Research ArticleMalaria

Targeting TLRs Expands the Antibody Repertoire in Response to a Malaria Vaccine

See allHide authors and affiliations

Science Translational Medicine  27 Jul 2011:
Vol. 3, Issue 93, pp. 93ra69
DOI: 10.1126/scitranslmed.3002135


Vaccination with an isolated antigen is frequently not sufficient to elicit a protective immune response. The addition of adjuvants to the antigen can increase the magnitude and breadth of the response generated, but quantification of this increase as a function of adjuvant has been intractable. We have directly determined the variation of the immunoglobulin G variable-chain repertoire of an entire organism as a function of vaccination. Using the well-established Plasmodium vivax antigen, PvRII, and massively parallel sequencing, we showed that the use of a Toll-like receptor (TLR) agonist in the vaccine formulation increased the diversity of the variable region sequences in comparison to the use of an oil-in-water emulsion adjuvant alone. Moreover, increased variable domain diversity in response to the use of TLR agonist–based adjuvants correlated with improved antigen neutralization. The use of TLR agonists also broadened the range of polymorphic variants against which these antibodies could be effective. In addition, a peptide microarray demonstrated that inclusion of adjuvants changed the profile of linear epitopes from PvRII that were recognized by serum from immunized animals. The results of these studies have broad implications for vaccine design—they may enable tailored adjuvants that elicit the broad spectrum of antibodies required to neutralize drifted and polymorphic pathogen strains as well as provide a method for rapid determination of correlates of adjuvant-induced humoral immunity.


Efforts to develop new vaccines based on soluble, recombinant proteins have been ongoing for the last 3 decades—with limited success. The magnitude and duration of the adaptive response induced by these refined antigens on their own are often not sufficient to protect vaccinated individuals. It is becoming increasingly clear that the addition of adjuvants that add an innate danger signal to the vaccination greatly enhances the quality, strength, and duration of the immune response to the antigen.

The innate immune system serves as the body’s first defense mechanism against pathogens. Toll-like receptors (TLRs) are partially responsible for the specific recognition of certain pathogen-associated molecular patterns (PAMPs), a plethora of molecular groups that are common to invading microbes. With respect to humoral immunity, TLR signaling has been shown to induce naïve B cells to undergo proliferation and differentiation to produce antibodies in the absence of B cell receptor cross-linking (13), and studies have shown that TLR signaling in B cells can also affect the long-term B cell memory response (4). Dendritic cells, which are activated by TLRs as well as present antigen to activate specific and memory immune responses, may form a bridge between innate immunity induced by pathogen-associated molecules and adaptive immunity (3, 5, 6).

The human antibody repertoire is large; it has been estimated to be greater than 1011 molecules per individual (79). The breadth of response against a vaccination is dependent on the ability of initial signals to recruit diverse cells from the recirculating naïve B cell pool into the germinal center (GC), where further diversification of the antibody repertoire is produced by virtue of somatic hypermutation. This is a process by which the variable chains of antibodies mature by mutating at specific hotspots where antigen recognition is encoded on the DNA level (8, 1012). Very few soluble antigens initiate a strong antibody response on their own, and both the quality and the longevity of the antibody response are highly dependent on the initial signals from the innate immune system that activate B cells (4, 1315). This mechanism linking innate and adaptive immunity has been studied but remains incompletely defined (11, 12). Specifically, the quantitative effect of how certain TLR agonists expand the B cell repertoire on the level of immunoglobulin (Ig) variable sequence diversification remains to be evaluated.

The well-established malarial blood-stage antigen PvRII is derived from the receptor binding domain, region II, of the Plasmodium vivax Duffy binding protein; associates with the Duffy antigen receptor for chemokines (DARCs) on human red blood cells; and is thought to mediate invasion (16). Immunization with recombinant PvRII formulated with an appropriate adjuvant is intended to elicit antibodies that interfere with this association and thereby protect against vivax malaria (17). Earlier studies have shown that PvRII immunization formulated with the TLR4-targeted adjuvant AS02A performed better at eliciting blocking antibodies in comparison to formulations in an oil-in-water emulsion or on alum. These studies attributed this effect to the magnitude of antibody responses produced (16, 17).

