Ethics statement

For the 59 P. malariae samples, no specific consent was required because, in coordination with the Santé Publique France organization for malaria care and surveillance, the human clinical, epidemiological, and biological data were collected in the French Reference National Center for Malaria (CNRP) database and analyzed in accordance with the public health mission of all French National Reference Centers. The study of the biological samples obtained in the context of medical care was considered as non-interventional research (article L1221-1.1 of the French public health code) and only required the patient’s non-opposition during sampling (article L1211-2 of the French public health code).

Regarding non-human primates (NHPs) samples, 212 were collected in French Guiana. Those samples are registered in the collection JAGUARS (https://kwata.net/gestion-collection-biologique/, CITES reference: FR973A) managed by the Kwata NGO (accredited by the French Ministry of the Environment and the Prefecture of French Guiana, Agreement R03-2022-12-30-0007 and R03-2024-11-07-00036), hosted at the Institut Pasteur de la Guyane, supported by Prefecture de la Région Guyane and Collectivité Territoriale de la Guyane, and validated by the French Guianan prefectoral decree n°2012/110.

Regarding samples from Colombia, ethical approvals for the collection of eight fecal and six blood samples were obtained by Universidad de los Andes, the National Environmental Licensing Authority of Colombia (ANLA) and the Centers for Disease Control and Prevention (permits numbers: 2017025578-1-000, 2017043863-1-000, 2017065795-1-000, 2017013727-1-000, 2017052943-1-000, 2017081458-1-000, 2017108650-1-000).

Origin of samples and species identification

The NHP samples were predominantly from Alouatta macconnelli (n = 140) and Saguinus midas (n = 58), with lower representation of Ateles paniscus (n = 7), Ateles hybridus (n = 5), Alouatta seniculus (n = 5), Sapajus apella (n = 4), Cebus versicolor (n = 3), Saimiri sciureus (n = 2), Aotus griseimembra (n = 1), and Pithecia pithecia (n = 1).

For the all newly sequenced samples in this study (P. malariae and P. brasilium), genomic DNA was extracted using the Qiagen DNeasy Blood and Tissue Kit according to the manufacturer’s recommendations. Plasmodium brasilianum samples were identified by amplification of Plasmodium cytochrome b using nested PCR, as described in Prugnolle et al. [23]. The reaction products were visualized on 1.5% agarose gels stained with EZ-vision and sent for Sanger sequencing to confirm Plasmodium species (Eurofin MWG). This allowed the identification of 20 positive P. brasilianum samples.

Selective whole genome amplification (sWGA), library preparation and sequencing

For the 59 P. malariae DNA samples, selective whole-genome amplification (sWGA) was performed to enrich parasite DNA from samples with low parasitemia using the P. malariae-specific protocol developed by Ben-Rached et al. [24]. This technique preferentially amplifies the P. malariae genome from mixed DNA samples while reducing host DNA contamination. Given the close genetic relationship between P. malariae and P. brasilianum, the same approach was also applied to P. brasilianum samples. The primer set consisted of ten 8–10 bp primers (PM1–PM10), each containing phosphorothioate bonds at the 3′ end to prevent exonuclease degradation. Each primer was initially prepared at 100 µM and combined into a Primer Set 1 master mix to achieve a final concentration of 1.225 µM per primer in a 50 µL reaction. For each reaction (50 µL total volume), approximately 50 ng of genomic DNA was used as input. The reaction mixture contained 17.5 µL of Primer Set 1 master mix, 5 µL of 10 × phi29 enzyme buffer (New England Biolabs), 3 µL of phi29 DNA polymerase (30 U; NEB), 2 µL of 25 mM dNTP mix (ThermoFisher), nuclease-free water, and elution buffer (EB) adjusted to reach the final volume of 50 µL. Amplification was carried out using a “ramp-down” thermocycling program: the temperature was decreased stepwise from 35°C to 30°C (10 minutes per degree), followed by a 16-hour incubation at 30°C. Enzyme inactivation was performed at 65°C for 10 minutes, and reactions were held at 4°C until further processing. For each sample, the products of the two amplifications (one per primer set) were purified with AMPure XP beads (Beckman Coulter) at a 1:1 ratio according to the manufacturer’s recommendations and pooled at equimolar concentrations. Each sWGA library was prepared using the two pooled amplification products and the Nextera XT DNA kit (Illumina), following the manufacturer’s protocol. Then, samples were pooled, clustered, and sequenced on one lane of a Illumina Novaseq-6000 S4 with 2 × 150-bp paired-end reads (MGX Montpellier).

Short-read mapping, SNP calling, and data compilation

We generated whole genome-sequencing data for 59 sequenced P. malariae isolates and 20 newly sequenced P. brasilianum samples. These were added to a compilation of previously published P. malariae genomic datasets: 222 samples from Ibrahim et al. [20], 17 samples from Ibrahim et al. [16], four samples from Rutledge et al. [17], one sample from Ansari et al. [18] and one sample from Plenderleith et al. [19]. Three P. malaria-like genomes obtained from chimpanzees (Pan troglodytes) were included as an outgroup [17,19]. Short read archives (SRA) from the published studies were retrieved from NCBI (accession numbers provided in S1 Table).

