Background

Malaria in humans is caused by six Plasmodium species: Plasmodium falciparum, Plasmodium vivax, Plasmodium malariae, Plasmodium ovale curtisi, P. ovale wallikeri and Plasmodium knowlesi which, although zoonotic, is an important pathogen in humans in several regions of South East Asia [1, 2]. Whereas current research is focused on malaria caused by P. falciparum and increasingly also P. vivax, only relatively little effort has been dedicated to research on the other human malaria species. Infections with these species usually present with low parasitaemia; they can persist for long periods and sometimes remain asymptomatic. Besides mono-infections with one Plasmodium species, mixed infections within one individual occur [3, 4] and interaction between concurrent species—although not well characterized—may play a role in disease progression and outcome [5, 6]. In clinical care, species determination is commonly performed by light microscopy, which has limited sensitivity and specificity. Double, triple or even quadruple infections can be detected more reliably by molecular methods. In addition, presence of distinct genotypes (strains) of the same species cannot be discerned microscopically. An important proportion of naturally occurring infections consists of multiple genotypes and “multiplicity of infection” (MOI) refers to the number of different genotypes of one species infecting a single host [7]. For P. falciparum, MOI has been associated with several conditions, including age of the host, clinical severity, and transmission intensity [8]. Determination of within-host diversity may be a useful marker to assess the impact of interventions [9].

Genotyping of polymorphic genes such as merozoite surface proteins (MSPs) and glutamate-rich protein (glurp), by nested-PCR is a gold standard method for assessing MOI in P. falciparum infections. An alternative approach uses next generation sequencing (NGS) technologies, which allow high-resolution analyses of a heterogeneous mixture of the parasites within the host [10]. Compared to the standard method, NGS revealed an up to six times higher MOI of P. falciparum in a previous study [11]. By generating multiple reads per sample (usually between 100 and 10,000), this technique is very sensitive and able to detect minor alleles. Recent analysis of the conserved cytochrome b gene (cytb) using NGS, detected multiple genotypes infections (up to 4) including non-falciparum species in 10% of 437 samples collected in Cameroon [12].

Here, a metagenomics sequencing based approach was applied using three pan-Plasmodium primer sets for nucleus (18S rRNA), mitochondrion (cytb), and apicoplast (clpC) encoded genes to characterize the different Plasmodium species that infect patients with microscopically diagnosed non-falciparum malaria in the department of Tsamba-Magotsi, Gabon, a remote area of rural Central Africa [13]. By this approach, parasite diversity in 46 Gabonese symptomatic patients is described, that are sampled over a time period of 3 years and provide a baseline for research on the epidemiology and potential influence of non-falciparum malaria on disease burden in Central Africa.

Methods

Patients

Parasite DNA for this study was extracted from dried blood spots on filter paper of 46 patients with uncomplicated, microscopically diagnosed non-falciparum or mixed species malaria, which represents a minority of malaria cases in the region but was found in more than 20% in a cross-sectional survey in the area [13]. Blood samples from two clinical studies were used. From the first study (recruited 2008–2010) samples of 30 patients were included, details of patients’ characteristics are published elsewhere [14]. Briefly, patients with uncomplicated malaria, defined as symptoms and presence of P. ovale or P. malariae in thick blood smear, either as mono or mixed infection were included after informed consent was given. All patients received artemether-lumefantrine as anti-malarial chemotherapy and recovered from the infection. From the second study (recruited 2012–2013) 16 samples of patients that were included for molecular assessment of Plasmodium spp. were used. Both studies were approved by the regional ethics committee (Comité d’Ethique Régional Indépendant de Lambaréné) and followed the principles of the Declaration of Helsinki in its 5th revision.