We hypothesized that this enhancement was not simply a function of the quantity of antibody produced but also the quality of the response. We predicted that the use of TLR agonist–based adjuvants in the vaccine affected both the magnitude and the diversity of the humoral response and, by virtue of the latter, also affected function. To measure this at the molecular level, we implemented a long-read massively parallel signature sequencing (MPSS) approach to sample the very large number of variable sequences represented in the B cells of an organism after immunization (18). The induction of a more diverse set of antibodies with a vaccine containing TLR-based adjuvant formulations could translate into recognition of more polymorphic sequences, thus suggesting use of these types of adjuvants when neutralization of drifted variants is required, such as for influenza, malaria, and HIV.


Variable-chain antibody sequences generated from complementary DNAs

MPSS generated a data set of about 100,000 full-length variable-chain compositions identified with specific samples from mice immunized with one of the four vaccine formulations or the naïve saline control. Germline matches were calculated by performing BLAST searches of the interprimer DNA sequences against genomic DNA references spanning the mouse Ig heavy or light regions. Sequences identical to an entire V region exon were counted as germline. Sequence numbers for each of the three different Ig isotypes used in the study (heavy chain γ, heavy chain μ, and light chain κ) and their frequency in the MPSS data set across all experimental groups are shown in Table 1. Analysis of variable-chain distributions demonstrated that recovered IgG sequences have less complexity and greater depth compared to IgM and Igκ sequences because there are higher percentages of unique sequences present twice or more in the IgG data set. This is consistent with the low percentage of total sequences identical to germline for IgG (0.8%) compared to IgM (4.2%) and Igκ sequences (64%).

Table 1

Sequences generated using MPSS long reads. This table summarizes the number and frequency of complete variable domain region spanning open reading frames in the MPSS data set for the indicated Ig chain isotypes. Samples for sequencing were harvested 1 week after the third immunization. Listed in each row are the statistics for sequences occurring in the data set at least as many times are the minimum count number listed in the left column. The three constant region isotypes tested have been broken into separate entries, each with three values. The percentage of unique predicted protein sequences present twice or more is 30% for the IgG sequence data set compared to 12% and 19% for the IgM or Igκ data sets, respectively. The percentages across IgG isotypes in the entire data set are 80%, 5%, 5%, and 10% for IgG1, IgG2A, IgG2B, and IgG3, respectively. N.R. is the total number of distinct, nonredundant predicted protein sequences in the set. Total is the absolute count of all recovered V region sequences regardless of redundancy. Germline is the number of sequences that are identical in DNA sequence to a murine germline V region exon.

View this table:

We analyzed the large set of predicted protein sequences to identify not just single sequences, but clusters of related sequences that appear to be preferentially expressed in one group relative to another. Looking at clusters allows for greater statistical confidence in identifying differential expression because more significant probabilities can occur with larger numbers. Also, clustering sequences that differ only slightly, perhaps due to sequencing or polymerase chain reaction (PCR) errors, helps to compensate for experimental noise. Clustering was performed by creating a phylogenetic tree of the predicted protein variable domain sequences. This resulted in large trees with 9884, 37335, and 9389 leaf nodes for IgG, IgM, and Igκ chains, respectively. For each adjuvant group, a value representing the chance of the observed count of the sequences derived from that group compared to the naïve control group occurring at random was calculated at each node of the trees. This value is represented by an expected score (e-score). A histogram-style representation of the set of e-score shows the effect of adjuvant type on the number and significance of sample-specific phylogenetic tree nodes (Fig. 1). As a control, the inverse e-score, which represents the chance of the observed count of the sequences being derived from the naïve group compared to each experimental group, was calculated, and there were no nodes with a negative log e-score above the histogram cutoff value of 15 for the IgM inverse group, and a relatively small count of between two and four nodes for each bin with negative log e-scores less than 18 for the IgG inverse group, confirming the significance of the e-scores shown.