Short reads were trimmed to remove potential lingering adapters and preprocessed to eliminate low-quality reads (−quality-cutoff = 30) using the cutadapt program [25]. Reads shorter than 50 bp and containing “N” (i.e., ambiguous bases) were discarded (−minimumlength = 50 –max-n = 0). Cleaned paired-end reads were mapped to the P. malariae reference genome PmUG01 [17] using bwa-mem v0.7.17 [26] with default parameters. Duplicate reads were marked using the MarkDuplicates tool from the Picard tools v2.5.0 (broadinstitute.github.io/picard/) with default options. Local realignment around indels was performed using the IndelRealigner tool from Genome Analysis Toolkit (GATK [27], v3.8.0). Variants were called per sample using the HaplotypeCaller module in GATK with the parameter -stand_call_conf equals to a Phred-scaled confidence score of 10. Lastly, the different isolated variant call format (VCF) files were merged using the GATK module CombineGVCFs.

Core genome SNP filtering and final dataset composition

Combining previously published dataset (n = 248) with the 79 newly sequenced samples resulted in a total of 327 samples prior to filtering: 304 P. malariae samples, 20 P. brasilianum samples, and three P. malariae-like samples. All filtration steps are detailed in S1 Fig.

Samples with >75% missing data were removed (n = 69). As several parasite strains can infect the same host, the within-host infection complexity was assessed with the F_WS metric [28], calculated with vcfdo (github.com/IDEELResearch/vcfdo; last accessed July 2022). Samples with pluri-clonal infections, i.e., F_WS ≤ 0.85 were removed (n = 44) as in Ibrahim et al. [20] (S2A Fig). For samples that are the sole representatives of their population, the F_WS index was replaced with a homozygosity measure to evaluate the multiplicity of infection and avoid the bias introduced by an F_WS value of 0 for isolated samples (S2B Fig).

Highly related samples and clones can generate spurious signals of population structure, bias estimators of population genetic variation, and violate the assumptions of the model-based population genetic approaches used in this study [29]. The relatedness between haploid genotype pairs within each country was measured by estimating the pairwise fraction of the identity by descent (IBD) between strains within populations using the hmmIBD program [30], with default parameters for recombination and genotyping error rates, and using the allele frequencies estimated by the program. Within each country, isolate pairs that shared >50% of IBD were considered highly related. In each family of related samples, only the strain with the lowest number of missing data was retained, thus removing 33 samples (S2C Fig).

The final dataset included whole genome sequencing data of 181 samples from 29 countries, including 47 newly sequenced strains (Fig 1), and consisted in two P. brasilianum samples (Colombia n = 1, French Guiana n = 1), 179 P. malariae samples with four samples from South America (French Guiana n = 1, Brazil n = 3), eight from North Africa (Algeria n = 1 and Sudan n = 7), 61 from West Africa (Senegal n = 1, Gambia n = 1, Guinea n = 5, Sierra Leone n = 2, Ivory Coast n = 9, Mali n = 4, Ghana n = 14, Togo n = 1, Benin n = 2, Nigeria n = 24), 50 from Central Africa (Cameroon n = 21, Central Africa n = 4, Equatorial Guinea n = 1, Gabon n = 6, Congo n = 11, Angola n = 5), 40 from East Africa (Uganda n = 14, Kenya n = 10, Tanzania n = 12, Malawi n = 4) and 13 from Asia (India n = 1, Thailand n = 10, Malaysia n = 1, Papua New Guinea n = 1). Two P. malariae-like samples from Gabonese chimpanzees (Pan troglodytes troglodytes) were included as an outgroup. The mean sequencing depth ranged from 0.4x to 732x for P. malariae genomes and from 18.4x to 521.4x for P. brasilianum genomes (S1 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 1. Geographic origins of the 179 P. malariae isolates, two P. brasilianum, and two P. malariae-like isolates from African great apes.

Sample distribution by region: South America (n = 6, blue shades), North Africa (n = 8, purple shades), West Africa (n = 63, red shades), Central Africa (n = 50, orange shades), East Africa (n = 40, yellow shades), and Asia (n = 13, green shades). Sample counts are shown in brackets and with circle size. The countries involved in the sampling are shown in shades of grey. Darker shades indicate countries with newly sequenced genomes, while lighter shades indicate countries with previously published data. Bolded country names denote locations where new genomes were generated in this study. Sample counts from the literature are listed first in brackets, followed by newly sequenced samples. The base layer of the map was made with Natural Earth (naturalearthdata.com).

https://doi.org/10.1371/journal.pntd.0014176.g001

Most population genetic analyses were conducted using ANGSD v0.940 [31] or ANGSD suite software, only on the core genome regions as defined by Ibrahim et al. [16]. All sites with a base quality ≥20 were kept. For analyses that did not require invariant sites, only SNPs with a p-value = 1e-6 were retained. For the maximum likelihood (ML) tree, we used a filtered VCF with only bi-allelic SNPs with ≤50% missing data, a quality score ≥30, and a minimum and maximum depth coverage set between 10X and 182X. Moreover, singletons were removed to minimize sequencing errors.

