Background
Plasmodium vivax infected an estimated 8.55 million people in 2016 and is a significant contributor to global malaria morbidity, with the majority of
P. vivax cases occurring within South-East Asia [
1]. There remains a significant need for a vaccine against
P. vivax, and an understanding of the targets of natural immune responses following
P. vivax infection is likely to aid such an effort. A key challenge in vaccine development is the identification of specific antigens and epitopes that are targets of protective antibody responses. It is possible to use population genetic data to identify regions of proteins that are under immune-mediated selection pressure, which gives rise to balancing selection within that protein region. Tajima’s D is one test statistic that is often used to identify departures from a neutral model of selection, and has been applied to malaria genes both on a per-gene basis [
2,
3], or as a sliding window analysis along a gene [
4‐
7]. A number of studies have previously examined
P. vivax proteins such as apical membrane antigen 1 (
PvAMA1) and Duffy-binding protein (
PvDBP) for evidence of immune selection pressure using these approaches [
4,
8‐
11]. However, because a sliding window analysis is typically performed over the linear gene sequence, it does not take into account the impact of the three-dimensional (3D) structural constraints of the protein in the calculation of selection pressures. A new method that allows incorporation of protein structural information into tests for selection pressure has recently been described [
12], and has been applied here to two leading
P. vivax vaccine candidates:
PvAMA1 and
PvDBP.
AMA1 is a type I transmembrane protein present in all
Plasmodium species [
13]. It is localized to the parasite micronemes, and is released onto the surface of the merozoite prior to invasion of red blood cells [
14]. AMA1 binds to RON2 during the formation of the tight junction between parasite and host–cell membranes [
15,
16] and is a target of protective immune responses [
17‐
20]. The ectodomain of
Plasmodium AMA1 proteins is divided into three domains, termed Domains I (DI), II (DII) and III (DIII) (Additional file
1) [
21]. DI is considered to be the most polymorphic, and is also the site of RON2 binding [
22]. RON2 binds a conserved hydrophobic cleft that is surrounded by a number of highly polymorphic regions, the most notable being the C1L loop, a surface exposed loop with high variability that is suggested to define strain-specificity in anti-AMA1 responses in
Plasmodium falciparum infection [
23]. While DI is generally considered to be the most important for functional antibody responses, there is evidence that DII and DIII may also be targets of functional antibody responses in
P. falciparum [
8,
24]. A number of studies have investigated selection pressures on
PvAMA1. Evidence for balancing selection within Domain I has been observed in a Venezuelan population [
9], in a number of Papua New Guinean populations (Madang and Wosera, Madang and East Sepik) [
4,
8], an Iranian population [
10] and in a Peruvian population [
25]. In contrast, two other studies examining
PvAMA1 sequences in isolates from Korea [
26] and Myanmar [
27] did not find any evidence of balancing selection, but instead observed evidence of recent population bottleneck and expansion in those populations.
PvDBP is an important micronemal protein that binds to the Duffy antigen/receptor for chemokines (DARC) on human reticulocytes during invasion [
28,
29]. Whilst there is evidence that
PvDBP is not absolutely essential for invasion of reticulocytes [
30‐
32], Duffy-negative individuals are largely resistant to
P. vivax infection, and hence
PvDBP makes an attractive vaccine target [
33].
PvDBP is part of the erythrocyte-binding like (EBL) family of proteins, which include EBA175, EBA181 and EBA140 in
P. falciparum [
34].
PvDBP is the sole EBL family protein in
P. vivax [
35]. EBL family proteins are composed of a number of distinct domains, with Region II (RII) being a cysteine-rich Duffy-binding like (DBL) domain that is involved in binding to erythrocytes. EBL family proteins each recognize a different receptor via their respective DBL domains [
29,
36‐
38];
PvDBP binds to DARC via its DBL domain (RII) [
39]. During this process two
PvDBP molecules form a dimer around two DARC molecules [
39,
40]. RII of
PvDBP has been divided into a number of subdomains (subdomains 1–3) [
41] (Additional file
2), and it is subdomain 2 that contains both the dimer interface and DARC binding residues [
39]. Immune responses against
PvDBP have been associated with protection from clinical malaria in naturally exposed cohorts [
42,
43] whilst antibodies against
PvDBP RII epitopes have been found to inhibit both attachment of
PvDBP RII to erythrocytes [
44] and in vitro invasion of erythrocytes [
45]. With regards to immune selection pressure on
PvDBP, a study of 100 Sri Lankan isolates found no evidence of significant selection pressure on this region using Tajima’s D, dN/dS or Fu and Li’s D and F statistics [
11]. Another study examining genetic diversity of
PvDBP RII across multiple populations showed a significantly positive value of dN/dS in this region, suggesting that this region may be under immune selection pressure [
25].