Fig. 1

Effect of vaccine adjuvant on number of specific Ig V region sequences. Full-length variable domain Ig heavy-chain locus sequences recovered from the MPSS procedure were organized into a phylogenetic tree, and e-scores were calculated comparing the sequence counts from the SE, GLA-SE, R848-SE, and GLA/R848-SE experimental groups relative to naïve control samples at each node in the phylogenetic tree. The effect of adjuvant formulation on Ig variable sequence repertoire is represented as a histogram distribution of number of maximum specificity nodes binned according to their “significance score” defined as negative log of the e-score. The number of nodes in each specificity bin is shown in the figure for both IgG and IgM sequences. The analysis produced no nodes above the specificity score cutoff of 15 when applied to the Igκ data set.

Diversity of sequences induced in response to immunization

Not surprisingly, more sample-specific sets of related variable domain sequences are found in the immunized groups compared to the naïve control group. Note that without the addition of the TLR4 agonist (SE), many more IgM sequences arise with greater significance (lower log e-scores) compared to the IgG set; however, after combining with the agonist (GLA-SE), many more are found in the IgG set, suggesting that TLR agonists enhance class switching. With respect to the three chain types considered in the analysis, fewer group-specific sequences were seen for IgM compared to IgG, and no groups with an e-score of 1 × 10−15 or less were seen in the Igκ chain type.

Of the sample-specific IgG sequences, the highest diversity was observed in animals immunized with adjuvant containing both the TLR4 and the TLR7/8 agonists (GLA/R848-SE). This was closely followed by the TLR4 agonist-alone group (GLA-SE). Animals immunized with the TLR7/8 formulation (R848-SE) or the formulation vehicle alone (SE) showed the least diversity. Analyses of the IgM sequences showed that animals that had been immunized with the formulation vehicle have significantly lower e-scores when compared to other groups. When the same analysis was performed on the Igκ light-chain sequences, no sequence clusters with e-scores of less than 1 × 10−15 occurred, indicating that the heavy chain, not the light chain, is the major driver of differentiation in response to vaccination. The lack of significant sample-specific Igκ transcript sets was consistent with the greater number of germline sequences mentioned above and indicated that little differentiation occurred in response to vaccination within the κ light chain.

Overlap of variable-chain sequences

Molecular profiling of the antimalarial antibody response also allowed us to compare the overlap in sequence diversity elicited by different vaccine formulations. The TLR4/7/8 combination vaccine group contained the greatest number of sequences found only in that sample (1973), followed by the TLR4 group (722), with the most shared sequences found between these two groups (206) (Fig. 2). The TLR7/8 group, which had the fewest number of unique sequences (131) found only in that group, also had the least overlap with the TLR4 agonist–containing groups, although there was a significant overlap with the SE alone group. Some sequences were found in common in two or more experimental groups, which indicates that the same antibody was being expressed by different animals in response to the immunization. Because distinct immunizations in animals can lead to a number of shared sequences, certain IgGs may be high-quality binders that share common variable sequences and antigen recognition sites.

Fig. 2

Overlap of unique sample-specific IgG sequences between groups. Sets of predicted variable domain protein sequences in clusters with negative log peak node e-scores of 15 or greater for each experimental group were compared, and total numbers of sequences in common or exclusive to a group are shown in the respective overlapping regions. The cutoff of 15 is used because false-positive nodes (determined by the inverse e-score calculation) are low at this threshold. Treatments correspond to innate stimulation as follows: SE = no added TLR signal, vehicle control; GLA-SE = TLR4 agonist; GLA/R848-SE = TLR4,7/8 agonist; R848-SE = TLR7/8 agonist.

A three-dimensional depiction of the entire IgG phylogenetic tree is shown in Fig. 3. The model has two major branches with a smaller group of sequences joining the tree at a higher level than the node connecting the two major branches. These centrally located smaller groups are less closely related than the two main branches, as shown by the increased joining height above the X-Y plane. The two major branches were broken down further into distinct clusters of more closely related sequences. Sequences within spatially isolated clusters seemed to be derived from closely related V regions (all three major clans of variable domains are represented in various clusters), although the most abundant was the V1 type. In general, there was good separation of regions of the tree with significant e-scores between experimental groups as seen by the distribution of the e-score disks. However, in a few cases, such as the combined red- and yellow-marked cluster on the upper left side of the figure, the same sequences are significant to two experimental groups, indicating that there are common sequences between the groups that are specific relative to the naïve control.

