Genome-wide diversity and differentiation in New World populations of the human malaria parasite Plasmodium vivax

Background The Americas were the last continent colonized by humans carrying malaria parasites. Plasmodium falciparum from the New World shows very little genetic diversity and greater linkage disequilibrium, compared with its African counterparts, and is clearly subdivided into local, highly divergent populations. However, limited available data have revealed extensive genetic diversity in American populations of another major human malaria parasite, P. vivax. Methods We used an improved sample preparation strategy and next-generation sequencing to characterize 9 high-quality P. vivax genome sequences from northwestern Brazil. These new data were compared with publicly available sequences from recently sampled clinical P. vivax isolates from Brazil (BRA, total n = 11 sequences), Peru (PER, n = 23), Colombia (COL, n = 31), and Mexico (MEX, n = 19). Principal findings/Conclusions We found that New World populations of P. vivax are as diverse (nucleotide diversity π between 5.2 × 10−4 and 6.2 × 10−4) as P. vivax populations from Southeast Asia, where malaria transmission is substantially more intense. They display several non-synonymous nucleotide substitutions (some of them previously undescribed) in genes known or suspected to be involved in antimalarial drug resistance, such as dhfr, dhps, mdr1, mrp1, and mrp-2, but not in the chloroquine resistance transporter ortholog (crt-o) gene. Moreover, P. vivax in the Americas is much less geographically substructured than local P. falciparum populations, with relatively little between-population genome-wide differentiation (pairwise FST values ranging between 0.025 and 0.092). Finally, P. vivax populations show a rapid decline in linkage disequilibrium with increasing distance between pairs of polymorphic sites, consistent with very frequent outcrossing. We hypothesize that the high diversity of present-day P. vivax lineages in the Americas originated from successive migratory waves and subsequent admixture between parasite lineages from geographically diverse sites. Further genome-wide analyses are required to test the demographic scenario suggested by our data.

Introduction Plasmodium vivax is the human malaria parasite with the widest global distribution and accounts for nearly half of the combined malaria burden in South and Southeast Asia, Oceania, and Central and South America. Over one-third of the world's population is currently at risk of infection with this species, with 16 million clinical cases recorded each year [1]. Although P. vivax has most likely evolved from parasites that infect chimpanzees and gorillas in sub-Saharan Africa [2,3], it is nowadays rare in most of this continent, where human populations lack a key erythrocyte receptor for host cell invasion by blood-stage parasites, the Duffy antigen/ receptor for chemokines (DARC) [4]. Where both species coexist, P. vivax typically causes less severe cases and fewer deaths than P. falciparum, the most virulent human malaria parasite, but represents a major challenge for ongoing malaria elimination efforts worldwide [1].
The Americas were the last continent colonized by humans carrying malaria parasites, but the dates and routes of migration of P. vivax to the New World are still debated [5][6][7]. Archaeological evidence for infection with this parasite in indigenous, pre-Columbian populations is currently limited to a single report of P. vivax antigens being visualized by immunohistochemistry in the liver and spleen of South American mummies dating from 3,000 to 600 years ago [8]. Interestingly, specific antibodies failed to detect P. falciparum antigens in these same samples [8]. These findings are consistent with the hypothesis that P. vivax, but not P. falciparum, was brought to the New World by early human migrations from East Asia or the Western Pacific [5], but more specific molecular techniques are required to confirm them [9]. Nevertheless, present-day New World populations of P. vivax appear to be more closely related to extant African and South Asian parasites and now extinct European lineages than to East Asian and Melanesian strains [10][11][12], consistent with much more recent parasite migrations with European conquerors and African slaves during the colonial era [7,13].
Clinical isolates from Brazil are underrepresented in global genomic analyses of P. vivax [11,14]. However, this country contributes 37% of the malaria burden in the Americas, a region with over 20 million people at high risk of infection [1]. Obtaining large amounts of host-free P. vivax DNA from clinical samples from Brazil has been a major challenge for genome sequencing projects, because (a) blood-stage parasite densities are typically very low [15], (b) clinical blood samples are heavily contaminated with human DNA from leukocytes, and (c) methods for long-term in vitro propagation of P. vivax are neither practical nor widely reproducible [16][17][18]. Here, we combined an improved sample preparation strategy, for reducing human DNA contamination and increasing target parasite's DNA yield, with next-generation genome re-sequencing to examine a local population of P. vivax from the Amazon Basin of Brazil. Our nine high-quality genome sequences were compared to those previously obtained from four countries ( [11,14] to reveal local and regional patterns of diversity and differentiation in extant P. vivax populations from the New World.

