A checkpoint for autoreactivity in human IgM⁺ memory B cell development

To characterize the antibodies produced by IgM⁺ memory B cells, we purified single cells from peripheral blood of three healthy donors and analyzed their antibody heavy and light chains. In agreement with previous reports, we found that IgM⁺ memory B cells showed a significant increase in VH3 gene family representation and a decrease in JH6 gene usage when compared with mature naive B cells from the same donors (Fig. 1, A–C, and references 23–26). As expected, the decreased JH6 usage was associated with shorter CDR3s (Fig. 1, A–C and reference 25). The numbers of positively charged residues in IgH CDR3s, as well as Igκ and Igλ light chain V and J gene usage, were all comparable in naive and IgM⁺ memory B cells (Fig. S1). We conclude that antibodies expressed by IgM⁺ memory differ from naive B cells only in VH3 and JH6 usage. 94.7% of IgM⁺ memory antibodies were somatically mutated (124/131; Tables S1–S3) with a mutation frequency of 3.20% for IgVH, 1.62% for IgVκ, and 1.61% for IgVλ genes (Fig. 1, D and E). Replacement mutations were enriched over silent mutations in CDRs compared with framework regions (FWRs) 1–3 (Fig. 1 E). Thus, antibodies expressed by IgM⁺ memory B cells show signs of antigen-mediated selection.

The self-reactivity profiles of IgM⁺ memory and naive B cell antibodies from the same donors were compared in ELISAs using lysates of the human cell line HEp-2 (Fig. 2 A and reference 2 for GO and JB). On average, 19.7% of the antibodies from mature naive B cells from three donors JH (17.6%; Fig. 2 A), GO (22.8% and reference 2), and JB (16.7% and reference 2) were self-reactive (Fig. 2 A and Tables S4–S6). In contrast, only 2.1% of antibodies expressed by IgM⁺ memory B cells from the same individuals were self-reactive (0% for JH and JB, 4.8% for GO; Fig. 2 A and Tables S1–S3). This result was confirmed by indirect immunofluorescence assay (IFA) with HEp-2 cell–coated slides (unpublished data). Thus, the frequency of self-reactive antibodies expressed by IgM⁺ memory B cells was significantly lower than that measured in mature naive B cells (Fig. 2 A; P < 0.0001 for all donors combined; P = 0.028 for JH, P = 0.033 for JB, and P = 0.021 for GO). We conclude that B cells expressing self-reactive antibodies are efficiently excluded from the IgM⁺ memory B cell pool.

To measure the frequency of polyreactive antibodies in mature naive and IgM⁺ memory B cells, we performed ELISAs on ss/dsDNA, LPS, or insulin (Fig. 2 B). As previously reported, 5.4% (5/93) of the antibodies cloned from mature naive B cells from donors GO and JB showed low levels of reactivity with at least two of these antigens (2). Antibodies cloned from mature naive B cells of JH were not significantly different from the previously published result showing 8.3% polyreactivity (3/36, P = 0.685; Table S6). On average, 6.2% of mature naive B cell antibodies were polyreactive (8/129; Fig. 2 B and reference 2). In contrast, only 1% of antibodies cloned from IgM⁺ memory B cells were polyreactive (1/105, P = 0.044; Fig. 2 B). In summary, polyreactive and self-reactive antibodies comprising 5% and 20% of the mature naive repertoire, respectively, are rarely found in the IgM⁺ memory antibody pool.

Our experiments indicate that the transition to the IgM⁺ memory B cell compartment is associated with selection against self-reactivity. To determine whether selection occurs before or after somatic hypermutation, we reverted the somatic mutations of 27 IgM⁺ memory B cell antibodies to their unmutated germline form and tested the recombinant antibodies for self-reactivity and polyreactivity (Fig. 3, Fig. S2, and Table S7). Although an average of 19.7% of all antibodies produced by naive B cells were self-reactive (Fig. 2 A and reference 2), none of the 27 reverted antibodies displayed self-reactivity or polyreactivity, including the two self-reactive and one polyreactive IgM⁺ memory B cell antibodies (Fig. 3 and Table S7). We conclude that naive B cells expressing self-reactive antibodies do not contribute to the IgM⁺ memory B cell compartment. Furthermore, the rare IgM⁺ memory B cells that produce antibodies with low levels of self-reactivity acquire this reactivity by somatic hypermutation.

