Difficulties in eliciting broadly neutralizing anti-HIV antibodies are not explained by cardiolipin autoreactivity

Objective: In a recent report [Haynes et al. Science 2005; 308:1906-1908], difficulties in eliciting broadly neutralizing antibodies to HIV were linked to the binding of prototypic broadly neutralizing monoclonal antibodies to autoantigens and in particular, to the binding of two antigp41 antibodies, 2F5 and 4E10, to the autoantigen cardiolipin. We used a number of assays to understand whether 2F5 and 4E10 are autoreactive, polyreactive, or have a generalized affinity for lipids that may facilitate recognition of their membrane proximal epitopes.

Methods: 2F5 and 4E10 were evaluated for autoreactivity using diagnostic assays developed to detect serum antibodies associated with antiphospholipid syndrome (APS). As an indication of polyreactivity, we measured the binding of 2F5 and 4E10 to liposomal bilayers of differing composition using surface plasmon resonance (SPR) spectroscopy and to protein microarrays using biochip technology.

Results: 2F5 showed completely negative results in the APS and SPR studies, indicating that it is neither autoreactive nor absolutely requires phospholipid binding for epitope recognition. In contrast, 4E10 bound to more than one lipid and showed weak activity in the APS studies. The activity displayed by 4E10 more closely resembles that of antiphospholipid antibodies elicited during many infections than that of autoimmune APS antibodies, at variance with the notion that difficulites in eliciting 4E10-like antibodies can be attributed to tolerance mechanisms. The microarray studies further indicated that broadly neutralizing anti-HIV mAb are not exceptionally polyreactive.

Conclusion: These results suggest that autoantigen mimicry cannot be reliably invoked as a general mechanism for HIV immune evasion.

Introduction

Efforts to generate a protective humoral component for an HIV-1 vaccine have frequently drawn upon a few human mAbs whose hallmark is their ability to neutralize multiple HIV-1 subtypes. The two most broadly neutralizing mAb are 2F5 and 4E10 ^[1], directed against adjacent linear epitopes within the membrane-proximal external region (MPER) of the envelope glycoprotein gp41 ^[2,3]. These mAbs are potential HIV-1 therapeutics ^[4,5], and their highly conserved epitopes are currently being exploited as vaccine targets ^[6]. Recently, Haynes et al. ^[7] reported that 2F5 and 4E10 bind an endogenous phospholipid, cardiolipin (CL), and that 4E10 exhibits lupus anticoagulant activity. As anticardiolipin antibodies (aCL) and lupus anticoagulants (LA) are associated with antiphospholipid syndrome (APS) ^[8,9], this finding has raised concerns about using these mAbs as therapeutics, and has led Haynes et al. to hypothesize that MPER-based vaccines have failed because of autoantigen mimicry ^[7,10]. Explicitly, if MPER recognition is associated with CL autoreactivity, B cells directed against these viral epitopes may be subject to tolerance in healthy individuals. Alternatively, this lipid reactivity may relate to the ability of 2F5 and 4E10 to bind epitopes that lie very close to the viral membrane ^[11], particularly given evidence that these mAbs generally bind better to gp41 in the context of a membrane ^[12,13]. Therefore, we have reexamined the interaction between 2F5 and 4E10 and CL using a number of assays to better understand if these mAbs are autoantibodies in a classical sense or perhaps have some general affinity for lipids that may facilitate epitope recognition.

