Clinical expression and antigenic profiles of a Plasmodium vivax vaccine candidate: merozoite surface protein 7 (PvMSP-7)

This work was approved by the Institutional Review Board for Human Research of the Faculty of Medicine, Chulalongkorn University, Thailand (IRB No. 104/59). Written consent was obtained from all participants or from their parents or guardians prior to admission into the study.

For transcriptomic analysis, five human patients were recruited from each of two malaria endemic areas of Thailand: (i) Ubon Ratchathani, located along the North-eastern border between Thailand and Cambodia, and (ii) Yala, located along the Southern border between Thailand and Malaysia. Patients of any age who attended malaria clinic with fever but did not show severe malaria symptoms were recruited to the study. Hence, all patients were symptomatic and were asked to provide information on the days of fever they had experienced (Table 1). Patients were screened with microscopic examination by an experienced laboratory technician and ~ 600 μL of venous blood were collected from P. vivax-infected patients. Blood samples were preserved in RNAlater^® solution (Ambion, Grand Island, NY, USA) with a 1:1 ratio and stored at − 20 °C until processing. Whole RNA was extracted from 500 μL of each blood sample using a QIAamp RNA blood mini kit (Qiagen, Hilden, Germany), following the manufacturer’s recommendations and stored at − 80 °C until used. A nested rRNA polymerase chain reaction (PCR) capable of detecting all five human malaria species [34] was used to confirm the presence of P. vivax and the absence of other species.

For screening of the MSP-7 peptide microarray, plasma was collected from 64 vivax-malaria patients from Ubon Ratchathani province (North-eastern Thailand) between 2013 and 2016. Negative controls, i.e. malaria naïve donors (n = 20), were recruited from non-malaria endemic areas at Chulalongkorn hospital, Bangkok. These patients had not lived/spent time in malaria-endemic regions of Thailand, as far as was known. Approximately 2 mL of venous blood was collected into EDTA anticoagulant from patients with single-strain P. vivax infection, confirmed by nested rRNA PCR [34], and plasma was collected by centrifugation. These plasma were sorted into five experimental age cohorts: 0–14 years (n = 11), 15–29 years (n = 22), 30–44 (n = 19), 45–59 (n = 10) and 60–74 (n = 2). A master plasma sample was prepared for each age group by pooling 5 µL of plasma from each vivax-infected patient. All samples were preserved at − 80 °C until use.

Ten RNA samples were treated with DNase, followed by Epicentre Globin-Zero Gold kit to deplete rRNA and globin transcripts. RNA-seq libraries were prepared at the Centre for Genomic Research, University of Liverpool using the NEBNext Ultra Directional protocol. The RNA-seq libraries were sequenced on the Illumina HiSeq4000 platform in a single lane, generating ~ 30 million paired-end 150 bp reads for each sample. The quality of the libraries was affirmed by Agilent 2100 Bioanalyser. RNA-seq data have been deposited in the EMBL-EBI ArrayExpress database (http://​www.​ebi.​ac.​uk/​arrayexpress/​experiments) under Accession Number E-MTAB-6753.

The initial processing and quality assessment of the raw reads began with base calling and de-multiplexing of indexed reads using CASAVA version 1.8.2 (Illumina) to produce ten samples from one lane of sequence data in fastq format. The Illumina adapter sequences were trimmed from the raw fastq files by using Cutadapt version 1.2.1 [35]. The option “-O 3” was specified to trim off the 3′ end of any reads that matched to the adapter sequence over at least 3 bp. The reads were further trimmed to remove low quality bases, using Sickle version 1.2 with a minimum window quality score of 20. Before mapping the filtered reads to the reference genome, reads shorter than 10 bp were removed. Subsequently, all the filtered reads were mapped to the human genome (GRCH37) using TopHat2 [36]. The unmapped reads were then aligned to the PvP01 reference genome [37] using TopHat2. Following the alignment of reads to the P. vivax genome, the read counts of each gene were quantified with featureCounts [38]. Read counts were estimated by dividing the total number of mapped reads by the length of the gene, and then scaling the estimates to one million. The genome-wide coverage of each sample was determined using Qualimap 2 [39]. Transcript abundance, expressed as Read Count per Million bases (CPM) was estimated by DESeq2 [40]; this program normalizes data for both library size and compositional differences (e.g. the presence and absence of large abundance transcripts) using median ratios of log₂-transformed raw counts. To establish reproducibility among the transcriptomes and to identify possible low-quality outliers, Pearson correlations of transcript abundances (N = 5838) were estimated for each pair of samples in R version 3.4.3 and corrplot [41].

