To the Editor

In the May 2012 issue of your journal, we described an approach that uses proteomics and next-generation sequencing to identify antigen-specific antibodies directly from the serum of immunized animals, and we applied it to clone circulating antibodies to five different antigens from the serum of rabbits and mice1. Many technologies for isolating antibodies have been developed and applied to gain insight into the specific human antibody response to various pathogens, but none have directly addressed the serological response at the proteomic level2,3,4,5. In addition, recent evidence from a study conducted in mice suggests that not all memory B cells contribute directly to the serological response to a pathogen6. Thus, a method capable of interrogating the composition and complexity of the circulating antibody repertoire elicited to specific vaccines or pathogens is necessary. Here, we report the application of our strategy1 to the identification and cloning of high-affinity, antigen-specific human monoclonal antibodies directly from plasma of a donor vaccinated against hepatitis B virus (HBV). We also clone potent neutralizing human monoclonal antibodies against human cytomegalovirus (HCMV) from a healthy, naturally infected individual.

A key first step of our approach is the affinity purification of polyclonal material from a serum or plasma sample to enrich for monoclonal antibodies with desired functional properties. Earlier we have enriched for antibodies with high specific activity in various immunoassays such as enzyme-linked immunosorbent assay (ELISA), western blotting, immunofluorescence, flow cytometry and immunohistochemistry1. But the affinity purification can be tailored to select specifically for high-affinity interactions, potent neutralization or activity in other types of assays. Once the desired properties have been enriched, liquid chromatography tandem mass spectrometry (LC-MS/MS) is used to identify the monoclonal components of the purified fraction by matching to a reference database of antibody variable regions (V regions) produced by next-generation sequencing of the B cell immunoglobulin repertoire of the immunized animal.

In contrast to work in hyperimmunized laboratory animals, translating our approach to the isolation of circulating human antibodies faces several potential challenges. For instance, in our earlier work1 we isolated B cells from the spleen of sacrificed animals, but in humans, accessing immunological organs that are highly enriched in B cells, such as the spleen, lymph nodes or even bone marrow, may not be feasible. Furthermore, in most cases humans are not hyperimmunized or maintained in a controlled environment, as would be the case for laboratory animals. In some cases, naturally exposed or convalescent donors might be the preferred or only source of polyclonal serum and B cell–derived genetic material, and therefore high-titer or adequate quantity of material could be limiting. In such cases, the immunogen is not known and thus the appropriate antigen for isolating the desired antibodies needs to be characterized before screening.

As a proof of principle of our methodology (Supplementary Fig. 1), we first sought to isolate immunogen-specific antibodies from HBV-vaccinated donors. Chronic HBV infection affects >350 million people, a considerable percentage of whom succumb to hepatic failure or have a high risk of developing hepatocellular carcinoma7. HBV small surface antigen (HBsAg) is the main clinical marker indicating acute or chronic HBV infection, and the recombinant version of this protein has been widely used in vaccine formulations since 1986 (ref. 8). To investigate the serological immune response in HBV vaccine recipients, we screened plasma samples from volunteers who were recently immunized during an in-house HBV vaccination program for reactivity to recombinant HBsAg by ELISA. Of the six volunteers who had received their second HBV immunization dose 7 d before blood collection, donor C037 showed the strongest plasma γ immunoglobulin (IgG) reactivity to the antigen (step A in Supplementary Fig. 1, Supplementary Fig. 2). The vaccine-specific response of this donor was confirmed by ELISA using total IgG purified with Sepharose-conjugated protein G, in which the half-maximal effective concentration (EC50) was <10 μg ml−1 (Fig. 1a).

Figure 1: Affinity purification and identification of HBsAg-specific IgG from an HBV-vaccinated donor and of CMV-neutralizing IgG from a CMV-positive donor.
figure 1

(a) HBsAg ELISA titration of total IgG from HBV vaccinated (C037) and unvaccinated (C009) donors, and affinity-purified IgG from the vaccinated donor (red). (b) Phylogenetic tree (generated by aligning all 19 heavy chains using neighbor joining method for multiple sequence alignment by CLC Bio's genomic workbench software) of all 19 heavy chain variable sequences identified from affinity-purified IgG C037. Sequences that use VH1, VH3, VH4 and VH7 gene families are in green, pink, blue and gray, respectively. Scale represents number of substitutions per 100 residues (display was generated using Fig Tree; http://tree.bio.ed.ac.uk/software/figtree/). (c) Antigen-specific ELISA binding of protein G-purified IgG from HCMV-positive (C008) and CMV-negative (C009) donors, compared with binding activity of protein G-purified C008 IgG after depletion of AD4 binding IgG following AD4 affinity purification. Antigens for which ELISA binding was tested were recombinant AD4 and lysate of HCMV-infected and uninfected MRC5 cells. (d) Neutralizing activity of total plasma IgG from HCMV-positive (C008) and HCMV-negative (C009) donors and three neutralizing monoclonal antibodies isolated from C008. In a, c and d, samples were tested in duplicate; error bars indicate the deviation of each value from the mean.

