Epitope-based sieve analysis of Plasmodium falciparum sequences from a FMP2.1/AS02_A vaccine trial is consistent with differential vaccine efficacy against immunologically relevant AMA1 variants.

Abstract

To prevent premature dismissal of promising vaccine programs, it is critical to determine if lack of efficacy in the field is due to allele specific-efficacy, rather than to the lack of immunogenicity of the candidate antigen. Here we use samples collected during a field trial of the AMA1-based FMP2.1/AS02 _A malaria vaccine, which incorporates the AMA1 variant encoded by the reference Plasmodium falciparum 3D7 strain, to assess the usefulness of epitope-based sieve analysis for the detection of vaccine-induced allele-specific immune responses.

The samples used are from volunteers who received the malaria vaccine FMP2.1/AS02_A or a control (rabies vaccine), during a vaccine efficacy field trial, and who later developed malaria. In a previous study, P. falciparum DNA was extracted from all samples, and the ama1 locus amplified and sequenced. Here, a sieve analysis was used to measure T and B-cell escape, and difference in 3D7-like epitopes in the two treatment arms.

Overall, no difference was observed in mean amino acid distance to the 3D7 AMA1 variant between sequences from vaccinees and controls in B-cell epitopes. However, we found a significantly greater proportion of 3D7-like T-cell epitopes that map to the AMA1 cluster one loop (c1L) region in the control vs. the vaccinee group (p=0.02), consistent with allele-specific vaccine efficacy. Interestingly, AMA1 epitopes in infections from vaccinees had higher mean IC50, and consequently lower binding affinity, than epitopes generated from the control group (p=0.01), suggesting that vaccine-induced selection impacted the immunological profile of the strains that pass through the sieve imposed by the vaccine-induced protection. These findings are consistent with a vaccine-derived sieve effect on the c1L region of AMA1 and suggest that sieve analyses of malaria vaccine trial samples targeted to epitopes identified in silico can help identify protective malaria antigens that may be efficacious if combined in a multivalent vaccine.

1. Background

To best identify promising vaccines, and therefore avoid premature withdrawal or halting of their development, it is essential to assess whether the lack of efficacy observed during clinical trials is due to vaccine-induced allele-specific immune responses, which can potentially be overcome with multivalent vaccines, and not to poor overall immunogenicity or low protective power of the candidate antigens. Sieve analysis, introduced by Gilbert et al. [1], aims to determine whether a vaccine induces allele-specific immunity, thus leading to the preferential selection of non-vaccine variants. It compares genetic data generated from samples collected during vaccine trial follow-up from volunteers in the vaccine and control study arms to measure vaccine protection against different pathogen genotypes. Knowledge of the relative risk of each variant, i.e., its probability of escaping vaccine-induced protection relative to the genotype in the vaccine formulation, may inform the development of a vaccine that protects against genetically diverse organisms, such as Plasmodium falciparum [2].

Sieve analyses have been applied to human immunodeficiency virus vaccine clinical trials [3–8] and, to a lesser extent, in the malaria field [9, 10]. In a phase 2 vaccine trial targeting P. falciparum apical membrane antigen 1 (AMA1) [11], the AMA1 whole protein variant used in the vaccine, from the P. falciparum 3D7 reference strain, represented only 2.9% of the parasite population at the study site. This resulted in negligible vaccine efficacy, despite a high level of protection against parasites encoding an ama1 allele identical to 3D7 in the region of the cluster 1 loop (c1L) which is comprised of amino acid residues 196, 197, 199, 200, 201, 204, 206 and 207 [10, 12]. Likewise, monovalent vaccines based on the merozoite surface protein 2 (MSP2) [13], circumsporozoite protein (CSP) [9], and whole P. falciparum NF54 parasites, either radiation-attenuated or under chemoprophylaxis, have shown incomplete protection against infection by heterologous strains in controlled human malaria infection (CHMI) studies [14–16] and against clinical or severe malaria in the field [17], partly due to strain-specific immunity. These results highlight the need for field studies prior to vaccine design, so that the most prevalent antigen variants can be identified and used in vaccine formulations [18].

