|Home | About | Journals | Submit | Contact Us | Français|
During immune responses, antibodies are selected for their ability to bind to foreign antigens with high affinity, in part by their ability to undergo homotypic bivalent binding. However, this type of binding is not always possible. For example, the small number of gp140 glycoprotein spikes displayed on the surface of the human immunodeficiency virus (HIV disfavours homotypic bivalent antibody binding1–3. Here we show that during the human antibody response to HIV, somatic mutations that increase antibody affinity also increase breadth and neutralizing potency. Surprisingly, the responding naive and memory B cells produce polyreactive antibodies, which are capable of bivalent heteroligation between one high-affinity anti-HIV-gp140 combining site and a second low-affinity site on another molecular structure on HIV. Although cross-reactivity to self-antigens or polyreactivity is strongly selected against during B-cell development4, it is a common serologic feature of certain infections in humans, including HIV, Epstein-Barr virus and hepatitis C virus. Seventy-five per cent of the 134 monoclonal anti-HIV-gp140 antibodies cloned from six patients5 with high titres of neutralizing antibodies are polyreactive. Despite the low affinity of the polyreactive combining site, heteroligation demonstrably increases the apparent affinity of polyreactive antibodies to HIV.
Surprisingly, however, B cells can re-acquire poly- and self-reactivity during the germinal centre reaction7, and self-reactivity is also associated with serum antibody responses in several persistent infections, including vaccinia virus in mice8 and HIV, Epstein-Barr virus and hepatitis C virus in humans. This strange though relatively common feature of antibodies was first documented in studies of anti-hapten-specific myeloma proteins over three decades ago9; however, its role (if any) in the antibody response to a specific antigen has yet to be explored.
One possible function for polyreactivity would be to increase antibody affinity for a pathogen where simple homotypic bivalent ligation is not feasible. To test this idea we studied antibody response to HIV because its gp140 surface spike is present at a very low density of only 10–15 viral spikes per virion2,3,10,11. Homotypic bivalent binding is therefore unlikely, but an anti-HIV antibody might increase its overall apparent affinity to the virus by heterogenous ligand binding or heteroligation. In this model, one combining site would bind with high affinity to gp140, whereas the other would bind with low affinity to any of several other ligands on the viral surface.
We studied 134 unique anti-gp140 and 52 control non-gp140 reactive antibodies5 (Supplementary Table 1). All of the anti-HIV antibodies were compared with previously reported IgG antibodies cloned from the memory B cells of uninfected controls7, non-gp140 binding antibodies from the same patients5, and previously characterized broadly neutralizing anti-HIV antibodies 2F5, 4E10, 2G12 and b12.
To determine whether anti-HIV antibodies are polyreactive, we measured binding to single-stranded DNA, double-stranded DNA, lipopolysaccharide, insulin and keyhole limpet haemocyanin (KLH) by enzyme-linked immunosorbent assay (ELISA)4 (Fig. 1a, b, Supplementary Figs 1a and 2 and Supplementary Table 1). Seventy-five per cent of all anti-gp140 memory antibodies were polyreactive (P < 0.0001 compared with uninfected controls, Fig. 1b). Although the frequency of polyreactive antibodies varied between patients (from 59 to 82%), the difference between gp140 binding and non-binding antibodies was highly significant in all cases, whether compared with all antibodies or with those within an individual (Fig. 1b and Supplementary Fig. 1a). Cardiolipin binding was highly correlated with polyreactivity (Fig. 1c and Supplementary Figs 1b and 3, P < 0.0001 compared with control Fig. 1c and Supplementary Fig. 4). Similarly, anti-gp140 antibodies were more reactive in the HEp-2 ELISA than matched or historical controls (51% anti-gp140+ versus 33% gp140−, P = 0.033, Fig. 1d and Supplementary Fig. 1c). We conclude that the majority of all anti-gp140 antibodies studied are polyreactive and more likely to bind to cardiolipin and self-antigens in HEp-2 cell lysates compared with control antibodies.
To determine whether polyreactivity, cardiolipin or HEp-2 reactivity is associated with binding to specific epitopes on gp140, we segregated the ELISA results according to reactivity with gp120, gp41, the CD4-binding site (CD4bs), the CD4-induced site (CD4i), the variable loops and the gp120 core antigen (gp120core5). Anti-gp41 antibodies were more reactive than anti-gp120 in all assays (Fig. 2a). Among the anti-gp120s, antibodies to gp120core, CD4bs and the variable loops were more likely to be poly-, cardiolipin- or HEp-2 reactive than antibodies to CD4i (Fig. 2a).