Global population structure and genetic relationships

Principal Component Analysis (PCA) and the ancestry plots were computed with ANGSD and PCAngsd v0.98, after selecting only SNPs present in the core region of the P. malariae genomes and excluding SNPs with a minor allele frequency (MAF) ≤1%. The variants were linkage disequilibrium (LD) pruned to obtain a set of unlinked variants using ngsld v1.2 [32], with a threshold of r² = 0.5, a window size of 5kb and a step of 1kb. After LD-pruning, 210,703 SNPs were retained for 181 individuals. The two P. malariae-like outgroup samples were excluded from these analyses. The ancestry plots were estimated using PCAngsd for k value (number of clusters) ranging from 2 to 16. Then, pong v1.5 [33] was used to analyze the PCAngsd resulting outputs and compute ancestry proportions.

The ML tree was reconstructed with IQ-TREE v2.3.4 [34] using the best-fitted model determined by ModelFinder [35]. As the dataset with outgroup comprised 118,772 SNPs of the core genome and no invariant sites, the ascertainment bias correction was added to the tested models. The best inferred model was a general time reversible (GTR) model of nucleotide evolution that integrated unequal rates and unequal base frequency. The node reliability was assessed with Ultrafast Bootstrap Approximation [36] and the SH-aLRT test [37].

Characterization of population structure within Africa

To investigate fine-scale population structure within Africa, we restricted the dataset to African P. malariae isolates, as Africa represented both the most densely sampled region (n = 161) and the largest proportion of newly generated genome data in this study. Because one Sudanese sample clustered with American rather than African populations, it was excluded from these subsequent analyses. The final African dataset comprised 160 samples, in 22 countries (Fig 1).

Population structure within Africa was first explored using PCA and genetic ancestry plots computed with ANGSD and PCAngsd v0.98, after selecting only SNPs present in the core region of the P. malariae genomes and excluding SNPs with a minor allele frequency (MAF) ≤1%. The variants were LD-pruned to obtain a set of unlinked variants using ngsld v1.2 [32], with a threshold of r² = 0.5, a window size of 5kb and a step of 1kb. In total, 210,587 SNPs were used. These analyses revealed the presence of two major genetic clusters within Africa. The ancestry plots were therefore inferred with PCAngsd assuming K = 2. To assess potential geographic structuring, the relationship between ancestry coefficients and country longitude was evaluated using a Spearman rank correlation test [38].

To quantify genetic diversity within each of the two inferred African clusters, Tajima’s D [39] and nucleotide diversity (π) were measured using pixy v2.0.0 [40], with non-overlapping windows of 500 bp in the core genome, and only windows with ≥100 sites were kept.

Genome-wide patterns of genetic differentiation caused by reduced recombination between the two African clusters were characterized using estimates of F_ST and D_XY computed with pixy [40] in sliding-window of 5 kb windows and 500 bp steps. For F_ST, only windows containing at least 100 SNPs were retained, whereas for D_XY, windows were required to include a minimum of 1,000 sites. Genomic regions proximal to telomeres and centromeres exhibited elevated values due to edge effects (incomplete windows). Since the core genome used has only been defined on a few genomes [16], it is possible that this pattern is also due to the inclusion of telomeric or subtelomeric regions in the windows. The windows were therefore excluded from subsequent analyses. However, the patterns found in the genome with these two metrics might be caused by a reduction in recombination rate between the two clusters or because of changes in population demographic history, and may not necessarily result from positive selection [41]. Thus, lineage-specific tests of selection were subsequently performed.

Lineage specific positive selection detection

To identify genomic regions potentially under positive selection in each African cluster, the population branch-site (PBS) values [42] were calculated for each cluster, with pixy [40] using sliding windows of 5 kb and a step size of 500 bp. The outgroup populations were the alternative African cluster and Thai population. All windows with <100 SNPs were removed to avoid extreme values caused by a low SNP number. The 0.1% most extreme values were considered evidence that the genomic regions displayed signs of selection in the African cluster.

We also used haplotype-based tests (XP-EHH and Rsb [43,44]) between African clusters to identify signals of positive selection. These tests use haplotype length variations (i.e., linkage disequilibrium) to detect recent or ongoing selective events [43,44]. These tests are based on differences in the lengths of homozygous haplotypic segments across the genome [43,44] and therefore detect local differences in genetic diversity between the two clusters. XP-EHH and Rsb between the two African clusters were inferred using rehh [45] and 34,0174 SNPs. The significance threshold was set at −log (p-value) = 4, as recommended by Gautier et al. [45].

Once positive selection signals were detected, the identified genes were annotated using the general feature format (GFF) file GCF_900090045.1_PmUG01_genomic.gff (available from PlasmoDB) and the intersect function of BEDtools v2.31.1 [46]. Additional information was retrieved from PlasmoDB (plasmodb.org, accessed in December 2025).