To determine whether antibodies cloned from IgM⁺ memory B cells react with bacteria or T-I bacterial antigens, we performed ELISAs using Streptococcus pyogenes, a mixture of 10 Streptococcus pneumoniae serotypes (1, 3–6, 9V, 14, 18, 19, 23F); Enterococcus faecalis; Escherichia coli, a protein A mutant (SpA-) strain of Staphylococcus aureus (Fig. 4 A); or bacterial polysaccharide antigens in human vaccines PNEUMOVAX 23 (Streptococcus pneumoniae), PedvaxHIB (Haemophilus influenzae), and Menomune (Neisseria meningitidis) (Fig. 4 B).

As expected, polyreactive antibodies from mature naive (4.3%, 4/91) and IgM⁺ memory B cells (1%, 1/105) showed high levels of reactivity with whole bacteria and bacterial antigens (Fig. 4 and Tables S1–S6). In addition, we found a new group of antibodies that showed broad reactivity with bacteria or bacterial polysaccharide antigens, but not with DNA, LPS, or insulin (Fig. 4 and Tables S1–S6). These antibodies comprised 8.8% (8/91) of the mature naive but only 2.9% (3/105) of the IgM⁺ memory B cell repertoire (Fig. 4 C). When combined, the broadly bacterially reactive antibodies comprised 13.1% of the mature naive and 3.9% of the IgM⁺ memory B cell compartment; overall, IgM⁺ memory B cells show a lower frequency of broadly bacterially reactive antibodies than mature naive B cells (P = 0.0195).

Among 91 antibodies from naive B cells and 105 antibodies from memory B cells tested, we found only two antibodies with high levels of specific reactivity against bacterial antigens. Both antibodies were cloned from IgM⁺ memory B cells from donor JB (2/27, 7.4%) who had previously been vaccinated and both recognized polysaccharides from the serotype 11A vaccine strain of Streptococcus pneumoniae contained in PNEUMOVAX 23 (Fig. 4 B, top right; Tables S1–S3, and Fig. S3).

VH3 genes are overrepresented in IgM⁺ memory B cells and several members of the VH3 gene family bind to modified Staphylococcus protein A (mSpA) in an Fc-independent manner (27, 28). This association led to the suggestion that SpA binding is positively selected in IgM⁺ memory B cells (29). To determine if SpA binding is involved in selection, we compared VH3 antibodies cloned from naive and IgM⁺ memory B cells for mSpA binding by ELISA (Fig. 5, A and B, and Fig. S4). We found that 70.0% of VH3-positive antibodies from mature naive B cells showed mSpA binding. In contrast, only 54.3% of VH3-positive antibodies from IgM⁺ memory B cells were SpA reactive (Fig. 5, A and B, Tables S1–S3, and Fig. S4).

To further determine if non–SpA-reactive antibodies from IgM⁺ memory B cells had lost SpA binding as a result of somatic mutation, we reverted 27 of these antibodies into their unmutated counterparts. After reversal of V gene mutations, 10 out of the 27 reverted antibodies became strongly mSpA reactive (Fig. 5 C). Our results suggest that SpA binding is not a selective advantage because it is frequently lost in the transition between the naive and IgM⁺ memory B cell compartments as a result of V gene somatic hypermutation.

Antibody genes in IgM⁺ memory B cells are somatically mutated, and show signs of antigen-mediated selection with increased numbers of replacement mutations in IgH and IgL chain CDRs (5, 11, 23–25, 30–34). However, the development of somatically mutated IgM⁺ memory B cells does not require CD40L or CD40 and, therefore, these cells appear to be germinal center independent (33). Phenotypically, IgM⁺ memory B cells resemble MZ B cells and are thought to be their circulating counterparts (11). In support of this idea, IgM⁺ memory B cells are strongly reduced in splenectomized and asplenic patients and cannot be detected in infants, where the marginal zone is still populated with mature naive B cells (10, 35). In all of these cases and in a sub-group of common variable immunodeficiency (CVID) patients, loss of circulating IgM⁺ memory B cells is associated with a specific increase in susceptibility to infection with encapsulated bacteria (10, 11). Consistent with the idea that circulating IgM⁺ memory B cells play an important role in these infections, we found antibodies with high specificity for streptococcal polysaccharides enriched in this B cell compartment after immunization with PNEUMOVAX 23 (9–11).

