Rheumatoid Factor Specificity of a VH3-Encoded Antibody Is Dependent on the Heavy Chain CDR3 Region and Is Independent of Protein A Binding¹

Rheumatoid factors (RF)³, autoantibodies that bind to the Fc region of IgG, are commonly found in patients with rheumatoid arthritis (RA) and contribute to synovial inflammation (1). Although low affinity polyreactive IgM RF are commonly found in normal individuals immunized with tetanus toxoid or blood group Ags, the RF of RA patients are of higher affinity, are less polyreactive, are more heterogeneous in their molecular genetic origins and binding characteristics, and have often undergone class switching to the IgG isotype (2, 3).

We have recently undertaken molecular genetic characterization of a panel of monoclonal RF derived by EBV transformation of peripheral B cells of RA patients. Our analysis of the light chains of these Abs revealed that RF activity was primarily dependent on the heavy chain but that the light chain VJ junction also made a major contribution to RF specificity. Reactivity with the RF-associated anti-Id 4C9 was relatively independent of heavy chain usage and was mapped to the VJ junction of Abs using VκIIIa light chain genes (4).

Previous studies have shown that the heavy chain complementarity-determining region 3 (CDR3) region contributes to low affinity polyreactivity, including RF specificity of “natural” autoantibodies (5, 6). Similar findings were reported for a high affinity RF derived from a patient with type II mixed cryoglobulinemia (7). To determine whether this heavy chain region contributes to the specificity of high affinity RA-associated RF, we have identified a high affinity polyreactive IgM RF from a patient with RA (B′20), a low affinity polyreactive IgM RF from a patient with RA (RC1), and an idiotypically related IgM Ab (B9601) from a patient with systemic lupus erythematosus that is neither polyreactive with the panel of Ags tested nor RF positive (4, 8). All three lines use VH3 genes and two (B′20 and RC1) bind to the VH3 superantigen protein A.

To define the molecular basis for RF activity, protein A binding, polyreactivity, and 4C9 idiotypic specificity, heavy chains with site-directed mutations and CDR3 switches were generated and paired with the germline counterpart of the B′20 light chain derived from an idiotypically related cell line MF8. RF specificity of the Abs was found to be dependent on the heavy chain CDR3 region. Superantigen specificity was independent of CDR3 and was abrogated by a single amino acid change in FR3. Polyreactivity was lost as a result of class switching to IgA, but could be reconstituted by using immune complex lattices in an enhanced IgA ELISA. Under these circumstances, polyreactivity was also found to be dependent on the heavy chain CDR3 region. Our study points to the critical contribution of the somatically generated heavy chain CDR3 to high affinity RF autoantibody specificity of B′20 as well as to low affinity polyreactivity.

The parental IgM lines MF8, RC1, and B′20 were generated by EBV transformation of peripheral blood lymphocytes from three RA patients, and the B9601 line was generated by fusion of peripheral B cells of a patient with systemic lupus erythematosus to GM4782. The sequences and antigenic specificities of the cell lines have been previously reported and are summarized in Table I. RC1 is polyreactive, binding with low affinity to Fc (5 × 10⁻⁵ K_d mol/l), tetanus toxoid, rabbit IgG, and ssDNA (8). B′20 is also polyreactive, but binds with high affinity to human Fc (1.8 × 10⁻⁷ K_d mol/l) and lower affinity to rabbit IgG, ssDNA and tetanus toxoid (4, 8). Neither MF8 nor B9601 binds Fc (4, 8). Of the VH3-encoded lines, B′20 and RC1 bind to protein A, but the B9601 line does not. B′20, MF8, and B9601 use VκIIIa-encoded light chains and are positive for the 4C9 Id (8). The MF8 light chain is the germline counterpart of the B′20 light chain (8).