Fig. 3

Families of related sequences induced by TLR agonists. The 9884 nonredundant full-length variable domain protein sequences recovered for IgG were assembled into a bifurcating phylogenetic tree according to the neighbor-joining method. To display the tree in a single image, the tree has been folded and drawn in three dimensions, with individual sequences as the leaf nodes represented by colored squares on the X-Y plane and the branches extending upward in the z dimension proportionally to the evolutionary distance. The z axis is drawn pointing to the top and extending out of the plane of the page. The leaf colors blue, red, green, and yellow indicate that the sequence came from the emulsion only, GLA, R848, and GLA/R848 groups, respectively. The colors of leaf nodes whose sequence is present in multiple groups are blended proportionally, and leaf nodes whose sequences only appear in the naïve group are drawn in black. The nodes are connected by the lines whose height in the z axis is proportional to the distances between the nodes as determined by the neighbor-joining algorithm, and the distance in the X-Y plane and line intensity are determined by the increase of total sequences in the resulting parent node. Colored disks are drawn at nodes with areas proportional to the −log of the e-score of the significance of an experimental group count relative to the naïve group, and colored according to group in the same way as the leaf nodes.

Recognition of PvRII-derived linear epitopes

Because the TLR4 agonist–based adjuvants elicited more diverse IgG sequences and shared a number of variable regions (Fig. 2), we hypothesized that serum from these groups would react with regions of the PvRII antigen in a similar manner and that these adjuvants may recognize more areas of the PvRII protein. To test this, we generated a PvRII peptide microarray and probed it with serum from the various groups. As expected, the signal strengths for individual peptides between the TLR4 agonist groups and between the vehicle and TLR7/8 groups correlated well (Spearman r of 0.92 and 0.93, respectively), whereas there tended to be less correlation in signal strengths between the other groups (Table 2 and Fig. 4A). GLA-SE and GLA/R848-SE trended toward recognizing more linear peptides compared to other groups, but the difference was only statistically significant for the GLA-SE group versus the R848-SE group (Fig. 4B). This could be because the number of exposed epitopes did not allow for much expansion or the expansion is more specific for conformational epitopes that are not reflected in the peptide microarray.

Table 2

Peptide recognition as a function of adjuvant. The table shows the correlation coefficients in peptide recognition by sera from animals immunized with various adjuvant formulations. Intensity data on the chip of peptides recognized by PvRII were plotted with data from one adjuvant on one axis and data from another on the orthogonal. Spearman r values were computed for each set of axes and are shown.

View this table:
Fig. 4

Effect of TLR adjuvants on PvRII recognition. (A) Comparison between signal intensities on peptide microarray. An array with PvRII peptides was probed with sera from animals immunized with different TLR adjuvant/PvRII combinations. Peptide-specific antibodies were detected with fluorescent-conjugated, goat anti-mouse IgG, Fc-specific secondary antibody. The signal intensity for each spot from one adjuvant versus the intensities on the same spot using sera from a second adjuvant treatment is shown. The line is 45° to show where perfect concordance would be plotted. Right: GLA-SE versus GLA/R848-SE; left: GLA-SE versus R848-SE. (B) Number of peptides recognized. Recognition was defined by an increase in the average signal of duplicate peptide spots by a factor of ≥3 over that of the average signal from the serum of three naïve animals. Data are from two independent experiments. SE = vehicle; GLA-SE = TLR4 agonist; GLA/R848-SE = TLR4,7/8 agonist; R848-SE = TLR7/8 agonist. Data were analyzed by one-way ANOVA and Tukey’s post hoc test. GLA-SE is significantly different from SE alone (P < 0.05) or R848-SE (*P < 0.01).

