference peptides (72), were defined as buried and other residues were defined as solvent exposed. Data were mapped onto structures using ChimeraX (73).

Most analysis was carried out using R v 4.2.2 (63) via the RStudio environment (74). The following packages were used: Biostrings v 2.66.0 (75), broom v 1.0.3 (76), cowplot v 1.1.1 (77), furrr v 0.3.1 (78), future v 1.33.0 (79), ggpubr v 0.6.0 (80), ggtree v 3.4.1 (81), ggseqlogo v 0.1 (66), ineq v 0.2-13 (62), msa v 1.30.1 (82), openxlsx v 4.2.8 (83), rstatix v 0.7.2 (84), and Tidyverse v 1.3.2 (85). Local installations of ANARCI (60) and DSSP v 2.1.0 (70, 71) were used on the Boston University Shared Computing Cluster. Alphafold 3 was accessed via the web interface, https://alphafoldserver.com/ (69). ANARCI can also be accessed via a web interface, https://opig.stats.ox.ac.uk/webapps/sabdab-sabpred/sabpred/anarci/.

The frequency of mutations varies across LC sequences due to the biased mechanism of somatic hypermutation (13). Within each IGV_L gene, alignments of OAS sequences provide a reference for this variability. We constructed consensus matrices from all OAS LCs for each IGV_L gene, which define the frequency of each residue, including gaps, at all 127 IMGT positions. These matrices are visualized as heat maps in Figures 1A, B for IGKV1-33 and IGLV2-14, respectively. Results for other germline genes are shown in Supplementary Data Sheet 1 . Frequency data for the OAS matrices are provided in Supplementary Table 1 . The germline residue was present in > 90% of sequences at most positions. However, in positions that vary between alleles or are prone to mutation, the fraction of sequences with the most common residue was lower. These positions are frequent in CDRs but occur throughout the sequences. For the 20 IGV_L genes where at least five AL LCs were available, we generated additional consensus matrices for AL and MM LCs and calculated differences between pairs of matrices. Data for IGKV1-33 and IGLV2-14 are shown in Figure 1 , and data for other IGV_L genes is shown in Supplementary Data Sheet 1 .

To further evaluate the distribution of residue changes within sets of LCs, we generated sequence logos for AL, MM and OAS LCs, which show the frequency at which each residue was observed along the sequences ( Supplementary Data Sheet 1 ). Quantitation of the variation within and between sets of LC sequences showed that the positions that differ between AL and MM LCs are also most variable positions within the sequences ( Supplementary Data Sheet 1 ).

To investigate the relative contributions of within-gene and between-gene variability, we measured the pairwise distances between each of the consensus matrices and analyzed the resulting data using hierarchical clustering ( Supplementary Figure 3 ). This analysis showed that AL LCs were consistently more closely related to MM and OAS LCs derived from the same gene than to AL LCs derived from other genes.

We calculated the frequency at which every residue in AL or MM LC sequences was observed in the OAS consensus matrices. Histograms of these frequencies are shown in Figure 2A . Both AL and MM sequences show a bimodal distribution of residue frequencies. Based on these distributions, we defined a residue as “uncommon” if it occurred in < 10% of sequences from the relevant OAS consensus matrix. Residues which occur in ≥ 10% or more of OAS sequences were defined as “common”. Similar proportions of residues were uncommon in both AL and MM LCs ( Figure 2B ) and uncommon residues were observed throughout the LC sequences in both groups ( Figure 2 ). We examined the distribution of residues that occur in < 1% of OAS sequences (dark bars in Figure 2B ), which also occurred throughout the LC sequence.