Plasmodium brasilianum infections in South American NHPs

Among 226 samples collected from NHPs across two Latin American countries (14 Colombian, and 212 French Guianese), 20 (n = 4/14 from Colombia, and n = 16/212 from French Guiana), were positive for Plasmodium using the cytochrome-b based PCR assay [23] (S1 Table). Twenty Plasmodium-infected samples were identified across six NHP species (Alouatta macconnelli (n = 10/140), Saguinus midas (n = 3/58), Ateles hybridus (n = 1/5), Alouatta seniculus (n = 2/5), Aotus griseimembra (n = 1), and Cebus versicolor (n = 1/3)). Sequence analyses showed that all infections were attributable to P. brasilianum. All positive samples were then subjected to selective whole-genome amplification (sWGA) followed by whole-genome sequencing. However, after applying stringent quality filtering and SNP selection criteria, only two P. brasilianum samples (one from a Colombian A. seniculus and one from a French Guianan A. hybridus) retained sufficient high-quality SNPs to be included in the final dataset. Although the limited number of retained genomes precluded population genomic analyses of P. brasilianum, these samples were included to place African P. malariae diversity within the broader evolutionary context of the P. malariae–P. brasilianum complex.

Cryptic population structure within African populations

To explore the genetic relationships between P. malariae populations and with P. brasilianum, the population structure was characterized using complementary population genomic approaches, including principal component analysis (PCA), maximum likelihood (ML) phylogenetic tree, and model-based individual genetic ancestry inference.

The first three principal components (PCs) of the PCA revealed four major genetic clusters (Fig 2A). PC1 separated African isolates into two distinct clusters, whereas PC2 distinguished Asian populations from all others. Along PC3, South American P. malariae and P. brasilianum samples were separated from the remaining populations. Although the Asian and American clusters were clearly distinguishable in the ML tree (Fig 2B), the two African clusters were not. Overall, few nodes showed sufficient statistical support to be considered reliable (using thresholds established in the literature [36,37,47]), with the exception of those within the Asian and American clades. Furthermore, the two isolates of P. brasilianum were connected basally to the Latin American clade, although two samples of Brazilian P. malariae (PM_BRA_004 and PM_BRA_007) were actually included in the African diversity and not the Latin American (respectively * and ＊ on Fig 2B). One of the isolates (PM_BRA_004) is even very similar to a Nigerian isolate (PM_NGA_029), which can be seen on the ML Tree and PCA (Τ on Fig 2, and S3 Fig). Moreover, one Sudanese sample (PM_SDN_009) is genetically close to the American P. malariae clade (፠ on Fig 2).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. P. malariae and P. brasilianum genetic structure.

(A) Principal component (PC) analysis of 179 P. malariae and two P. brasilianum strains showing the first, second and third PCs based on the genotype likelihood of 210,703 unlinked SNPs. Plasmodium brasilianum are indicated by circles and P. malariae by squares. (B) Maximum likelihood phylogenetic tree of 179 P. malariae, two P. brasilianum and two P. malariae-like individuals. The tree includes two P. malariae-like strains used as outgroups. Note that the length of the outgroup branch (central line) has been truncated. Black dots at some nodes indicate highly supported nodes (both SH-aLRT ≥ 80% and UFboot ≥ 95%, following the thresholds established in the literature [36,37,47]). The symbols (፠,＊, and *) mark samples whose placements are discussed in the paper.

https://doi.org/10.1371/journal.pntd.0014176.g002

The genetic ancestry analyses were consistent with the PCA and the ML phylogeny. As observed in the PCA and the ML tree, the Asian and American clusters were well defined, emerging from K = 3 for the Asian cluster and K = 5 for the American cluster (Fig 3 and S4 Fig). In addition, two Brazilian P. malariae samples clearly clustered with African populations (* and ＊ on Fig 3A), in agreement with the PCA and ML tree. Similarly, the Sudanese sample appeared to be genetically closer to the American cluster (፠ on Fig 3A).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Ancestry of P. malariae and P. brasilianum.

(A) Individual genetic ancestry assuming K = 3, K = 6 and K = 8 genetic clusters estimated using PCAngsd (see S4 Fig for the other K values). K = 6 is the best K according to the method of Meisner and Albrechtsen [48] (see S3 Fig). The symbols (፠,＊, and *) mark samples whose placements are discussed in the paper. (B) Individual genetic ancestry assuming K = 6 genetic clusters estimated using PCAngsd displayed as pie charts on the world map. The base layer of the map was made with Natural Earth (naturalearthdata.com).

https://doi.org/10.1371/journal.pntd.0014176.g003

At a finer scale within Africa, two clusters were also detected (Fig 3). However, they were less clearly defined than in the PCA (Fig 2A), as many isolates showed admixed genetic ancestry. This admixture may explain the lack of clear separation between the African clusters in the ML tree, as well as the low statistical support observed for the nodes within the African diversity (Fig 2B). Moreover, no clear geographic structuring was apparent between the two African clusters (Fig 2B), which appear to coexist in the same regions and recombine. Because no obvious biological or geographic criteria allow these clusters to be distinguished, they were referred hereafter according to the colors used in the ancestry plots: cluster R (red) and cluster Y (yellow).