Inhibition of DARC binding

Having established that the adjuvant affects the antibody repertoire, we went on to test the ability of the various sera to inhibit DARC binding, which is a measure of the ability of the plasmodium surface Duffy binding protein to bind to the surface of the red blood cell. Cohorts of mice that had been immunized with either TLR4 or TLR4/7/8 agonist(s) showed the greatest overall binding inhibition (Fig. 5A). In terms of cross-strain reactivity, sera from each cohort were tested for the ability to inhibit binding to variant PvRII alleles derived from Papua New Guinea (PNG) field isolates, which account for greater than 50% of the polymorphisms seen in PNG (19). The use of the TLR4 agonist alone or a combination of TLR4 and TLR7/8 agonists tended to make the groups more similar in their inhibition, whereas there were significant differences between groups containing the TLR7/8 alone or the emulsion vehicle (Fig. 5B). As a correlative comparison, the total numbers of IgG peak e-score nodes equal or less than 1 × 10−15 are shown in Fig. 5C as a measure of the complexity of the recovered sample-specific V domain sequences. The formulations that elicited the most complex IgG variable region repertoire generated sera with the greatest ability to inhibit binding and were better at recognizing drifted variants of PvRII.

Fig. 5

Diversity of antibody sequences correlates to inhibition of PvRII-DARC binding and to drifted antigen recognition. (A) DARC binding inhibition induced by different vaccine formulations. Sera were titered by ELISA for binding to PvRII, and the ability to inhibit DARC binding was determined. The GLA-SE and GLA/R848-SE formulations inhibited significantly more DARC binding than the R848-SE formulation; additionally, the GLA/R848-SE formulation was more powerful than the SE only formulation. (B) Recognition between polymorphic PvRII alleles. Sera raised by immunization with the Sal I allele of PvRII formulated with different adjuvants were tested for inhibition of binding of the homologous PvRII allele (Sal I) as well as heterologous PvRII alleles (P, O, AH, C, and T) to DARC at a dilution of 1:4000. Significance calculations within each formulation demonstrated no significant differences within the GLA-containing formulations, whereas the SE or R848-SE formulations had significant differences in the ability to recognize polymorphic alleles. (C) Diverse antibody sequences for each adjuvant group. The number of significant sequences with a negative log e-score of >15 was computed for each group and is shown as a bar graph to relate to the inhibition and cross-neutralization graphs shown in (A) and (B). For inhibition plots, means are plotted with error bars indicating SD of the mean. *P < 0.05; **P < 0.01; ***P < 0.001, ANOVA with Tukey’s post hoc test.


Recent problems with attaining a sufficient supply of flu vaccines as well as the landmark approval by the Food and Drug Administration of Cervarix, a vaccine that uses a TLR4 agonist adjuvant, have highlighted the ability of adjuvants to provide higher-quality immune responses to vaccine antigens. Here, we investigated whether the use of specific TLR agonists in a vaccine could enhance the diversity of the antibody repertoire. We used a well-established blood-stage malaria vaccine candidate (17, 20) in a mouse model to elicit an immune response and used massively parallel sequencing technology to sequence millions of bases of DNA representing the Ig variable regions from the immunized animals. This information was used to look at adjuvant effects on Ig classes and subtypes and assembled into phylogenetic trees depicting antibody families. We then measured the ability of the sera to recognize a range of linear epitopes, inhibit binding of the protein to its pathogen receptor, and interfere with binding to proteins from drifted strains.

Our data show that the extent of diversification of the antibody repertoire is controlled by the adjuvant(s) used to initiate and enhance the immune response. However, the precise role of innate signals in B cell recruitment to and activation in the GCs is uncertain. This process could be dependent on the expression of TLRs by the B cell itself, but other antigen-presenting cells (APCs) and helper cells likely play a central role. Considering the critical role of T helper–expressed costimulatory molecules like ICOS and CD40L for the initiation of GC reactions, it is possible that the effect observed is, in part, the result of increased induction of CD4+ T cell help. Although CD4+ T cells lack TLR4, indirect recognition of TLR stimulation via APC-secreted inflammatory molecules like tumor necrosis factor–α (TNF-α) and type I interferons may alter their activation and/or surveillance. In addition, TLR expression on follicular dendritic cells within the GC has been shown to be crucial in somatic hypermutation, suggesting that these dendritic cells—when functioning as APCs in the GC—are important in this process (3). TLR activation may contribute to a more diverse antibody response through multiple mechanisms.