In this study, selection pressures on
PvAMA1 and
PvDBP Region II were examined in the context of protein structure, using a newly developed tool called BioStructMap [
12]. BioStructMap enables the application of a 3D sliding window over a protein structure. This allows incorporation of protein structural information into tests such as Tajima’s D or nucleotide diversity that are traditionally performed as a linear 2D sliding window over a protein or nucleotide sequence. A previous study identified a discontinuous region of
PfAMA1 bordering DII and DIII that had a strong signature of balancing selection when considering spatially derived Tajima’s D [
12]. Given that other studies have identified DI of
PvAMA1, rather than DII or DIII, as being under balancing selection, it was considered that incorporation of protein structural information might yield additional insights into other regions under immune selection pressure. Genomic sequences from a number of populations were analysed, and spatially-derived nucleotide diversity and Tajima’s D were examined using protein structural information for
PvAMA1 and
PvDBP Region II. Structural patterns of nucleotide diversity were similar across all populations examined, with Domain I of
PvAMA1 having the highest nucleotide diversity and displaying significant signatures of balancing selection (Tajima’s D > 0). Nucleotide diversity for
PvDBP was highest bordering the dimerization and DARC-binding interface, although there was less evidence of immune selection pressure on this antigen.
Discussion
In this study, patterns of nucleotide diversity and selection were examined over the protein structures for the
P. vivax antigens
PvAMA1 and
PvDBP. A number of major observations stand out from this work. Firstly, patterns of diversity on both
PvAMA1 and
PvDBP were remarkably similar across multiple geographic populations, despite phylogenetic trees for both
PvAMA1 and
PvDBP sequences suggesting a level of clustering according to geographic location. The only exception to this for
PvAMA1 was the South Korean population which displayed evidence of a recent bottleneck and expansion. This similarity in patterns of diversity is important when trying to extend conclusions made from studies from single geographic locations to a worldwide population, and these observations suggest a universality with regards to major epitopes on these antigens. It is also interesting to note that highly polymorphic residues for both
PvAMA1 and
PvDBP tended to fall within regions surrounding, but not a part of, ligand binding interfaces. For AMA1, RON2 binds in a hydrophobic cleft in DI, and polymorphic residues fall on one side of this hydrophobic cleft, but generally not within residues that make contact with the RON2 peptide. Similarly with
PvDBP, contact with DARC occurs primarily via subdomain 2, and the most polymorphic regions were near the DARC binding and dimerization interface. In contrast, the residues directly involved in the
PvDBP binding interface were highly conserved. These results are suggestive of two things. Firstly, the residues that make up the key binding interfaces in these two antigens have limited capacity for polymorphic variation due to functional constraints, as has been previously suggested by other studies [
39,
63]. This makes them attractive vaccine targets, as potential epitopes within these binding sites would have very limited antigenic diversity, and are also less likely to undergo extensive mutations to evade immune responses. Secondly, the high degree of polymorphism around these interfaces suggests that antibody responses that target these polymorphic sites are capable of inhibiting parasite invasion. This inhibition is likely the result of steric hindrance preventing receptor binding and/or dimerization. Future efforts could involve epitope focusing techniques [
66,
67] to direct antibody responses to these key conserved interfaces.
For
PvAMA1, we observed balancing selection primarily on DI in all populations examined. Additionally, while the 3D sliding window approach highlighted additional residues under balancing selection as compared to a linear sliding window approach, nearly all of these regions fell within DI. This agrees with a number of other studies in which
PvAMA DI is the only domain found to be under significant balancing selection [
4,
8‐
10]. This is in contrast to selection pressures observed on
PfAMA1, in which both DI and DIII have been observed to be under balancing selection [
5‐
9,
68]. Previous work of ours has applied spatially-derived Tajima’s D calculations over a
PfAMA1 structure and identified strong balancing selection in a region bordering DII and DIII [
12], lending further evidence to DIII being under immune selection pressure in
P. falciparum but not
P. vivax AMA1. The biological reasons for such a difference are unclear, as AMA1 has a conserved role between
Plasmodium species, although it is possible that DIII of
PvAMA1 is less immunogenic than the corresponding
PfAMA1 domain due to structural or sequence differences between the two antigens.
