Profiling the humoral immune response to infection by using proteome microarrays: High-throughput vaccine and diagnostic antigen discovery

Results

An approach has been developed that allows ≥384 genes derived from genomic template from a microorganism to be expressed and printed onto microarray chips very rapidly (in as little as 3 days for most genes), and the chips can be used to scan the humoral immune response from vaccinated or infected animals and humans. The overall approach involves PCR, recombination cloning, in vitro transcription/translation, and microarray printing.

In Vivo Recombination Cloning in E. coli. A linear T7 vector encoding an N-terminal His tag and a C-terminal HA tag was generated by extensive restriction digestion, followed by PCR. ORFs from vaccinia virus genomic DNA were amplified by using gene-specific primers containing 33 nucleotide extensions complementary to the ends of the linear T7 vector. Mixtures of PCR-amplified ORFs and linear T7 vector were introduced into chemically competent E. coli, resulting in antibiotic-resistant cells containing plasmid with the PCR fragment directionally inserted in-frame. This high-efficiency recombination cloning method enables hundreds of PCR fragments amplified from genomic template to be cloned rapidly.

Linear Vector. The plasmid used to generate the linear recombination vector is shown in Fig. 5, which is published as supporting information on the PNAS web site. This vector contains a T7 promoter, a prokaryotic ribosome-binding sequence (Shine-Delgarno), followed by ATG start codon, a 10× His sequence, a spacer sequence in front of the first codon of the ORF to be cloned, a BamHI site, and a T7 terminator. The vector was double-digested at the BamHI site to eliminate residual circular vector, because incompletely digested vector creates background colonies that lack insert. This linearized vector was amplified further by PCR to dilute out any residual circular vector and to generate inventory of the linear recombination vector.

PCR of ORFs from Genomic Template. Each ORF was amplified from genomic template by using gene-specific primers (Fig. 5). The 5′ oligonucleotide contained 53 nucleotides, of which 33 nucleotides constitute the 5′ extension and 20 nucleotides make up the gene-specific sequence. The 3′ custom oligonucleotide also contains 53 nucleotides, of which 33 nucleotides comprise the 3′ extension and 20 nucleotides are specific to the gene of interest. A gel showing a set of PCR products amplified from vaccinia is shown in Fig. 1A . For genes shorter than 1,000 bp, the success rate for getting the predicted PCR product was >99%. For these short genes, failures could be recovered by ordering new primers, indicating that most of the failures were caused by errors in olignucleotide synthesis. Longer genes (>2,000 bp) can be produced under different PCR conditions favoring amplification of longer products.

Transformation and Overnight Culture. Transformation of the DH5α competent cells was accomplished with a mixture of PCR fragment and linear vector in a molar ratio of 1:1, with 50 ng of total DNA used in the transformation. The competent cells were transformed, grown overnight, and observed for turbidity due to bacterial growth. Under these conditions, cloning efficiency was >90%, but if the cells were plated for colony selection on the day of transformation, the observed success rate was lower. The rate of successful transformation progressively declined as the total DNA used for transformation was reduced to 25 and 10 ng (data not shown). The process was streamlined by omitting colony selection before using the QIAprep Spin Miniprep kit. Fig. 1B shows a “cracking gel” (phenol-lysed bacteria showing total nucleic acid) from overnight cultures by using the PCR fragments shown above. The top band on these gels (band 1) is genomic DNA, the bottom two bands (bands 3 and 4) are 23S and 16S rRNA, and the central band (band 2) is the plasmid formed by recombination with linear vector and PCR fragment. Empty vector is included on this gel for reference. Of the 42 plasmids shown in Fig. 1, only 1 (E9L) lacks insert of the appropriate size. To calibrate the efficiency of the overall system, a test set of genes from the bacterium Francisella tularensis were amplified, cloned, and expressed. The data show (Table 1, which is published as supporting information on the PNAS web site) that, of 1,933 genes attempted, 96% were successfully amplified and 93% were successfully cloned.

In Vitro Transcription and Translation. The proteins encoded on recombinant pXT7 plasmids, with or without prior cloning by colony selection, were expressed in an E. coli-based cell-free in vitro transcription/translation system that was supplemented with T7 RNA polymerase. The in vitro transcription/translation reactions (≤25 μl) were incubated for 5 h at 30°C, and the unpurified reactions were resolved on SDS/PAGE gels. The gels were routinely checked for expression by immunodot blots or Western blots probed with anti-His tag Ab. The Western blots (Fig. 6, which is published as supporting information on the PNAS web site) show expression of the His-tagged products of the predicted molecular weights.