A FMP2.1/AS02_A vaccine trial was conducted in 2009 in Bandiagara, Mali, to assess the efficacy of a monovalent 3D7-based malaria vaccine targeting the AMA1 protein against infection and/or clinical episodes by circulating P. falciparum strains, in young children [10]. Although the vaccine did not induce protection against clinical malaria episodes [10], the primary endpoint, subsequent exploratory analyses showed that there was a protective effect directed against homologous (genetically identical) strains with respect to specific residues in the AMA1 cluster 1 loop [10, 12]. These studies identified the presence of allele-specific protection induced by the FMP2.1/AS02_A vaccine by estimating the incidence of the vaccine strain in the two treatment arms, based on an 8 amino acid residue haplotype composed of positions known a priori to be the target of protective antibodies [19–21], evaluated the contribution to protection of individual residues of the protein ectodomain, and estimated the relative risk of infection associated with different haplotypes [10–12]. Here, we aim to apply epitope-based sieve analyses on an agnostic basis, assuming no a priori knowledge of residues associated with protection, to determine its applicability to the identification of protective epitopes in poorly studied antigens. To this end, we conducted detailed sieve analyses based on the bioinformatically-identified CD4+ and CD8+ T cell epitopes, as well as B cell epitopes. Hence, we use the same well-known dataset as a system to study the usefulness of a sieve analysis as described by Gilbert and colleagues [22] to identify vaccine-induced epitope escape and differential binding affinity in P. falciparum. These analyses also include an evaluation of differences in physicochemical properties between control-derived and vaccinee-derived parasite sequences at the whole protein level [3]. Our analyses show that sieve analyses based on in silico-predicted epitopes are powered to detect allele-specific efficacy in c1L. Moreover, these analyses show a difference in the proportion and binding affinity of CTL epitopes derived from sequences generated from FMP2.1/AS02_A vaccinees and the control group.

2. Materials and methods

2.1. Ethics statement.

The parent study was approved by the Institutional Review Boards of the University of Sciences, Techniques and Technology of Bamako, Mali; the University of Maryland Baltimore; the Walter Reed Army Institute of Research; and the United States Army Surgeon General. The exploratory study being conducted here is part of the listed analyses planned in the parent study.

2.2. Sequence origin.

As described previously, 400 children were randomized 1:1 to receive either the FMP2.1/AS02_A malaria vaccine or the rabies vaccine as a control in an efficacy, safety and immunogenicity trial [23]. Children were immunized at days 0, 30, 60 and followed for six months, starting two weeks after the last immunization (Day 75), for malaria clinical episodes (defined as a fever equal to 37.5°C or greater and a parasite density of 2,500/microliter or greater) and/or a positive thick smear. Dried blood spots (DBS) were collected during each active and passive visit at the health clinic. From these DBS, P. falciparum DNA was extracted and the whole ama1 gene was sequenced [10]. A total of 609 Pfama1 sequences were generated from the two treatment arms during the first 6 months of follow-up. Of these sequences, 464 were single and/or predominant infections, as defined by secondary peak heights less than 50% of the primary peak at polymorphic codons [12]. The analyses conducted in this study are based on these 464 sequences.

2.3. Statistical analyses.

Vaccine-induced selective pressure was assessed by comparing the incidence of non-conserved residues during study follow-up. If present, this pressure should lead to measurable differences between the two treatment arms in amino acid distribution, including in the frequency of 3D7-type epitopes.

2.3.1. Testing for differences in allele frequency distribution.

To determine if the frequency of the 3D7 allele in the vaccine arm differs from expectation, allele distribution between the two study arms were investigated with Wright’s F_ST, as implemented in DNASP v6 [24], with significance assessed by permutation.