DNA sequences of the heavy and light chain genes of these lines have been previously reported (8, 9). B′20 and B9601 derive from different VH3 germline genes but differ from each other in the V_H region by only 9 amino acids (Fig. 1). B′20 has two mutations from its most homologous germline VH3 gene (3-23/DP47/VH26) and B9601 has only one. RC1 uses a different VH3 gene (DP46/hv3005). Recombination of the B′20 heavy chain with the B9601 light chain results in an Ab with RF specificity, demonstrating that it is the B9601 heavy chain that is not permissive for RF activity (4). The B′20 light chain has two somatic mutations in the V_L region but is otherwise homologous to the germline-encoded (L16/Humkv328) MF8 light chain. Recombination of B′20 with the MF8 light chain results in an Ab that has slightly higher affinity for Fc than the original B′20 (4) and identical protein A binding (data not shown). For this reason the germline-encoded MF8 light chain was used in all the recombination experiments described in this report.

Amplification of the B9601 and B′20 heavy chain genes was conducted utilizing PCR as previously described (4). Six heavy chain constructs were generated, comprising the original B′20 cell line DNA, and chimeras of the B′20, B9601, and RC1 heavy chain genes as shown in Figure 1. The FR3-J chimera was constructed by recombinant PCR as previously described. Two fragments, one corresponding to the B′20 FR1-FR3 and one corresponding to the B9601 FR3-J, were generated from the cloned B′20 and B9601 heavy chain genes, respectively. The fragments for the 5′ and 3′ end of the chimera were designed so that there was a 30-bp overlap in FR3. The fragments were mixed in a 1:1 molar ratio and annealed by heating to 94°C, slowly cooling to 37°C, and then extending for 7 min at 72°C. For the recombinant PCR, the B′20 leader sequence primer and the B9601H 3′ primer were utilized to PCR the full-length heavy chain gene.

Because the initial data with this construct showed that the B′20 FR3-J region was responsible both for RF specificity and protein A binding, additional chimeras were generated with alterations in these domains. For these experiments mutations and domain switches were introduced by elongating the B′20 V_H fragment with successive 3′ oligonucleotides containing the replacements of interest. Two rounds of overlapping PCR using successive 3′ primers were performed to create a B′20 heavy chain, with the V segment of B′20, the CDR3 segment of B9601, and J segment of B′20 (named B9601D). Recombinant PCR was then used as above to introduce a serine into position 82a of FR3 of this construct (B9601-FR3*-D). To examine the effect of CDR3 region length on RF specificity, the B9601 CDR3 region was also shortened by removing the last 4 amino acids to make B9601-D*. Three rounds of PCR were performed to clone a heavy chain with the FR1-CDR2 region of B′20, the FR3 of B9601, and the DJ region of B′20 (named B9601-FR3). Finally, to examine the effect of the CDR3 region on polyreactivity, the CDR3 region of RC1 was switched into B′20 making RC1-D (see Fig. 2).

The heavy chain gene from B′20 and the light chain gene from MF8 were cloned in the PAH 4834 IgA heavy chain and PAG 4828 κ light chain vectors respectively and cotransfected into the NSO-Bcl-2 cell line as previously described (4, 10). The IgA construct does not allow us to examine any specificities that depend on the increased relative avidity of the pentameric IgM, but we have previously shown that the B′20/MF8 IgA construct retains Fc binding. After 8 to 10 days the clones were screened for IgA and κ light chain production by ELISA as previously described (4). All mutant heavy chain constructs were sequenced and then cloned into the PAH heavy chain vector. Mutant cell lines were generated as above by cotransfection of the mutant constructs with the MF8 light chain.

The amount of IgA in all the supernatants was quantitated using an ELISA as previously described. An IgA myeloma protein of known concentration was used as control (4).

RF activity was measured by ELISA. Because the constructs are of the IgA and not the IgM isotype, we used an assay for RF activity as previously described in which aggregated IgG or isolated Fc was used as substrate and the IgA molecules from the transfectant supernatants were preincubated with F(ab′)₂ anti-IgA before incubation with the IgG substrate. The assay using aggregated IgG as substrate is able to detect lower avidity Abs than the assay using Fc as substrate and was easily able to detect binding of an IgA construct where the affinity of the parental IgM was 1.3 × 10⁻⁶K_d mol/l (4). Supernatants from the transfected cell lines normalized to concentrations of IgA from 10 to 0.625 μg of IgA/ml were preincubated with 1/5000 dilution of peroxidase conjugated F(ab′)₂ goat anti-human IgA (Accurate Antibodies, Westbury NY) at 37°C for 1 h, and the immune complexes were then applied to IgG or Fc coated plates at room temperature for 2 h (4). The plates were then washed and developed with ABTS substrate (Kirkegaard & Perry, Gaithersburg MD).