Principal component analysis (PCA) was performed on the log-transformed, normalized CPM values for 5838 genes estimated by DESeq2 for ten patients. The PCA consistently placed the ten samples into four groups, which were adopted subsequently to identify differentially expressed genes (DEGs) in a pairwise manner, using DESeq2 [40]. Genes with a false discovery rate (FDR) threshold below 0.05 after correcting for multiple tests were deemed significantly differentially expressed. Additionally, differential expression analysis was also performed analysis with EdgeR [42], but the DEGs and PCA plot did not deviate substantially from the DESeq2 results. Thus, the latter was used in downstream analyses.

The DEGs identified for each patient group were imported into coseq, an R package to identify clusters of co-expressed genes [43]. The purpose was to associate differentially-expressed PvMSP-7 genes with other genes that are known to be developmentally regulated. A robust association with a known life-stage marker would then allow developmental regulation within the PvMSP-7 family to be inferred. Coseq uses Gaussian mixture models to cluster all the genes based on the proportion of normalized counts in their expression profiles. The DESeq2 data were fitted to a Gaussian mixture model on either arcsine- or logit-transformed normalized profiles. One hundred clusters were tested in each transformation with the following commands; arcsin_transformed < − coseq(counts, K = 2:100, model = “Normal”, transformation = “arcsin”) and logit_transformed < − coseq(counts, K = 2:100, model = “Normal”, transformation = “logit”). To choose accurately between two transformation models, coseq calculated the corrected integrated completed likelihood (ICL) values from these two models and selected the number of clusters and preferred model-transformation based on the highest corrected ICL value.

To infer the functional consequences of differential gene expression between patient groups, enrichment analysis was carried out on functional terms associated with the DEGs, using PlasmoDB release 35 [44]. First, heat-maps were plotted for the DEGs from each patient group comparison, relating Euclidian distance-based dendrograms of all DEG expression profiles and each patient group. These heat-maps neatly divided DEGs into clades based on their expression profiles and these were then subjected to enrichment analysis for both Gene Ontology (GO) terms [45] and Kyoto Encyclopaedia of Genes and Genomes (KEGG) pathways [46]. Functional terms and pathways were deemed significant where the adjusted p-value was < 0.05. Second, the genes belonging to clusters returned by Coseq were also subjected to the same enrichment analyses.

A custom peptide microarray was designed that included the entire, predicted amino acid sequences of all 13 PvMSP-7 paralogs. The microarray contained 15-mer peptides of each PvMSP-7 gene with an overlap of 11 amino acids, printed in duplicate: a total of 2346 peptides. The slides were manufactured by PepperPrint (Heidelberg, Germany) and coated with a poly(ethylene glycol)-based graft copolymer with a thickness of 13.5 nm and an additional three amino acid linker (β-alanine, aspartic acid, and β-alanine), which ensure optimal epitope orthogonal attachment and presentation. Each microarray slide consisted of three identical array copies, each framed by flag anti-HA (YPYCVPDYAG, 52 spots) as a quality control measurement. The signal intensities of these control peptides indicate the spot uniformity and binding specificities.