2.3.2. Comparison of CTL and B cell epitope sequences between study arms.

Using the most common human leukocyte antigen (HLA) frequencies in Mali (HLA-A*23:01, HLA-A*30:01, HLA-B*35:01, DRB1_1304, DRB1_0701, DRB1_1101, DRB1_0120) [25, 26], the T-cell epitope prediction algorithm IEDB [27] and NetMHCpan-4.0 [28] were used to search the 3D7 AMA1 sequence for potential CD4+ and CD8+ epitopes 12 amino acids in length. A putative epitope cutoff was set with binding strengths corresponding to an IC50 of at most 50 nM indicating a strong binding epitope [29, 30]. The ABCpred server [31], which has an estimated accuracy of 65.93% [32], was used to predict linear B-cell epitopes using a threshold of 0.85 [33]. A BLAST search was conducted in sequences analyzed in this study to identify sequences homologous to the AMA1 epitopes in 3D7 from infections in FMP2.1/AS02A vaccinees and controls. Those sequences were considered matched if they did not differ from the 3D7 strain in any of the amino acid residues of the epitope. Epitope binding affinity measures the concentration of added epitope required to bind half of the protein (IC50) and is a critical vaccine candidate selection parameter used to select epitopes with the smallest measured value of IC50. Epitope binding affinity was compared between FMP2.1/AS02_A vaccinees and rabies vaccinees generated epitopes, to test the null hypothesis that the means were equal, using a Mann-Whitney test. Moreover, the combined genetic distance to 3D7 of predicted AMA1 epitopes from each infection and/or clinical episodes was used to compare FMP2.1/AS02_A and control groups by means of Mann-Whitney test. Finally, a difference in the two treatment arms was assessed by comparing the proportion of ‘strong’ binder (threshold cutoff of 50 nM) [34] epitopes (SPMTLDEMRHFY, SYIATTALSHPI, KRNSMFCFRPAK, YWEHPYQNSDVY, FLTPVATGNQYL, FLTPVATVNQDL, YIATTALSHPIE, IIIENSNTTFLT) in the two groups using a chi-square test.

2.3.3. Assessment of epitopes average pairwise similarity scores.

Twelve amino acids-long CD4+ and CD8+ epitopes generated above were aligned using BioEdit [35] and MEGA X [36]. Pairwise amino acid distances between the 3D7 AMA1 protein epitopes and each of the AMA1 proteins epitopes from FMP2.1/AS02A vaccinees or the control group were measured in MEGA X [36]. The amino acid substitution type, and the p-distance model were used to generate the distance metrics. The rate of substitution at each position was considered uniform. The distribution of the mean amino acid p-distances between the 3D7 strain epitopes and AMA1 protein epitopes from FMP2.1/AS02_A vaccinees and likewise between 3D7 epitopes and AMA1 protein epitopes in the control group were compared using the Mann-Whitney test with a 2-sided p-value.

2.3.4. Evaluation of PfAMA1 physicochemical properties.

To assess differences in physicochemical properties in alleles that escape the FMP2.1/AS02A and control groups [3], amino acid sequences were characterized by three indices: charge (number of negative and positive residues), aliphatic index, and average hydropathicity [37] during study follow-up. The Expasy program (https://web.expasy.org/cgi-bin/protparam/protparam) was used to compute the physical and chemical parameters of each Pfama1-encoded sequence and the resulting values were used to assess the null hypothesis that the means of the three parameters in the two treatment arms are equal using a Mann-Whitney test.

3. Results

As part of a previous study, we generated Pfama1 sequences from 464 infections containing a single or a predominant parasite genotype (213 in FMP2.1/AS02_A group and 251 in the rabies group) from asymptomatic or clinical malaria episodes that occurred during the first six months of follow-up, starting at day 75 after the first immunization. Among these sequences, 151 were isolated from patients with repeated clinical malaria with 66 observed in the FMP2.1/AS02_A vaccine group and 85 in the rabies group [12]. This are the data used in the present study.

3.1. Assessment of AMA1 protein similarity scores and physicochemical properties.

The impact of FMP2.1/AS02_A vaccine on the distribution of ama1 whole sequences among vaccinees and controls was assessed using all sequences (n=464). To determine if there was a selective pressure imposed by immunization with the 3D7 AMA1 protein, resulting in different parasites genotypes in the FMP2.1/AS02_A and control vaccine groups, the allele distribution between the two study arms were investigated with Wright’s fixation index, F_ST. A comparison including sequences from all infections from the two treatment arms during follow-up shows a lack of genetic differentiation between the two groups (F_ST~0). When we use only sequences from clinical malaria episodes, then we see a slight differentiation in the allele pools from FMP2.1/AS02_A vaccinees and the control group (F_ST=0.003), but the difference is not significant (p=0.37).