Protein A binding was detected by ELISA. Protein A (Sigma, St. Louis MO) was coated to microtiter wells at 1 μg/ml in PBS. After blocking, the normalized supernatants were added to the wells, followed by peroxidase conjugated F(ab′)₂ anti-κ Ab (Accurate Antibodies) and substrate as above.

4C9 reactivity was assayed by ELISA as previously described (4).

Assays for polyreactivity were performed by ELISA. The following Ags were used in ELISA assays as previously described: ssDNA, tetanus toxoid, rabbit IgG, and BSA (8). The transfectant supernatants were preincubated with anti-IgA and applied to the plates for 2 h at room temperature as above. In addition, an enhanced assay was used in which optimal concentrations of protein A purified transfectant IgA (from protein A positive lines) and anti-IgA were determined for maximal binding of B′20/MF8 to ssDNA coated plates. The optimal molar ratio was found to be 1:2.7. In these assays, protein A purified transfectant IgA at 1 μg/ml was preincubated with peroxidase conjugated F(ab′₂ anti-IgA at 1.6 μg/ml and then applied to plates coated with ssDNA for 2 h, followed by washing and development with 2,2′-azino-di[3 ethyl benzthiazoline sulfonate] (ABTS) substrate.

Initial experiments with the FR3-J construct comprising FR1-CDR2 of B′20 and FR3-J of B9601 showed that it had completely lost specificity both for Fc and protein A, demonstrating that the B′20 heavy chain FR3 or CDR3 region was responsible for these antigenic specificities. Further mutations were therefore designed to examine which of these regions was responsible for Fc and Protein A binding. These are illustrated in Figure 2.

The results of RF activity of the transfectants using aggregated IgG or Fc substrate are shown in Table II. Results are shown at a concentration of 2.5 μg/ml of IgA, the concentration at which the RF positive Abs reached the plateau of their Fc binding curves. Only two of the constructs were able to bind to IgG and Fc, namely the original B′20/MF8, as previously described (4), and the B9601-FR3 that differs from B′20 by only 2 amino acids at positions 82a and 93 of FR3. The binding curves of these two Abs were similar (data not shown). All the other heavy chain constructs containing the B9601 CDR3 region or the RC1 CDR3 region were RF negative (<10% maximal binding) even at maximal concentrations or using the enhanced assay described above. Binding to rabbit IgG followed the same pattern as binding to human IgG.

These results show that RF specificity is dependent on the B′20 CDR3 region. The B9601-D* construct was then generated to examine the effect of CDR3 length on RF specificity. This construct has a deletion of the last 4 amino acids of the B9601D, yielding a CDR3 region that differs from B′20 by 5 of the 8 amino acids. This construct also failed to exhibit any RF activity.

Protein A binding was shown to be independent of the CDR3 region and could be abrogated by a single amino acid substitution at position 82a in FR3. This result was shown by a comparison of the positive B′20 construct with the negative B9601-FR3 that differs only by 2 amino acids in FR3 and by a comparison of the positive B9601D with the negative B9601-FR3*-D that differs only by the Asn-Ser substitution at position 82a (Fig. 1 and Table II).

An examination of 4C9 reactivity yielded surprising results. All constructs using the B′20 CDR3 region were positive for 4C9 expression. However, all constructs that used the 12 amino acid B9601 CDR3 were negative for 4C9 reactivity. In contrast, substitution of the B′20 CDR3 region with either the RC1 CDR3 region or the B9601 CDR3 region that had been shortened by 4 amino acids was permissive for 4C9 reactivity (Table II).

Analysis of the B′20 heavy chain cotransfected with the MF8 light chain (B′20/MF8) or with the B′20 light chain (B′20/B′20) and of the RC1-D Ab using the standard ELISA showed that none was polyreactive (data not shown), demonstrating that the class switch to IgA resulted in sufficient loss of relative avidity to abrogate low affinity polyreactivity. We next tested protein A purified B′20/MF8, RC1-D, and B9601D Abs in an enhanced assay in which optimal molar ratios of IgA:anti-IgA had been determined for maximal binding of B′20/MF8 to ssDNA. This finding reflects optimal immune complex lattice formation. Under these conditions, B′20/MF8 showed binding to ssDNA whereas B9601D and RC1-D did not bind (Table II).

