A panel of recombinant proteins from human-infective Plasmodium species for serological surveillance

Malaria is an infectious disease that remains a global health problem causing an estimated 219 million clinical cases resulting in 435,000 deaths in 2017 [1]. The disease is caused by parasites of the genus Plasmodium and several species are known to regularly infect humans. The vast majority of deaths occur in sub-Saharan Africa and are caused by Plasmodium falciparum, whereas outside of Africa, Plasmodium vivax is responsible for over half of all malaria infections leading to significant morbidity and mortality [2]. Much less is known about the other human-infective Plasmodium species, Plasmodium malariae, Plasmodium ovale and Plasmodium knowlesi both in terms of their global distribution and clinical impact. Plasmodium knowlesi, a parasite typically found in macaques, is a significant cause of human malaria in Southeast Asia, causing a spectrum of disease ranging from mild to fatal infections [3]. Malaysia has the highest incidence of P. knowlesi malaria with over 6700 cases reported in the last 2 years compared to only 85 cases of indigenous human malaria (unpublished data from the Ministry of Health, Malaysia).

Diagnosis of Plasmodium infections and epidemiological surveillance is important for guiding the distribution of resources into intervention measures and establishing their clinical impact over time [4]. Methods to measure the prevalence of Plasmodium infections include microscopy, rapid diagnostic tests (RDTs) and PCR-based approaches, each differing in their sensitivity, infrastructure requirements, and ability to diagnose the different species. Serological assays can provide a historical record of infection and because of the specificity of antibody-antigen binding, could also potentially discriminate between different Plasmodium spp. infections. Host antibodies appear rapidly after initial infection and can persist for months and even years after the parasites have been cleared [5, 6]. Serological screening has been applied in epidemiological settings to detect parasite exposure, evaluate transmission trends of malaria [7‐10], and identify antibody-based correlates of protection [11, 12]. It is also used in blood donation centres, where, due to the increase in international travel and migration, the need for serological diagnosis is becoming more important to reduce the risk of transfusion-transmitted infections. Currently, many centres assess these risks using patient questionnaires which is generally unsatisfactory; moreover, the limitations and costs of the currently available serological tests often make implementing these assays economically unattractive [13].

Many antibodies recognise epitopes that are only formed in the context of an antigen in its native conformation [14]. To detect these antibodies, it is vitally important that the proteins used are correctly folded so that they faithfully form these epitopes. Expressing Plasmodium proteins in a soluble recombinant form has proved challenging, perhaps because of the high A:T content of the genome and lack of recognisable protein domains in many Plasmodium proteins [15]. This problem is especially acute for parasite proteins that are secreted or embedded in membranes because these proteins additionally contain structurally critical post-translational modifications, such as disulfide bonds that are not typically added by many commonly used expression systems. Recently, a method of expressing large panels of recombinant Plasmodium proteins was developed that retained many of their biochemical functions [16]. Central to this approach was the use of a mammalian expression system which increases the chances that appropriate post-translational modifications are correctly added to ensure proteins adopt their native fold. For antigens expressed using this method a large fraction—and in some cases all—the immunoreactivity to antigens was heat-labile, demonstrating that antibodies that recognise conformational epitopes represent a major component of the humoral response [16]. Previously, this approach was used to create libraries of soluble recombinant merozoite cell surface and secreted proteins that encompass the entire ectodomain from both P. falciparum [16, 17] and P. vivax [18]. Using sera from patients living in endemic regions, several of these proteins were found to be highly immunoreactive and might therefore be useful target antigens in serological assays [8, 12]. Expanding the antigen panel to include the other parasite species infecting humans would be especially valuable if they could be used to determine exposure to the different species of Plasmodium. Here, the recent availability of high-quality genome sequences from the three other human-infective Plasmodium parasites: P. knowlesi, P. ovale and P. malariae were used to extend the available panel of proteins and were tested for reactivity to sera from individuals infected with different Plasmodium parasites.

With the aim of identifying antigens that could be used for serological markers of infection for Plasmodium parasites that infect humans, 12 proteins from P. falciparum and 10 proteins from P. vivax which were previously shown to be highly immunoreactive to sera from patients living in endemic regions were selected [12, 18] (Table 1). From these proteins, 8 that were produced at high levels in the expression system (CyRPA, GAMA, MSP10, MSP4, MSP5, P12, P38, P41) were selected to identify the orthologous proteins from the genome sequences of P. knowlesi [22], P. ovale and P. malariae [23, 24] (Table 1). The P92 orthologue from P. knowlesi and 2 paralogues from both the MSP3 and MSP7 multigene family from P. ovale and P. malariae were also selected since orthologues of these proteins are highly immunoreactive in P. falciparum (Table 1). Protein expression plasmids encoding the entire ectodomains for the selected genes were made by gene synthesis, and proteins were expressed as soluble recombinant proteins by transfecting HEK293 cells. As expected, protein expression levels varied considerably, and most were expressed at usable levels at the expected size (Fig. 1a, Table 1). Exceptions included the P38 orthologue from all 3 species, P12 from P. malariae and P. ovale, and PmCyRPA, which were all repeatedly expressed at low levels in independent transfections. PoMSP3.5, PoMSP7.8 and PkP41 showed evidence of some proteolytic processing (Fig. 1a). In summary, orthologues of proteins in the genomes of P. knowlesi, P. ovale and P. malariae that are highly immunoreactive to sera from patients with P. falciparum and P. vivax infections were identified and expressed as soluble recombinant proteins for serological screening.