Amplification and 454 sequencing

Capillary blood collected on filter paper from patients prior to anti-malarial treatment was processed for DNA extraction using QIAamp DNA blood mini kit (Qiagen) according to the manufacturer’s specifications. For malaria species and genotype identification using NGS, we designed three Plasmodium genus-specific primer sets from the conserved region flanking the highly polymorphic nucleotide sequence of the 18S rRNA, the mitochondrial cytochrome b (cytb) and the apicoplast caseinolytic protease C gene (clpC). All primers were 5′-fused to universal tail sequences. Using 2.5 µl of DNA extract, target specific PCR was carried out by using Phusion High-Fidelity PCR master mix (Finnzymes) for 35 cycles. The 454 MID kit (Multiplicom) was used to perform subsequent PCR addition of a 454-adaptor sequence linked to multiplex identifiers (MID) in order to discriminate the patients in following analysis steps. Here, the 100 times diluted first round PCR product was used as template and amplified for 20 cycles. All reactions were carried out using a Biometra T2 professional Thermocycler. The corresponding primer sequences and annealing temperature are given in Table 1. PCR amplicons were purified using AmPure XP kit (Agencourt) according to standard procedures (Roche Technical Bulletin No. 2011-007). Quality and purity of amplicons were checked using the Agilent DNA 1000 assay kit on a 2100 Bioanalyzer (Agilent Technology) and subsequently quantified using the Quant-iT Picogreen dsDNA reagent (Invitrogen) on a Fluoroskan Ascent microplate Fluorometer (ThermoScientific). Based on individual DNA concentration, each amplicon was diluted to 107 molecules/µl stock solutions. Amplicons were pooled in equimolar concentration to generate a single library and further processed following the GS Junior emPCR LibA method (Version April 2011) for emulsion PCR (emPCR) using a low copy per bead ratio (0.25 cpb). 500,000 DNA enriched beads were loaded onto a GS Junior Picotiter plate following the GS Junior sequencing manual (Version April 2011) and sequencing was performed in both, forward and reverse direction using the GS Junior Titanium sequencing kit.

Table 1 Primers used in this study
Full size table

Sequence analysis

Figure 1 summarizes the bioinformatics pipeline used to process Plasmodium sequence reads generated by GS Junior. The sfffile program (SFF Tools, Roche) was used to split raw sequence data based on multiplex identifier (MID). Low quality and short reads (< 200 bp) were excluded prior to analysis. Sequencing errors (PCR noise) and homopolymer stretches were corrected with Acacia [15]. Chimeric sequences were detected by using the software Uchime [16] and excluded from further analysis. Both programs were run with default parameters. High-quality filtered reads were mapped to a local reference database comprising 18S rRNA, cytb and clpC gene sequences of Plasmodium spp. downloaded from GenBank (Table 2). Unmapped sequences were further analyzed using BLAST searches against the NCBI nucleotide database and the Plasmodium database (Plasmodb) [17]. Single nucleotide polymorphisms (SNPs) in the analyzed genes were determined using the probabilistic variant detection method-implemented in the CLC Genomics Workbench 5 (CLC Bio, Aarhus, Denmark). Haplotypes were determined with DnaSP [18] using the following thresholds: SNP frequency equal or greater than 10% as well as per gene read coverage ≥ 10 fold. Plasmodium species genotypes were identified by querying consensus sequences against the NCBI GenBank database. All SNP positions were reported as absolute positions in the best matching NCBI reference sequence. Polyclonal infections were identified based on SNPs in the variable region (V5) of the 18S gene, giving the number of genotypes infecting one host (referred to as MOI). The number of genotypes obtained by this marker typically leads to an underestimation of polyclonality and characterization of genotypes based on more polymorphic genes might better estimate the scale of multiple infections. However, this would make comparisons between species more difficult as different genes would be evaluated.

Fig. 1
figure 1

Bioinformatics pipeline. Flowchart describing bioinformatic pipeline. The pipeline is divided into a quality control part, mapping and variant calling part. * used with default settings

Full size image
Table 2 Plasmodium spp. reference sequences used for 454-reads mapping
Full size table

Results

Sequence statistics from patients

The 46 patient samples were analyzed in batches of four sequencing runs on a Roche 454 GS Junior sequencer. The median number of raw reads assigned to each sample was 3829 (range 1563–11,590). However, after removal of chimeric and low-quality reads, the final informative sequences available for each sample was 3165 (range 655–9091). The sequence coverage for each locus per sample is reported in the Additional file 1. Differences in coverage between loci and overall batch to batch variation in the number of raw reads were observed. However, there is no evidence that this substantially influences the results, as the overall coverage per sample is expected to be high enough to detect minor genotypes.

Species discrimination by deep sequencing

Each filtered read was mapped to the target genes of the reference species for identification. Mixed-species malaria infections identified by the metagenomics approach were also detected by conventional nested-PCR in combination with gel electrophoresis [14]. By NGS, it was possible to achieve a more detailed resolution of the species diversity that included identification of nine novel genotypes and polyclonal infections using the same gene in all six investigated Plasmodium spp.