Considerable overlap was found in the sequence sets generated from TLR4 agonist–based adjuvants. Affinity maturation may produce optimal variable region “solutions” that are represented by the homologous Ig variable sequences found in different organisms. Indeed, the potential commonality of certain Ig sequences between individuals may allow for the development of variable region–dependent diagnostics that may be used as immune correlates for protection in response to antibody-based vaccines. These new diagnostic techniques could complement simple IgG titers in determining whether individuals have been protected by the vaccine.

The enhanced diversity of sequences and overlap of variable regions corresponded to functional readouts. TLR4-based sets correlated in the ability to recognize linear peptides as well as in the ability to better inhibit binding of PvRII to the pathogen receptor DARC. Although this inhibition assay is thought to correlate with efficacy against P. vivax (20), it is important to stress that this is an in vitro model around which there is some debate as to how well it will predict field performance of the vaccine. Although there was significant enhancement in the abilities of sera from the TLR4-containing adjuvant groups in neutralizing binding, the number of peptides that the expanded repertoire bound trended upward but was only significant for the TLR4 versus TLR7/8 pair. This could be due to the number of peptides available on the surface of the antigen, not allowing much room for expansion or because expanded conformational recognition cannot be seen by the linear peptide array, but is reflected in the DARC binding functional assay.

Variation and polymorphism of blood-stage antigens is a key problem in the eradication of malaria. The increased breadth of antibodies after vaccination in the presence of a TLR4 agonist inhibited binding of proteins derived from “drifted” malaria strains. Thus, adding adjuvants stimulating innate pathways may help overcome failure of antibody-mediated protection from certain vaccines in cases where there are multiple variant organisms or antigen drift such as influenza. Although the total number of variable domains may be vast, on the order of greater than 35 billion sequences (9), our data show that in response to a particular antigen, specific variable domains are elicited often enough that they can be detected in different animals at a reasonable frequency. The emergence of a practical means such as MPSS for profiling adaptive immunity at the molecular level will benefit studies of immunity and can be used for detection, diagnosis, and treatment of a wide variety of diseases.

Materials and Methods


BALB/c mice were immunized with PvRII (25 μg) (International Centre for Genetic Engineering and Biotechnology) in a stable emulsion alone (SE; an oil-in-water–based formulation similar to MF59) or TLR4-based adjuvant, GLA-SE (20 μg) or TLR7/8-based adjuvant, R848-SE (20 μg), or an adjuvant containing a combination of these two agonists (GLA/R848-SE) (IDRI; 3M Pharmaceuticals) (20 + 20 μg). Mice were immunized and boosted at weeks 3 and 6. Samples were harvested 1 week after the third immunization to perform MPSS and 4 weeks after the third immunization for sera to perform titers and inhibition assays. Bone marrow for samples to analyze by MPSS was harvested 4 weeks after the third immunization.