The distribution of residue frequencies was visualized graphically as a “frequency profile” for two amyloidogenic LCs, known as AL-09 and Pat-1, which are derived from the IGKV1-33 and IGLV2-14 genes, respectively (43, 44) ( Figure 3 ). The frequency of each residue within the OAS repertoire alignments is shown. Baden and coworkers identified N34I and Y87H (N40I and Y103H in IMGT numbering) as destabilizing residue changes in AL-09 in vitro, relative to its germline sequence (43). These two residues were observed at lower frequencies within the OAS alignment for IGKV1-33 (0.81% and 0.77%, respectively) than the rest of the AL-09 sequence. Similarly, Kazman and coworkers identified L81V (IMGT L94V) as highly destabilizing to Pat-1 in vitro (44). This residue is the least common of all Pat-1 residues in the OAS alignment for IGLV2-14 (0.55%).

To identify residue changes that are enriched or depleted in AL LCs compared with other LCs, we measured the proportion of sequences that harbor an uncommon residue at each position in all AL, MM and OAS LC sequences, segregated according to precursor gene. We calculated ORs and 95% confidence intervals for observing an uncommon residue at each position for AL vs. MM and AL vs. OAS LCs. Data at each position in LCs derived from IGKV1-33 and IGLV2-14 are shown for the AL vs. MM ( Figure 4A ) and AL vs. OAS ( Figure 4B ) comparisons. Equivalent results for other IGV_L genes are shown in Supplementary Data Sheet 1 , and details of all positions identified are reported in Supplementary Table 2 . Without correction for multiple comparisons, six positions in IGKV1-33, but no positions in IGLV2-14, were significantly enriched for uncommon residues in AL vs. MM LCs (p < 0.05, orange symbols). Two positions in IGLV2-14 harbor uncommon residues less frequently in AL LCs than in MM LCs (p < 0.05, purple symbols). For the comparison between AL and OAS sequences, 26 positions among IGKV1-33 LCs and 7 positions among IGLV2-14 LCs were significantly enriched for uncommon residues (p < 0.05, orange symbols). Three positions in IGKV1-33 LCs and two positions in IGLV2-14 LCs were under-represented in uncommon residues for the AL vs. OAS comparison (p < 0.05, purple symbols).

Analysis of the 20 precursor genes where at least 5 AL sequences were available showed that the number and location of positions enriched for uncommon residues differ substantially between genes ( Figure 5 ; Supplementary Table 2 ). Uncommon residues were over-represented in the AL vs. OAS comparison at 62 positions, or 125 gene/position combinations, encompassing all four FRs and three CDRs. Uncommon residues were under-represented in a further 20 positions, or 25 gene/position combinations. Notably, few of these positions were common between precursor genes, with 30 positions observed in only a single gene. The V_L-domain N- and C-termini were the most frequently enriched positions (6 and 9 genes, respectively), although this may reflect systematic biases due to sequencing strategies. Other positions enriched in uncommon residues in multiple genes were positions 52 (6 genes), 85 (5 genes), 97 (6 genes) and 103 (5 genes). We correlated the variability at each position with its degree of enrichment in the AL vs. OAS comparison and observed that uncommon residues were most frequently enriched at positions that were conserved in OAS LCs ( Supplementary Figure 4 ). Excluding the N- and C-termini, 85 of 110 positions that were enriched for uncommon residues in AL vs. OAS were within FRs ( Figure 5 ; Supplementary Figure 4 ).

We mapped these positions onto the available crystal structures of isolated V_L-domains corresponding to IGKV1-33, IGLV2-14 and IGLV6-57 (43, 44, 68) ( Figure 6 ). Structural mapping of these data for other IGV_L genes, using computational models of the isolated V_L domains, is shown in Supplementary Figure 5 . Positions enriched for uncommon residues occur in a variety of structural contexts, including both solvent-exposed and buried residues; those in hydrogen-bonded secondary structures and unstructured loops; and residues that interact with the LC’s heavy chain partner. These details are provided in Supplementary Table 2 . Again, there were no clear patterns of which positions were enriched for AL-associated residues

We next used this data to reexamine positions and residue changes that have previously been identified as associated with AL amyloidosis. A total of 48 positions were studied ( Tables 1 – 3 ). Additional information for these positions is provided in Supplementary Table 3 . We divided the residue changes into three groups, associated with all IGKV-derived LCs, specific IGKV genes and specific IGLV genes.