Each microarray was treated with 1% bovine serum albumin (BSA) blocking buffer for 30 min at room temperature. To determine the background intensity, one microarray was then incubated with goat anti-human IgG (H + L) DyLight680 antibody (Rockland Immunochemical Inc., USA) that was diluted 1:2500 in staining buffer (phosphate-buffered saline (PBS) with 0.05% Tween20 and 10% blocking buffer). Clean microarrays were then incubated overnight at 4 °C with 10 µL pooled plasma from one of five patient age cohorts (see above), diluted 1:50 with staining buffer. The peptide microarray was washed three times with standard buffer (PBS with 0.05% Tween20, pH 7.4) using an orbital shaker at 140 rpm. The microarrays were washed in 1 mM Tris dipping buffer (pH 7.4) to remove all contaminants from the array surface and dried by tapping the edge of the slide on tissue paper. To ensure that all microarrays were responding correctly, all steps were repeated with the Cy3-conjugated anti-HA control antibody supplied by PepperPrint (Heidelberg, Germany). A sixth microarray was assayed to provide a negative control consisting of naïve human plasma taken from individuals in malaria-free areas.

Microarrays were illuminated on an Agilent G2565CA microarray scanner, with the AgilentHD red colour single channel at 10 µm resolution and output to a 16-bit greyscale tiff files. The mean and median of local background and foreground fluorescence intensity values for all peptide spots were quantified using PepSlide analyser (SICASYS Software, Heidelberg, Germany). Correction for background intensity and normalization for variation between and within arrays were performed using the LIMMA package [47] implemented in R version 3.4.3. Background correction was performed on each array by subtraction of local background estimates from each foreground intensity value [48]. This step was performed to eliminate the effects of non-specific binding across the arrays. Spot intensities were also normalization for between-array variation using the quantile principle in LIMMA [49]. The quantile function compensates for systematic measurement errors between-arrays, producing a more uniform intensity distribution and leaving only the true biological differences in the datasets.

Statistical analysis was performed using the LIMMA package [47]. Normalized data were fitted to a linear model and peptides with significant responses to antibody, i.e. normalized fluorescent intensity significantly greater than background levels, were determined in all experimental groups by comparing the log transformation of expression intensity. A simple Bayesian model was used to estimate significant peptides between the experimental group and negative control [47, 49]. The summary statistics were computed by the eBayes() function, returning a list of significantly expressed peptides were displayed. Benjamini and Hochberg’s method or false discovery rate (FDR) was used to correct the p-value for multiple tests. FDR < 0.05 were considered statistically significant.

RNA sequencing generated ~ 30 million 150 bp paired-end reads for each of ten clinical isolates (Table 2). The samples were collected from field hospitals in remote areas and suffered a variable degree of degradation prior to sequencing; RIN values were consistent but low (between 2 and 2.8); moreover, sequence coverage was variable and relatively low for some isolates. For each isolate, all reads were aligned to the human genome (GRCh37) and then the unmapped reads were aligned to the P. vivax reference genome (PvP01; [37]). Typically, more than 70% of reads were derived from human, while reads mapping to P. vivax ranged between 0.78 and 22.38%. Mean genome coverage varied widely among the ten isolates, between 1.07× and 78.52× (mean = 18.8×). Isolate UBT3089 has the highest mean coverage of 78.52×, while five samples have a value < 10×. Global transcript abundance, expressed in Read Count per Million bases (CPM) and normalized for gene length and read number, was calculated for all genes using DESeq2 after mapping the data to the reference genome with TopHat2 (Additional file 1). To evaluate the reproducibility of the results in the light of this variation in coverage, Pearson correlation coefficients were calculated between each sample for gene-level CPM estimates; such coefficients ranged from 0.59 to 0.91 (mean = 0.72).

