Research Article

Association of Malaria Parasite Population Structure, HLA, and Immunological Antagonism

See allHide authors and affiliations

Science  20 Feb 1998:
Vol. 279, Issue 5354, pp. 1173-1177
DOI: 10.1126/science.279.5354.1173


Host-parasite coevolution has been likened to a molecular arms race, with particular parasite genes evolving to evade specific host defenses. Study of the variants of an antigenic epitope ofPlasmodium falciparum that induces a cytotoxic T cell response supports this view. In African children with malaria, the variants present are influenced by the presence of a human leukocyte antigen (HLA) type that restricts the immune response to this epitope. The distribution of parasite variants may be further influenced by the ability of cohabiting parasite strains to facilitate each other's survival by down-regulating cellular immune responses, using altered peptide ligand antagonism.

There is increasing evidence that HLAs are subject to ongoing selection pressures by infectious pathogens, supporting the theory that natural selection by such parasites plays the central role in maintaining major histocompatibility complex (MHC) polymorphism (1, 2). However, an understanding of the precise nature of this process has been impeded by a lack of field data from natural host-parasite populations on interactions between polymorphic antigens of the host and variable components of the parasite. The life cycle of the malaria parasite Plasmodium falciparum provides an opportunity to measure the outcome of a specific immunological interaction between genetic variants of a parasite and the MHC type of its host. This possibility arises from the occurrence of an HLA class I–restricted cytotoxic T lymphocyte (CTL) response against the parasite at the early liver-stage of infection (3, 4). The response occurs directly after inoculation of malaria sporozoites by an infectious mosquito; analysis of parasites in the blood of infected individuals reveals which parasite variants (here termed “strains”) have survived this potential in vivo selection step and have reached the subsequent blood-stage of the life cycle.

Studies of CTL responses to variant epitopes in human immunodeficiency virus (HIV) and hepatitis B virus have shown that some naturally occurring variants of these epitopes may specifically down-regulate the CTL response through altered peptide ligand (APL) antagonism, at least in in vitro assays (5, 6). Thus, the simultaneous presence of the variant antagonist epitope appears to deliver an altered signal to the responding T lymphocyte that induces nonresponsiveness, or limited responsiveness, to its target agonist epitope.

Antagonism of malaria CTL epitopes. CTLs from individuals exposed to endemic malaria recognize epitopes in a variety of pre-erythrocytic antigens ofP. falciparum, and indirect evidence indicates that these CTLs may play a role in protective immunity (7). CTLs restricted by the commonest HLA class I molecule in The Gambia, HLA-B35, often recognize a polymorphic epitope in a variable region of the circumsporozoite (CS) protein. Of the four allelic variants that are prevalent among strains of parasite in The Gambia, two (cp26 and cp29) are CTL epitopes that bind HLA-B35 (4); the other two (cp27 and cp28) are not epitopes and failed to bind to HLA-B35 in in vitro assembly (binding) assays (8). The cp26 and cp29 peptides are octamers that differ only at the second amino acid position, yet each interferes with activation of memory CTLs by the other (Fig.1). In CTL responses to these malaria epitopes, cp29 was able to antagonize cp26-specific CTLs, and cp26 was able to antagonize cp29-specific CTLs, even when present on different antigen-presenting cells and at low antagonist:agonist ratios. A panel of cp26- and cp29-specific CTL lines generated from both malaria-naı̈ve (Fig. 1A) and malaria-immune (Fig. 1, B through D) individuals were antagonized at the effector level. We also found that cp26 and cp29 could mutually interfere with the induction in vitro of primary CTL responses by the other peptide epitope in individuals unexposed to malaria. Hence, T cell responses toP. falciparum epitopes may be added to the list of immune responses shown to be subject to APL antagonism in vitro. However, the importance of the phenomenon in vivo and its possible impact on host-parasite evolution have been uncertain. Thus, we proceeded to examine the distribution of the allelic epitopes cp26 and cp29 among Gambian children with P. falciparum malaria.