Several features of antibodies have been associated with polyreactivity, including the length, charge and hydrophobicity of the IgH complementarity-determining region 3 (CDR3). However, these features have never been examined in the context of polyreactive antibodies that specifically recognize a particular antigen12,13. We found that none of these features were highly correlated with polyreactivity in the anti-gp140 antibodies (Fig. 2b).
Anti-gp140 antibodies are highly somatically mutated5. To examine the relationship between somatic mutation and antibody specificity, we reverted the somatic mutations in 20 randomly selected anti-gp140 antibodies and tested them for specific reactivity and polyreactivity. Gp140 binding was undetectable in 12 out of 20 reverted antibodies (Fig. 3a). The eight antibodies that retained gp140 binding showed lower binding activity in ELISA, consistent with a loss in binding affinity (Fig. 3a, Supplementary Fig. 5 and Supplementary Table 2). When tested for neutralizing activity, five of the eight reverted antibodies that retained residual gp140 binding showed complete loss of activity (Supplementary Tables 2 and 3). The other three showed loss of breadth and potency (Supplementary Table 3). We conclude that somatic hypermutation enhances the affinity and neutralizing activity of the anti-gp140 antibodies cloned from memory B cells of the selected patients with high serum titres of broadly neutralizing antibodies.
To examine the relationship between polyreactivity and somatic mutation, we measured the polyreactivity of the reverted antibodies (Supplementary Table 2). Fifteen of the 20 anti-gp140 somatically mutated antibodies were polyreactive before reversion (Fig. 3b, c). After reversion, 12 were still polyreactive (Fig. 3b, 3c and Supplementary Figs 2d and 5). Of the five that were not polyreactive before reversion, two were polyreactive after reversion. Thus 70% of the antibodies tested were polyreactive before entering the germinal centre as opposed to 5% in the overall naive repertoire4,6. We conclude that polyreactive antibodies are positively selected during the anti-HIV antibody response even before the germinal centre reaction and before somatic mutation; this reactivity is preferentially retained throughout affinity maturation despite hypermutation.
Polyreactive anti-gp140 antibodies would be positively selected if they bind more strongly to HIV than non-polyreactive antibodies (Fig. 4a and Supplementary Data 1). To determine whether heteroligation increases the affinity of anti-HIV antibodies we performed surface plasmon resonance (SPR) experiments (Fig. 4, Supplementary Figs 6 and 7 and Supplementary Table 4). We selected KLH to use as a representative low-affinity ligand because its degree and spectrum of reactivity in ELISA was intermediate in the polyreactivity ELISAs and because it is readily coupled to the SPR chips with gp120 (Supplementary Figs 2 and 6a). Chips bearing high densities of gp120 were designed to allow for bivalent antibody binding. In contrast, chips bearing low densities of gp120 should only allow for monovalent antibody binding. Finally, mixed chips derivatized with a combination of a small amount of gp120 and high-density KLH were designed to evaluate heteroligation by high-affinity monovalent binding to gp120 and simultaneous low-affinity binding to KLH (Fig. 4a and Supplementary Fig. 8).
The three polyreactive and two non-polyreactive antibodies tested bound to the high-density gp120 chips, with values of high-affinity Kaappranging from 5.6 × 108 to 4.5 × 1011 M−1 and, as expected, the reverted antibody 2–1262R bound with lower affinity to gp120 at higher density than its mutated counterpart (Fig. 4b and Supplementary Table 4, Kaapp 1.3 × 109 versus 4.5 × 1011 M−1, respectively). Also as expected, the values of Kaapp measured on the chips derivatized with the low-density gp120 were significantly lower than on the high-density gp120 chips (Fig. 4b, Supplementary Fig. 7a and Supplementary Table 4) and there was no binding by the polyreactive anti-gp41 controls to the low-density gp120-derivatized surface (Supplementary Fig. 6d). The three polyreactive antibodies bound to KLH with low affinities, with values of Kaapp ranging from 3.1 × 103 to 4.6 × 104 M−1, whereas binding by the non-polyreactive antibodies was too low to calculate. Interestingly, the combination of low amounts of gp120 with high-density KLH on a mixed surface resulted in an increase in polyreactive antibody affinity (Kaapp) of up to nearly 25-fold over the gp120 alone at low density (Fig. 4b). In contrast there was no such effect for the non-polyreactive anti-gp120 and polyreactive but non-gp120-reactive controls (Supplementary Fig. 7a and Supplementary Table 4). The net values of Kaapp produced by heteroligation were intermediate between monovalent and divalent homotypic binding to gp120 (Supplementary Table 4). Fabs did not show enhanced binding on the mixed chips (Supplementary Table 5 and Supplementary Fig. 7b, c). In addition, polyreactive but not non-polyreactive antibodies were inhibited by KLH when the antigen was present at low concentrations (Supplementary Fig. 9). Finally, the intact antibodies bound strongly to the virus in ELISA, whereas binding by Fab fragments of the same antibodies in ELISA was difficult to detect (Fig. 4c and Supplementary Fig. 10a). As expected, binding to the virus was inhibited by KLH at concentrations ranging from one to two orders of magnitude higher than gp120, and higher concentrations of KLH were required to inhibit binding of anti-HIV antibodies that showed higher levels of polyreactivity (Fig. 4d and Supplementary Figs 6a and 10b, c).