Nondenatured proteins from the cell-free reactions could also be detected on immunodot blots (Fig. 2A ). Membranes were probed with either anti-His tag Ab, anti-HA tag Ab, VIG, or VIG plus 10% E. coli lysate. The anti-His and HA tag Abs show no cross-reactivity with other proteins in the in vitro reactions and, therefore, were used routinely for monitoring the expression of large numbers of reactions. Of 112 different expressed proteins, only 4 were negative for both the His and HA tags (marked with an asterisk). To evaluate the overall efficiency of expression without prior colony picking, 390 F. tularensis genes that were prepared from overnight cultures were expressed in vitro. The reactions were spotted onto nitrocellulose and probed with either anti-His or anti-HA Ab. The results (Table 1) show that only 7% were HA and His negative. It is apparent that VIG has high titers of anti-E. coli Abs that mask any reactivity to vaccinia proteins. However, the addition of E. coli lysate to VIG reduces this background to a level such that the detection of specific vaccinia proteins by VIG is possible.

Microarrays and Serological Screening. E. coli lysate treatment of serum was also effective to reduce E. coli background reactivity on microarrays. A pilot microarray consisting of vaccinia proteins probed with VIG, with and without E. coli lysate, is shown in Fig. 2B . The effect of high titers of anti-E. coli Abs, as detected in dot blots, is also obvious on microarrays. This high background that is present also in the control preparations masks specific reactivity to vaccinia proteins. Addition of 10% E. coli lysate to VIG before probing the microarray reduced the E. coli background, revealing the specific reactivity.

Fig. 3 shows results from an array consisting of 376 reactions comprising 194 different ORFs, which we estimate to represent >95% of the complete vaccinia virus proteome. This array was screened with human VIG or sera from mice before and after vaccination with vaccinia virus. Naïve nonimmunized mice lack reactivity against all of the proteins on the array, but sera from vaccinia virus-immunized mice react with a subset of the antigens on the array. Unlike naïve mice, nonimmunized humans react with a subset of 13 antigens on the array, which are thought to represent nonspecific cross-reactions by Abs to other environmental antigens. After immunization with vaccinia virus, another subset of reactive antigens develops. Quantification of these data, plus vaccinia-immune macaque sera, is shown in Fig. 4. VIG recognizes 14 vaccinia-specific proteins. Similar profiles are also seen in sera from macaque and mice. Although there are species-specific responses (for example, A3L or A4L in mice only), there are many proteins recognized in common by humans and either animal model and 11 proteins recognized by all three species (Table 4, which is published as supporting information on the PNAS web site). These antigens would be priority candidates for the preclinical testing of a vaccine for use in humans. Overall, responses to viral structural proteins dominate the response, with more than one-half of these antigens being envelope proteins. The proteins that were seropositive included those with and without transmembrane domains or signal peptides, and isoelectric point ranges from 4 to 10. Moreover, several of these proteins have been reported to produce humoral responses in animals and humans, whereas others have not. Thus, the methodology described here identifies immunologically reactive antigens, not all of which would be identified by conventional predictive approaches. Data obtained with the arrays are in agreement with immunoblots that we have reported (5). Notably, in vaccinated humans, we see a rapid and strong increase in reactivity against a subset of dominant antigens (H3L, D13L, and A10L) after boosting many years after primary immunization.

Discussion

The PCR cloning method described relies on the principle that E. coli cells are able to recombine homologous sequences with high fidelity and speed (6). Several articles have appeared that suggest the usefulness of in vivo recombination in E. coli for PCR cloning (3, 6–9). Most of these approaches rely on the use of electroporation of certain strains of E. coli that are RecA-negative and RecE- and RecT-positive. Hence, the method has been referred to as ET recombination cloning. These reported methods are <90% efficient and require >40 bp of sequence homology between each end of the PCR fragment and the corresponding homology regions of circular or linearized vector. For the method reported here, linear DNA fragments were generated by PCR to include sequences on the 5′ and 3′ ends that are homologous with end sequences on a linearized vector. When the PCR fragments and the linear vector are mixed and transformed into competent E. coli, endogenous bacterial recombinase activity is able to join the two DNA fragments resulting in a circular plasmid. This method allows large numbers of PCR fragments to be cloned into bacterial or mammalian expression vectors in a high-throughput manner.

The method also uses chemically competent DH5α E. coli cells and relatively short 33-bp homologous sequences for recombination. Parrish et al. (10) reported an in vivo recombination approach for cloning PCR products by using an enzyme-restricted linear vector and ordinary chemically competent E. coli cells. The method, which used a 24-bp homologous sequence, showed 75% cloning efficiency but required colony picking to identify recombinant clones. The method reported here gives >90% cloning efficiency, which eliminates the necessity of the colony-picking step. Using the clonal mixture without colony selection streamlines the high-throughput cloning procedure and reduces the impact of PCR errors in the amplified ORFs. Several additional steps also contribute to robust performance. First, precautions were taken to eliminate residual circular plasmid from the linear vector. This removal is accomplished by double digestion of the circular vector template followed by PCR amplification to produce the linear vector. Second, a relatively high concentration of DNA was used during transformation. We found 50 ng of total DNA at a 1:1 molar ratio of PCR fragment/linear vector to be optimal. Last, the transformed competent cells were allowed to grow overnight before DNA isolation. This step allows time for a relatively low number of transformed cells to expand.