Methods

Antibodies IgG₁ 2F5 ^[2] and IgG₁ 4E10 ^[14] were used in this study. As per Woof et al. ^[15], antibodies were ultracentrifuged once at 100 000 × g for 15 min prior to use in all methods excepting protein biochips, in an attempt to remove potential antibody aggregates. For protein biochip experiments, IgG₁ 2G12, b6, b12, Synagis, and Z13e1 were also included ^[3,14,16,17]. IgG₁ 2F5, 2G12 and 4E10 were provided by H. Katinger. Synagis was provided by D. Richman. Recombinant IgG₁ b12, b6, and Z13e1 were expressed in Chinese hamster ovary-K1 cells in glutamine-free custom formulated Glasgow minimum essential medium (MediaTech Cellgrow, Herndon, Virginia, USA) to contain 2 mM non-essential amino acids, 1 mM sodium pyruvate, 500 μM L-glutamic acid, 500 μM L-asparagine, 30 μM adenosine, 30 μM guanosine, 30 μM cytidine, 30 μM uridine, 10 μM thymidine, and supplemented with 5% Ultra Low Bovine IgG Fetal Bovine Serum (Invitrogen, Carlsbad, California, USA), 100 U/ml penicillin, 100 μg/ml streptomycin, and 50 μM L-methionine sulfoximine (Sigma-Aldrich, St. Louis, Missouri, USA) in 10-layer cellstacks (Corning, New York, New York, USA). Antibodies were purified using Protein A affinity matrix (GE Healthcare, Piscataway, New Jersey, USA), and then dialyzed against phosphate-buffered saline (PBS).

As previously described ^[18], recombinant gp41 was produced as a fusion protein expressed in Escherichia coli that links the ectodomain of gp41_JR-FL (residues 535-681) to the C-terminus of maltose binding protein. Wild-type gp120JR-CSF was produced under contract at Advanced Product Enterprises (Frederick, Maryland, USA) in stably-transfected Drosophila cell lines. The recombinant protein was purified by Ni²⁺-NTA chromatography. Purity was verified by SDS-PAGE and Coomassie-blue staining.

The following materials were purchased commercially: Costar high-binding ELISA plates (Corning No. 3690) from Fisher Scientific (Pittsburgh, Pennsylvania, USA), Imject ovalbumin and alkaline phosphatase conjugated goat antihuman IgG F(ab')₂ from Pierce (Rockford, Illinois, USA), positive control serum with a high titer of IgG aCL antibodies from Louisville APL Diagnostics (Seabrook, Texas, USA), fatty acid free bovine serum albumin (BSA) from Roche (Indianapolis, Indiana, USA), bovine heart CL in ethanol and adult bovine serum (ABS) from Sigma-Aldrich. For liposome preparations, heart CL, egg L-α-phosphatidlycholine (PC), and ovine wool cholesterol, all in chloroform, were purchased from Avanti Polar Lipids Inc (Alabaster, Alabama, USA). Surface plasmon resonance (SPR) spectroscopy was performed using a Biacore 2000 system and Biacore L1 sensor chips (Biacore, Piscataway, New Jersey, USA). Unichip AV-400 protein biochips were purchased from Protagen Inc (Dortmund, Germany). with recombinant gp41 and gp120JR-CSF spotted onto the biochip at a starting concentration of 10 pmol/μl (20 fmol per spot) and serially diluted over five orders of magnitude. Cy3 conjugated goat antihuman IgG F(ab')₂ and BSA were purchased from Jackson ImmunoResearch Laboratories Inc (West Grove, Pennsylvania, USA) and Sigma-Aldrich, respectively.

ELISA

The aCL ELISA was adopted from Harris et al. ^[19] Explicitly: CL in 30 μl 90% ethanol or recombinant gp41 or ovalbumin in 30 μl PBS were coated overnight at 4°C. Coating density was modified to 1.35 μg/well as per Haynes et al. ^[7] Plates were washed twice with 100 μl PBS before blocking with 75 μl 10% ABS-PBS at room temperature for 1 h. Plates were then washed twice with 100 μl PBS before adding 50 μl of primary antibody or positive control serum in 10% ABS-PBS (starting concentration of primary antibody and positive control serum were 1 μM or 1: 50, respectively) and incubating at room temperature for 1 h. Plates were then washed three times with 100 μl PBS before adding 50 μl of secondary antibody [antihuman F(ab')₂] diluted 1: 1000 in 10% ABS-PBS and incubating at room temperature for 1 h. Plates were then washed three times with 100 μl PBS before developing with 50 μl of p-nitrophenyl phosphate at room temperature as per manufacturer's instructions. Absorbance was recorded at 405 nm.