A minimum of three high quality reads had to be assigned to at least one of the target loci to be included in the analysis. There were 44 (96%) patients with P. falciparum, 15 (33%) with P. malariae, 7 (15%) with P. ovale curtisi, and 8 (17%) with P. ovale wallikeri infections. Different parasite combinations in mixed infections and up to four species in the same host were detected. Of 46 samples, 24 contained only P. falciparum, another two contained only P. malariae or P. ovale wallikeri. The remaining samples contained double (n = 15), triple (n = 2), and quadruple (n = 3) Plasmodium species infections (Table 3).

Table 3 Number of Plasmodium spp. infections detected by deep sequencing
Full size table

Plasmodium falciparum infections

Plasmodium falciparum specific sequence reads were identified in all but 2 (96%) of the study samples. Based on the 18S A-type gene sequence, four novel genotypes (GenBank KJ170098—KJ170101) were identified, all are a single nucleotide polymorphism when compared to the reference sequence (GenBank JF681166) as shown in Table 4. No other sequences with these SNPs were found when a similarity search against GenBank was performed. Co-infections with other species were found in 20 patients (45%).

Table 4 Polymorphic Nucleotide positions compared to GenBank best-hits reference sequences
Full size table

Plasmodium malariae infections

A total of 15 infections were observed with the quartan malaria parasite, one as mono and the remaining ones as co-infection with other species. Based on the 18S rRNA gene, the parasites were further characterized as P. malariae-Asian type 1 (n = 1) and P. malariae-Asian type 2 (n = 14) [19]. Two unique genotypes (GenBank KJ170105, KJ170106) similar to Plasmodium cf. malariae type 2 were obtained (GenBank AF488000, 99% identity). Each genotype is defined by either a single nucleotide substitution or insertion compared to P. malariae-Asian type 2 (Table 4) (see comments below).

Plasmodium ovale wallikeri and P. ovale curtisi infections

Plasmodium ovale wallikeri and P. ovale curtisi were identified in the dataset supported by reads matching to specific 18S, cytb, or clpC sequences of the two P. ovale species (Additional file 1). In addition, mixed infections of P. ovale curtisi and P. ovale wallikeri were observed in four patients (Table 3).

All of P. ovale curtisi genotypes identified from the samples were similar to previously submitted sequences. Three novel genotypes of P. ovale wallikeri were identified based on 18S rRNA gene sequence polymorphisms (GenBank KJ170102—KJ170104). One of them possesses a single nucleotide substitution at position 1128. Genotype 2 and Genotype 3 showed six and two nucleotide substitutions when compared to the corresponding reference sequence (GenBank AB182493), respectively (Table 4).

Detection of P. ovale curtisi 18S rRNA gene variant

In three of the samples a fraction of reads (range 10–15 reads) could not be mapped to any reference sequence. A sequence identity search against the NCBI database initially did not show significant similarity with respect to query coverage and maximum identity with any of the Plasmodium spp. 18S sequence. A 1078 bp length sequence was constructed (GenBank KJ170108), spanning the variable regions 4 and 7 along with the 454 amplicon target (V5) by PCR and Sanger sequencing with pan-Plasmodium and specific primers, respectively (Fig. 2). The generated sequence showed 100% similarity to the draft genome of P. ovale (P. ovale Blast Server, Sanger Institute) and a recently discovered 18S rRNA gene variant (GenBank KF696378) of P. ovale curtisi [20].

Fig. 2
figure 2

Schematic representation of the sequencing approach by Sanger method to determine the new 18S rRNA gene type sequence of P. ovale spp. Conserved pan-Plasmodium 18S primers (rPLU6 and rPLU5) were first used to amplify the gene spanning three variable regions and two conserved blocks. The nucleotide sequences upstream and downstream of the 454 sequencing target region (V5) of the gene were determined by direct sequencing of the initial product using p-AVL-F and p-AVL-R primers. Partial 18S gene (1078 bp length) was obtained by assembling 454 reads with the sequences obtained using p-AVL-F and p-AVL-R primers

Full size image

Multiplicity of infection

An overview of the MOI based on the variable region (V5) of the 18S gene for the different Plasmodium species is given in Table 5. The relative frequency of each genotype per sample based on the number of reads is plotted in Fig. 3. Multiple P. falciparum genotype (MOI) infections were detected in five patients with a maximum of two genotypes per sample. All P. falciparum monoinfections carried a single genotype, except one patient (MID15) who carried two genotypes. Multiple genotype infections were found in two of the seven P. ovale curtisi infected samples (MOI of 3). Among eight samples infected with P. ovale wallikeri, six harboured multiple P. ovale wallikeri genotypes (MOI of 2–3). Analysis of the cytb and clpC reads showed 100% similarity to the published reference sequences. Nine samples contained multiple P. malariae genotypes (MOI of 2–4). The number of different genotypes per sample was greater for P. malariae when compared to the other species, showing up to four genotypes within one sample.