To determine whether polyreactivity enhances antibody binding to HIV, we performed SPR experiments comparing soluble gp120 and the virus. We included 1–863, which is not polyreactive, but like 2–1262 also binds to the CD4-binding site5, and 3N-1185, which does not bind to gp120 but is polyreactive (Supplementary Table 1 and Supplementary Figs 6a and 7a). The highest affinity for soluble gp120 was shown by 1–863, followed in order by 2–1262, 4–341R and 2–1262R (Fig. 4f, Supplementary Fig. 10d and Supplementary Table 5). In striking contrast, 2–1262 had the highest relative affinity for the virus, followed by 4–341R, 2–1262R and finally the non-polyreactive 1–863 antibody (Fig. 4e, g, Supplementary Fig. 10d and Supplementary Table 5). Also consistent with heteroligation, competition SPR experiments using the virion and HIV-mimic liposomes showed enhanced antibody dissociation from the virus in the presence of the liposomes (Supplementary Fig. 11). We conclude that polyreactivity enhances the relative affinity of anti-gp120 antibodies to the HIV virion.
Antibodies specific for conserved regions of the HIV spike protein have the ability to neutralize the virus and prevent infection in non-human primates14,15, and it is generally agreed that such antibodies would be an important component of any successful anti-HIV vaccine16–18. Our data show that somatic hypermutation is essential to increased antibody affinity, neutralizing potency and breadth.
Antibodies are traditionally considered to engage antigen in a monospecific bivalent manner, which enhances apparent affinity by decreasing the rate of dissociation from the ligand. Our experiments elucidate an unanticipated mechanism for increasing antibody affinity in cases when homotypic bivalent binding is not possible. Heteroligation appears to be a conserved feature of immune recognition, because a related mechanism is used by the T-cell receptor to facilitate signal transduction19. Although heteroligation increases apparent affinity and is positively selected during the immune response to HIV, this feature is not required for neutralization; for example b12 and 2G12 show little if any polyreactivity20 (Fig. 1a and Supplementary Fig. 6a).
Heteroligation improves antibody affinity when homotypic bivalent binding is disfavoured, but it may be particularly difficult to achieve because only a small fraction of the B cells in the naive repertoire are polyreactive4. This paucity of polyreactive B cells in the naive repertoire may in part explain the delayed and only occasional development of high-affinity broadly neutralizing antibodies during HIV infection. Finally, the existence of heteroligation and its contribution to anti-HIV antibody affinity suggests that mimicking the low-density viral antigens encountered during natural infection should be considered as a means to enhance anti-HIV immunization.
Anti-gp140, reverted anti-gp140 and control antibodies cloned from the peripheral blood IgG memory B cells from six HIV-infected patients5 were tested for polyreactivity, HEp-2 cells and other antigens by ELISA as previously described4,21.
Antibody affinities to gp120, KLH or both were measured by SPR using a Biacore T100 (Biacore). Antibody binding to HIV virus was tested by ELISA using BaL virus as antigen and by bead-based flow cytometry assay using a green fluorescent protein (GFP)–Gag BaL virus22. Competition experiments were performed with different concentrations of purified YU2-gp120 and KLH proteins. The relative affinity of the selected IgG antibodies to BaL virus22 was measured by SPR using IgG-immobilized surfaces.
Anti-gp140 and control antibodies were produced from peripheral blood IgG memory B cells isolated from six HIV-infected patients as described5,29. Twenty randomly selected anti-gp140 antibodies were reverted to their germline sequence by PCR and expressed as previously described7,28.