The modified aCL ELISA used to assess β2 gp-I dependency (as described in the main text), was performed essentially as the aCL ELISA with the following modification: 1% fatty acid free BSA-PBS was substituted for 10% ABS-PBS as the blocking and dilution buffer as per Sthoeger et al. ^[20].

Lupus anticoagulant assays

Normal human plasma was prepared from blood collected in 3.2% sodium citrate tubes by double centrifugation (2000 × g, 15 min). Antibodies were spiked into plasma to achieve final concentrations of 50 μg/ml (0.33 μM) and 200 μg/ml (1.33 μM) while maintaining the ratio of plasma: PBS. UV spectroscopic measurements of the diluted antibody stocks indicated that the final concentrations were lower than anticipated (Fig. 1c). Each antibody sample was prepared in quadruplicate and stored at -80°C until the time of analysis. Duplicate samples were sent to either Laboratory Corporation of America (Burlington, North Carolina, USA; LabCorp) or to the University of California at San Diego Medical Center (UCSD) for evaluation in the following phospholipid-dependent coagulation assays: partial thromboplastin time assay that is sensitive for LA (PTT-LA) and dilute Russell viper venom time assay (dRVVT) at LabCorp or activated partial thromboplastin time assay (aPTT) and dRVVT at UCSD. If a sample's coagulation time was prolonged with reference to a negative control, it was 'reflexed' or automatically submitted for secondary analysis. For the dRVVT assays, this consisted of both a 'mixing' and 'confirm' test, whereas for the aPTT and PTT-LA tests this included only a mixing test. Mixing tests identify false positives that arise because a test plasma possesses a deficiency in essential coagulation factors by 'mixing' normal plasma with the test plasma at a volume ratio of 1: 1. However, mixing studies can also correct prolonged coagulation resulting from weak antiphospholipid antibodies (aPL) by diluting the total aPL in a test sample beyond the sensitivity range of an assay ^[21]. In dRVVT confirm tests, excess PL are added to quench aPL in the test sample.

Fig. 1

Liposomes

Typically, 1 mg total lipid was pipetted into a glass tube in one of the following ratios: 4: 1 (w: w) PC: cholesterol for control liposomes or 1: 7: 2 (w: w: w) CL: PC: cholesterol for CL-containing liposomes [for cryo-electron microscopic (EM) studies, liposomes were made using 5 mg total lipid]. Lipid mixtures were placed under a steady flow of nitrogen and then under an ultra high vacuum for 2 h to evaporate chloroform. Lipids were resuspended in 1 ml PBS to give a final lipid concentration of 1 mg/ml. This solution was placed in a 65°C water bath for 5 min before being alternately placed into a dry ice-ethanol bath or 65°C water bath for 3 min each. After five freeze-thaw cycles with intermittent vortexing, the resulting lipid solutions were passed 21 times through a 0.1 μm polycarbonate filter using a Mini-Extruder from Avanti Polar Lipids Inc. The size and shape of the resulting particles were immediately characterized by dynamic light scattering and cryo-EM and were used in SPR experiments the same day.

Surface plasmon resonance

Liposomes were diluted to 0.25 mg/ml in PBS and flowed over an L1 sensor chip in PBS buffer at a rate of 5 μl/min until saturation was attained (i.e. no increase in signal was observed). Immobilization levels were consistently around 11 000 resonance units for control liposomes and 10 000 resonance units for CL-containing liposomes. Immobilized liposomes were equilibrated in PBS overnight at a flow rate of 20 μl/min; this did not affect immobilization levels and served as a measure of surface stability. Antibodies at 1000 nM were diluted twofold and flowed over the surface of the immobilized liposomes in PBS buffer at a rate of 20 μl/min starting with PBS alone and continuing with increasing antibody concentrations. The surface was regenerated with a short pulse of 0.5 M NaCl-PBS. BIAevaluation software was used to subtract the signal contributed by PBS.

Dynamic light scattering

In order to ensure comparable quality between independent liposome preparations, the hydrodynamic radii of newly extruded liposomes were routinely assessed using a Dynapro Titan instrument from Wyatt Technology Corp., Santa Barbara, California, USA at a final lipid concentration of 5 μg/ml.