In mice, MZ B cell development is promoted by transgenic expression of B cell antigen receptors with low levels of self-reactivity or polyreactivity (15–20). For example, MZ B cells but few follicular B cells develop in transgenic mice expressing a polyreactive VH81x heavy chain (15, 16). Mouse MZ B cells are optimally anatomically positioned to react with blood-borne pathogens and expression of low affinity broadly reactive antibodies by these cells would be consistent with a protective role early in infection (13, 14).

Human IgM⁺ memory B cells are enriched in VH3 antibodies, some of which bind SpA (23–25, 27, 28). Based on sequence information, it was proposed that this might reflect selection into the IgM⁺ memory/MZ B cell compartment for cells reactive with common blood-borne pathogens (29). However, we find that VH3 antibodies cloned from IgM⁺ memory B cells are not more SpA reactive than VH3 antibodies from mature naive B cells and, in many cases, have lost SpA reactivity by somatic mutation. In addition, broadly bacterially reactive antibodies are selected against during IgM⁺ memory B cell development. These observations were surprising because they appear to be at odds with the transgenic mouse experiments showing that B cells expressing polyreactive antibodies are selected into the MZ B cell compartment (15–20). However, the reactivity of marginal zone B cells has never been measured in normal mice or humans and experiments with anti–hen egg lysozyme (HEL) transgenic mice show that the development, half-life, and anatomic distribution of self-reactive B cells is altered in the absence of competition (36, 37). For example, anti-HEL transgenic B cells develop normally in the presence of HEL and fill the follicular B cell compartments; however, the same B cells fail to enter the follicular compartment when exposed to HEL in the presence of normal B cells (36, 37). Thus, the repertoire of authentic MZ B cells developing in the presence of a complete B cell repertoire may differ from those that arise in transgenic mice in the absence of competition. Alternatively, MZ B cell selection might differ from mouse to human, or the circulating IgM⁺ memory B cell repertoire could represent a subset of cells entirely different from the phenotypically similar MZ B cells in the human spleen.

Most newly arising antibodies are self-reactive, but many of these are removed from the repertoire during B cell development in the bone marrow and the periphery (1, 2). Nevertheless, 20% of mature naive B cells in healthy humans express antibodies with low levels of self-reactivity (2). Despite the prevalence of self-reactive B cells in the naive B cell compartment, these potentially autotoxic antibodies are not likely to make a significant contribution to the circulating repertoire because naive B cells do not secrete high levels of antibodies. In contrast with naive B cells, IgM⁺ memory B cells are readily induced to proliferate in an Ig-independent manner and differentiate into antibody-secreting plasmablasts (38–41); therefore, any autoantibody producing IgM⁺ memory B cells would more likely be autotoxic. Our experiments suggest that, under physiologic conditions, this possibility is averted by a third checkpoint for B cell tolerance between the naive and IgM⁺ memory compartment imposed before B cells undergo somatic hypermutation.

The study was performed in accordance with institutional review board–reviewed protocols of The Rockefeller University and all samples were obtained after signed informed consent. Single cell sorting was performed as described previously (2, 42). In brief, peripheral blood B cells were enriched by incubation with RossetteSep (Stem Cell Technology, Inc.) followed by Ficoll-Paque Plus (GE Healthcare) gradient centrifugation. Enriched B cells were stained with anti–human CD19-allophycocyanin, anti–human CD10-PE, anti–human CD27-FITC, and anti–human IgM-biotin (BD Biosciences), and biotinylated anti–human IgM antibody was visualized using Streptavidin-PeCy7 (Caltag Laboratories). Single mature naive B cells (CD19⁺CD10⁻IgM⁺CD27⁻) and IgM⁺ memory B cells (CD19⁺CD10⁻IgM⁺CD27⁺) were isolated by FACS into 96-well plates containing 4 μl of 0.5 × PBS, 10 mM DTT, 8 U RNAsin (Promega), and 3 U Prime RNase Inhibitor (Eppendorf) using a FACSvantage (Becton Dickinson). All samples were immediately frozen on dry ice and stored at −70°C.