Fab fragments were produced from mutated and reverted anti-gp140 IgG antibodies by papain digestion using Fab preparation kit (Pierce). Their purity was checked on R-250 blue-stained 4–12% NuPAGE gel (Invitrogen).
IgGs were isolated from serum from HIV, and from patients with systemic lupus erythematosus and healthy humans5,21 using Protein G sepharose beads (GE Healthcare) according to the manufacturer’s instructions.
Antibodies and serum IgGs were tested for polyreactivity, cardiolipin, HEp-2 cell, KLH (Sigma), gp140 and gp120 reactivity by ELISA as previously described4,21. Antibodies were considered polyreactive when they recognized at least two structurally different antigens out of the four tested: single-stranded DNA, double-stranded DNA, insulin and lipopolysaccharide4. Threshold values for reactivity were determined by using control antibodies mGO53 (negative), eiJB40 (low positive) and ED38 (high positive) for polyreactivity ELISAs. The polyreactive ELISA used to detect antibody binding to various antigens is a sensible assay that cannot allow an absolute measurement of antibody affinities to the substances tested. Additional positive and negative control sera for HEp-2 reactivity were used as per manufacturer’s suggestion (INOVA Diagnostics)4,27.
High-binding ELISA plates (Costar) were coated with low- and high-density gp120 at a concentration of 0.5 µg ml−1 and 5 µg ml−1, respectively. Plates were blocked with 2% BSA-0.05% Tween 20-1 mM EDTA-PBS and incubated for 2 h with IgG antibodies at 1 µg ml−1 in 1:2 serially diluted KLH-containing PBS (with a KLH concentration from 0.09 to 5.55 µM). Plates were developed as previously described4,21. Data shown are representative for at least three independent experiments.
Virus neutralization was measured using a TZM-bl cell assay as previously described5.
HIV-mimic liposomes were produced by mixing 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE), 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoserine (POPS), brain sphingomyelin and cholesterol dissolved in chloroform (Avanti Polar Lipids) at a molar ratio of 9.35:19.25:8.25:18.15:45.00 (ref. 29). The mixture was dried in the fume hood under a gentle stream of nitrogen and then by a high vacuum overnight. The lipid pellet was suspended in PBS above the transition temperature for 1 h by vigorous mixing and five repeated freeze/thaw cycles. The suspension was sonicated in a sonicator bath (two cycles of 15 min with a 10-s pulse and a 15-s pulse-off) and passed through a 100-nm polycarbonate membrane (15 times) using a mini-extruder (Avanti Polar Lipids). Before SPR experiments, the liposome suspension was sonicated again (one cycle as described above) above the transition temperature.
All experiments were performed with a Biacore T100 (Biacore) in HBS-EP+ running buffer (Biacore) at 25 °C. Samples were analysed in kinetic experiments performed at least in duplicate. KLH, YU-2 gp120 and IgGs were immobilized on CM5 chips (Biacore) by amine coupling at pH 4.5. The concentration of gp120 was 0.0125 mg ml−1 or 0.125 mg ml−1 for low and high immobilization levels resulting in 100 RU and 10,000 RU, respectively. KLH was immobilized at 0.25 mg ml−1, resulting in 15,000 RU for high-density KLH surface. For the mixed surface, gp120 was at 0.0125 mg ml−1 and KLH at 0.25 mg ml−1 KLH, resulting in a total immobilization level of 15,000 RU. IgGs were immobilized at 0.1 mg ml−1, resulting in 5,000 RU. For kinetic measurements on the gp120- and KLH-derivatized chips, IgGs or Fabs were injected through flow cells in at least five different concentrations (1,379 nM, 689.7 nM, 344.8 nM, 172.4 nM, 86.21 nM, 43.1 nM, 21.55 nM, 10.78 nM for IgGs, and 1000 nM, 500 nM, 250 nM, 125 nM, 62.5 nM for Fabs) in HBS-EP+ running buffer (Biacore) at flow rates of 100 µl min−1 with 2-min association and 5-min dissociation. For the HIV-binding analyses on the IgG-immobilized surfaces, purified BaL virus22 was injected through flow cells at a p24 concentration of 78.1 nM (with four successive 1:2 dilutions) in HBS-EP+ running buffer at a flow rate of 40 µl min−1 with 2 min association and 5 min dissociation. The purified gp120 at 43.10 nM used as control and liposomes at 1,333 µg min−1 (with four successive 1:2 dilutions) were injected through flow cells under the same conditions. Competition experiments for BaL virus binding to the IgG-immobilized surfaces using liposomes were performed using the same experimental settings