Cryo-electron microscopy

Liposomes were diluted to a final lipid concentration of 1 mg/ml in PBS. Four μl of this solution were adsorbed to plasma cleaned (plasma cleaner Model 1020, Fischione Instruments Inc., Export, Pennsylvania, USA) holey carbon grids, rapidly blotted with filter paper and flash-frozen in liquid ethane slush using a Vitrobot vitrification device (FEI Co., Hillsboro, Oregon, USA). Images of frozen hydrated liposomes were recorded at 29 000 × (0.385 nm pixel) on a Gatan UltraScan 4000 CCD camera (Gatan Inc., Pleasanton, California, USA) with Leginon software using a Tecnai F20 FEG microscope (FEI Co.) operating at 120 kV and a Gatan cryoholder (Model 626).

Protein biochips

All solutions were prepared fresh and filtered (0.22 μm) prior to use. The following protocol was adopted as per the manufacturer's instruction: chips were blocked in 2% BSA-PBS at room temperature for 1 h. Each mAb (500 μl), diluted to 2 μg/ml in blocking buffer, was applied directly to the chip and incubated under a coverslip for 2 h at room temperature. To estimate the signal contributed by the secondary antibody alone, an additional chip was incubated with 500 μl blocking buffer during this step. PBS was used to flush the coverslip from the surface of the chip. Chips were washed for 15 min in 0.5 M NaCl, 0.1% Tween-20 PBS with shaking. Three washes were performed before the Cy3 conjugated anti-human IgG F(ab′)₂, diluted 1: 500 in blocking buffer, was applied directly to the chip and incubated under a coverslip for 1 h at room temperature. The chips were again washed three times before being rinsed with distilled water and centrifuged at 800 × g for 4 min at room temperature. After the addition of the Cy3-conjugated antibody, all subsequent steps were performed in the dark.

The chips were scanned using an excitation wavelength of 550 nm, an emission wavelength of 570 nm and a resolution of 10 μm. Multiple scans were performed per chip with a 30-40% laser power setting and 30-62% photomultiplier (PMT) gain. The resulting TIFF were processed using Imagene 6.0 software. For each spot, signal mean and background mean values were generated. For each antigen, the signals contributed by individual spots were first corrected for background and then averaged together.

Results

Figure 1a summarizes the results obtained when 4E10 and 2F5 approach μM concentrations in a diagnostic aCL ELISA. It should be noted that this ELISA has been developed specifically to reveal interactions between serum antibodies and CL and/or protein-CL complexes and eschews many standard ELISA protocols such as detergent washing steps ^[22]. 2F5 and 4E10 bind recombinant gp41 with half-maximal binding occurring in the low nanomolar range, but do not bind ovalbumin. Notably, 2F5 exhibits no significant CL reactivity, whereas 4E10 does. However, the interaction between 4E10 and CL does not saturate even at 1 μM, indicating that it is at least two orders of magnitude weaker than the interaction between 4E10 and gp41, although quantitative half-maximal antibody concentrations could not be determined against CL for this range of antibody concentrations.

The aCL and LA associated with clinical manifestations of APS, such as thrombosis and pregnancy morbidity, are aPL directed against phospholipid-binding serum proteins that complex anionic phospholipids, rather than the lipids themselves ^[8]. aPL are also associated with a number of infections, for example HIV-1, malaria, syphilis ^[23]. In contrast to the 'autoimmune' aPL described above, these 'infectious' aPL do not generally recognize phospholipid-binding serum cofactors, rarely coincide with thrombosis or pregnancy morbidity ^[8,23] and are readily elicited as evidenced by their appearance in multiple infections. We adopted a modified aCL ELISA to evaluate whether 4E10 exhibits autoimmune or infectious type CL specificities, by removing the most abundant CL-binding serum cofactor, β2 glycoprotein-I (β2 gp-I), from the assay. As apparent in Fig. 1b, 4E10 binds CL in the absence of β2 gp-I, whereas the positive control serum does not. This distinction indicates that 4E10 has an affinity for CL alone - a characteristic of infectious, rather than autoimmune, aCL.

