We used the data sets from two studies where monoclonal Abs were cloned from different subpopulations of healthy human B cells. Thus, we analyzed sequences of 398 Abs characterized in the study of Shehata et al. (27). These Abs were cloned from naive, IgM memory, and IgG memory B cells and from long-lived plasma cells (LLPCs). In addition, we evaluated sequence correlates of polyreactivity of 240 monoclonal Abs cloned from IgA memory B cells, described in the study of Prigent et al. (37). In both studies, Abs were cloned without selection for a particular antigenic specificity and expressed exclusively as IgG1 for functional analyses. Of note, the two studies used different approaches for the estimation of Ab polyreactivity. Thus, Shehata et al. used a technique referred to as polyspecificity reagent assay (PSR), where Ab reactivity to proteins available in human cell lysate is measured by flow cytometry and presented as a gradually increasing numeric score (27). Prigent et al. assessed Ab polyreactivity by ELISA (37). To use the semi-quantitative data presented in the work of Prigent et al. in our correlation analyses, we first assigned numeric scores of reactivity based on the number of recognized antigens by a given Ab (see Methods).

In previous works, V region sequence patterns that determined polyreactivity were evaluated by using pools of Abs that originate from different B-cell subpopulations. Moreover, some of these studies used Abs directed to pathogens (influenza virus or HIV-1) or Abs designed to be used as therapeutics. Here, we aimed to assess the sequence traits of V_H and V_L regions that correlate with Ab polyreactivity using Abs without pre-defined antigen specificity. We also considered it important, in addition to bulk analyses, to segregate the Abs as based on their origin (B-cell types from which they were isolated) and perform the statistical analyses individually.

We applied Spearman non-parametric correlation analysis to assess the sequence patterns of V_H and V_L domains associated with polyreactivity of human Abs. The obtained data from correlation analysis for the V_H domain are depicted in Figure 1 . These data demonstrated that Ab polyreactivity significantly correlates with increased or reduced frequencies of certain amino acid residues in the V domain. Strikingly, distinct patterns of significant correlations were observed between the groups of Abs originating from different B-cell subpopulations ( Figure 1 ). Thus, polyreactivity negatively correlated with the number of V_H somatic mutations for Abs originating from IgM+ B cells, but no significant correlation was found in other groups. In addition, a positive correlation between the length of CDR H3 and polyreactivity was observed only for Abs cloned from IgA+ B cells. The polyreactivity of Abs from three B-cell subtypes—naive, IgG+ memory and plasma cells, correlated with the presence of a higher number of basic amino acid residues (Arg, His, and Lys) in CDR H2 (naive B cells and LLPC) and CDR H3 (IgG+ memory B cells) and within entire V_H region (LLPC). However, no increased number of basic amino acid residues in any part of the V region was associated with the polyreactivity of Abs isolated from IgM+ and IgA+ memory B cells. Reduced number of acidic amino acid residues (Asp and Glu) in the entire V_H region or in CDR H1 and CDR H2 was associated with a higher polyreactivity in the case of Abs cloned from all types of memory B cells but not from naive B cells and LLPC ( Figure 1 ). Abs cloned from each B-cell subpopulation demonstrated unique patterns of positive or negative correlates of polyreactivity with the number of polar amino acid residues in V_H domain. The presence of polar residues was especially important for determining the polyreactivity of IgG+ memory B cells ( Figure 1 ). Thus, a higher number of polar amino acids (as a physiochemical category) in CDR H1 and H2 and in the entire V_H was positively correlated with promiscuous antigen binding of Abs. Statistically significant correlations were also found for the number of Gln residues in the entire V_H and Cys residues in CDR H3 ( Figure 1 ). Contrary, polyreactivity negatively correlated with the presence of Thr in CDR H1 in Abs cloned from naive B cells. Our data demonstrated that the polyreactivity of Abs isolated from IgA+ memory B cells significantly correlated with the presence of different hydrophobic residues (Val, Ile, Leu, and Met) in the entire V_H region or in CDR H1 and H3 loops ( Figure 1 ). This prominent pattern was not observed in Abs isolated from other types of B cells. On the contrary, the polyreactivity of Abs from IgG+ memory B cells was negatively correlated with the number of hydrophobic amino acids (as a physiochemical category) in the entire V_H domain or CDR H2. The increased number of hydrophobic residues in V_H of Abs isolated from IgA+ B cells resulted in a significant correlation of polyreactivity with augmented hydrophobicity index (GRAVY) of all CDR loops and entire V_H domain ( Figure 1 ). Notable patterns of sequence correlations were also observed regarding the presence of aromatic amino acid residues in V_H domain. Thus, the number of aromatic amino acids (as a group) or of Tyr in V_H or in CDR H3 positively correlated with enhanced polyreactivity of Abs cloned from IgM+ B cells ( Figure 1 ). Positive correlation between polyreactivity and aromatic amino acids was also found for Abs from LLPC, but in this case, the presence of any aromatic residues in CDR H1 or Phe in CDR H3 reached significance. Notably, reverse tendencies were observed for Abs isolated from IgG+ and IgA+ B cells, where the presence of aromatic residues or Phe negatively correlated with the antigen-binding polyreactivity ( Figure 1 ). The number of Gly residues significantly correlated with polyreactivity only for Abs that originated from naive B cells. Thus, a lower number of Gly in the V_H region and CDR H2 was associated with higher polyreactivity. However, a reverse tendency was observed for Gly residues in the case of CDR H1 ( Figure 1 ).