Although individuals in malaria endemic areas develop antibodies to
PvAMA1 [
69‐
71], there are no comprehensive studies on how these antibodies interact with the different domains of
PvAMA1. However, AMA1 is functionally conserved across
Plasmodium species, and there is evidence that
PvAMA1 is functionally equivalent in a
P. falciparum transgenic line in which
PfAMA1 is replaced by
PvAMA1 [
72]. As such, comparisons can be drawn with antibody studies on
PfAMA1. Dutta et al. [
24] generated a panel of monoclonal antibodies (mAbs) to
PfAMA1 and observed that their strain specificity and functional activity was determined by the diversity of the epitope sequence. The limited diversity in
PvAMA1 DIII observed in this current study aligns with the observations that mAbs to
PfAMA1 DIII were the most strain transcending. Similarly, we observed that the polymorphic face of
PvAMA1 DI had the highest diversity, in line with the observation by Dutta et al. that mAbs that bound the polymorphic face of
PfAMA1 DI were strain specific, compared to the others that bound the conserved face [
24]. Importantly, mAbs that bound to the conserved face of
PfAMA1 still showed strong growth inhibitory activity suggesting that epitopes on the conserved face can be targets of neutralizing antibodies, despite being under less immune pressure than immunodominant polymorphic regions.
For
PvDBP, one of the regions of high diversity across all populations was a previously identified epitope within subdomain 2 termed the DEK epitope [
33]. Others have observed that the DEK epitope is highly polymorphic and immunodominant [
44]. However, due to the polymorphic nature of this epitope, cross-strain specificity is a concern when creating a
PvDBP RII-based vaccine. Recent work has characterized the location of several conserved epitopes within
PvDBP RII that are the target of inhibitory mAbs 2D10, 2H2 and 2C6 [
73,
74], and all of these conserved epitopes fall within subdomain III, which had the lowest overall nucleotide diversity in this current study. This highlights some of the limitations of using population-level genomic data for identification of functionally important targets of antibody responses—the possibility that conserved regions may contain potential epitopes that are the targets of inhibitory antibodies cannot be excluded.
Several attempts have been made to divert immune responses away from these highly polymorphic regions of
PvDBP RII and towards conserved epitopes. One attempt to focus away from polymorphic regions involved mutating residues in the DEK epitope to reduce its immunogenicity, and these DEKnull mutants induce antibodies that bind
PvDBP and can inhibit the interaction with Duffy Binding Ligand [
33]. More recently, further epitope focusing techniques have been employed with
PvDBP RII, with one strategy involving mutation of all polymorphic residues to alanine, threonine or serine residues [
75]. This ‘DEKnull-2’ recombinant
PvDBP RII construct was shown to elicit broadly neutralizing antibodies following mouse immunization, with some naturally exposed individuals also shown to recognize the conserved epitopes on this construct [
75]. Other recent efforts towards the development of a
PvDBP RII vaccine include a Phase 1a trial of a prime-boost viral-vectored vaccine that demonstrated both safety and immunogenicity, with cross-strain inhibition demonstrated for the single heterologous strain tested [
76].
Given the success using epitope focusing techniques for
PvDBP RII, we suggest that such an approach could be applied to AMA1 to focus antibody responses towards conserved epitopes on the silent face of DI or within DIII. This would involve mutating major polymorphic residues to reduce the immunodominance of epitopes within the polymorphic face of DI; the most polymorphic residues identified in this study (Additional file
11) could serve as starting point for this work. Alternatively, it has been shown for
PfAMA1 that immunization with multiple heterologous strains of
PfAMA1 is capable of inducing strain transcending antibody responses [
24], and this approach could also be applied to
PvAMA1. Other epitope-focusing approaches also exist, including the use of small protein scaffolds to mimic native epitopes [
66,
67], although these might be challenging given the discontinuous nature of many potential epitopes within AMA1. The approaches used in this work could also be applied to other antigens such as
PvRBP2b, which has recently been identified as a ligand for reticulocyte invasion via binding to transferrin receptor 1 (TfR1) [
77].