Phospholipid-dependent coagulation assays, known as lupus anticoagulant assays, are also routinely used to evaluate patient sera for aPL; aPL that exhibit anticoagulant activity (i.e., prolonged coagulation) in vitro correlate with thrombosis in vivo ^[8]. To evaluate 2F5 and 4E10 for anticoagulant activity, each antibody was spiked into normal plasma at concentrations exceeding 1 μM and analyzed by multiple tests at more than one facility to maximize sensitivity and account for inter-laboratory variability, respectively. The results of these studies indicate that 2F5 has no anticoagulant activity, whereas 4E10 prolonged coagulation in three tests at 1.2 μM (Fig. 1c). Samples with prolonged coagulation were submitted to additional 'confirm' and 'mixing' tests (described in Methods) to verify the nature of the observed anticoagulant activity. Mixing tests confirmed that the prolonged coagulation observed for 4E10 was not a result of coagulation factor deficiencies in the plasma. Confirm tests demonstrated that the anticoagulant activity of 4E10 is phospholipid dependent.

We also characterized the binding of 2F5 and 4E10 to CL that is organized in bilayers by coupling SPR spectroscopy and liposome technology. We measured 2F5 and 4E10 binding to immobilized control liposomes and liposomes containing CL (Fig. 2a) and found that 2F5 binds neither liposome. 4E10 binds to both control liposomes and CL-containing liposomes. Interestingly, better relative binding is noted to control liposomes at low antibody concentrations and to CL-containing liposomes at high antibody concentrations, as indicated by the arrows in Fig. 2b. While the reasons for this phenomenon are unclear, it does not appear to be attributable to structural differences between control and CL-containing liposomes as a combination of dynamic light scattering (DLS) and cryo-elecron microscopy methods revealed that both lipid compositions generate liposomes of similar shape and size (see Supplementary Figure I online).

Fig. 2

To more comprehensively assess the levels of general cross-reactivity exhibited by 2F5 and 4E10, we evaluated their binding to a substantially larger pool of antigens in a microarray format. We included for comparison the broadly neutralizing anti-HIV human mAbs b12 and 2G12, directed against the envelope glycoprotein gp120, Z13e1, also directed against a linear epitope within the MPER, a well characterized 'non-neutralizing' anti-HIV mAb directed against gp120, b6, which exhibits only modest activity against highly neutralization-sensitive viruses ^[1], and a widely-used mAb for prevention of acute respiratory syncytial virus infection in children, Synagis ^{[3,16,17,24,25]}.

Figure 3 illustrates the results obtained when each mAb is incubated on the surface of a protein biochip containing over 400 recombinant proteins expressed in E. coli. The mAbs bind only a small fraction of the proteins in the array and then with relatively modest signals in comparison to the positive control proteins (see Supplementary Figure II online). For example, after correcting for the secondary antibody binding signal, we find that the top 'hit' for each mAb is < 5% of the maximum antibody binding signal against antihuman IgG - with the exceptions of 2F5 and Z13e1 whose top hits against glyceraldehyde phosphate dehydrogenase represent 17% and 80% of their respective maximum antibody binding signals. Aside from these exceptional reactivities, the cross-reactivity profiles of the different mAbs cluster together, suggesting that the broadly neutralizing mAbs are not unusually cross-reactive.

Fig. 3

Discussion

The results presented above reinforce the conclusion, also drawn by others ^[7,13], that in addition to being a highly specific anti-HIV antibody (as evidenced here by nanomolar binding to gp41 and a lack of cross-reactivity in protein microarray studies), 4E10 also possesses a low affinity for lipids. This is further supported by additional ELISA data, not shown here, which shows 4E10 binding to immobilized phosphatidylserine. In contrast, the ELISA results for 2F5 differ from those reported by both Haynes et al. ^[7] and Sanchez-Martinez et al. ^[13], which also differ quantitatively from each other. These differences most likely reflect differences in ELISA protocols, which are given for comparison in the Supplementary Material online. It should be noted that the assay used here corresponds to the only clinically approved assay. Furthermore, the trends observed in the aCL ELISA correlate well with the trends observed in SPR experiments, including additional experiments showing 4E10 binding to PS-containing liposomes.