A smaller subset within the expanded panel of recombinant proteins which were immunoreactive needed to be identified as suitable serological markers of infection. To determine which of the P. ovale and P. malariae proteins were immunoreactive, the responses to IgG, IgM and IgA immunoglobulins were tested using sera pooled from over 800 adults living in Malawi [19] where there is active transmission for both species [35]. For P. malariae, P41, MSP5, and MSP10 were selected as they were the most immunoreactive and expressed at acceptable levels (Fig. 1b); MSP4 was not selected because although immunoreactive, was repeatedly expressed poorly. Using the same criteria, the P. ovale proteins P41, MSP3.5 and MSP10 were selected (Fig. 1b). To identify the most suitable antigens for P. knowlesi, the immunoreactivity to the panel of proteins was tested using sera derived from 10 Malaysian patients with PCR-confirmed single P. knowlesi infections, and P12, P38 and MSP10 were chosen (Fig. 1c). From these experiments, a panel of 22 proteins were selected that contained 7 proteins from P. falciparum, 6 from P. vivax, and 3 from each of P. malariae, P. ovale and P. knowlesi (Table 1).

Using this panel of 22 immunoreactive proteins, the antibody responses to a specific antigen from a particular Plasmodium species were examined to determine whether there was any detectable cross-reactivity with antigens from other species. Ideally, this would make use of sera from human patients which have diagnosed mono-infections for each Plasmodium species. To increase the chances that patients had been exposed to a single species, serum samples were obtained from patients of European origin who had contracted malaria by visiting malaria endemic regions and in which the infecting Plasmodium species had been confirmed by PCR. For the more common parasites, a reasonable number of samples were obtained: P. falciparum (n = 26), P. vivax (n = 17), but access to serum samples of travellers’ malaria with diagnosed infections for the rarer parasites was limited: P. ovale (n = 7), P. malariae (n = 3). For P. knowlesi, responses in the 10 Malaysian patients with PCR-confirmed single P. knowlesi infections were used. These patient sera were screened against the panel of all 22 antigens from five Plasmodium species. Two patients exhibited a broad reactivity to antigens from two species: a Swedish traveller diagnosed with P. vivax but whose serum exhibited broad reactivity across antigens from three Plasmodium species (P. vivax, P. falciparum, P. knowlesi) and one of the Malaysian patients diagnosed with P. knowlesi but whose serum reacted strongly with four out of six P. vivax antigens; this patient was subsequently identified as a migrant worker from Indonesia where P. vivax is endemic. Because it is likely that these patients had prior exposure to another Plasmodium species other than their diagnosed infection, these patients were removed from this analysis. The pairwise correlation of the responses to each of the antigens in the panel was examined in the remaining patients (Fig. 2a). For P. falciparum and P. vivax, where more samples were available, the antibody responses to most antigens within that species were positively correlated, as expected (Fig. 2a). There were two antigens that showed significant cross-reactivity between their orthologues: the HPzz proteins from P. falciparum and P. vivax, and MSP5 from P. vivax and P. malariae. Furthermore, responses to P. malariae P41 correlated with P. ovale MSP3.5. This suggested that responses to these proteins may not be suitable to diagnose exposure for these species, and so were omitted from further analyses. There was no evidence that sequence identity between orthologous proteins correlated with cross reactivity. For example, the amino acid sequence identity between P. falciparum and P. vivax HPzz proteins was only 39%, and yet responses showed strong evidence of cross-reactivity (Fig. 2b). By contrast, the sequence identity between P. knowlesi and P. vivax P12 was much higher (72%), and yet patient responses showed little evidence of cross-reaction (Fig. 2c). While this analysis must come with the caveat that the number of serum samples from travellers with malaria is limited, especially for the rarer parasites, P. ovale and P. malariae, antigens that appeared potentially cross-reactive between species were excluded, and this apparent cross-reactivity did not correlate with sequence identity between the orthologous proteins.