Finally, P. brasilianum isolates seem genetically closer to cluster R, whereas American P. malariae samples seemed more closely associated with cluster Y (Fig 3).

Genomic evidence for divergence and selection in the two sympatric cryptic African clusters

Although all isolates originated from the same continent, two genetic clusters were identified within African P. malariae populations (Figs 2 and 3). We therefore investigated the presence of genetic substructure within African P. malariae. Because one Sudanese sample (PM_SDN_009) clustered with American rather than African populations (Figs 2 and 3), it was excluded from subsequent analyses. A PCA restricted to African samples (n = 160) confirmed the separation into two distinct genetic clusters (Fig 4A). In contrast, ancestry inference using PCAngsd revealed a less sharply defined pattern, with several individuals exhibiting mixed ancestry between the two clusters. Despite the absence of a clear geographic partitioning, cluster Y was more prevalent in East and North Africa, whereas cluster R was more frequently observed in West Africa (Fig 4B). Notably, isolates from both clusters were detected across multiple regions of Africa, indicating broad geographic overlap.

Download:

• PNG
  larger image
• TIFF
  original image

Fig 4. Genetic characterization of two P. malariae clusters in Africa.

(A) Principal component analysis (PCA) of 160 African P. malariae isolates, showing the first two principal components derived from genotype likelihoods at 210,587 unlinked SNPs. (B) Individual genetic ancestry of 160 African P. malariae strains assuming K = 2 genetic clusters estimated using PCAngsd, represented as pie charts on a map of Africa. K = 2 is the best K according to the method of Meisner and Albrechtsen [47] (see S5 Fig). The base layer of the map was made with Natural Earth (naturalearthdata.com).

https://doi.org/10.1371/journal.pntd.0014176.g004

To further characterize the two African clusters, we retained only samples predominantly associated with a single cluster, defined as individuals showing more than 70% ancestry for one cluster in the ancestry analysis. This resulted in 74 samples in cluster R, 70 samples in cluster Y and 16 monoclonal infection samples were classified as admixed, with less than 70% ancestry assigned to either cluster (S2 Table).

We investigated potential geographic structuring by testing the relationship between ancestry coefficients and longitude. Although significant correlations were detected (R = −0.26 for cluster Y and R = 0.27 for cluster R; p < 0.001), longitude accounted for less than 7% of the variation in ancestry coefficients. Moreover, visual inspection revealed substantial overlap between the two clusters across Africa, with both clusters occurring throughout most of the sampled longitudinal range (Fig S6). These results suggest a weak east–west gradient in cluster frequencies rather than pronounced geographic structuring.

The genome-wide distributions of nucleotide diversity (π) and Tajima’s D [39] differed significantly between the two clusters (Wilcoxon signed-rank tests, p-value < 0.001 for both statistics). However, a Wilcoxon effect size test [49] revealed that the difference was very small, even negligible (r < 0.1) [50]. Specifically, π was slightly higher (medians of 2.16 × 10^-4 for cluster Y and 3.09 × 10^-4 for cluster R) and Tajima’s D was marginally lower (medians of -1.107 for cluster Y and -1.163 for cluster R) in cluster R than in cluster Y (S7 Fig).

To better understand the origin of these two clusters and identify genomic features underlying their differentiation, we performed genome-wide scans of F_ST and D_XY. Genomic regions where the recombination is less efficient between lineages, named islands of differentiation, are expected to exhibit elevated values of both F_ST and D_XY [51,52]. No region showed high values for both metrics. However, numerous F_ST peaks suggest differentiation driven by selection specific to each lineage (S8 Fig).

Given the genetic differentiation between the two African clusters, they may be subject to distinct selective pressures. We therefore assessed cluster-specific signals of adaptation by estimating the population branch statistic (PBS), an F_ST-based metric, for each African cluster using comparisons with the alternate African cluster and a Thai population. Windows with outlier PBS values (top 0.1%) were interpreted as regions showing elevated differentiation specific to the focal African cluster. We identified 489 outlier windows in the cluster Y and cluster R, corresponding to positive selection in the CDS of 203 genes (cluster Y) and 112 genes (cluster R) (Fig 5 and S3 Table). We also investigated signatures of recent or ongoing positive selection between the African clusters using haplotype-based approaches [43,44]. In cluster Y, XP-EHH and Rsb analyses identified 414 and 340 significant SNPs, respectively, located within the coding sequences of 5 and 7 genes (S9 Fig and S3 Table). In cluster R, XP-EHH and Rsb analyses detected 766 and 815 significant SNPs, respectively, across 28 and 35 genes (S9 Fig and S3 Table). For both clusters, all genes identified as significant by XP-EHH were also detected by Rsb, whereas several genes showed evidence of selection exclusively with Rsb (S9 Fig and S3 Table). Notably, in cluster R, three genes exhibited consistent signatures of selection across all three tests (S3 Table).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 5. Evidence of selection signals in African P. malariae.