Figure 1

Reciprocal CTL antagonism in PBMC cultures by two naturally occurring P. falciparumCS protein variants. See (8, 21). (A) Inhibition of lysis of cp26- or cp29-pulsed cells by cp29 or cp26 at an antagonist:index peptide ratio of 1:1. Effector CTL lines were generated from malaria-naı̈ve donors BR and PTE as described (26) and used at an E:T ratio of 20:1. Specific lysis is shown at 4 hours for BRL1 and BRL2, and at 18 hours for BRL3, BRL4, and PTEL1. (B) Inhibition of lysis of cp26 pulsed cells by cp29 at antagonist:index peptide ratio of 0.3:1-0.03:1 (27). (C) Inhibition of lysis of labeled (hot) cp26-pulsed cells by cp29 variant pulsed onto unlabeled (cold) targets. As a control, FluB35 peptide pulsed cold targets were used. The assay shown was harvested at 4 hours. Effectors were PBMC from donor Z22 or Z25 stimulated with cp26 14 days before and used at an E:T ratio of 50:1. Similar levels of antagonism were again observed using cells from this donor in a subsequent study 12 months after this assay. (D) Inhibition of lysis of hot cp29-pulsed cells by cp26 pulsed onto cold targets. Effectors were peripheral blood mononuclear cells (PBMC) from donor Z87 stimulated with cp29 14 days before and used at an E:T ratio of 100:1. The assay shown was harvested at 8 hours. In this and previous assays (4), PBMC from donor Z87 restimulated with cp29 were noncrossreactive with cp26.

P. falciparum population structure. Table1 shows the frequencies of the allelic variants of the CS gene in parasite DNA from Gambian children with clinical malaria. Parasite DNA was amplified from peripheral blood, so these parasites have survived clearance at the earlier liver-stage of infection. Infection by more than one variant was found in 41% of the children, similar to previous studies (9). The distribution of parasite strains differed markedly from that expected under random mating without selection. In particular, cp26 and cp29 were found together more than twice as frequently, as expected from their prevalence in the host population (P < 1 × 10–8). Pairing of cp26 with cp29 (with or without cp27 or cp28 as well) was present in children with HLA-B35 [observed, 55; expected, 12; at 50% parasite rate (PR) in the population] and without HLA-B35 (observed, 144; expected, 26; at 50% PR). There was no difference in the distribution of strains between severe (n = 482) and mild (n = 313) malaria cases.

Table 1

Frequency of infection with cp26 to cp29 strains in 795 Gambians with malaria. The expected frequency of single and mixed infections with each of the strains, assuming random mixing, are shown for different values of the parasite rate (PR) in the population. The actual PR in the locality from which the malaria cases were sampled (from August to November) is in the range of 10 to 50%. Analysis of the distribution showed cp26 and cp29 present together significantly more often than expected, whatever the PR (*, χ2 = 10–50; **, χ2 = 50–100; ***, χ2 > 100,P < 10–20) (23). Methods are described in (24).

View this table:

We then assessed whether the presence of HLA-B35 affected the distribution of the four allelic types in the host population (Table2). We found cp26 and cp29 more frequently in children with HLA-B35 (P = 0.012). This association was independent of numerous potentially confounding factors, and similar analysis of the possible influence of 12 other class I and class II alleles on the distribution of these parasite variants showed no significant association with these host alleles.

Table 2

Frequency of cp26 and cp29 in individuals with HLA-B35 compared to the rest of the population. There is an increased occurrence of cp26 and cp29, both together and separately, in individuals with HLA-B35. This association was found in both individuals with mixed [P = 0.02, OR = 1.54 (1.05–2.27)] or single infections [P = 0.006, OR = 2.3 (1.20–4.36)]. HLA type was determined by PCR as described (4). Analysis is described in (25).

View this table:

Several lines of evidence suggest that the observed cohabitation results from a functional effect of the cp26, cp27, cp28, and cp29 variation itself, rather than of other parasite genes. Genetic recombination occurs during the sexual stage of the P. falciparum life cycle, and field studies (10) do not show evidence of clonality. Furthermore, analysis of a polymorphic region of the CS gene only 180 base pairs 5′ to the epitopes studied (11) revealed in this parasite population only very weak linkage disequilibrium between these alleles. The strongest allelic association (Table 1) is between cp26 and cp29, suggesting that this is the primary interaction. Finally, the differing frequencies of the variants that are epitopes for HLA-B35 in hosts with and without this HLA type support a mechanism involving the antigen-presenting function of this molecule.