The diversity of the preimmune Ab repertoire is generated by several mechanisms that depend on the assembly of complete Ig coding sequences from multiple gene segments. The random joining of different V, D, and J heavy chain segments or V and J light chain segments together with random association of heavy and light chains, leads to the production of a large number of Ig heavy and light chains from a much smaller number of V, D, and J genes (11). In addition, the joining of the coding ends of the rearranged gene segments is imprecise due to base additions, base losses and out of frame joining (11). Marked diversity is generated in the heavy chain CDR3 region primarily due to junctional diversity and N region addition at both the V-D and D-J joins, as well as the possible use of multiple gene segments from atypical D region genes (DIR genes), D-D joining, and D inversion (12). Thus the CDR3 regions, particularly that of the heavy chain, contribute the greatest diversity to the Ig molecule.

Two somatic mechanisms can generate autospecificity in the preimmune B cell repertoire. First, particular V/L combinations may be autoreactive even though the individual heavy and light chains when used in a different pair are not. This finding has been clearly shown by Radic et al. (13) for anti-DNA Abs that use the same heavy chain and different light chains, and by us for RF (4). Second, junctional diversity of both heavy and light chains might generate an autoreactive specificity. The latter mechanism is supported by experiments showing that the incidence of natural autoantibodies is markedly diminished in mice with restricted ability to generate junctional diversity due to absence of the tDt enzyme (14).

Our laboratory has previously studied these mechanisms for generating RFs using a gene expression system that allows us to recombine the wild-type or mutant heavy and light chains of Ig derived from RF producing B cell lines. In this system, the V regions of interest are cloned into expression cassettes containing either the IgA or κ light chain constant regions, and the heavy and light chain constructs are cotransfected into a nonsecreting mouse myeloma cell line (4). We have found that the IgA constructs retain the specificity of the parental lines, which can be studied without the contribution of the pentameric IgM constant region to relative avidity. We have previously studied four related B cell lines in great detail. Initial studies showed that RF specificity of these lines is highly dependent on the heavy chain used, but that V/L pairing is also important in generating autospecificity (4). Using the B′20 Ab, we previously showed that pairing of the B′20 heavy chain with a number of different VκIIIa-encoded light chains resulted in a panel of Abs with varying affinity for Fc and that RF specificity and affinity was highly influenced by the somatically generated light chain CDR3 region (4).

In this report we have further analyzed the contribution of the heavy chain domains to RF specificity. Studies of low affinity polyreactive “natural autoantibodies” that are thought to represent those B cells that escape negative selection in the bone marrow have shown that a major contribution to antigenic specificity is donated by the heavy chain CDR3 region. This result has led to a paradigm shift regarding polyreactive self specificity. Rather than being a germline-encoded phenomenon that is dependent on the sequence of the V_H or V_L region, it is instead a somatically generated phenomenon that results from VDJ junctional diversity during formation of the preimmune repertoire (5, 6).

RF from patients with RA have clear differences from those derived from patients with B cell malignancies and those derived from normal individuals, both in their gene usage and in their fine specificity. RFs from patients with B cell malignancies and from normal individuals have a relatively restricted pattern of germline heavy chain gene usage, and VH1 genes are more often used than VH3 genes (15, 16). RA-derived RFs more often use VH3 genes and use a less restricted repertoire. For example, the 3-23 gene that encodes the B′20 heavy chain has not been observed among normal RFs (16). Fine specificity studies have shown that normal and malignancy associated RFs recognize IgG1, -2, and -4, but rarely IgG3, whereas RA-derived RFs often recognize IgG3 (2, 17). Finally, the affinity of normal RFs for IgG is lower than that of RA-derived RFs (16). Examination of the patterns of mutation of VH1 and VH3-encoded RFs has shown further differences. Whereas some high affinity VH1-encoded RA-derived RFs appear to be generated by somatic mutation of lower affinity Abs (18, 19), there is not a preferential accumulation of CDR replacement mutations in VH3-encoded RFs in RA patients (16). These findings suggest that recombination events contribute to high affinity RF specificity, and that in RA there may be defects in negative selection of such RFs (16).