as for the BaL virus binding analyses except that increasing concentrations of liposomes (from 20.8 to 1333 µg ml−1) were mixed with the purified virus (p24 at 78.1 nM). The sensor surface was regenerated between each experiment with a 30-s injection of 10 mM glycine-HCl pH 2.5 at a flow rate of 50 µl min−1. Off rate (Kd (s−1)), on rate (Ka (M−1 s−1 and RU−1 s−1 for bivalent binding Ka2)) and binding constants (KD (M) or KA (M−1 and RU−1 for bivalent binding KA2)) were calculated after subtraction of backgrounds (binding to a control flow cell and signal of the HBS-EP+ running buffer) using Biacore T100 Evaluation software using the kinetic analysis and the bivalent model (IgGs on high-density gp120 and KLH surfaces), 1:1 binding model (IgGs and Fabs on low-density gp120 surface) or heterogeneous model (IgGs on mixed surface). The KA1 value was used as an estimation of apparent KA (KAapp) because KA1 represents most of the binding31. KD of the Fab binding to KLH and gp120/KLH mixed surfaces were determined by Scatchard analysis using nonlinear regression (one-site binding model). The relative immobilization level of gp120 on the mixed versus low-density gp120 surfaces was estimated by the specific binding of the non-polyreactive anti-gp120 1–863 antibody to gp120 ligand on both chips. Maximal binding capacity (Bmax) of 1–863 to each surface was determined by Scatchard linear plotting, and the relative immobilization level of gp120 protein on the mixed surface was given by the Bmax ratio ((Bmax low gp120/high KLH / Bmax low gp120) × 100) (Supplementary Fig. 8). Fold increase attributed to heteroligation was calculated using the following formula: KAapp mixed low gp120/high KLH / (KAapp KLH + KAapp low gp120). To compare the antibody affinities to BaL virus, we calculated a relative KA (KArel BaL) using an arbitrary concentration of 100 nm of virus for 43.1 mM of p24 protein. For the analysis of the gp120 binding (secondary binding) to the IgG antibodies bound to gp120 immobilized on the low- and high-density chips (100 and 10,000 RU, respectively) (primary binding), the antibodies were injected through flow cells (IgG concentrations indicated above) in HBS-EP+ running buffer at flow rates of 30 µl min−1 with 2-min association and 160-s dissociation. Purified YU-2 gp120 (at a final concentration of 450 nM) was then injected in HBS-EP+ running buffer at a flow rate of 30 µl min−1 with 2-min association. The sensor surface was regenerated as described above.
BaL and GFP–Gag-labelled BaL viruses were produced, purified and concentrated at 2–2.5 µg ml−1 of p24 protein as previously described22. BaL-strain-coated ELISA plates were incubated with IgG and Fab at 1.5 µM and four consecutive 1:2 dilutions. IgG and Fab bindings were detected with peroxydase-conjugated goat anti-human IgG F(ab′)2 antibodies (Jackson ImmunoResearch).
Using an HIV bead-based flow cytometry assay, varying concentrations of YU2-gp120 and KLH were used to compete for the antibody binding (IgG concentration of 25 nm) to GFP–Gag BaL virus. GFP–HIV-antibody complexes bound to human IgG1 capture beads (BD Pharmingen) were detected by GFP fluorescence emission using a FACS Calibur flow cytometer (Becton Dickinson). Relative mean fluorescence index was calculated by dividing the sample mean fluorescence index by the mean fluorescence index given by negative control mGO53. Data shown are representative for at least two independent experiments.
P values for Ig gene repertoire analyses, analysis of lengths, positive charges, hydropathy of IgH CDR3 and antibody reactivity were calculated by 2 × 2 or 2 × 5 Fisher’s exact test. Polyreactivity and anti-cardiolipin ELISA OD405 nm values and number of virus strains neutralized and off-target reactivity were compared using Spearman’s correlation test.
We thank J. R. Mascola and R. T. Wyatt for discussion and supplying gp140 and gp120 proteins. This research was supported by the Rockefeller University, the National Institutes of Health (NIH 1 PO1 AI081677), the International AIDS Vaccine Initiative and the Bill and Melinda Gates Foundation. T.J.H. was supported by the National Institutes of Health (R01 AI047770). M.J.Z. and H.W. were supported the German Research Foundation (GRK1121). B.D.W. and M.C.N. are Howard Hughes Medical Institute investigators.
Supplementary Information is linked to the online version of the paper at www.nature.com/nature.
The authors declare no competing financial interests.