(A & B) Manhattan plots of population branch statistic (PBS) values, an F_ST-like statistic that measures allele frequency differentiation relative to two reference populations. For each African P. malariae cluster, we compared it with the other African cluster and Thai samples as outgroups. Red points indicate the top 0.1% of PBS values, suggesting higher allelic frequency differentiation compared to the reference populations, and thus indicating significant evidence of positive selection that affects these regions. A corresponds to Cluster Y and B to Cluster R.

https://doi.org/10.1371/journal.pntd.0014176.g005

More specifically, 21 of the identified genes are known or predicted to be involved in interactions between the parasite and its hosts (in bold in S2 Table). Genes putatively associated with interactions with primate hosts, including AP2-L [53], ROM1 [54], SPELD [55], and STP1 [56] showed signals of selection in cluster Y, whereas 6-cysteine proteins (P36 and P52 [57–59]) appeared to be under selection in cluster R. In addition, several genes displaying signatures of selection are implicated in interactions with the mosquito vector, including AP2-O [60] and CAP93 [61] in cluster Y and TREP [62] in cluster R. Two genes showing signatures of selection in cluster Y, GAP50 and ADF2, are predicted to be involved in interactions with both primate and mosquito hosts [63,64]. Two annotated genes implicated in interactions in both hosts, MAEBL [65,66] and AMA1 [66–68], appeared to be under selection, in cluster R and cluster Y, respectively. Notably, AMA1 is among the most extensively studied Plasmodium antigens and plays a key role in erythrocyte invasion while being a major target of host immune responses [66–68]. The detection of selection signals at this locus may therefore indicate that immune-mediated selective pressures contribute to differentiation between the two cryptotypes. Notably, members of the PHIST gene family exhibited signals of selection in both clusters. This family has been implicated in a range of biological processes, including cytoadherence, gametocytogenesis, host cell modification, and the production of extracellular vesicles [69,70]. However, none of the PHIST genes identified here have been functionally characterized in P. malariae (S3 Table). Furthermore, signals of selection at the DHFR-TS gene, which is associated with resistance to antimalarial treatment in Plasmodium [71,72], were detected exclusively in cluster R (Fig 5B). In total, we identified 352 genes exhibiting signatures of positive selection in African P. malariae, including multiple candidates involved in interactions with both vertebrate hosts and mosquito vectors.

Circulation of P. brasilianum in NHPs

A total of 226 American NHPs from Colombia (n = 14) and French Guiana (n = 212) were screened for Plasmodium infection. Cytochrome-b PCR screening identified 20 P. brasilianum infected individuals (8.9%), including four from Colombia and sixteen from French Guiana. Infections were detected across six host species: A. macconnelli (10/140), S. midas (3/58), A. hybridus (1/5), A. seniculus (2/5), A. griseimembra (1/1), and C. versicolor (1/3). No infections were detected in the remaining four screened species (A. paniscus, S. apella, S. sciureus and P. pithecia) but this may be due to the small sample size for these species (from n = 1 to n = 7). The detection of infections across multiple primate genera, including spider and howler monkeys, as well as smaller-bodied taxa, is consistent with previous reports indicating that P. brasilianum circulates among phylogenetically diverse American primates [7–9,11]. Although sample sizes varied among species, the presence of infection in several host lineages further supports the existence of a multi-host sylvatic transmission cycle within forest ecosystems.

Despite the successful detection of infections, the recovery of adequate genomic data has proven challenging. After sWGA and stringent SNP filtering, only two P. brasilianum genomes were retained for population genomic analyses (S1 Fig and S1 Table). This limitation likely reflects the typically low parasitemia observed in natural infections and high host DNA contamination, underscoring the technical challenges inherent to studying malaria parasites in wildlife hosts. Although the limited number of high-quality P. brasilianum genomes prevented detailed population genomic analyses, the NHP screening component remains important because it documents ongoing sylvatic circulation and provides evolutionary context for interpreting the diversity observed within the P. malariae–P. brasilianum complex. Nevertheless, the positioning of these two P. brasilianum genomes within the broader P. malariae diversity confirms their close genetic relationship with American P. malariae isolates. Moreover, the affinity of P. brasilianum with African cluster R suggests possible historical connectivity between African and American parasite populations, potentially linked to past human-mediated dispersal events. Nevertheless, the limited number of high-quality P. brasilianum genomes currently available constrains inference. Future studies should prioritize the generation of well-covered genomes from diverse American monkeys and geographic regions to clarify the origin, diversification, and timing of introduction of this lineage in Latin America.