An evaluation of PfAMA1 physicochemical properties, in particular average charge, average aliphatic index, and average hydropathicity, at the whole protein level, did not differ when sequences from the control group were compared to the FMP2.1/AS02_A vaccine group (p>0.4, p>0.5 and p>0.2, respectively).

3.2. B- and T-cell epitopes sieve effect.

3.2.1. Sieve effect on epitope distance to 3D7.

Using IEBD, NetMHCpan-3.0, ABCpred, and the most frequent MHC variants in Mali, we identified 50 strong- and average-binders among 12 amino acid-long predicted epitopes across the 3D7 AMA1 protein with polymorphic residues across samples (Table 2 and Figure 2). Eight of these epitopes were classified as ‘strong’ binder epitopes. As epitopes with high measured values of IC50 may have low affinity and specificity, these analyses were conducted using ‘strong’ binder epitopes; we conducted two separate analyses, one with only the sequences generated from individuals who experienced clinical malaria episodes during follow-up (n=151), and another with all sequences (n=464). The mean genetic distance of each sequence to 3D7 was defined as the mean epitope genetic distance across all predicted epitopes. Although sequences from FMP2.1/AS02_A vaccinees were more distant to the 3D7 reference sequence than were sequences from the control group when all malaria episodes were considered, the difference was not statistically significant (Table 1, p = 0.30). However, when the analyses were limited to only clinical malaria episodes (n=151), the epitope SPMTLDEMRHFY, detected by the HLA genotype HLA-A*0101, which is in a very polymorphic region of the ama1 gene (the cluster 1 loop region), was differentially distributed between the two treatment arms. Correcting for false discovery rate using the Benjamini Hochberg method [38], the data indicate that the number of 3D7-type epitopes was significantly higher in the control vaccine group compared to the FMP2.1/AS02_A vaccinee group (11% vs 3%, p=0.02, Figure 1). Overall, there were more 3D7-like T-cell epitopes with strong binding affinity in sequences from the control group compared to the FMP2.1/AS02_A group.

Table 2.

Potential high binding (IC50<50nM) CD8+ T cell, CD4+ T cell, and B cell epitopes in AMA1. Polymorphic codons in c1L are highlighted in bold.

Allele	Predicted epitope	IC50 (nM)	MHC restriction	Binding affinity
T-cell epitopes
			Class I
HLA-A*23:01	LYCVLLLSAFEF	54.2		Strong
HLA-A*23:01	SYIATTALSHPI	118		Strong
HLA-A*23:01	MFCFRPAKDISF	852,7		Weak
HLA-A*23:01	YMINFGRGQNYW	1035		Weak
HLA-A*30:01	KRNSMFCFRPAK	80.2		Strong
HLA-A*30:01	KVCPRKNLQNAK	336.9		Weak
HLA-A*02:01	FLTPVATGNQYL	26.5		Strong
HLA-A*02:01	LLLSAFEFTYMI	43.5		Strong
HLA-A*35:01	YWEHPYQNSDVY	12.9		Strong
HLA-A*35:01	SPMTLDEMRHFY	331.5		Weak
HLA-A*01:01	SPMTLDEMRHFY	88.5		Strong
HLA-A*01:01	MTLDEMRHFYKD	472.3		Strong
			Class II
DRB1_1304	IIIENSNTTFLT	19.01		Strong
	IIIASSAAVAVL	26.07		Strong
	EMVSNSTCRFFV	32.67		Strong
	CLINNSSYIATT	42		Strong
	TILMVYLYKRKG	71.3		Weak
	IASSAAVAVLAT	82.83		Weak
	GKGIIIENSNTT	85.53		Weak
DRB1_0701	MFCFRPAKDISF	57.07		weak
	SYIATTALSHPI	63.83		weak
DRB1_1101	ILMVYLYKRKGN	17.86		Strong
	MVYLYKRKGNAE	42.65		Weak
	FTYMINFGRGQN	62.66		Weak
DRB1_0120	FLPTGAFKADRY	11.43		Strong
	IFNVKPTCLINN	14.31		Strong
	NKKIIAPRIFIS	16.34		Strong
	YIATTALSHPIE	17.20		Strong
	IIIENSNTTFLT	21.37		Strong
	ASMIKSAFLPTG	23.01		Strong
	FKNKNASMIKSA	29.83		Strong
	MFCFRPAKDISF	31.19		Strong
	CLINNSSYIATT	38.15		Strong
	PTCLINNSSYIA	38.76		Strong
	CHILYIAAQENN	38.95		Strong
	YTYLSKNVVDNW	44.09		Strong
	LVFELSASDQPK	48.13		Strong
c1L T-cell epitopes				Score
	SPMTLDDMRLLY	20		Strong
	SPMTLDDMRRFY	658		Weak
	SPMTLDDMRVLY	64		Weak
	SPMTLDHMRDFY	122		Weak
	SPMTLDHMRDSY	479		Weak
	SPMTLDQMRHFY	142		Weak
	SPMTLNGMKDFY	2995		Weak
	SPMTLNGMRDLY	2182		Weak
B-cell epitopes				Score
	SYIATTALSHPI			0.88
	RNSMFCFRPAKD			0.86
	GEDAEVAGTQYR			0.85