Plasmodium vivax infrapopulations in clinical settings usually comprise variable mixtures of different parasite life stages, which, depending on the precise compositional balance, could be reflected in stage-specific expression patterns. Unfortunately, the proportions of parasite developmental stages could not be determined from microscopic slides as these were not retained by the field hospitals. Instead, the relative proportions of parasite stages were characterized by quantifying the expression of stage-specific markers. Ten patients were separated into groups based on a principal component analysis of genome-wide transcript abundance, which was independent of age and days of patent infection (i.e. fever). The ten patients clustered into four groups using the expression profiles of 5838 genes produced by the DESeq2 analysis (Fig. 1; Additional file 1). Together, the two largest principal components account for 69% of the total variation in transcript abundance. These groups do not reflect coverage, so YL3113 can be clearly distinguished from YL3111, YL3112 and YL3114, despite these four samples having a comparable low coverage; similarly, UBT3089 and UBT3091, which had the highest sequence coverage values do not cluster together. This reassures that the variation seen is not simply an artefact of the degree of degradation within the samples. The study focused on patient group 4 because these patients had a distinct MSP-7 expression profile (see below). Patient group 4 included two patients that had experienced longer patency (i.e. 7 days), than patients in groups 1–3, who had experienced fever for 2–3 days. These four patient groups were the basis for pairwise comparison of PvMSP-7 expression and analysis of differential gene expression. The relatedness among the ten parasite isolates was assessed to rule out the possibility that relatedness might explain the patient groups. 42,988 biallelic single nucleotide polymorphisms (SNPs) were extracted from mapping data for each RNA-seq dataset. Principal component analysis of these SNPs shows that isolates cluster by geographical origin, and do not recapitulate the patient groups (Additional file 2). Therefore, affinities among isolates based on transcript abundance are not simply a reflection of genetic relatedness or geographical origin.

To investigate the expression of PvMSP-7 during diverse clinical infections, and to examine variation in transcriptional profile within the family, normalized transcript abundance in each isolate for the 12 paralogous genes was plotted as a heat map, shown in Fig. 2. From the relationships among isolates, seen across the top of Fig. 2, it is immediately clear that there is substantial variation in PvMSP-7 expression among patients, which only partly coincides with patient groups, i.e. genome-wide expression profile. There are also consistent differences among PvMSP-7 paralogs; 7A and 7F display higher abundance than all others (mean log2 normalized read counts of 7.76, 7.19, and 6.28, respectively). Even in patients with otherwise low PvMSP-7 expression, such as YL3112 and YL3114, these two paralogs maintain the same high level of expression. Conversely, 7E, 7G, and 7J were uniformly rare transcripts (mean log2 normalized read counts of 1.44, 0.81 and 0.33, respectively); the latter is perhaps unsurprising since PvMSP-7J is a pseudogene [13]. Other paralogs are only transiently abundant, for example, 7H and 7I (mean log2 CPM of 4.03 and 3.58 respectively), were highly abundant in patient group 4 only (mean log2 CPM of 8.65 and 7.29 respectively); indeed, in these two patients that had experienced longer patency, their expression was comparable to 7A and 7F. Hence, when differential expression analysis was performed on normalized transcript abundance values (data shown in Additional file 3), significant differences in the abundance of PvMSP-7H were found (mean log2FC = 4.86; mean p_adj = 0.026) when patient group 4 was compared with all other groups; and likewise for 7I (mean log2FC = 3.84; mean p_adj = 0.024) when patient group 4 was compared with groups 2 and 3, and 7C (log2FC = 6.48; p_adj = 0.0002), when compared with patient groups 1 and 3. Finally, MSP-7L (log2FC = 3.64; p_adj = 0.007), 7K (log2FC = 4.54; p_adj = 0.03) were differentially expressed when patient group 4 was compared with group 3 only. In contrast, no significant differences in PvMSP-7 expression were found in pairwise comparisons among patient groups 1–3.

Given that critical differences were seen in expression profile among PvMSP-7 paralogs in distinct patient groups, it was necessary to better understand the global gene expression landscape in these groups, to help explain the PvMSP-7 patterns. Having identified differentially expressed genes (DEGs) in pairwise comparisons of patient groups using DESeq2, the log2 normalized read counts for all DEGs in each comparison were plotted in heat-maps. The highest number of DEGs (1493; Additional file 3) was observed between patient groups 3 and 4, and these are plotted in Fig. 3. The heat-map relates gene expression profiles (rows) with clinical isolates (columns); the vertical dendrogram clusters genes based on their expression profile, the horizontal dendrogram clusters isolates based on their global expression profile. The heat-map is divided into sectors based on clades observed in vertical dendrogram, and Gene Ontology enrichment analysis was performed on each sector.