Hidden structure in Africa: discovery of two recombinant cryptotypes of P. malariae

At a global scale, we identified two well-defined genetic groups corresponding to Asia and South America (Fig 2A and 2B), consistent with the literature [20]. In contrast, our study revealed a previously unnoticed population structure within Africa. While Ibrahim et al. [20] noted a potential substructure among African isolates, this pattern was not formally supported. The discrepancy between these two studies likely reflects methodological differences. In particular, the removal of closely related samples in our dataset reduced the risk of inflated structure signals and biased estimators of genetic variation, which can arise from uneven sampling of related infections [29], a step that was incorporated in the present study. In addition, no technical factors were found to explain the observed clustering. Each cluster contained samples originating from different studies, including both low- and high-coverage datasets, comparable levels of within-host infection complexity, and overlapping sampling periods (S2 Table).

To our knowledge, this study provides the first evidence for the presence of two distinct genetic clusters within African P. malariae, designated as clusters R and Y (Figs 2–4). These two clusters appear to coexist in sympatry and recombine across multiple African regions, with substantial geographic overlap (Fig 4B). This pattern is reminiscent of the “cryptic” population structure recently described in P. falciparum, where specific genetic backgrounds can be detected across Africa despite extensive recombination and broad geographic overlap. For instance, Miotto et al. [22] identified a low-frequency, continent-wide P. falciparum cluster (AF1) characterized by a shared multi-locus genetic background, termed a cryptotype, that persists despite outbreeding with local parasites. More broadly, African P. falciparum also shows macro-scale population structure (West/Central/East) revealed by genome-wide clustering [14,73]. The identification of cryptotypes in both P. falciparum and P. malariae raises the possibility that hidden population structure may be more common in African malaria parasites than previously appreciated. By analogy with these observations, we refer to the two African P. malariae “cryptic” clusters identified here as “cryptotypes”. This term is defined here as previously unrecognized genetic lineages that can be identified through genome-wide population structure analyses but that still recombine. Unlike the sympatric species P. ovalecurtisi and P. ovalewallikeri [74], the two P. malariae cryptotypes identified here show evidence of admixture and ongoing recombination (Fig 3A and S2 Table) and therefore do not constitute distinct species or subspecies. Rather, they represent differentiated genetic backgrounds that coexist across Africa despite continued gene flow. Nevertheless, the possibility that these clusters represent lineages in the early stages of speciation cannot be excluded. The current dataset does not provide sufficient resolution to accurately estimate recombination rates between the two clusters, which might occur at relatively low frequencies.

The persistence of two differentiated genetic backgrounds despite evidence of recombination suggests that the forces generating or maintaining these cryptotypes may be sufficiently strong to counteract the homogenizing effect of gene flow. Similar patterns have recently been reported in African P. falciparum [22] and may result from balancing selection, local adaptation, ecological specialization, or assortative transmission through distinct host or vector communities. Although the present data do not allow these alternative hypotheses to be formally distinguished, they provide a useful framework for interpreting the coexistence of these two sympatric and recombining cryptotypes.

One possible (but currently untested) explanation for their distribution could be ecological heterogeneity, such as the presence of tropical forest-associated transmission systems [75,76]. Indeed, cluster R seems more prevalent in countries covered by tropical forests (e.g.,: Gabon, Cameroon, Central Africa, Benin, Togo, Ivory Coast and Guinea). However, ecological conditions are likely intertwined with other factors that may influence parasite population structure, including mosquito species composition, vector abundance, transmission intensity, pathogen community composition, and the intensity of malaria control interventions [77,78]. African malaria transmission is maintained by numerous Anopheles species whose geographic distributions and ecological preferences vary substantially across the continent [79,80]. Such variation in vector communities could impose different selective pressures on parasite populations, particularly on genes involved in mosquito invasion, development, and transmission. The identification of cluster-specific selection signals in several genes implicated in vector-parasite interactions is consistent with this hypothesis. These factors therefore represent important future investigations to better characterize the distinct ecological and epidemiological contexts associated with each cluster. However, testing these hypotheses would require more finely resolved sampling. Because most isolates are associated only with country-level geographic information, these hypotheses could not be formally evaluated. Furthermore, our sampling covers several years (S1 Table), and the tropical forest coverage highly transformed into tropical savanna changed between 2010 and 2017 [81], which further complicates direct ecological interpretations. A better understanding of how parasites shift their distribution and move across populations could greatly contribute also from a One Health perspective.

The hypothesis of vector-driven ecological heterogeneity is consistent with our observation that P. brasilianum samples are genetically closer to cluster R, whereas American P. malariae samples are more closely related to cluster Y (Figs 2A and 3). Because P. brasilianum is adapted to NHPs, its transmission might be sylvatic [12]. This pattern might therefore reflect adaptation to forest-associated environments, maybe with a different transmission dynamic, although further ecological and entomological data would be required to evaluate this hypothesis.

African cluster-specific adaptation signals comprise host-interaction genes

Genome-wide scans aimed at identifying regions showing cluster-specific signals of positive selection did not reveal genes with clear links to environmental adaptation (S7 Fig). However, our results point toward processes more closely tied to within-host biology and host–parasite interactions rather than adaptation to external abiotic conditions.