It is of interest that one previous study has examined a high affinity somatically mutated VH1-encoded monoclonal RF from a patient with type II mixed cryoglobulinemia and compared it with a low affinity germline-encoded polyreactive RF that uses the same heavy and light chain genes. In this study, switching of the CDR3-JH region from the high affinity into the low affinity Ab was sufficient to confer high affinity RF activity, whereas polyreactivity was dependent both on CDR3-JH and a germline-encoded residue at position 5 of FR1 (7). Because previous studies have not been performed with prototypic RA-derived RFs, we wished to determine whether similar structural correlates would be found using the B′20 Ab.

For our studies the B′20 Ab was very useful for several reasons. First, we consider it a prototypic RA-derived RF because it is of high affinity with a K_d falling in the midrange of those described for other high affinity RFs (16), it recognizes IgG3 in addition to IgG1, -2, and -4, and its heavy chain is encoded by a VH3 gene not observed among normal RFs. Second, the two somatic mutations in the B′20 light chain do not influence its RF specificity, allowing us to examine the specificity of the B′20 heavy chain when paired with its germline-encoded light chain (4). Third, the heavy chain has only two amino acid differences from the most homologous germline gene, one in FR1 and one in CDR2. Last, the B′20 heavy chain is highly homologous to the B9601 heavy chain that encodes a RF negative Ab (8).

Using chimeras of B′20 and B9601, we have found that the B′20 CDR3 region is instrumental in conferring the RF specificity of the B′20 heavy chain, a finding analogous to that for natural autoantibodies and for the previously described high affinity VH1-encoded RF. The RC1 CDR3 region that derives from a low affinity RF was unable to substitute for the B′20 CDR3 region in conferring RF specificity. Further analysis of the B′20 CDR3 region shows that it is derived from several sources (Fig. 3). First, there are three templated “P” additions at the 5′ end followed by a region of 9 nucleotides that cannot be assigned, and then D6-13 (DN1) gene in germline configuration (20). Although 7 nucleotides in the unassigned region are homologous to the DIR1 gene (21), it is not possible to confidently assign this region to DIR1 (20); in view of the preponderance of G+C residues in this region, it is most likely the result of “N” addition.

Our previous studies have shown that although RF activity of B′20 is maintained when it is recombined with several VκIII-encoded light chains, the 10 amino acid light chain germline-encoded CDR3 region of B′20 is also important in conferring RF activity of B′20 (4). It is therefore tempting to speculate that the B′20 Ab is a high affinity autoantibody generated by junctional diversity in the preimmune repertoire and that there is a defect in negative regulation of such Abs in RA. However, due to the marked heterogeneity of the human heavy chain CDR3 region it is not possible to know with complete certainty whether somatic mutation has occurred in this area. In addition B′20 has two mutations from the most homologous germline gene found by GenBank search (3-23/DP47). These do not confer RF specificity in the absence of the B′20 CDR3 as shown by the B9601D construct. The B′20 FR1 mutation at position 5 results in a substitution (Leu for Val) that is found in many VH3 germline genes and thus may be an allelic difference. We did not evaluate the contribution of the CDR2 mutation (Ser to Gly at position 56) to affinity for Fc in these studies.

It has been suggested previously that both the length and sequence of the CDR3 region are important in conferring RF specificity (22, 23). Shortening of the B9601 CDR3 region to 8 amino acids showed that RF activity is dependent on the 5 amino acid B′20 CDR3 stretch APYSS at positions 97–100A that form the tip of the CDR3 loop. Replacement of this stretch with the more hydrophobic and negatively charged VLEWL from B9601 results in loss of RF specificity. Interestingly, this area of the CDR3 region is 60% homologous to the CDR3 regions of the monoclonal VH3-encoded RFs Pom and Lay that both express GPYVS in this region (24). Further mutational analysis of this area will allow us to more precisely map the Fc binding area of B′20.