Our data also revealed that sequence features of framework regions (FW) of V_H can also contribute to the polyreactivity of Abs. Importantly, we found that the patterns of significance in FW regions also depended on the origin of Abs ( Figure 1 ). As FW regions are less frequently subjected to somatic mutations, the observed differences may reflect biased use of V_H genes in the case of Abs with extended antigen-binding polyreactivity.

Our analyses showed that several sequence features of the V_L domain also correlate with polyreactivity of Abs ( Figure 2 ). Again, different sequence patterns of correlation were detected depending on the B-cell origin of the studied Abs, albeit these differences were less prominent as compared to those observed in V_H ( Figures 1 , 2 ). An important observation from these data is the presence of a positive correlation between the size of the entire V_L region or CDR L1 and polyreactive antigen binding in Abs cloned from three different B-cell subpopulations, naive, IgG+ memory, and LLPC. Although the data did not reach statistical significance, the same tendency was present in the case of IgM+ memory B cells. These data suggest that the length of CDR L1 is an important determinant of polyreactivity of Abs. Interestingly, statistical significance was present between the lower numbers of Gln residues in CDR L1 and antigen-binding polyreactivity in all B-cell subpopulations except in naive B cells ( Figure 2 ). Of note, this CDR L1 is characterized by the highest variability in length among CDRs in V_L region. Another common sequence trait positively correlating with polyreactivity of different B-cell subpopulations is the presence of a higher number of aromatic amino acids in all CDR loops. Such correlation reached significance for Abs cloned from naive-, IgG+, and LLPC ( Figure 2 ). An important general observation of the correlation analyses of V_L domain is that the region that has the most significant capacity to determine the promiscuous antigen binding of Abs, from different B-cell subpopulations is CDR L1 ( Figure 2 ). Similarly, as the observation for V_H, the sequence features in FW regions correlating with polyreactivity of Abs showed particularities as dependent on the B-cell type from which Abs were cloned.

Taken together, the data from Spearman correlation analyses of V_H and V_L performed on Abs categorized based on their B-cell origin unravel numerous sequence patterns significantly correlating with the antigen-binding promiscuity. Notably, unique patterns were present in different groups of Abs.