To enrich for vaccine-specific IgG from plasma for analysis by LC-MS/MS, we used a purification method based on magnetic microbead separation (step B in Supplementary Fig. 1) as described in Supplementary Methods. Elutions containing affinity-purified polyclonal antibody were neutralized and tested for antigen-specific binding activity by ELISA, showing an enrichment of 150-fold when compared with the protein G–purified starting material (Fig. 1a).

To maximize identification of the V regions of the purified polyclonal mixture using LC-MS/MS, Fc regions were removed by digestion with IdeS (step C in Supplementary Fig. 1), an IgG-degrading cysteine proteinase from Streptococcus pyogenes9, and the dimeric form of antigen-binding fragments (F(ab′)2) was gel purified (Supplementary Fig. 3), digested in-gel separately using chymotrypsin, LysC or trypsin (step D in Supplementary Fig. 1), and then extracted for analysis by LC-MS/MS as described in Supplementary Methods. Identification of V-region peptides was maximized by analysis of samples from each digest in duplicate using a 72-min gradient for six LC-MS/MS runs per purified antibody resulting in 29,000 spectra per run.

To map MS/MS spectra to V-region sequences, a custom sequence reference database of antibody V regions from the donor is necessary (step E in Supplementary Fig. 1). Influenza vaccine–specific plasmablasts and memory B cells in circulation peak between days 7 and 14 (ref. 5). Thus, we reasoned that 7 d after HBV vaccine immunization would be an ideal time to collect memory B cells or plasmablasts for building a reference database. We isolated memory B cells from peripheral blood mononuclear cells of donor C037 using magnetic cell sorting to generate a cDNA library. In parallel, we also generated a cDNA library from antigen-enriched B cells, which were sorted using biotinylated HBsAg immobilized to magnetic streptavidin beads. The cDNA libraries from these cell pools were amplified by PCR using human γ-, κ- and λ-chain-specific fusion primers (Supplementary Table 1), followed by sequencing of heavy and light chain V regions using a Roche 454 FLX+ instrument.

After two independent sequencing runs of each library, we obtained 882,000 and 862,000 sequences that cover the entire V region from libraries of memory B cells and antigen-specific cells, respectively. Among these sequences, 132,373 unique γ-chain CDR3 sequences came from the memory B cell library but only 7,258 unique γ-chain CDR3 sequences came from the HBsAg-enriched B cell library. A smaller CDR3 sequence diversity was observed in the antigen-enriched B cell pool compared with the memory B cell pool, indicating as we expected that the antigen-enriched B cell repertoire is more restricted. We combined the pool of identified chains from both libraries into a single analysis to maximize the discovery of antibody V regions. Using this database, we confidently mapped 3,305 spectra to V-region sequences from donor C037 with a false discovery rate of 2%.

We next identified high-confidence V-region sequences containing unique CDR3s (see Supplementary Fig. 4a,b for example of a high-confidence heavy chain identified from HBV-vaccinated donor C037) using described methods1 (Supplementary Methods). In this manner, we selected, synthesized and cloned 24 γ-, 20 κ- and 10 λ-chain V-region sequences (see Supplementary Table 2 for percentage coverage of CDR3 and V region, and total number of mapped peptides for chains that yielded functional antibodies; see Supplementary Tables 3 and 4 for identification of peptide spectral matches corresponding to V-region sequences of heavy chain 3 and light chain 53 from donor C037).

To produce recombinant monoclonal antibodies, heavy and light chains were combinatorially paired and transiently expressed in HEK293 cells in a 96-well format as described1. We screened the recombinant antibodies by ELISA and identified 37 unique heavy and light chain pairs (comprising 19 γ, 12 κ and 9 λ chains) that bound specifically to recombinant HBsAg. This result indicates that our criteria for selecting antibody chains were effective: 19 of 24 γ chains, 12 of 20 κ chains and 9 of 10 λ chains contributed to antigen-specific monoclonal antibodies when paired.

The V(D)J usage (Supplementary Table 5) and phylogenetic analysis of the positive heavy chain V regions (Fig. 1b) demonstrate that the identified antibody clones are of diverse B cell origin. Moreover, the high frequency of somatic hypermutation (Supplementary Table 5) in both the heavy and the light chains suggests that these antibodies have undergone substantial affinity maturation. By competition ELISA, we determined that these antibodies could be classified into at least four distinct epitope groups (data not shown). Furthermore, using overlapping linear peptides that span the entire length of the HBsAg adw subtype sequence, we mapped the epitope of one antibody (monoclonal antibody (15+91)) to within 10 residues and another (monoclonal antibody (17+69)) to 20 residues within the neutralization-sensitive, antigenic loop (Supplementary Fig. 5). The region of amino acids recognized by these monoclonal antibodies overlaps epitopes of described neutralizing antibodies10. Taken together, these results suggest that the monoclonal antibodies we obtained likely react against relatively diverse epitopes within HBsAg.