Table 5 Multiplicity of Plasmodium spp. infections
Full size table
Fig. 3
figure 3

Relative frequency of each Plasmodium spp. genotype within one infected host. a Relative frequency of P. falciparum genotypes (defined by the 18S A-type gene variation); b relative frequency of P. malariae genotypes. c Relative frequency of P. ovale curtisi and P. ovale wallikeri

Full size image

Reads mapping to cytb and clpC showed 100% sequence identity to the reference sequences for all species (KC175316 and DQ642846 for P. falciparum; KJ930413 and KP050439 for P. ovale wallikeri; KP050432, KP050446 for P. ovale curtisi; HQ842634 and AB649418 for P. malariae).

Discussion

Gabon is highly endemic for malaria with perennial transmission of P. falciparum and co-endemicity of P. malariae and P. ovale [21, 22]. Gabon has a low population density (6.7 persons per square km in 2015) [23], but population mobility is high, and a large fraction of inhabitants live close to or within non-cultivated areas, that contain diverse and intact wildlife [24]. Originally, the study was designed to assess the efficacy of artemisinin combination therapy (artemether-lumefantrine) in uncomplicated non-falciparum and mixed species malaria [14]. Species identification was based on microscopy and verified by species-specific nested PCR [25]. Here, a metagenomics approach was designed based on high throughput sequencing to re-analyze the diversity of Plasmodium spp. Two studies were conducted recently using NGS to explore the evidence of zoonotic transmission of Plasmodium spp. in humans [12, 26]. These studies were designed to differentiate Laverania species of apes and humans based on the diagnostic single nucleotide polymorphism found in the mitochondrial DNA (mtDNA). The current study expanded the target genes to three genes including the 18S rRNA and clpC gene to improve accuracy and assess the diversity of Plasmodium populations in blood samples of malaria patients. The 18S rRNA gene is one of the most commonly used targets in the molecular diagnosis of Plasmodium spp. including a widely used nested PCR and a number of real-time PCR assays [25, 27]. Due to the high copy numbers (5–10 copies), and the presence of both highly conserved and variable regions, 18S genes are well-suited and frequently used molecular targets for the detection and discrimination of Plasmodium species. Until to date, no study has applied these genes to identify Plasmodium spp. using NGS.

As expected, results obtained by conventional species-specific PCR were reproduced upon DNA re-extraction and analysis by deep sequencing. Beyond this, this analysis gives a snapshot in time; showing the breadth of co-infections and an unbiased estimate of infection intensity of the different Plasmodium spp. in a cohort of patients with uncomplicated malaria. P. malariae is widespread in Africa but rarely characterized by molecular techniques. Based on morphological differences and sequence polymorphisms compared to genuine P. malariae (Uganda CDC isolate), two possible subtypes were identified in Asia: P. cf. malariae type 1 and P. cf. malariae type 2 [19]. It would be interesting to investigate if the Asian and the Ugandan type differ in their sensitivity to artemisinins, since a reduced sensitivity of an infection in Uganda has been observed [28, 29] but all patients in this study responded well to the treatment [14]. Only few studies have looked at MOI of non-falciparum species. It has been shown in one study that P. malariae infections in Malawi, often consisted of multiple genotypes per infected individual and showed a surprisingly similar pattern when compared to P. falciparum [30]. Up to five genotypes were detected from a single sample by multilocus genotyping based on microsatellite markers in asymptomatic carriers [31]. There are also up to four different genotypes of P. malariae in two patients and three different genotypes of P. ovale wallikeri and P. ovale curtisi each in two patients in the here presented study based on the 18S region, respectively. Many of the P. falciparum infections were on the contrary caused only by one genotype when judged by polymorphisms in the 18S rRNA gene. These results are surprising as one would assume a lower MOI in P. malariae and both P. ovale species as the reported prevalence in this region is low. However, results are also in line with earlier findings from Cambodia for P. ovale [32], reflecting the possibility that these parasite species are more prevalent than previously thought. Evidence suggests that P. malariae is not completely species-specific and also prevalent in non-human primates [29, 33], having a larger pool of hosts. In addition, infections with these species might be more chronic and long-lived so that multiple genotypes could accumulate in one host. It has to be considered that the marker used to define a genotype in this study is not as polymorphic as markers commonly used to define genotypes; for example microsatellites, or genes like msp1, msp2 and glurp for P. falciparum [34]. It is expected that the number of genotypes are larger if more polymorphic markers had been used. The advantage of our approach is that homologous genes between the different species can be compared and one can get an impression on the population structure of the co-infecting species. Particularly, because there is no evidence that selection leads to different mutation rates in this gene region for the different species, this can be one approach to compare diversity between species. These results highlight the potential of amplicon-based high throughput sequencing combined with adequate polymorphic markers to obtain reliable molecular characterization of other non-falciparum species, where only limited data are available. Despite short-read length (< 400 nt) generated by the NGSs technique, significant dimorphism of the targeted genes allowed accurate assignment of reads between P. ovale curtisi and P. ovale wallikeri. By using a multiple locus approach, co-infections of the two P. ovale species in four patients were detected, adding yet further evidence that the two species do not recombine in nature. The primer for the cytb gene was selected from a region not including the ape-specific SNPs leading to potential inability to detect non-human species. However, there was no evidence for non-human Plasmodium species infections based on the analysis of the other two genes.