Sequence sample preparation

Complementary DNA (cDNA) was generated from peripheral blood samples taken 6 days after the second immunization and at the terminal time point 10 weeks after the initial immunization. In addition, bone marrow cDNA was also generated at the terminal time point. cDNAs were synthesized on oligo(dT)-coated magnetic beads (Invitrogen) with SuperScript III reverse transcriptase according to the provided protocol (Invitrogen). Samples from three individuals from the same group were pooled, and the sequences were combined when calculating the e-score values. These samples were used as templates for PCRs spanning the variable domain regions of the Ig heavy-chain locus, as well as the κ light-chain variable domain regions. The annotated 2.7-Mb region of the murine Ig heavy-chain locus from the National Center for Biotechnology Information was cross-referenced with the International Immunogenetics (IMGT) Web resource, which contains a database of functional murine V regions and associated data. This list of potential V regions was culled by the requirement of an in-frame splice site proximal to the upstream leader sequence. The generation of these hypothetical transcripts was confirmed by sequence searches of mouse expressed sequence tag databases. These V regions were then separated into their respective clans and families, their sequences were aligned, and minimal sets of primers with 18-base exact sequence identity to the template were designed that cover each potential V region. The primers were designed to conform to PCR primer design rules including the absence of 3′ complementary regions in the primer set, the absence of single-base repeats of four or more, and a percent GC content in the range of 45 to 60. Complementary reverse primers covering each potential IgG isotype were also designed with identical 3′ ends and pooled in equal amounts for use as the reverse primer. Because there is only a single IgM isotype, only one reverse primer was required for the IgM PCR. In addition to the 18-base-long region of identity to the template, an additional 8-base “barcode” was added to the ends of each of the primers. This allows for multiplexing of the samples in a single sequencing reaction mixture. Forward primers sharing the same six bases at the 3′ end were combined in equimolar amounts and used in the same PCR. Six sets of upstream primers were sufficient to cover the set of V regions. The same method was used to generate primers for the light-chain loci, resulting again in a set of six forward primer pools. PCRs designed to minimize crossover artifacts were performed with Phusion Flash high-fidelity PCR master mix (New England Biolabs), 50 μM deoxynucleotide triphosphates, and 0.5 μM total forward and reverse primers. First, six cycles of a one-way PCR were performed with just the forward primers and the following cycling times: 98°C for 5 min and six cycles of 98°C for 15 s, 57°C for 15 s, and 72°C for 40 s. Then, the cDNA-coated beads were magnetically removed and 1 μl of the one-way reaction was used to program a 10-μl forward and reverse primer reaction with the following cycle times: 98°C for 5 min and 17 cycles of 98°C for 10 s, 68°C for 15 s, and 72°C for 10 s. Three microliters of the resulting reaction was used as a template in a 30-μl “reconditioning” PCR, with the same cycling parameters as the previous PCR but with final extension of 72°C for 40 s. The PCRs from the same cDNA samples were pooled, the DNA was purified with a Qiagen MinElute PCR purification kit (Qiagen), and equal amounts of each were mixed to create the sequencing sample.

Long-read MPSS

The DNA sample generated by the above protocol was used to generate a titanium sequencing library according to the internal protocol from Roche. The IgG and IgM samples were used for a one-half plate sequencing reaction, and the Igκ sequences were similarly generated in a separate titanium sequencing run.

Calculation of significance

The raw MPSS sequences were screened with the following criteria: (i) the presence of matching forward and reverse primer sequences with barcodes corresponding to a combination used with a cDNA sample, (ii) an open reading frame spanning the forward and reverse primers, and (iii) a minimum interprimer sequence length of 250 bases.

To identify sequences that are overrepresented in experimental immunized samples compared to naïve control samples, we performed the following analysis on the set of sequences: First, variable domain predicted protein sequences from the IgG, IgM, or Igκ isotypes were compared to each other using the Smith-Waterman algorithm with a PAM250 substitution matrix (21, 22). Then, the sequences were organized into a phylogenetic tree with the Scoredist method to convert the substitution matrix similarity score into a distance score, and then the neighbor-joining algorithm was applied to construct a phylogenetic tree (2325). A value representing the chance of the observed distribution of the sequences in the samples occurring at random, the e-score, was calculated for each node and leaf element of the tree with the following formula, where n is the number of occurrences of a sequence in both experimental and control sample sets, and m is the number of occurrences in the experimental sample set. P is the probability at random of a sequence occurring in the experimental set determined by the ratio of samples in the experimental set to samples in the control set, and C(n,k) is the binomial coefficient, n!/(k!(nk)!):k=mnC(n,k)Pk(1P)(nk)