The 1493 DEGs identified between patient groups 3 and 4 may be segregated into five sectors (i to v). Genes in sector i (n = 252) and ii (n = 456) were significantly less abundant in group 4 and enrichment analysis showed that these genes are associated with Nucleic acid binding, ATP-dependent peptidase activity, helicase activity, tRNA processing, nitrogen compound metabolism, and RNA metabolism. Genes in sector iv (n = 192) were also more abundant in group 3 and GO terms associated with these related to macromolecular complex, eukaryotic translation initiation factor 3 complex, Chaperonin-containing T-complex, small molecule metabolic process, and protein folding. Conversely, genes in sector iii (n = 403) and iv (n = 190) were significantly more abundant in group 4 isolates. These genes were enriched for GO terms associated with Rhoptry, cell surface, protein–DNA complex, nucleus, regulation of metabolic process, DNA binding, and protein binding. PvMSP-7K, 7I, 7H, 7C, and 7L were included among the sector iii genes.

Comparisons of patient group 4 with 1 and 2 identified fewer DEGs but consistent patterns of differential expression. A heat-map of the 251 DEGs detected between Groups 4 and 1 is shown in Additional file 4. Sector i contains genes that are significantly less abundant in group 4 (n = 125); these are associated with GO terms for Cell surface, Maurer’s cleft, host cell surface binding, and Uridylyltransferase activity, and implicate such genes as reticulocyte binding surface protein (PvP01_00004240; log₂ FC = 9.27, p_adj = 2.97e⁻¹⁵), tryptophan-rich protein (PvP01_0504200; log₂ FC = 7.32, p_adj = 0.001) and a small heat shock protein HSP20 (PvP01_0518800; log₂ FC = 5.17, p_adj = 0.034). Sector ii also contains genes that are expressed more in group 1 isolates (n = 71); here, GO terms are associated with crystalloid, cell surface, and host cell cytoplasm. Sector iii (n = 22) and sector iv (n = 33) contain genes that were more abundant in group 4 isolates. GO terms enriched within this gene set concern Rhoptry, cell surface, and proteolysis, and implicate known invasion-related genes such as high molecular weight rhoptry protein 3 (PvP01_0703800; log₂ FC = 6.38, p_adj = 0.006), high molecular weight rhoptry protein 2 (PvP01_0727900; log₂ FC = 5.81, p_adj = 0.0004), 6-cysteine protein (PvP01_1136400; log₂ FC = 6.16, p_adj = 0.008) and merozoite surface protein MSA180 (PvP01_0814200; log₂ FC = 7.25, p_adj = 9.46e⁻⁵). The comparison of patient groups 2 and 4 is shown in Additional file 5. 351 DEGs were identified and these segregate into five sectors in the heat-map. No distinct expression patterns are observed within sector i (n = 99) and ii (n = 49). Genes in sector iii (n = 47) and iv (n = 142) were less abundant in group 4 isolates. They are associated with functional terms for purine metabolism and host cell surface binding.

While the analysis of differential expression shows that these clinical isolates have different transcriptional landscapes, and clearly associate expression of PvMSP-7C, 7H and 7I with patient group 4 only, relatively few DEGs contribute to the enrichment analysis and so enriched functional terms have limited capacity to explain the transcriptional differences. Among the DEGs, various specific markers of parasite development were observed, such as schizont egress antigen-1 (SEA1; PVP01_0607000), which was more abundant in patient group 4 than 3 (log₂ FC = 4.92; p_adj = 9e⁻⁰⁶) and gametocyte development protein 1 (GDV1; PVP01_0734100), which was most abundant in patient group 3 than 4 (log₂ FC = 3.73; p_adj = 0.0001). Thus, to explore developmental composition as an explanation of transcriptional differences, associations between PvMSP-7 expression profiles and such specific markers were sought using co-expression analysis.