It is noteworthy that the Abs cloned from IgA+ memory B cells (n=240) were analyzed for polyreactivity by ELISA (13), whereas the binding promiscuity of the rest of Abs (n=398) was assessed by PSR assay (27). We realized that the use of two distinct approaches can introduce bias in data, and direct comparison of the results should be interpreted with caution. However, previous analyses demonstrated that PSR and ELISA methods significantly correlate in their capacity to identify polyreactive Abs (21, 22). The validity of our results was also supported by the fact that there were considerable differences in the sequence patterns of V_H determining polyreactivity in groups of Abs that were assessed by an identical polyreactivity assay (PSR). For example, note the differences between Abs from IgG+ memory B-cell group (n=146) and Abs from LLPC (n=143) ( Figure 1 ).

Next, we elucidated the sequence correlates of V regions that determine antigen-binding polyreactivity in an integrated set of Abs originating from different B-cell subpopulations. We excluded from this bulk repertoire only Abs originating from IgA+ B cells, as their polyreactivity was assessed by a different technique. By performing the statistical analyses of a panel of 398 Abs, a number of significant correlations were observed ( Figure 3 ). Thus, Ab polyreactivity in bulk repertoire correlated with the presence of a higher number of Arg residues in CDR H3 and a lower number of acidic amino acids (especially Glu) in the entire V_H domain. Polyreactivity was also negatively correlated with Thr in CDR H1 and positively associated with the number of Gln residues in the whole V_H. No significant correlation was observed between the number of hydrophobic amino acids in CDR loops of V_H domain and the polyreactivity of Abs. Notably, higher polyreactivity of Abs correlated with an increased number of aromatic residues (Phe and Trp) in CDR H3 ( Figure 3 ). The number of aromatic residues in FR1 was also significantly elevated in Abs with higher antigen-binding polyreactivity.

The correlation analyses revealed even a larger number of significant correlations of V_L sequence characteristics with the polyreactivity in bulk Ab repertoire ( Figure 3 ). Thus, a positive correlation was observed between the length of V_L region and particularly of CDR L1 and promiscuous antigen binding. In addition, the number of polar amino acids in the V_L or in CDR L1 and L3 correlated positively (Tyr and Thr) or negatively (Gln) with the Ab polyreactivity. The presence of the hydrophobic amino acid Val in the entire V_L and in CDR L1 and L2 was similarly significantly associated with increased polyreactivity of Abs. However, the presence of Ile in whole V_L and Ala in CDR L2 had a significantly negative impact on the Ab promiscuity ( Figure 3 ). Similarly, as in the case of V_H, polyreactivity of Abs in integrated Ab repertoire positively correlated with the presence of aromatic amino acid residues in CDR L1 (Tyr), CDR L2 (Trp), and in CDR L3 (as type). Polyreactivity of Abs also significantly correlated with an increased number of Gly residues in V_L as well as in CDR L3 ( Figure 3 ).

Our analyses demonstrated that the FW regions of light-chain V regions have a more decisive role in determining polyreactivity in integrated Ab repertoire as compared with the FW region of heavy V regions ( Figure 3 ). This observation suggests that a larger set of V_L genes may be associated with the polyreactivity of Abs.

Collectively, these results revealed several significant correlates of Ab polyreactivity in bulk immune repertoire consisting of Abs from distinct B-cell subpopulations. These data showed certain prominent sequence patterns associated with polyreactivity—lack of acidic residues in V_H and presence of aromatic amino acids in CDRs of both V_H and V_L. The sequence pattern correlating with polyreactivity in bulk repertoire differed from the patterns in Abs from a particular B-cell subtype. Moreover, the strength of correlations was diminished, suggesting the presence of multiple pathways for the achievement of polyreactive antigen binding.