Sample collection
Parasite samples were collected between November 2012 and June 2013 in eastern Acre and southern Amazonas, Amazon Basin of Brazil, close to the borders with Peru and Bolivia. Malaria epidemiology in the study sites has been characterized in detail elsewhere [15,19]. Venous blood samples (10 ml) were collected from eight adult patients attending malaria clinics in the town of Acrelândia, Acre (9˚43' S, 66˚53' W), and one adult patient living in the farming settlement of Remansinho, Amazonas (9˚40' S-9˚43' S, 66˚52' W -66˚59' W), 120 km east of Acrelândia (Fig 1). P. vivax infection was diagnosed by on-site microscopy and later confirmed by qPCR as described below.

Leukocyte depletion and sample cryopreservation
We adapted BioR 01 Plus leukocyte-depletion filters (Fresenius Kabi, Bad Homburg, Germany; S1A Fig) to process 10 to 50-ml volumes of venous blood in our field laboratory in the Amazon [20]. The per-unit cost of these filters in Brazil is around US$ 25. We first cut off under sterile conditions, with a scissor, the tubing that connects the filtering device to the 400-ml blood storage bag and to the adapter (S1B and S1C Fig). S1D Fig shows how the filtering device was used in a laminar flow hood; briefly, a 10-ml syringe was used to apply blood treated with acid citrate dextrose anticoagulant, while a second 10-ml syringe was adapted to the end of the remaining tubing to recover the filtered, leucocyte-depleted material, which was transferred to 50-ml sterile centrifuge tubes. No priming with saline was required. After the filtration process, the leukocyte depletion device was washed through with at least twice the volume of RPMI medium as the original blood sample to recover red blood cells (RBCs) that had been retained in the filter and tubing. Next, the mixture of filtered blood and RPMI medium recovered in 50-ml tubes was centrifuged at 800 g for 5 minutes and the supernatant (plasma plus RPMI medium) was removed with a sterile Pasteur pipette.
For cryopreservation, the RBC pellet was resuspended dropwise in Glycerolyte-57 solution (Fenwall, Fresenius Kabi), at the proportion of 1.66 ml of Glycerolyte for each 1 ml of cell pellet, under gentle agitation. One ml aliquots of the RBC-Glycerolyte mixture were transferred to screw-capped cryovials and placed in a Nalgene Mr. Frosty freezing container https://doi.org/10.1371/journal.pntd.0005824.g001 (ThermoFisher Scientific, Waltham, MA) that was kept at -80˚C for 24 hours. The next day, cryovials were plunged in liquid nitrogen for long-term storage. Samples were shipped to São Paulo in liquid nitrogen, for subsequent schizont maturation.
Short-term in vitro culture for P. vivax schizont maturation and chloroquine resistance testing For sample thawing, cryovials were removed from the liquid nitrogen and maintained for 1 min at room temperature (20-25˚C) and 1 min at 37˚C. The 1-ml samples were then transferred to 50-ml centrifuge tubes, gently mixed with 200 μl of 12% NaCl solution, and let stand for 1 minute. Next, 10 ml of 1.6% NaCl were gently added, the mixture was gently agitated, and centrifuged at 180 g for 8 min [21]. After removing the supernatant, the RBC pellet was washed twice in incomplete McCoy's 5A medium supplemented with glucose (0.5% w/v), HEPES (25 mM), and hypoxantine (0.005% w/v), and resuspended in complete McCoy medium (as above but supplemented with 25% AB+ heat-inactivated human serum) to a final hematocrit of 2%. Short-term culture in vitro was carried in flat-bottomed dishes placed in a gas chamber with controlled O 2 and CO 2 levels that was kept at 37˚C for up to 48 hours. Parasite growth and maturation were monitored as described [17].
Chloroquine resistance (CQR) was evaluated using an ex-vivo schizont maturation assay in selected P. vivax samples with > 1,000 parasites/μl of blood and > 50% ring stages at the time of thawing [22]. There is no consensus regarding the 50% inhibitory concentration (IC 50 ) indicative of CQR in P. vivax; suggested cut off values range between 100 nM [23] and 220 nM [24], but all IC 50 values in our samples were < 50 nM (Table 1).
DNA isolation and quantification of human and P. vivax DNA with realtime PCR DNA templates were isolated from 200-μl aliquots of either whole venous blood (before leukocyte removal) or RBC pellet (after leukocyte removal) using QIAamp DNA blood kits (Qiagen, Hilden, Germany). To estimate the relative proportion of human and parasite DNA, we used SYBR Green qPCR targeting single-copy genes coding for human topoisomerase III and P. vivax aldolase. Each 15 μl qPCR mixture contained 2 μl of template DNA, 7.5 μl of 2× Maxima SYBR Green qPCR master mixture (Fermentas, Burlington, Canada) and 0.3 μM of each of the primer pairs, Pvaldo-F (GAC AGT GCC ACC ATC CTT ACC) plus Pvaldo-R (CCT TCT CAA CAT TCT CCT TCT TTC C) and Top3-F (CAT GTT TGA GCT GAG CCT GA) plus Top3-R (CCA CAC CAC ACC CCT AAC TT). Standard curves were prepared with serial tenfold dilutions of a plasmid containing both target sequences to allow for copy number quantitation (number of amplicons/μl of blood). We used a Step One Plus Real-Time PCR System (Applied Biosystems, Foster City, CA) for PCR amplification with a template denaturation step at 95˚C for 10 min, followed by 40 cycles of 15 sec at 95˚C and 1 minute at 60˚C, with fluorescence acquisition at the end of each extension step. Amplification was followed by a melting program consisting of 15 sec at 95˚C, 15 sec at 60˚C, and a stepwise temperature increase of 0.2˚C/sec until 95˚C, with fluorescence acquisition at each temperature transition. All reactions were made in triplicate. We measured parasite DNA enrichment as the parasite: human copy-number ratio after filtering divided by parasite:human copy-number ratio before filtering. To estimate the proportion of host and parasite DNA in each sample, we considered the amplicon copy numbers and the genome size of humans (3.2 Gb) and P. vivax (26.8 Mb; Sal-I assembly); the DNA content in each copy of the human genome corresponds approximately to that of 119 copies of the P. vivax genome.