Two structurally distinct types of 18S rRNA have been reported in many Plasmodium species [35]. In P. falciparum, type A and type S has been described with up to 11% difference in the sequences between the two types [36]. The existence of paralogous 18S rRNA genes in P. malariae, P. ovale wallikeri and P. ovale curtisi genome is not well described until now. The new type of 18S rRNA gene sequence from this study, together with the recent report [20], clearly suggests that at least two different forms exist in P. ovale curtisi and P. ovale wallikeri [20]. A real-time PCR based assay targeting this new sequence showed high sensitivity and specificity and can be used for the differential diagnosis of P. ovale species infections (unpublished observation).

The main limitation of techniques that involve PCR is the inherent risk of contamination and the NGS approach is no exception. Thus, an essential aspect of sample preparation for sequencing is the careful adoption of experimental strategies aimed at minimizing cross-contamination. In addition, amplicon-based NGS are prone to errors such as chimeras, a well-known issue in metagenomics analyses of environment bacterial communities. Several measures to minimize and avoid these underlying problems at each step, including a non-template control to check contamination of reagents during each target amplification, a separate work station for DNA extraction, PCR assay set-up, and post-PCR processing are necessary.

The true burden of disease caused by P. malariae and the P. ovale species is not known as these species are mostly underdiagnosed despite their worldwide distribution [29]. Malaria caused by these parasites is generally more benign when compared to P. falciparum. Infections present often with a low level parasitaemia that is difficult to detect by microscopy, but might cause a more chronic illness associated with anaemia [37]. Recent reports reveal a higher than expected prevalence of these species in many African countries when diagnosed by PCR [38], going in line with the deep sequencing result. Currently, deep sequencing methods may not be applicable in field settings but can be helpful in longitudinal epidemiological studies to investigate the emergence and change in composition of plasmodial species. The study population was highly selected and represents only a minority of all malaria cases in the area. Microscopic species differentiation is not reliable, particularly when parasite density is low. Therefore, it is unlikely that the appearance of new plasmodial species is noted without molecular techniques. Diversity of the two P. ovale species, and characterization of the P. malariae population should be taken into consideration for the design, endpoints and feasibility of malaria control strategies, e.g. longer follow up or separate treatment of hypnozoites. Fortunately, artemisinin combination therapy was efficacious in all patients of the present study but emergence of novel zoonotic species or isolates that may be inherently less responsive to current treatment regimens shall be detected early, to provide a specific treatment regimen, especially when severe malaria can occur, as in the case of P. knowlesi infections. The impact on malaria vaccine development shall even be stronger since cross-protection between Plasmodium species is limited [39] and mechanisms of high-grade protection are often not known but likely to be different for each parasite species.

Conclusions

Parasite diversity of Plasmodium species in naturally acquired malaria is larger than expected. The population structure, especially of non-falciparum species, needs further assessments to better understand the prevalence and biology of these parasites. Metagenomics analysis by deep sequencing provides a tool for this and boosts the understanding of naturally acquired malaria. It will be exciting to expand such investigations to other malaria-endemic regions and larger cohorts.