Antibodies were cloned as described previously (2, 42). In brief, single cell cDNA was synthesized in the original sort plates in a total volume of 14.5 μl (4 μl sort lysis solution + 10.5 μl RT-PCR reaction mix containing 150 ng random hexamer primer [pd{N}₆]; GE Healthcare), 0.5% vol/vol Igepal CA-630 (NP40; Sigma-Aldrich), 7 mM DTT, 4 U RNAsin (Promega), 7.5 U Primer RNase Inhibitor (Eppendorf), 1.2 mM each dNTP, and 50 U Superscript II RT (Invitrogen). cDNA was synthesized at 42°C for 55 min. Individual IgH (μ) and IgL chain (κ or λ) gene rearrangements were amplified in two successive rounds of PCR (50 cycles each) using primers as described previously (2, 42). Ig genes were amplified from 3.5 μl cDNA or first PCR product in 40 μl of total reaction volume with 0.2 μM 5′ and 3′ primers or primer mixes, 312.5 μM each dNTP and 1 U Hotstar Taq DNA polymerase (QIAGEN). All PCR products were purified (Qiaquick; QIAGEN), sequenced, and analyzed by Ig BLAST comparison with GenBank. Digested PCR products were purified and cloned into human IgG1, Igκ, or Igλ expression vectors (2, 42). Plasmids were sequenced to select clones with inserts identical to the original PCR product.

Recombinant antibodies were produced as described previously (2, 42). In brief, 293A human embryonic kidney cells were cultured in DMEM supplemented with 10% ultra-low IgG FCS (Invitrogen) and cotransfected with 12.5 μg of IgH and IgL chain encoding plasmid DNA by calcium phosphate precipitation. Cells were washed with serum-free DMEM for 8-12 h after transfection and thereafter cultured in DMEM supplemented with 1% Nutridoma SP (Roche). Supernatants were collected after 7 d of culture. For self-reactive HEp-2 ELISAs, antibodies were purified with protein G–Sepharose (GE Healthcare).

Streptococcus pneumoniae strains DCC1355 (serotype 19), DCC1335 (9V), DCC1420 (23F), DCC1490 (14), DCC1714 (3), DCC1850 (6), AR314 (5), AR620 (1), GB2017 (18), GB2092 (4), and Enterococcus faecalis V12 were grown in Brain Heart Infusion broth (Sigma-Aldrich). Streptococcus pyogenes D471 was grown in Todd-Hewitt broth (Sigma-Aldrich). Staphylococcus aureus protein A-negative Foster strain 8325–4 spa::Ebr was grown in super broth (10 g/liter MOPS, 30 g/liter tryptone, 20 g/liter yeast extract). Escherichia coli was grown in Luria-Bertani medium. Cell counts from log-phase bacteria cultures were determined before treatment with 50 μg/ml mitomycin-C for 1 h at 37°C. Whole bacteria were suspended at a concentration of 10⁸/ml PBS for ELISA.

To measure reactivity of antibodies with individual antigens, microtiter plates (COSTAR 9018) were coated with 50 μl/well of 10 μg/ml of ssDNA, dsDNA, LPS (Sigma-Aldrich), or 5 μg/ml of recombinant human insulin (Sigma-Aldrich) or 10 μg/ml of human bacterial polysaccharide vaccines PNEUMOVAX 23 (Merck & Co.), PedvaxHIB (Merck & Co.), or Menomune (Aventis) or 10 μg/ml of modified protein A. To measure antibody reactivity with specific bacteria, microtiter plates were coated with 100 μl/well of 100 μg/ml poly-l-lysine, 25 μl/well of bacteria solution, and 25 μl/well of 0.2% gluteraldehyde were added at room temperature for 20 min. Plates were centrifuged at 1,500 g for 20 min and washed with PBS. To block aldehydes, 50 μl/well of 0.1 M l- lysine in PBS was added and wells were washed with PBS.