Our immunological findings on APL antagonism suggest that the cohabitation of cp26 and cp29 could result from these strains facilitating each other's survival through mutual antagonism of CTL to the other variant at the liver stage of infection. But can this mechanism account for the cohabitation observed in individuals both with and without HLA-B35?

A model of antagonism. To address this question quantitatively, we developed a mathematical model (12) to explore the dynamics of two parasite strains within two different host classes. In individuals with HLA-B35, each strain (that is, cp26 or cp29) elicits short-lived pre-erythrocytic strain-specific infection-blocking immunity. Individuals without HLA-B35 develop short-lived pre-erythrocytic non–strain specific immunity (acting equally against cp26 and cp29) upon infection with either strain. We assume that there is very little cross-reactivity between immune responses to cp26 and cp29, as indicated by the data on secondary CTL responses (Fig. 1) (4).

We incorporated APL antagonism into this system at both the effector and induction level. HLA-B35 individuals who are already immune to a given strain (for example, cp26) would, in the absence of effector-level antagonism, become infected with the other strain (cp29) when subjected to simultaneous inoculation. The degree of effector-level antagonism may thus be represented by a parameter, α, which determines the probability that an individual immune to a given strain will become infected with both strains upon simultaneous inoculation, as a result of the existing CTL being antagonized. Induction level antagonism is incorporated by assuming that no immunity develops after simultaneous inoculation with cp26 and cp29.

Figure 2 shows how the distribution of parasites within the two host classes changes in the model with increasing effector-level antagonism, when there is total antagonism at the induction level. We assume in this example that cp29 is more likely to induce protective immunity than cp26, consistent with the observations that cp29 binds better than cp26 to HLA-B35 and induces primary CTL more readily (8, 13). Two nonintuitive results are apparent. Firstly, a small difference in duration of immunity to cp26 and cp29 in a small segment of the host population (the 30% with HLA-B35) can translate into a large difference in the relative population prevalence of the two strains. This is because the two strains are in direct competition with each other within the non–HLA-B35 host population as a result of strain-transcending immunity (acting equally against cp26 and cp29) (14). Second, APL antagonism can act to structure the parasite population such that there is a preponderance of mixed infections. A high degree of antagonism is required at the effector level as well as at the level of induction of CTL responses; the latter alone is insufficient to generate the observed levels of mixed infections (15). Importantly, the altered parasite population structure is reflected in both the HLA-B35 and the non–HLA-B35 populations in proportions commensurate with the levels of immunity induced by the relevant epitopes. Levels of infection with cp26 and cp29 in non–HLA-B35 hosts, for whom these are not the immunodominant epitopes, can either be higher or lower than in HLA-B35 hosts who do respond in an epitope-specific manner (Fig. 2).

Figure 2

Output of the model. The proportions of hosts infectious for cp26, cp29, and both, in hosts with HLA-B35 and without HLA-B35 is dependent on the degree of effector-level antagonism. As this increases, there is a nonlinear increase in cohabitation. There is also a reversal from lower to higher levels of cp26 and cp29 variants in HLA-B35 compared to non-HLA-B35 hosts. In the absence of antagonism, there is more infection in the non-HLA-B35 hosts because their average duration of immunity to any strain is effectively only 0.5 years (γ = 0.01) compared to 1.25 years for cp26 (γcp26 = 0.025) and 5 years for cp29 (γcp29= 0.1) in the HLA-B35 hosts. However, as antagonism increases, the advantage this provides to parasites within the HLA-B35 population outweighs the effects of higher levels of immunity, leading to levels of infection that are higher than in the non-HLA-B35 population, which responds more weakly to other epitopes in parasites bearing these variants. Other parameter values used are as follows: β = 50, σ = 10, δ = 0.975, μ = 0.02 (12).