To determine whether the panel of proteins could be used in serological assays to establish exposure to different human-infective Plasmodium species, a larger panel of serum samples from individuals with diagnosed infections that corresponded to both ongoing acute cases of imported malaria (n = 81), and endemic malaria from a cohort of Tanzanian adults (n = 21) were tested. For P. knowlesi, sera from the 10 patients diagnosed with P. knowlesi infections from Kapit Hospital in Malaysian Borneo were used. The antibody responses between the different sources were investigated by segregating the samples into three different categories: travellers of European origin (n = 53), travellers with an origin in a malaria-endemic area (n = 28), and individuals residing in an endemic area (n = 31). The responses to each antigen corresponding to the species diagnosis was plotted within each category, and while differences between the different patient groupings were observed on a per antigen basis (Additional file 2: Fig. S1a), when averaged across the panel, no systematic differences were observed (Additional file 2: Fig. S1b). The converse analysis was performed whereby the responses to the antigens corresponding to the species other than that in the diagnosis were plotted, and again segregated according to the different categories. As expected, there was little response across the antigen panel, although it was clear that the Tanzanian endemic samples showed evidence of exposure to other species other than the one for which they had a diagnosed ongoing infection, especially for P. falciparum, but also P. ovale (Additional file 2: Fig. S1c). Immunoreactivity to both HPzz orthologues from both P. falciparum and P. vivax and PfMSP1 were observed which suggested antigen cross-reactivity, as had been already established for HPzz (Fig. 2), or, in the case of PfMSP1, there was a higher background signal, including from the uninfected controls (Additional file 2: Fig. S1c).

A logistic-regression classifier was trained to combine the immunoreactivities to the antigens from each species into a prediction of prior exposure. Sera from both travellers and individuals living in endemic regions for each species were used together with unexposed control samples and iteratively and randomly split 50:50 into training and testing sets, with approximately equal representation of diagnoses for each species. For each Plasmodium species, a separate model was trained to assign a binary outcome indicating if a patient was infected by that species or not. A score near zero indicates that the model assigns a low probability of the patient being infected with the given Plasmodium species, while scores near one indicate likely infections. Because patients from endemic regions are very likely to have had prior exposure to species other than the one they were actively diagnosed with, the positive training set was defined as patients only with a confirmed species diagnosis by microscopy or PCR.

The performance of each diagnostic model was evaluated, and the random sampling procedure into training and testing sets was iteratively repeated. To quantify the models’ performance, the averaged receiver operating characteristic (ROC) curves for predicting exposure to each species (Fig. 3a) was first calculated. Moderate predictive power (area under the curve (AUC) ≥ 0.7) was achieved for all species, and especially for P. vivax, P. falciparum and P. knowlesi where exposure could be predicted with good accuracy (AUC ≥ 0.8). The relatively small number of available diagnosed samples for the rarer parasites P. ovale (n = 18) and P. malariae (n = 17), limited the performance of their classifiers, as well as in the case of P. malariae, the reliance on just a single antigen (PmMSP10) after removing those that showed evidence of cross-reactivity.

Diagnostic models were used to determine a likelihood of prior exposure of the patients to each of the different Plasmodium species (Fig. 3b). In general, the models returned a positive prediction for the parasite species that agreed with the diagnosis, and the sera from unexposed controls being negative. The classifiers performed less well on those patients with co-infections and those diagnosed by PCR rather than microscopy (Fig. 3b). As expected, several individuals showed strong evidence of prior exposure to a different Plasmodium parasite species other than the one for which they were positively diagnosed. For example, some individuals positively diagnosed with P. ovale infections, also showed evidence of prior exposure to P. falciparum and P. vivax.

Following the encouraging performance of the antigen panel in diagnosing infections by P. knowlesi, a larger number of serum samples comprising 50 P. knowlesi malaria patients and 66 uninfected controls from the same region in Malaysia were evaluated. Using the logistic regression model to combine the responses against all three antigens (PkMSP10, PkP12, PkP38), P. knowlesi infections could be identified with high confidence (AUC > 91%) (Fig. 4a). If a threshold is set to a model score of above 0.5, then 82% of P. knowlesi-infected patients are correctly diagnosed at a false-positive rate of 3% (Fig. 4b).