Theoretically, cell-free protein-expression systems offer many advantages over in vivo approaches, such as a simple and straightforward protocol and small reaction volumes applicable to high-throughput applications, the ability to be used with both circular and linear DNA templates, and the ability to express toxic proteins. In our experience, >90% of the target proteins were expressed at a level of 5–20 μg/ml (data not shown). Of the protein present in these reactions, 99% is from E. coli lysate and only 1% is from the target protein. Nevertheless, these crude reactions can be spotted directly onto nitrocellulose microarrays without purification or concentration at levels sufficient to capture detectable antigen-specific Abs from the serum of vaccinia-immunized humans and animals.

Because the in vitro expression system that we used contains E. coli lysate, our serological screening method depends on the competitive blocking of anti-E. coli Abs that are present in human sera that otherwise overwhelm any specific antigen detection. For mouse sera, which lack endogenous E. coli reactivity, there is no difference in the vaccinia-specific reactivity profile with or without E. coli lysate treatment, indicating that lysate pretreatment does not deplete vaccinia-specific reactivity (data not shown). However, we are also generated microarrays and profiling immune responses against prokaryotic organisms such as F. tularensis and M. tuberculosis that might exhibit more cross-reactivity to E. coli proteins. Nickel-coated slides could be used to selectively capture the His-tagged proteins and eliminate E. coli proteins as an alternative way to reduce E. coli background reactivity. Eukaryotic expression systems, which avoid E. coli, also can be used if posttranslational modification of the antigen is desirable. We are also aware that HA-tagged proteins could generate background reactivity from individuals previously exposed to influenza. However, VIG, which is pooled from multiple donors, shows no evidence of such background, and vaccinia-specific signals have been detected in nearly all of the ≥20 individual human sera that we have screened so far.

One of the most difficult tasks in developing a recombinant protein subunit vaccine or DNA vaccine is the identification of the antigens that will stimulate the most effective immune response against the pathogen, particularly when the genome of the organism is large (11). A comprehensive way to accomplish this task would be to obtain each of the structural, metabolic, and regulatory antigens of the pathogen and test their protective immunity individually or as mixtures in the vaccine (12). Although this approach may work for small viruses encoding several antigens, it is not practical for large viruses like smallpox, or bacteria, which encode hundreds or thousands of antigens. Our high-throughput platform for serological screening allows screening of the proteins without the risk of losing them through purification. We have found that sera reveals reactivity skewed toward vaccinia envelope proteins. Despite the simplicity of this method, its nonrandom detection of antigens gives us confidence that this method provides biologically meaningful data. Although here we have used secondary Abs to detect IgA/IgM/IgG, the detection of specific Ig isotypes is readily achieved with the appropriate isotype-specific secondary Abs (13).

The method in its current format is designed to detect protein antigens only. Clearly, other surface antigens that contain lipid or polysaccharide are not detected. However, this format is not likely to be a serious limitation if the goal is to identify protein antigens that can be expressed in E. coli and formulated for use as subunit vaccines or as antigens in diagnostic tests. The significance of the vaccinia proteins detected by VIG and sera from human vaccinees has implications for rational vaccine design and diagnostics. However, general observations can be made about the properties of the antigens detected. First, the presence of hydrophobic transmembrane domains of the vaccinia envelope proteins did not cause major insolubility problems that prevented them from being printed successfully onto the arrays. Although we are aware that the small sample of proteins presented here may not be typical of transmembrane proteins in general, at the time of writing this article, we have expressed >190 proteins from vaccinia and >1,700 proteins from F. tularensis. Of these proteins, >90% were detected by anti-His and/or anti-HA Abs on the dot blots. Second, because the proteins are printed directly from the in vitro transcription/translation reactions, conformations may be preserved that would otherwise be lost in Western blots. Third, we believe that there is a wider applicability of this technique for the identification of strongly immunogenic proteins. For example, some of the vaccinia proteins detected by VIG or immune sera are not found on the virion surface and, therefore, cannot be the targets of neutralizing Abs. In this study, we used a proteomic approach to identify the targets recognized by vaccinia neutralizing sera. These data lead us to a different set of candidate antigens than those investigated for subunit vaccines that were chosen by predictive means. In particular, we found several instances of Abs to nonenvelope proteins that would not have been identified by any predictive algorithm. These antigens, which are probably released from infected cells, are unlikely to elicit neutralizing Abs. Whether or not their immunogenicity contributes to protection in other ways remains to be determined. However, we are interested in the possibility that these Abs are acting as markers for recognition of these antigens by T cells.

In summary, we have developed a simple and robust high-throughput proteome-synthesis process that is useful for determining antigen-specific Ab profiles from vaccinated or infected animals and humans. More widespread application of this general approach could contribute to basic infectious disease research and more rapid discovery of antigens for vaccines and diagnostics applications.