We note that there is considerable day-to-day variability in the ELISA data for immobilized CL, which is expected. A number of multiple laboratory consensus studies have reported on the inter- and intra-method result variability associated with the aCL ELISA, but despite numerous efforts to introduce standardized protocols and controls, the problem remains unresolved ^[26-28]. As mentioned previously, Goldberg et al. ^[29] find that most of the CL lost during an aCL ELISA occurs after the blocking step. They postulate that this results from phospholipid-binding serum proteins in the blocking buffer binding to and dissociating phospholipids from the ELISA plate surface. A process as stochastic as this could explain the inherent variability of this assay, particularly as none of the proteinaceous antigens produce such characteristically variable results in our assays. Lipid oxidation may also contribute to ELISA variability above and beyond the sources of error associated with ELISA assays in general (e.g., inconsistency in antigen coating from well to well).

The LA results suggest no concerns with respect to therapeutic use of 2F5 but caution with respect to 4E10. In this regard, it is also probably wise that 4E10 be examined in terms of in vivo fetal resorption for insight as to whether it may pose a specific safety concern to expectant mothers. However, it should be noted that 4E10 is only weakly active in these assays at very high concentrations that are only likely to be achieved transiently in most therapeutic modalities. Indeed, both 2F5 and 4E10 have been administered to humans without reported adverse events ^[4].

Notwithstanding the glyceraldehyde phosphate dehydrogenase reactivity of 2F5 and Z13e1, which warrants further investigation, the broadly neutralizing anti-HIV mAbs are not exceptionally cross-reactive when assessed in a microarray format. The general level of cross-reactivity is similar to that for Synagis, an antibody that has been used effectively in the clinic for some time. Moreover, a number of studies have confirmed the epitope specificity of these antibodies ^{[11,12,24,25,30-34]}. In the case of 2F5 and 4E10, epitope specificity was elegantly demonstrated in two independent studies where neutralization resistant SIV ^[30] and HIV-2 viruses ^[34] were rendered neutralization sensitive upon engineering of the core epitope sequences of these antibodies into the MPER of SIV and engineering of the HIV-1 MPER into HIV-2. Therefore, failure to induce broadly neutralizing antibodies from epitope-based immunogens more likely reflects factors such as a failure to recapture the precise epitope presentations that elicited the mAbs in vivo and/or in the case of the MPER mAbs, conformational constraints placed on antibodies recognizing epitopes very close to the membrane, rather than a failure to correctly identify broadly neutralizing epitopes.

Taken altogether, these finding gave no further indication that 2F5 and 4E10 are auto-reactive and so continued testing for autoreactivity was not considered to be a valuable approach.

In conclusion, 2F5 shows completely negative results in the ELISA, LA and SPR studies, indicating that recognition of the MPER does not absolutely require CL or PL recognition. 4E10 does exhibit some generalized affinity for lipids; however, this reactivity is more characteristic of infectious aPL than autoimmune aPL. Since infectious aPL are readily elicited in a number of infections, it is therefore not immediately apparent that difficulties in eliciting 4E10-like antibodies are attributable to tolerance mechanisms.

Table 1

Acknowledgements

We sincerely thank D. Richman for Synagis; J. Quispe and C. Potter at the National Resource for Automated Molecular Microscopy for assistance with cryo-EM samples; M. Thompson and T. Spillman at LabCorp and K. Donnelly and D. T. Le at the UCSD Medical Center for lupus anticoagulant assays; J. Hoffmann for assistance with protein biochip scanning and data analysis; P. Dawson and J. Williamson for helpful discussions.

Sponsorship: Some of the work presented here was conducted at the National Resource for Automated Molecular Microscopy which is supported by the National Institutes of Health though the National Center for Research Resources' P41 program (RR17573). Financial support was provided by NIAID (AI32292, DRB; AI069993, MBZ; AI42267, LT) and by the International AIDS Vaccine Initiative.