To obtain further information about molecular correlates of polyreactivity, we next focused our analyses only on Abs demonstrating the most prominent antigen-binding promiscuity. To this end, we selected the top 5 polyreactive Abs from each B-cell subpopulation. Table 1 presents a summary of the characteristics of the variable regions of the Abs. These data revealed that the most polyreactive Abs did not have any specific bias in the lengths of CDR loops or their isoelectric points. Thus, CDR H3 and CDR L1 regions displayed the highest variability of lengths among different polyreactive Abs, i.e., 8–26 and 6–12 residues for CDR H3 and CDR L1, respectively ( Table 1 ). In silico calculated isoelectric points of CDRs also varied in a very broad range between different Ab molecules ( Table 1 ). For example, in the case of the most diverse and important region for antigen recognition (CDR H3), the calculated pI values of the strongly polyreactive Abs were in the range between 3.67 and 9.14. It is noteworthy that substantial variability in the length of CDR loops and the pI values was observed between Abs within all groups, except in the case of Abs isolated from IgA+ B cells ( Table 1 ). These data indicate that a substantial level of antigen-binding promiscuity can be achieved by alternative topological and physicochemical features of the antigen-binding sites.

Furthermore, to gain more detailed information about the topology of the antigen-binding sites of polyreactive Abs, we applied the structural modeling algorithm Rosie-2 (Rosetta Commons). The side views of the most probable structures of V regions of Abs are shown in Figure 4 . These models demonstrated that the highly polyreactive human Abs have diverse topologies of their variable regions. One apparent feature observed in a part of the Abs is the presence of long and protruding CDR H3 ( Figure 4 ). This structural feature was especially evident in polyreactive Abs from naive B cells (ADI-45498), IgM+ memory B cells (ADI-45499, ADI-45497), and LLPC (ADI-47276 and ADI-47265) and well corresponded to the length of CDR H3 ( Table 1 ). Interestingly, the structural models also indicated that some Abs have projecting only CDR L1 (ADI-45435, ADI-45430, and ADI-47275), or in some cases, both CDR H3 and CDR L1 are protruding (ADI-45435). Despite this obvious structural feature of variable regions of some of the polyreactive Abs, the structural models also showed that another fraction of highly polyreactive Abs did not display abnormally protruding CDR loops ( Figure 4 ). Indeed, the topology of their binding sites is similar as the one depicted for most of the protein-binding Abs available in Ab structural databases. This observation further substantiated our results from correlation analyses, suggesting that polyreactive Abs use diverse pathways for the achievement of broad antigen-binding promiscuity.

Polyreactivity has been associated with specific physiochemical characteristics of antigen-binding sites of Abs such as the presence of patches of positive charges on the molecular surface. To elucidate whether this is the case in the human Abs isolated from different B-cell compartments, we depicted the distribution of charges on the surface of antigen-binding sites of a selected set of Abs characterized with outstanding polyreactivity ( Figure 5 ). We also calculated three-dimensional Coulombic surface electrostatic potential of variable regions of the selected Abs ( Figure 6 ). These data demonstrated that different polyreactive Abs display distinct patterns of distribution of charged or non-polar residues in their binding sites ( Figure 5 ) and that the energy and space distribution of their electrostatic potentials differs substantially ( Figure 6 ). Some of the most polyreactive Abs, indeed, demonstrated extensive positively charged antigen-binding sites (for example, ADI-45498, ADI-46713, ADI-45499, ADI-47301, ADI-47300, and ADI-47303); moreover, the data revealed that a predominant fraction of Abs had antigen-binding surfaces deprived of negative charges. Nonetheless, there were highly polyreactive Abs with balanced positive and negative charged patches or in some cases; Abs had antigen-binding sites that carry little charged residues or even predominantly negative charges ( Figures 5 , 6 ).