Table 1 and Figure 7 show residue changes proposed to be associated with amyloidosis in all IGKV-derived LCs. All changes yielded an uncommon residue at the specified position. Eight of 14 residue changes were significantly enriched in AL vs. OAS LCs. Five residue changes were significantly enriched in AL vs. MM LCs, of which four were also significantly enriched compared to OAS LCs. However, the frequency of each residue change in AL LCs was low, accounting for at most 12% of IGKV-derived AL LCs. The most significantly enriched residue change in this group was gain of an asparagine residue at position 86 (X86N), which was previously identified as a site for N-glycosylation (38, 49, 50).

Table 2 and Figure 8 show residue changes identified in specific IGKV genes, and Table 3 and Figure 9 show residue changes identified in specific IGLV genes. The tables show the frequency of the specified substitution and the figures show the frequencies of both the specified residue and all uncommon residues. Equivalent calculations for any uncommon residue at each position are provided in Supplementary File 1 . Of the 34 residue changes, 32 yielded uncommon residues at that position. Seventeen of the 25 previously identified residue changes in IGKV genes are significantly enriched (OR > 1, FDR < 0.05) among AL vs. OAS LCs ( Table 2 ). Any uncommon residue was significantly more frequent in AL than OAS LCs at 10 positions ( Figure 8 ). Three residue changes in IGLV genes were significantly enriched among AL vs. OAS LCs ( Table 2 ), but uncommon residues were only significantly enriched at position 25 of IGLV6-57 ( Figure 9 ). No positions were enriched for specified or uncommon residues in the AL vs. MM comparison.

Our prospective analysis identified position 25 in IGLV6-57 as the site of an amyloid-associated residue change, in agreement with previous studies (46, 86–90). This residue change is associated with a germline-encoded polymorphism, rather than a somatic hypermutation: the IGLV6-57 *01, *03 and *04 alleles encode an arginine residue (R25), whereas the IGLV6-57*02 allele has a C>G nucleotide transversion that encodes a glycine residue at position 25 (G25). The presence of G25 has been proposed to be associated with AL amyloidosis (46, 86–90), which is supported by our analysis ( Table 3 ; Figure 9 ). R25 forms a cation-π interaction that stabilizes the native V_L domain, while G25 is destabilizing in the context of the V_L-domain and the full-length LC (46, 91). We therefore investigated the distribution of residues at position 25 in AL, MM and OAS LC sequences derived from IGLV6-57 as a function of the assigned allele.

Among 137 AL LCs, 33 were G25 (24%), 99 were R25 (72%) and five had other residues at this position ( Supplementary Figure 6 ). None of the 9 IGLV6-57-derived MM LCs, and 5,958 of 90,110 IGLV6-57-derived OAS LCs harbored G25 (6.6%). The OR for non-argenine residues between AL and MM was 2.32 (p = 0.42). The ORs between AL and OAS were 2.81 for non-argenine residues (p = 6.17 × 10^-8) and 3.65 for glycine residues (p = 9.67 × 10^-11). All AL LCs with G25 were assigned to the IGLV6-57*02 allele ( Supplementary Figure 6 ). Of the 36 IGLV6-57*02 AL LC sequences, 33 harbored a glycine at position 25, two had alanine and one had a gap when aligned to IMGT numbering. The single MM LC assigned to IGLV6-57*02 had alanine at position 25. OAS LCs were assigned only to the *01 and *02 alleles and all sequences with G25 were derived from the *02 allele.

We repeated this analysis on 46 positions within 15 IGV_L genes where there were residue differences between alleles ( Supplementary Figure 7 ). The five other IGV_L genes studied had either a single allele or no residue differences between the alleles. In many cases, there was substantial heterogeneity at these positions. However, none of these positions showed a significant difference between AL and MM, other than IGLV6-57 R25G, after correction for multiple testing.