For individual sequences at the leaf nodes of the phylogenetic tree, the sample counts are simply the number of times that sequences occur in the sample set, whereas for branch nodes of the tree, the counts are the sum of occurrences of the sequences beneath the branch. A final parameter that is incorporated into the calculation is a number cutoff value for counts of individual sequences in a sample. If the cutoff value is set to one, then an individual sequence is counted just once per sample no matter how many times it may occur. At the other extreme, if the cutoff value is infinite, then no matter how many times a sequence occurs in a sample, each occurrence is treated as an independent event for the purposes of the calculation. Because the occurrence of a sequence resulting from the experimental protocol is neither completely independent nor completely dependent, an intermediate cutoff value of 30 was used for calculations. To reduce redundancy while reporting node e-scores, we used only nodes of local peak significance, those with less significant child, and parent nodes to create Fig. 2.

Peptide microarrays

PvRII peptides were generated as 15-nucleotide oligomer overlapping 11 amino acids, functionalized with OAC group on the terminal amine (Sigma). Peptides and controls were spotted onto glass slides by Arrayit. Peptides were printed in duplicate side by side, and consecutive sequences were distributed randomly to avoid edge effects. Controls included fluorescent markers as well as whole protein to assay specific and nonspecific interactions. Twenty-four individual arrays were printed on a single slide, allowing simultaneous testing of samples. Sera were normalized for PvRII-reactive IgG antibodies by diluting in a 1% bovine serum albumin (BSA)/phosphate-buffered saline (PBS) solution. Arrays were processed with Arrayit Solutions and were detected with goat anti-mouse IgG, Fc-specific, DyLight 549 (Jackson ImmunoResearch). Arrays were scanned with the GenePix 4000b microarray scanner.

Inhibition of PvRII-DARC binding by mouse sera raised against recombinant PvRII formulated with different adjuvants

Sera from mice immunized with recombinant PvRII formulated with different adjuvants were tested for inhibition of PvRII-DARC interaction at different dilutions with an enzyme-linked immunosorbent assay (ELISA)–based functional binding assay as described earlier (19). Batches of recombinant PvRII corresponding to the Sal I strain as well as heterologous PvRII alleles (P, O, AH, C, and T) were produced and used for binding assays as previously described (19). Mouse sera were tested for inhibition of binding of both homologous as well as heterologous PvRII to DARC at a dilution of 1:4000.

Statistical analysis

Statistics on data from serum samples were performed on the large data set comparing the adjuvant groups as a function of PvRII variant with the R Statistical Computing Package (26) version 2.8.1; analysis of variance (ANOVA) was performed with Tukey’s honestly significant difference post hoc test to compensate for the large number of comparisons. For smaller data sets, GraphPad Prism version 5.00 for Windows (GraphPad Software, was used to perform one-way ANOVA with Tukey’s post hoc test to determine significance; similarly, the package was used to calculate Spearman’s r in Table 2.


  • * These authors contributed equally to this work.

References and Notes

  1. Acknowledgments: We thank X.-X. Tan (SeqWright, Houston, TX) for performing the MPSS procedure. We also thank J. Vergara for technical support in analysis of peptide arrays. Funding: This work was supported in part by a grant from PATH Malaria Vaccine Initiative to C.E.C. C.E.C. is a recipient of the TATA Innovation Fellowship, Department of Biotechnology, Government of India, and is a member of the EVIMalaR Network of Excellence supported by the European Commission. This research was supported in part by grant #42387 from the Bill and Melinda Gates Foundation. Author contributions: S.R.W. performed the computations on sequences. All authors contributed to experimental design and execution. Competing interests: D.C. and S.G.R. are named inventors on patent filings PCT/US2007/021017 (US-2008-0131466) and PCT/US2009/045033 (US-2009-0181078) pertaining to GLA formulations. The other authors declare that they have no competing interests.
  • Citation: S. R. Wiley, V. S. Raman, A. Desbien, H. R. Bailor, R. Bhardwaj, A. R. Shakri, S. G. Reed, C. E. Chitnis, D. Carter, Targeting TLRs Expands the Antibody Repertoire in Response to a Malaria Vaccine. Sci. Transl. Med. 3, 93ra69 (2011).

View Abstract

Navigate This Article