To confirm and broaden these observations, we further analyzed the structural and physicochemical characteristics of polyreactive Abs that bind to an identical epitope. A large number of broadly neutralizing HIV-1 Abs (bNAbs) are known to manifest antigen-binding polyreactivity (12, 13, 52). The Abs b12, 12a21, 3BNC117, CH103, 45-46m2, and VRC07 all bind to the same region on the HIV-1 envelope protein, i.e., gp120 CD4 binding site and were reported to be polyreactive (12, 13, 52, 53). Importantly, the structures of Fab regions of these Abs have been determined by X-ray analyses (54–58). Our comparative analyses of the topology and the physiochemical properties of antigen-binding sites of a set of polyreactive bNAbs demonstrated that there is not a common structural pattern that could explain their promiscuous antigen binding. Thus, b12, 45-46m2, and VRC07 have long and protruding CDR H3 loops ( Figure 7 ), whereas other polyreactive bNAbs have short and less exposed CDR H3. The polyreactive bNAbs also substantially differ in their three-dimensional electrostatic potentials and in the distribution of charges in their antigen-binding sites ( Figure 7 ). Thus, Abs 45-46m2 and VRC07 have predominantly positively charged antigen-binding sites with a potent positive electrostatic potential. However, the positive patches could not explain the polyreactivity of b12, 12a21, and 3BNC117, as these Abs have a more balanced distribution of the positive and the negative charges in their antigen-binding sites ( Figure 7 ). These data corroborate the results from our analyses of the physicochemical properties of the in silico modeled structures of V regions of polyreactive Abs ( Figures 4 – 6 ). Moreover, they indicate that even among the polyreactive Abs targeting the same region in antigen, there could be distinct molecular mechanisms for the attainment of promiscuous antigen binding.

Collectively, the data from this part of the study strongly suggest that the polyreactive Abs do not use universal topological or physiochemical qualities that determine the binding to multiple unrelated antigens. Rather, variable regions with diverse properties can endow Abs with binding promiscuity.

We used data sets from two studies (27, 37) where polyreactivity of human monoclonal Abs was assessed. The study of Shehata et al. published the sequence information of all monoclonal Abs. The sequence information for Abs in the study of Prigent et al. was kindly provided by Dr. Hugo Mouquet (Institute Pasteur, Paris, France). Since the polyreactivity in the study of Prigent et al. was displayed in semi-quantitative value, we assigned numerical values based on the number of recognized antigens from the ELISA panel. Since the polyreactivity in the study of Prigent et al. was displayed in a semi-quantitative value, we assigned a polyreactivity score as based on the number of antigens for which each antibody was considered as reactive by authors (score ranking from 0 to 4 antigens)

The identification of CDR regions and assessment of the number of amino acid replacements in V domains was performed by IMGT/V-Quest alignment software (https://www.imgt.org/). The GRAVY index was determined with the GRAVY Calculator (https://www.gravy-calculator.de).The polyreactivity of Abs was correlated with different features of variable regions—the number of somatic mutations in V_H and V_L domains; the length of CDR loops; the number of charged, polar, aromatic, and hydrophobic residues in the CDR and FW regions; and the frequency of individual amino acid residues in the CDR and FW regions. The correlation analyses were performed by non-parametric Spearman’s rank-order analysis using GraphPad Prism v.10 software (La Jolla, CA). A significant correlation was considered only with p value ≤ 0.05.

For modeling structures of the V regions of selected Abs that manifested the highest level of antigen-binding polyreactivity, we applied Ab structure modeling algorithm—RosettaAntibody3 (71–74). The sequences of V regions were submitted to the Rossie-2 module of the Rosetta online server (https://rosie.rosettacommons.org/). Relaxed three-dimensional models with lowest energy of coupled V_H and V_L domains were visualized by using the Chimera UCSF Chimera package. The same software was used for coloring the electrostatic charges of the molecular surfaces of the antibodies. Chimera is developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco (supported by NIGMS P41-GM103311) (75). The three-dimensional Coulombic electrostatic potential was calculated and visualized by using Swiss-PdbViewer v. 4.1, http://www.expasy.org/spdbv/ (76).

Structural coordinates of Fab fragments determined by X-ray crystallography were obtained from the protein data bank (https://www.rcsb.org/) for the following polyreactive HIV-1 bNAbs: b12, 12a21, 3BNC117, CH103, 45-46m2, and VRC07 (PDB files: 2NY7, 4JPW, 4JPV, 4JAM, 4JKP, and 4OLU, respectively). The structural configuration and the electrostatic properties of V regions of these Abs were visualized with the same software packages as indicated before.