Antibody concentrations in supernatants and after purification were determined by anti–human IgG1 ELISA using human monoclonal IgG1 as standard (Sigma-Aldrich). For polyreactivity and bacterial ELISAs, supernatants were adjusted to a starting antibody concentration of 1 μg/ml and with three subsequent 1:4 dilutions in PBS. Samples were considered negative if the OD₄₀₅ did not exceed a threshold value at any of the four dilutions in at least two independent experiments as indicated in each graph. In all assays, threshold values were set using the negative control antibodies mGO53, low positive control eiJB40, and high polyreactive antibody ED38 (2, 43). HEp-2 cell self-reactivity ELISAs used QUANTA Lite ANA ELISA plates (INOVA Diagnostics) and purified antibodies at a concentration of 10 μg/ml with three subsequent 1:4 dilutions in PBS. Positive and negative controls for HEp-2 ELISAs included control sera from patients and healthy individuals (INOVA Diagnostics) as well as ED38 and were included in every experiment. Consistent with our previously published results, the negative and low positive serum was used to determine the cut-off OD₄₀₅ for reactivity in each experiment (2). Cut-off OD₄₀₅ levels vary among experiments depending on the absolute OD₄₀₅ values and antibody concentrations, but allow us to compare the data among individual experiments. Antibodies were considered reactive if they showed reactivity levels above the cut-off OD₄₀₅ in at least two independent experiments and if the results could be confirmed by IFA on HEp-2 cell–coated slides (unpublished data). All ELISAs were developed with horseradish peroxidase–labeled goat anti–human IgG Fc antibodies (Jackson ImmunoResearch Laboratories) and horseradish peroxidase substrates (Bio-Rad Laboratories). OD₄₀₅ was measured with a microplate reader (Molecular Devices).

Antibodies for reversion were chosen randomly and are listed in Table S7. Mutated IgH and IgL chain genes were reverted to their germline sequence by two separate PCR reactions for the V gene and the (D)J gene followed by a third overlap PCR to fuse the two PCR products (44). Germline V genes were amplified from previously cloned Ig genes from unmutated naive B cell templates using FWR1 sense and FWR3 antisense primers. Germline (D)J sequences were generated with sense primers spanning the 3′ end of the V gene and the CDR3 region and antisense primers in the J gene. A second 20-cycle PCR reaction fused the two overlapping V and (D)J PCR fragments to generate unmutated germline antibody sequences (Fig. S2). All unmutated IgH and IgL chain PCR products were sequenced before and after cloning into the corresponding expression vectors to confirm the lack of mutations. For all clones' mutated and unmutated IgH and IgL chain CDR3 regions, Ig gene repertoire and reactivity profiles are summarized in Table S7.

p-values for Ig gene repertoire analysis, analysis of positive charges in IgH CDR3, and antibody reactive were calculated by 2 × 2 of 2 × 5 Fisher's Exact test or Chi-square test. p-values for IgH CDR3 length were calculated by two-tailed Student's t test.

Fig. S1 shows positive charges in IgH CDR3, IgV, and IgJ gene usage for Igκ and Igλ light chains from mature naive and IgM⁺ memory B cell antibodies. Fig. S2 shows the PCR-mediated reversion strategy of mutated IgM⁺ memory B cell antibodies into their unmutated germline forms. Fig. S3 shows specificity with serotype 11A polysaccharides from Strep. pneumoniae for PNEUMOVAX 23–reactive antibodies from donor JB. Fig. S4 shows ELISA reactivity with mSpA for VH3 family antibodies from IgM⁺ memory B cells and their unmutated counterparts. Tables S1–S6 show IgH and IgL chain characteristics and antibody reactivities with ssDNA, dsDNA, LPS, insulin, HEp-2, mSpA, Staph. aureus (A-), Strep. pyogenes, Strep. pneumoniae, Ent. faecalis, E. coli, PNEUMOVAX 23, PedvaxHIB, and Menomune from IgM+ memory B cells and mature naive B cells of donors JB, GO, and JH. Table S7 shows IgH and IgL chain gene repertoire, reactivity profiles, and mutated and unmutated CDR3 sequence of antibodies selected for reversion.

We thank Dr. G.J. Silverman, Dr. C.S. Goodyear, Dr. V.A. Fischetti, Dr. J.M. Loeffler, Dr. D. Ho, and Dr. Y. Huang for reagents and advice, and all members of the Nussenzweig laboratory and Dr. E. Besmer for discussion and critical reading of the manuscript.

This work was supported by grants from the National Institutes of Health (to M.C. Nussenzweig and S. Yurasov), the Dana Foundation (to H. Wardemann), and the Naito Foundation (to M. Tsuiji). M.C. Nussenzweig is a Howard Hughes Medical Investigator and S. Yurasov is a New York Chapter Arthritis Foundation Young Scholar.

The authors have no conflicting financial interests.