Figure 2.

Location of predicted peptides on Plasmodium falciparum Apical Membrane Antigen-1 (PfAMA1). B-cell epitopes are in black pattern rectangle. HLA Class II alleles DRB1*01:20 epitopes are in solid cyan, DRB1*07:01 epitopes in solid black, DRB1*11:01 epitopes in solid chocolate and DRB1*13:04 epitopes in solid orange small rectangles. MHC class I alleles HLA-A*23:01 epitopes are emphasized in solid red, HLA-A*30:01 epitopes in solid blue, HLA-B*35:01 epitopes in solid light green, and HLA-A*02:01 epitopes in solid purple rectangles. Numbers are amino acids positions in PfAMA1 of the 3D7 strain. The blue line inside the long rectangle is the most polymorphic amino acid of the protein (position 197)

Table 1.

Genetic distance^a between the AMA1 sequence used in the FMP2.1/AS02_A malaria vaccine (3D7 variant) and the AMA1 sequences isolated from Plasmodium falciparum infections from children immunized with the FMP2.1/AS02_A vaccine or with the control rabies vaccine.

	FMP2.1/AS02A Vaccine	Control Vaccine	p-value
Global sieve effect
Whole protein mean	0.02	0.02	0.31
Local sieve effect
C1L mean	0.10	0.10	0.6
Epitope-level sieve effect
Epitope mean	0.02	0.02	0.30

^aGenetic distance was estimated according to amino acid p-distance.

Figure 1. Prevalence of strong, medium, and low binding affinity epitopes in FMP2.1/AS02_A and control groups.

The incidence of vaccine-type high binding epitope was significantly lower in the AMA1 vaccine group compared to the control group. No difference was observed in the low and medium binding epitopes.

3.2.3.1. Sieve effect on epitope binding.

A list of strong-binding polymorphic epitopes identified by the most frequent MHC super family/allele in Mali was generated (Table 2 and Figure 2). For each MHC allele, the binding affinity of each AMA variant was defined as the sum of binding affinities across all epitope predicted to be identified by that MHC allele [3]. In addition, strong binding polymorphic B-cell epitopes (SYIATTALSHPI, RNSMFCFRPAKD, and GEDAEVAGTQYR) were predicted using the ABCPred server. A comparison through a Mann-Whitney test of the average binding affinity scores of variants in the FMP2.1/AS02_A and rabies groups was not significant (p>0.5) for any MHC allele.

To assess whether the FMP2.1/AS02_A vaccine may induce T-cell escape, with few T-cell epitopes found in the vaccine group, we compared the proportion of strong binder epitopes generated using the 50 nM threshold in the two treatment arms. We did not find evidence of a significant difference in the proportion of SPMTLDEMRHFY, SYIATTALSHPI, KRNSMFCFRPAK, YWEHPYQNSDVY, FLTPVATGNQYL, FLTPVATVNQDL, YIATTALSHPIE, and IIIENSNTTFLT in the two treatment arms (p>0.56 in all comparisons).