We identified multiple genes showing signatures of selection that are putatively involved in interactions with hosts, including both mosquito vectors and vertebrate hosts, with patterns that differ between the African clusters (Fig 5 and S2 Table). Malaria vector species are unevenly distributed across Africa [79,80]. In parallel, human genetic backgrounds vary across the continent [82]. Additional host factors, including the distribution of Duffy negativity and other erythrocyte polymorphisms known to influence malaria susceptibility [83], may further contribute to heterogeneous selective environments across Africa. Consequently, the observed differentiation in genes involved in host interactions may provide insights into the factors underlying the separation of the two clusters. These findings raise the possibility that each cluster is adapted to specific mosquito vectors and human host populations. More generally, future studies investigating diversity patterns in antigen genes involved in immune acquisition and immune evasion may provide additional insights into the role of host immunity in shaping the divergence and persistence of these cryptotypes. Unfortunately, the available metadata do not allow these hypotheses to be investigated further. As the dataset is derived from anonymous travelers, detailed information regarding the specific locations visited within each country, as well as the genetic and health backgrounds of the individuals, is unavailable.

Although many candidate loci remain functionally unknown in P. malariae, their putative roles have been extrapolated by orthology-based inferences from studies in other Plasmodium species infecting primates (P. falciparum, P. knowlesi) or rodents (P. berghei). Among genes without functional annotation in Plasmodium, some genes may have played a role in African P. malariae adaptation to the mosquito and/or primate host. Thus, functional follow-up, as experimental validation will be essential to move from statistical signatures to mechanistic understanding of P. malariae adaptation.

In cluster R, the DHFR-TS gene shows signs of differentiation. This signal has been previously reported and functionally validated in P. malariae [20]. Indeed, as in P. falciparum, this gene is involved in resistance to antimalarial treatment [20,72,84]. Importantly, this represents the only signal associated with drug resistance detected in our study. This finding raises the hypothesis that the differentiation between the two African clusters may partly reflect adaptation to selective pressures imposed by antimalarial treatments. Indeed, treatment use varied over time and among African countries [85,86]. However, regional heterogeneity subsists within countries [86,87], which may explain the presence of both clusters, at varying frequencies, across the same country. Certain regions may provide more favourable conditions for one cluster over the other, depending on the malaria control strategies implemented. Future investigations of these clusters would need more precise geographic data together with detailed information on local malaria control strategies to adequately test this hypothesis. Beyond their evolutionary significance, the identification of two sympatric cryptotypes may also have implications for malaria surveillance and elimination efforts. If these cryptotypes differ in transmission ecology, vector associations, infection persistence, susceptibility to antimalarial interventions, or interactions with other Plasmodium species, they may respond differently to control measures. Although the present study does not allow these possibilities to be formally assessed, our results highlight the importance of considering within-species diversity when designing and evaluating malaria elimination strategies.

Several signatures of selection identified in Africa were shared between this study and Ibrahim et al. [20], including the genes DHFR-TS, DPAP3, EIF3L and a hypothetical merozoite protein (PmUG01_10046700) (see S3 Table). However, numerous selection signals were detected exclusively in one study or the other. Differences between the two studies may reflect both analytical and conceptual choices. Ibrahim et al. [20] employed only approaches based on identity-by-descent and haplotype length, whereas our analyses relied on both the PBS, which is derived from allele frequency differentiation (F_ST), and haplotype length. Haplotype-based methods are generally more sensitive to recent selection but may fail to detect older selective events due to the breakdown of linkage disequilibrium through recombination. Our approaches may capture recent and older selection signals. In addition, Ibrahim et al. [20] treated African P. malariae as a single population, whereas our analyses explicitly accounted for the presence of two distinct African genetic clusters and focused on cluster-specific signals of adaptation.

In conclusion, although P. malariae remains a neglected malaria parasite, it poses a tangible public health risk and represents an additional challenge to malaria control and elimination efforts. By assembling and analyzing the largest and most comprehensive whole-genome dataset for this species to date (to our knowledge), we provide new insights into the global genetic diversity of P. malariae. The identification of two recombinant and sympatric African genetic clusters adds to the current views of P. malariae transmission and diversity across the continent. This complexity is particularly relevant given that each cluster exhibits distinct signatures of adaptation to human hosts and mosquito vectors, even though the ecological or evolutionary drivers of their divergence still remain unclear. Our findings underscore the need for larger, geographically finer-scale sampling to better resolve the evolutionary dynamics of P. malariae. Indeed, hidden population substructure can hinder malaria control strategies by obscuring transmission pathways, masking local adaptation, and potentially influencing responses to interventions such as drug treatment or vector control. If the two cryptotypes identified in the present study differ in biologically relevant traits, including transmission ecology, infection persistence, or susceptibility to interventions, their coexistence could further complicate malaria elimination efforts. Thus, by revealing previously unrecognized population structure in Africa, this study highlights the importance of integrating population genomic data into malaria surveillance, control, and elimination programs to ensure that control and elimination strategies account for the full diversity of malaria parasites circulating in endemic regions.