Evolutionary, immunological, and parasitological implications. We have shown previously that HLA type affects a child's risk of developing severe malaria in this West African population (1). Here, we report that HLA class I antigens may affect the strain of parasite found in children with malaria. Both observations are compatible with an effect of HLA class I–restricted CTLs acting at the liver stage of P. falciparum infections. However, the present observations have particular implications for the coevolution of host and parasite diversity. The different distribution of allelic types of the CS gene in children with and without HLA-B35 provides evidence of an ongoing selective effect of HLA antigens on the strain of parasite causing blood-stage malaria infections. This, in turn, will influence the strain of parasite that is transmitted to further individuals and, ultimately, the population prevalence of the strains. Over time, such small effects are sufficient to markedly affect parasite frequencies. The cohabitation of the alleles cp26 and cp29 can be explained by mutual down-regulation of each other's CTL responses at the liver stage of infection. Although this will occur only in HLA-B35–positive individuals, the mathematical model indicates that, over time, the cohabitation should spread to individuals of all HLA genotypes, consistent with the observed lack of a detectable effect of HLA genotype on the extent of cohabitation. Thus, host HLA type may affect the parasite allele distribution, but the converse also applies. The observed antagonistic interactions of parasite alleles implies that geographical variation in parasite allele frequencies may affect the local magnitude of HLA associations with infectious diseases and thus the selection pressure exerted by that parasite on HLA frequencies.

Plasmodium falciparum strains may be structured at a primary level by immune responses against a polymorphic and antigenically varying blood-stage antigen, PfEMP1 (16). The observations on HLA-B35 and CS variants we present may thus reflect a particular substructure of the parasite population that results from interaction between a less effective liver stage–specific immune response and its target epitope. This raises the possibility that several such substructures of association between particular host and parasite variants might be observed as the result of each parasite locus evolving away from the specific host molecule regulating immune responses to it.

The discovery of APL antagonism has been of considerable immunological interest, in part because it might allow interventions to specifically down-regulate pathological immune responses. Although it has been observed for both HLA class I– and class II–restricted T cell responses, and in CTL responses to hepatitis B and HIV-1 infections (5, 6), the in vivo significance of antagonism and the potential evolutionary advantage of this type of putative immune escape mechanism have been unclear. As discussed elsewhere (17), it might appear more efficacious for pathogens to simply evade an immune response by mutations that prevent binding of epitopes to MHC molecules rather than by those that produce APL antagonism. A novel feature of the antagonist peptide epitopes displayed here is that the antagonism is mutual. The antagonist is effective at very low antagonist:agonist ratios and, unusually, inhibits the polyclonal CTL response (18) to the agonist. These unusual features suggest that this pair of antagonists might reflect natural selection in the malaria parasite of variants that are mutually inhibitory. The data on the CS allele distributions argue that antagonism is employed as an escape mechanism in natural pathogen populations. They also suggest that malaria parasites may manifest strains with mutually antagonistic epitopes because the use of an antagonist is a more efficient method of escaping an immune response than avoidance of MHC binding: the mutually antagonistic variants cp26 and cp29 were found more frequently in HLA-B35 individuals than were the nonbinders cp27 and cp28. We speculate that, by maintaining binding, cp26 and cp29 may prevent the emergence of an immunodominant response to another HLA-B35–restricted epitope in P. falciparum such as the ls8 epitope in liver-stage antigen-1 (4).

Some malaria vaccine programs are aimed at inducing CTL responses against epitopes expressed by the liver-stage parasite (19). Our results imply that inclusion of all allelic peptide variants in such vaccines may not offer a solution to the well-recognized problem of antigenic polymorphism in malaria: inclusion of a variant epitope might even lead to antagonism of naturally acquired immunity to other variants, resulting in increased susceptibility.

This study has identified a specific association between variable components of a parasite and its mammalian host. Combined genetic, immunological, and mathematical analyses of further such examples from natural populations should provide a molecular understanding of the mechanisms driving host-parasite coevolution.

  • * These authors contributed equally to this research article.

  • To whom correspondence should be addressed. E-mail: adrian.hill{at}


View Abstract

Navigate This Article