Finally, to determine the effect of epitope binding affinity on the occurrence of clinical malaria in the two treatment arms, the average binding affinity of epitopes generated from sequences in malaria clinical episodes was compared between the FMP2.1/AS02_A and control vaccinees using a Mann-Whitney test. Among all the eight ‘strong’ binding epitopes identified, and after correcting for multiple comparisons, the data show that the mean IC50, across HLA genotypes, of the strong binding epitope (SPMTLDEMRHFY) and its variants generated from the first clinical malaria episode of the FMP2.1/AS02_A vaccine group was significantly higher than those of the same epitope and its variants generated from the control vaccinees (817 nM vs 89 nM, p=0.01; Table 3 and Figure 3). Therefore, immunization with FMP2.1/AS02_A imposed differences between the vaccinee and the control arms of the study in terms of the immunological properties of the AMA1 protein encoded by the parasites that infected each group.

Table 3.

Mean binding affinity (nM) of predicted CTL epitopes from Plasmodium falciparum infections occurring in children immunized with the FMP2.1/AS02_A malaria or control vaccine.

	FMP2.1/AS02A Vaccine	control	p-value
All clinical episodes
Mean (nM)	1058	937	0.23
N	66	85
Clinical episodes after day 75
Mean (nM)	1096	1015	0.36
N	52	64
First clinical episode
Mean (nM)	817	89	0.01
N	6	16

Figure 3. Mean binding affinity of strong binding predicted CTL epitopes from Plasmodium falciparum infections in FMP2.1/AS02A and control groups.

The mean binding affinity of the strong binder epitope (SPMTLDEMRHFY) generated from the FMP2.1/AS02_A vaccine group was significantly higher than the mean binding affinity of control group epitopes in first clinical malaria episode. No mean difference was observed during all clinical episodes and clinical episodes occurring after day 75.

4. Discussion

This study illustrates a proof of concept to evaluate the ability of sieve analysis based on in silico-predicted epitopes to detect vaccine-induced selection that we know exists, and hence to identify regions of the protein-coding gene that are important in parasite immune escape when no a priori knowledge exists of epitopes associated with vaccine efficacy. While previous studies on the allele-specificity of malaria vaccines have solely focused on the impact of individual amino acids on disease risk relative to the vaccine strain [10, 12], here we apply the concept of epitope-based sieve analysis, using sequences from vaccine trials, and its analytical methods, developed by Gilbert et al. in 1998 [1], to assess the efficacy of a candidate malaria vaccine against a polymorphic pathogen and detect sieve effect in a novel type of data partition. This concept has been used extensively in the field of HIV during the past 15 years [2, 4, 5, 22]. In a phase 3 trial of the RTS,S vaccine which is based on the circumsporozoite protein (CSP), a sieve analysis was conducted to assess the differential vaccine efficacy against 3D7-matched variants relative to specific CSP regions [9]. Although the analysis was limited to whether or not the variant present at the time of infection was identical to (i.e., matched) the vaccine strain, the percent amplicon match, which compared the number of vaccine-type amplicons, was significantly different in the two treatment arms [9] indicating a sieve effect.

Here, we take advantage of the availability of whole ama1 sequence data generated during the FMP2.1/AS02_A vaccine trial [10] to conduct a sieve analysis of vaccine-induced selection. Sieve analyses focused on amino acids frequencies have shown that the FMP2.1/AS02_A vaccine was efficacious mainly against homologous strains relative to amino acids in the cluster one loop (c1L) [10, 12]. Identifying which peptides/epitopes play an important role in vaccine escape may help in the design of a cross-protective and/or multivalent malaria vaccine by focusing development on particular region of the protein.