To further characterize these antigen-specific antibodies, we assessed the relative binding activity of all HBsAg-specific monoclonal antibodies by ELISA (Supplementary Fig. 6). Next, antibodies with higher relative binding activity were subjected to affinity measurements using a Biacore T200 instrument (GE Healthcare). We found 23 heavy and light chain pairs with affinities with Kd values ranging from 517 pM to 613 nM (Table 1, monoclonal antibodies with unique heavy chains; Supplementary Table 6, additional monoclonal antibodies). The highest-affinity monoclonal antibody was C037(3+53) (binding kinetics in Supplementary Fig. 7a). The antibodies we isolated using our stringent purification method could represent only a fraction of the total HBsAg-specific antibodies found in circulation owing to purification conditions that are too stringent for lower-affinity antibodies; limited B cell source for generation of the next-generation sequencing reference database for mass spectrometry; and/or the limit of detection of the mass spectrometry instruments. Nevertheless, the diversity of high-affinity antibodies identified by our proteomic approach from a single donor shows that HBV vaccination elicits a humoral response that triggers expansion of a diverse B cell response. This leads to circulating antibodies with high affinity for the HBV vaccine.

Table 1 Monoclonal antibodies to HBsAg isolated from HBV-vaccinated donor C037

We next sought to identify neutralizing antibodies elicited against a pathogen from the plasma of a naturally infected or exposed individual. We focused on HCMV owing to its high prevalence in healthy adults and its clinical importance for immunocompromised individuals, solid organ transplant recipients and newborns, who may contract the virus congenitally11.

In contrast to the HBV vaccine experiments in which we isolated a wide spectrum of antigen-specific antibodies, we designed our purification procedure for HCMV to enrich specifically for neutralizing antibodies. Antibody titer to envelope glycoprotein B (gB) in HCMV-infected individuals is correlated with neutralizing activity against the virus12, and antibodies specific to a single discontinuous external region of gB known as antigenic domain 4 (AD4; residues 121–132 and 344–438, Supplementary Fig. 8) can effectively inhibit HCMV infection of a variety of cell types including fibroblasts and endothelial, epithelial and dendritic cells13. We therefore used the AD4 domain of gB to purify antibodies with the aim of identifying neutralizing human IgG against HCMV.

Healthy volunteers were screened for HCMV-neutralizing activity in their plasma using an in vitro microneutralization assay (Supplementary Fig. 9a)14. Plasma from one donor, C008, exhibited half-maximal inhibitory concentration (IC50) titer of 1:500 dilution (Supplementary Fig. 9b). Total plasma IgG from donor C008 was then purified using protein G, and showed very strong binding reactivity to AD4 and to total HCMV-infected cell lysate by ELISA (Fig. 1c). IgG purified from an HCMV-negative donor (with no detectable anti-HCMV IgG), C009, showed no reactivity in either assay (Fig. 1c).

To identify antibodies to HCMV from donor C008, we purified antibodies using recombinant AD4 and confirmed by ELISA that most of the AD4-specific activity was adsorbed but that binding activity to other HCMV proteins was unaffected (Fig. 1c). AD4-purified antibodies were analyzed by LC-MS/MS, and high-confidence heavy and light chain V regions (Supplementary Table 7) were identified (details in Supplementary Methods; see Supplementary Fig. 4c,d for example of high-confidence heavy chain identified from donor C008, and Supplementary Tables 8 and 9 for identification of V-region peptide spectral matches corresponding to V regions of heavy chain 1 and light chain 29) and combinatorially expressed as described above. Fifteen heavy and light chain pairs (comprising five γ, three κ and ten λ chains) showed AD4-binding specificity by ELISA (data not shown). Five of ten γ chains, three of four κ chains and ten of ten λ chains selected led to productive antibodies, suggesting that the selection method based on coverage worked.