Our study of the polyreactivity of the B′20 Ab yielded similar results. Binding to rabbit IgG followed that for human IgG suggesting that this specificity is also contributed by the heavy chain CDR3 region. Inhibition studies (data not shown) demonstrated that binding of B′20 to ssDNA is inhibitable by IgG but not vice versa, suggesting that the binding sites for these two Ags are similar but the affinities are different. The loss of avidity of the IgA construct compared with the original IgM construct resulted in loss of the low affinity polyreactivity of the B′20 Ab in our standard ELISA, but not of high affinity Fc binding. This is different from the findings previously reported by Ichiyoshi and Casali (5), who found that polyreactivity of an IgG variant of a natural moderate affinity autoantibody was only slightly less efficient than that of its IgM counterpart (2- to 10-fold depending on the Ag). It is unclear whether the IgM they studied was pentameric; however, SDS-PAGE analysis of B′20 has shown that it is always pentameric (data not shown). Our studies show that loss of polyreactivity of IgG compared with pentameric IgM Abs might be simply a result of class switching if the starting affinity of the polyreactive Ab is low. Using an enhanced assay in which purified transfectant IgA was preincubated with anti-IgA at a 1:2.7 molar ratio to generate an immune complex lattice, we were able to reconstitute ssDNA binding of the B′20 IgA construct and show that switching of the B′20 heavy chain CDR3 with either 9601D or RC1-D resulted in loss of this specificity. Thus the B′20 CDR3 region is important both for high affinity autoreactivity and for low affinity polyreactivity.

Our study of superantigen specificity of the B′20 Ab showed clearly that protein A binding is abrogated by the nonconservative replacement of the usual Asn by a germline-encoded Ser at position 82a of FR3. Protein A binding is a function of VH3-encoded Abs, and the amino acids that contribute to the protein A binding site of VH3 Abs has been previously extensively mapped. Although a number of sites on VH3 are postulated to contribute to protein A binding, the three main areas implicated are positions 9–27 in FR1, position 57 in CDR2 (25), and positions 74–77 or 74–84 in FR3 (24, 25). These regions are adjacent and form part of a β-pleated sheet on the lateral aspect of the Fab region away from the CDR regions (24). The Ser at position 82a is encoded by the B9601 germline gene and falls outside the FR3 74–77 region implicated by Silverman (25), but within the extended FR3 74–84 area implicated by Sasso and colleagues. Position 82a was postulated by Hillson et al. (24) to be a candidate residue involved in protein A binding. It is of note that none of 57 protein A binding VH3 regions previously reported have a serine at this residue, whereas 3 of 9 protein A negative lines do (24, 25). Thus there is a subset of VH3 genes that does not encode for Protein A binding specificity.

Our examination of 4C9 reactivity of the mutant panel yielded surprising results. We have previously shown that 4C9 reactivity is dependent mostly on the use of a VκIIIa-encoded light chain with a permissive VJ join (4). 4C9 reactivity of the MF8 light chain was retained upon recombination with a number of different heavy chains. However recombination of the MF8 light chain with the B′20 mutants containing the 12 amino acid B9601 CDR3 region abrogated 4C9 reactivity. Shortening of the CDR3 region by 4 amino acids to generate a CDR3 region the same length as the B′20 CDR3 region partially restored 4C9 reactivity. Substitution of the 14 amino acid RC1 CDR3 region was also permissive for 4C9 reactivity. These studies demonstrate the complex effects of heavy chain folding and structure on the configuration of the light chain and show that 4C9 reactivity, like RF specificity, is generated during B cell ontogeny both by junctional diversity and by combinatorial diversity.

Our studies of the B′20 Ab suggest that RFs of moderately high affinity might be generated by combinatorial and junctional diversity in the preimmune repertoire. This binding specificity is independent of superantigen specificity that is dependent on the germline VH3 region. Abnormal bone marrow release, activation, and expansion as well as mutation of autoreactive B cells might all contribute to the generation of a high titer RF response in patients with RA. Low affinity polyreactivity may be lost through class switching that generates a lower avidity Ab. This finding may in part explain why normal individuals rarely exhibit an IgG RF response despite being able to mount significant low affinity IgM RF responses during routine immunizations.

We thank Drs. H. Keiser, C. Putterman, and B. Diamond for critical review of the manuscript.