Malaria continues to be a significant global health problem although cases have dropped in recent years due to the deployment of effective intervention measures [1]. Spurred on by these successes, the malaria control policies of governments and international agencies are moving towards the goal of eradication, which will require more sensitive detection and tracking of the different species in endemic regions. The application of serological screening may have an increasingly important role as transmission declines because the longevity of antibody responses should provide a historical record of exposure rather than being limited to the detection of current infections [36]. In areas of declining transmission, where there are increasing incidences of sub-microscopic infection, models using data from serological surveillance have been shown to provide precise estimates of parasite transmission [10]. Because of the asexual amplification of malaria parasites in the blood, those antigens expressed during these blood stages are likely to be good choices for diagnostic antigens because they are known to be particularly immunodominant. Several serological assays have been developed to detect exposure to Plasmodium parasites although few have attempted to distinguish infections by the different species [37]. This has mainly been due to the lack of a genome sequence from the two parasites, P. ovale and P. malariae, making the cloning and expression of many candidate proteins from these species difficult. Recent sequencing of P. ovale and P. malariae [23, 24] has now permitted the expression of several candidate proteins from these parasites. Here, a panel of antigens was compiled that are likely to be expressed by the blood stages using the recent availability of the genome sequences of all five main species of Plasmodium parasites that infect humans, together with the approach of expressing extracellular parasite proteins in a functionally active form using a mammalian expression system. After removing those proteins which showed evidence of cross-reactive responses, it was found that immunoreactivity to the antigens from P. ovale, (MSP10, P41), P. malariae (MSP10) and P. knowlesi (MSP10, P12, P38), together with logistic regression modelling had moderate power to predict prior exposure to these species, and performed especially well for P. knowlesi.

Expressing Plasmodium proteins is known to be technically challenging, possibly due to the unusual codon bias or highly repetitive amino acid sequences often found in Plasmodium proteins [15, 38]. Most studies expressing recombinant antigens for serological assays use prokaryotic expression systems (especially Escherichia coli) or cell-free systems, such as wheat germ extracts [39‐41]. While these expression systems have the advantages of being cost effective, high yielding and widely available, they may not be suitable for expressing extracellular proteins because they usually require a reducing environment which would interfere with the formation of structurally critical disulfide bonds [16, 42]. While protein refolding procedures can be used, they are usually complex, time consuming, and have uncertain outcomes which are often hard to determine if they have been successful [43]. When used for serological screening, proteins that do not adopt the native conformation are unlikely to be useful in detecting antibodies that recognise conformational epitopes, potentially reducing sensitivity. By using a mammalian expression system to produce Plasmodium proteins, it was previously shown that these recombinant proteins can retain conformational epitopes, and shown that for the majority of antigens, a large fraction of the immunoreactivity to sera from Plasmodium-exposed patients is heat labile, demonstrating the benefit of using proteins that retain native folding for serological assays [16]. Plasmodium blood-stage proteins expressed using this approach have been useful in sero-epidemiological studies for P. falciparum [12, 44] and P. vivax [8]. One likely consequence of using conformational epitopes in serological assays, which are aimed at distinguishing between species, is that the percentage sequence identity between orthologous proteins is unlikely to be a reliable indicator of cross-reactivity. In the current study, a likely cross-reactivity was observed between the P. falciparum and P. vivax orthologues of the HPzz protein between patients diagnosed with these parasites, despite there being only ~ 40% amino acid shared sequence identity. By contrast, very little cross-reactivity was observed in patient serological responses between the P. vivax and P. knowlesi P12 orthologues which are relatively well conserved, sharing > 70% amino acid sequence identity. One likely possibility is that the informative epitopes for these proteins are mainly composed of the precise arrangement of solvent-exposed amino acids, making overall linear sequence identity less important, whereas repeats of just a few amino acids, for example the repetitive “NANP” motifs in the circumsporozoite protein can be highly immunogenic [45].

A panel of three recombinant antigens (PkMSP10, PkP12, PkP38), was found to accurately detect exposure to P. knowlesi using a serological assay. This compares well with a recent study where the authors also selected candidates that were orthologous to known immunoreactive proteins in P. falciparum, and selected four antigens that were associated with P. knowlesi exposure [46]. Together, these studies will contribute towards the further development of accurate serological assays for this parasite which is becoming an increasing public health concern in Southeast Asia, particularly in Malaysia where it is now replacing P. falciparum and P. vivax as the major cause of malaria [47, 48].

The use of plasma from native Europeans who had contracted travellers’ malaria and had been diagnosed accurately increased the chances that they had only been exposed to a single species which was useful for establishing which responses to different antigens might be cross reactive. However, accessing large numbers of these samples was difficult, especially for the rarer parasites, P. malariae and P. ovale, and therefore it is not certain that these patients had been previously exposed to other species of Plasmodium. Together with other assays that have been described for serological diagnosis of Plasmodium infections [37, 46, 49], this panel of proteins will contribute to an assay that will be a useful tool for establishing prior exposure to different species of Plasmodium parasites. Surveying more patient samples in longitudinal cohorts in a variety of transmission settings will be necessary to establish whether the responses to the proteins described here are different in patients living in different endemic regions, and vary according to the age and exposure of the patient.

The recent availability of genome sequences for Plasmodium parasites was used to create a panel of recombinant proteins corresponding to immunoreactive blood stage proteins from five human-infective Plasmodium species. This panel of proteins will provide a basis for developing serological assays to determine exposure to the different species of parasite for serological surveillance and diagnostics.