The correlation of expression profiles for 1493 DEGs across patient groups 4 and 3 produced 14 transcript clusters (Additional file 6). Of these, five clusters (including 740 DEGs) are of interest because they describe a dynamic that clearly distinguishes the patient groups; these are shown in Fig. 4. The top panel (Fig. 4a) describes a cluster that is more abundant in patient group 4 than group 3. This cluster included PvMSP-7K, 7I, 7H and 7C, reflecting the differential expression analysis, in which log₂ fold change ranged from 4.21 to 6.48. These MSP-7 transcripts were co-expressed with several stage-specific genes with known functions in erythrocyte invasion, such as early-transcribed membrane protein (ETRAMP, PvP01_0618300), schizont egress antigen-1 (SEA1, PvP01_0607000), rhoptry neck protein 4 (RON4, PvP01_0916600). Figure 4b describes a transcript cluster that included MSP-7L (log₂ FC = 3.64) and other invasion-related genes, such as plasmepsin V (PMV, PvP01_1231100), merozoite organizing protein (MOP, PvP01_0715400), and rhoptry neck protein 5 (RON5, PvP01_0517600).

Contrasting with these two clusters, Fig. 4c–e displays transcriptional profiles that are most abundant in patient group 3, declining in group 4. These clusters include other stage-specific markers indicative of different parasite life stages, such as gamete antigen 27/25 (PvP01_0422700), sporozoite and liver stage tryptophan-rich protein (TryThrA, PvP01_0532600), and liver specific protein 3 (PvP01_0405000). Besides these, various multi-copy gene families typical of asexual-stage cells are co-expressed, such as tryptophan-rich protein (TRAG36, PvP01_0119200), Plasmodium interspersed repeat (PIR, PvP01_0816000), Plasmodium exported protein (EXP, PvP01_0300700) and erythrocyte membrane-associated antigen (EMAA, PvP01_0103700). Pearson’s correlation coefficient (r) was estimated for each stage-specific marker in relation to MSP-7H. Overall, positive correlations (r > 0.90) were observed for schizont/merozoite-stage expression gene while, negative correlations (r > − 0.85) were seen with sporozoite, liver, and gametocyte stage markers. MSP-7I and 7C genes enriched in patient group 4 gave consistent results when correlated with stage-specific genes.

The correlation of expression profiles for 251 DEGs across patient groups 4 and 1 produced seven transcript clusters (Additional file 6); four are shown (including 208 DEGs) in Additional file 7. In this comparison, PvMSP-7H and 7C were significantly differentially expressed (log₂ FC = 5.31, 7C and 6.33, respectively). These two PvMSP-7 paralogs were placed in different clusters (Additional file 7a and b), although weakly positively co-expressed (r > 0.60). They were significantly co-expressed with various invasion-related genes, including PIESP1, PMV, rhoptry neck protein 3 (RON3, PvP01_1469200), serine-repeat antigen-1 (SERA, PvP01_0417100), subtilisin-like protease 3 (SUB3, PvP01_1026800), and high molecular weight rhoptry protein 3 (RhopH3, PvP01_0703800). Genes negatively associated with PvMSP-7H and 7C included two gametocyte stage-specific markers (gamete release protein (PvP01_0115300) and gamete antigen 27/25 (PVP01_0422700).

The correlation of expression profiles for 351 DEGs across patient groups 4 and 2 produced seven transcript clusters (Additional file 6); four clusters (comprising 176 DEGs) are shown in Additional file 8. These clusters present a similar picture to that above. Expression of MSP-7H and 7I was positively correlated with SEA1 (Additional file 8b), but contrasted with expression of liver specific protein 3 (Additional file 8c), and gamete antigen 27/25 (Additional file 8d), as well as various Plasmodium exported proteins and tryptophan-rich proteins associated with asexual stages.