The 15 anti-AD4 antibodies were further characterized for binding affinity by Biacore T200 measurements and tested for in vitro neutralization activity against HCMV. These antibodies had affinities ranging from 278 pM to 7.76 nM (Table 2, binding kinetics constants of antibodies with unique heavy chains; Supplementary Table 10, V(D)J gene usage and frequency of somatic hypermutations; Supplementary Table 11, list of all other antibodies to AD4). Seven heavy and light chain pairs (with heavy chains 1, A1 and A2) showed neutralizing activity with IC50 values ranging from 0.04 μg ml−1 to 25 μg ml−1 (Table 2, Supplementary Table 11 and Fig. 1d, neutralization curves of most potent neutralizers for each γ chain). The in vitro neutralizing activity of monoclonal antibodies C008(1+29), C008(1+27) and C008(A2+B1), with IC50 values of <0.04 μg ml−1, <0.07 μg ml−1 and <0.3 μg ml−1, respectively, is among the most potent reported so far for antibodies that inhibit HCMV infection of fibroblasts13,15, with Kd values of 695 pM, 549 pM and 6.44 nM, respectively (Table 2 and Supplementary Table 11).

Table 2 Monoclonal antibodies to AD4 (HCMV gB domain) isolated from HCMV-positive donor C008

In summary, we present two separate proof-of-concept cases to demonstrate our proteomics-based antibody discovery approach to isolate and clone antiviral monoclonal antibodies directly from human circulation. In vaccine research, antibody titers measured by ELISA are often used to identify and/or map neutralizing epitopes and to establish correlates of immune protection16. More recently, high-throughput single-cell cloning strategies have been applied to investigate the memory and plasma cell pool in vaccinated or naturally exposed donors to reflect a specific humoral response5,17. However, none of these techniques allow direct interrogation of the circulating antibody population. In the first proof of concept, we took advantage of the serological response of an HBV vaccine recipient to rapidly obtain monoclonal antibodies to the recombinant HBsAg vaccine. Some of these vaccine-specific antibodies have high affinities that are comparable to those of recently identified HBV-neutralizing antibodies2, and two of them bind to described neutralizing epitopes10. Our method could be used to track changes in the vaccine-specific circulating antibody repertoire over time and examine their correlation with the evolving functional characteristics of individual antibody components.

Characterization of biochemical properties and biological activity of isolated antibodies during purification is a critical step that helps focus the proteomics process on the identification of monoclonal antibodies with desired functional properties. We demonstrate this principle in the second proof of concept, namely the isolation of HCMV-neutralizing human monoclonal antibodies from a naturally infected donor. To accomplish this task, we first screened for donors' plasma with potent neutralizing activity in vitro (Supplementary Fig. 9a). Using plasma from one such donor and a purification strategy restricted to the use of AD4 domain of gB from HCMV13, we isolated monoclonal antibodies to HCMV with high affinities (up to 278 pM) and potent neutralization activity (IC50 values as low as 0.04 μg ml−1). In future experiments, one could envision discovering additional neutralizing monoclonal antibodies from the same donor by using additional components of gB13 or other HCMV glycoproteins18 for the affinity purification step. In this way, one could reconstitute a combined pool of potent neutralizing, fully human monoclonal antibodies. Considering that pooled HCMV hyperimmune globulin preparations are still the only available antibody-based HCMV-specific therapy, a neutralizing mixture made by recombinant human monoclonal antibodies would provide an improved clinical tool for passive immunotherapy against HCMV.

Manipulating purification conditions upfront facilitates the isolation and identification of antigen-specific human monoclonal antibodies with various biophysical or biochemical characteristics such as acid resistance, high heat tolerance, specific association and/or dissociation rates, ability to compete with a specific ligand, binding to a protein domain or a combination of any of these properties. We tracked antigen-specific binding activity to monitor enrichment of the desired polyclonal fraction (Fig. 1a,c). Such enrichment before mass spectrometry analysis enhances the probability of reconstituting functional heavy and light chain matches by combinatorial pairing, and we speculate that some of the matches may correspond to cognate pairs.

Finally, functionally validated antibody sequences identified using this approach can be used as a guide for further mining of additional clonally related antibody chains from the next-generation sequencing database generated from the same donor19 (Supplementary Fig. 10). These additional antibody chains could have been missed owing to their very low affinity or very low abundance in serum, or because they were encoded by memory B cells, that did not contribute to the serological response at the time the blood sample was drawn. In addition, we cannot rule out that other specific antibodies enriched through purification could not be identified because they were expressed only by plasma cells in the bone marrow or other lymphoid organs and thus their sequences may be absent in the cDNA sequence databases from circulating B cells.

Whether the goal is to identify a broad antibody pool against a whole protein antigen or a more restricted set of neutralizing antibodies against a smaller domain, these results demonstrate that our proteomics approach is applicable in humans and thus may be useful to address questions in humoral immunity and facilitate the development of human antibody therapeutics.

Author contributions

S.S., S.A.B., W.C.C. and R.D.P developed the methodology, designed experiments, analyzed data and wrote the manuscript. S.A.B. did bioinformatic analyses. S.S., S.A.B., L.P., J.G.B., R.K.R., X.Z., J.S.W. and S.M.S. did experiments.