Comprehensive sieve analyses which take into account differences in polymorphism, binding affinity and physicochemical properties in epitopes across treatment arms of the vaccine trial allow us to evaluate the effect of vaccination on breakthrough sequences. Analyses at the whole protein level demonstrated no differences in the mean genetic distance when the two treatment arms were compared [12]. Here, analyses were conducted on twelve-residue epitopes, predicted using available algorithms and information on major MHC classes found in Mali. This epitope length was chosen because there are only nine residues in the epitope-binding core and anti-CSP monoclonal antibodies targeting twelve-residue epitopes were effective in protecting mice against malaria in a passive immunization [39].

We compared the selection induced by the FMP2.1/AS02_A vaccine on AMA1 epitopes to the selection type occurring in the natural parasite population (control vaccine group) in order to assess the distribution of ‘strong’ binder epitopes in the two treatment arms. Although all B-cell epitopes and most T-cell epitopes from the FMP2.1/AS02_A and rabies treatment were randomly distributed in the treatment arms, the data indicate that in malaria clinical episodes, the SPMTLDEMRHFY epitope covering ama1 c1L region was differentially allocated, likely because of selection by the FMP2.1/AS02_A vaccine. The SPMTLDEMRHFY epitope which was experimentally validated in an ELISpot assay [40] was shown to be more represented in the rabies group compared to the FMP2.1/AS02A vaccine group. This region of the protein [19, 41] has been shown to play an important role in the binding of AMA1 antibodies to monoclonal antibodies 1F9 [20]. Residues located in this region are very polymorphic consistent with their role in immune-escape. These findings highlight the important role this epitope may play in immune evasion. The fact that, after correcting for multiple measurements, more 3D7-type SPMTLDEMRHFY epitope was observed in the rabies group highlights a sieve effect, as has been observed in HIV [2] and malaria [9].

Epitope binding affinity measures the strength of the interaction between a protein and its ligand (epitope). This parameter is used to identify epitopes that are selective and specific to a target. Epitopes with low IC50 have been used to predict and design epitope-based vaccines [33]. Our analyses indicate a low mean IC50 of specific AMA1 epitopes generated from the rabies group compared to the FMP2.1/AS02_A vaccine group. This finding is concordant with the high sequence similarity of these epitopes with epitopes generated from the 3D7 reference and suggest a T cell-mediated selection in post-vaccination infections. As AMA1 is expressed in both the pre-erythrocyte and blood stages, immune response is mainly through CD4+ and CD8+ T cells [42], this T-cell response is indicative of a cell-mediated response occurring before the erythrocytic stage. The pressure induced by FMP2.1/AS02_A could have led to T cell escape in immunologically relevant regions of the proteins. Alternatively, the outcome of the analyses may be the result of the physical overlap between B cell epitopes and predicted T cell epitopes detected during blood stage infection.

Overall, these data indicate that the design of a cross-protective AMA1 vaccine may be complex due to the role played by relevant polymorphic amino acids and epitopes distributed along the protein-encoding gene. This work demonstrates the need to integrate comprehensive sieve analyses in vaccine trials targeting polymorphic antigens to avoid discarding candidates that could be used in specific regions of the world where the potential vaccine variant exists in high frequency. The results of this study also highlight the potential of using pathogen protein sequence data collected during vaccine trials to identify epitopes that could be used to select antigen variants or parasite strains for a multivalent cross-protective vaccine.

Some limitations of this study include the restriction of our analyses to only the most common HLA allele families in Mali; a powerful validation study could be conducted by contrasting the epitope sequence of each breakthrough infection with the HLA genotype of the respective subject in which it occurred. Moreover, functional assays were not conducted to confirm the role of HLA/epitope pairs in malaria pathogenesis. Even though the number of HLA allele families used was limited, we were able to find agnostically epitopes that cover c1L, a region that plays a critical role in AMA1 immunogenicity.

In conclusion, in silico epitope detection resulted in the identification of putative T cell epitopes overlapping the same region detected in earlier studies. Furthermore, these analyses based on predicted epitopes show a differential distribution of 3D7 epitopes between control and vaccine arms. The findings suggest that similar analyses can be used to identify protein regions associated with vaccine induced selection. Finally, the detection of the presence of allele-specific efficacy can be used to confirm the protective nature of an antigen, hence preventing it from being summarily discarded from the list of potentially useful vaccine antigens;