From the differential expression and co-expression analyses, it is apparent that MSP-7H and 7I are positively co-expressed with the schizont stage-specific marker SEA1, and various other invasion-related genes. This indicates that isolates from patient group 4 are distinguished from other groups by a greater proportion of schizonts in the parasite infrapopulation, relative to other life stages. Accordingly, MSP-7H and 7I have a negative association with liver-stage and gametocyte-stage markers.

Immuno-reactive linear B-cell epitopes were identified by screening plasma from naturally-infected individuals with a customized PvMSP-7 peptide microarray, which consisted of 1173 15-mer peptides spanning the complete coding sequence of 12 PvMSP-7 paralogs (see Additional file 9). The peptide microarray was incubated with plasma from patients with single P. vivax infection, pooled into five age groups. The purpose of separating plasma into these arbitrary age divisions was not to examine the effect of age per se, but to ensure that the epitopes identified were significantly more responsive than naïve samples regardless of patient age. A secondary goat anti-human IgG antibody with conjugated fluorophore (DyLight 680) was applied to determine the plasma reactivity profile for each peptide. Figure 5 shows that plasma from the 30–44 age group displayed immuno-reactivity to the most epitopes (Fig. 5c), while the fewest responses was observed from the 45–59 age-group (Fig. 5d). A negative control comprised plasma from malaria-naïve individuals; some fluorescence was observed from the negative control, but these responses were relatively weak. Furthermore, responses from the naïve control did not generally coincide with immune-reactive peptides bound by plasma samples. By comparing the spots between the white lines in Fig. 5a–f, the characteristic response seen in infected plasma (Fig. 5a–e) is not reproduced in the naïve control (Fig. 5f).

Fluorescent intensity values were corrected for background intensity (i.e. non-specific binding by secondary antibody), and then also for inter- and intra-array variation (Additional file 10). Normalized intensity values were then compared to corresponding responses from the naïve control using the LIMMA package [47] to determine the significance of each response. The number of significantly immunogenic peptides found among 12 PvMSP-7 paralogs was 238, but these belonged disproportionately to specific paralogs and specific plasma age groups. Figure 5h arranges these putative B-cell epitopes in order of decreasing spot intensity. Four PvMSP-7 paralogs [PvMSP-7A (21%), 7B (18%), 7K (12%) and 7L (20%)] contained a large proportion of predicted epitopes; six other PvMSP-7 proteins (PvMSP-7C, 7E, 7F, 7G, 7H, and 7J) each contributed less than 5% of putative epitopes, while no significant responses were detected against PvMSP-7D.

Figure 5g shows the number of putative epitopes by plasma age group as a Venn diagram. Age group of 30–44 displayed the highest number of the significant intensity values (n = 123), followed by 0–14 (n = 121), 15–29 (n = 110), 60–74 (n = 83), and 45–59 (n = 71). The responses for the 45–59 age cohort are weak by comparison with others, and not, on appearance, substantially different to the naïve cohort. The lack of response from the 45–59 age cohort cannot be explained by sampling artefact, given that it consisted of 10 individuals sampled at different times. Nevertheless, although plasma from the 45–59 age cohort are not generally more responsive than the naïve plasma, all of the ‘universal’ epitopes identified below were specifically recognized by these infected plasma, and to a degree significantly greater than naive plasma.

Of the 236 spots that responded significantly to infected plasma, 13 putative epitopes were detected in all age groups (Table 3). These universal epitopes were found in five PvMSP-7 proteins (PvMSP-7A, 7B, 7H, 7I, and 7L), but 6/13 were derived from PvMSP-7A. These universal epitopes are not always among the most immuno-reactive (Fig. 5h), but 6/13 occur among the 10% most responsive peptides and three of these are derived from MSP-7A. The position of these putative epitopes in relation to predicted protein secondary structure is shown in Fig. 6.