DNA library preparation and whole-genome sequencing
Parasite DNA templates were quantified by fluorometry using a Qubit 3.0 fluorometer (Invitrogen, Carlsbad, CA) and sequenced using Ion Torrent Personal Genome Machine (PGM) and Ion Proton platforms (Life Technologies, Foster City, CA) at the Unit of Computational Genomics, Laboratory of Bioinformatics, National Laboratory of Scientific Computation, Brazil. Separate libraries were prepared for each sequencing platform, using 1 μg of template DNA per isolate. DNA samples were sheared using the Bioruptor UCD-200 TS (Diagenode, Liege, Belgium) sonication system until fragment sizes of 200 bp (for Ion PGM libraries) or 150 bp (for Ion Proton libraries) were obtained. Libraries were prepared using the Ion Xpress Plus Fragment Library kit, with Ion Xpress Barcode adapters according to the Ion Xpress Plus gDNA Fragment Library Preparation protocol (Life Technologies). Size selection was performed on E-Gel SizeSelect 2% agarose gels using the E-Gel iBase Power System (Thermo-Fisher Scientific). Emulsion PCR was done on the Ion OneTouch 2 system (Life Technologies) with the Ion PGM Template OT2 200 kit or the Ion PI Template OT2 200 kit version 2 for Ion PGM and Ion Proton, respectively, following the manufacturer's instructions (Life Technologies). Ion PGM libraries were loaded on Ion318 chips v2 and sequenced using the Ion PGM Sequencing 200 kit v2; Ion Proton libraries were loaded on Ion PI chips v2 and sequenced using the Ion PI Sequencing 200 kit v2 (Life Technologies). All samples were sequenced on both platforms; sequence reads (150-200 bp) from two runs in each platform were merged into a single fastq file per sample.

Additional genomic data sets
To place our genomic data in a regional context, we reanalyzed raw paired-end Illumina reads from 107 additional P. vivax clinical isolates from the Americas. Fastq files were downloaded from the Sequence Read Archive (SRA) of the National Center for Biotechnology Information, United States, and processed in the same way as our newly obtained sequences. Three clinical isolates from Brazil had sequence data generated on an Illumina Genome Analyzer II platform at the Welcome Trust Sanger Institute (Hinxton, Cambridge, UK), as part of the P. vivax genome variation project coordinated by the MalariaGEN network [14]. Other isolates-20 from Brazil, 34 from Peru, 31 from Colombia, and 19 from Mexico-had whole-genome sequence data generated on an Illumina HiSeq 2000 platform at the Broad Institute of MIT and Harvard (Cambridge, MA, USA), as part of the International Centers of Excellence for Malaria Research (ICEMR) program [11]. All isolates from Brazil sequenced in these two previous studies were collected in endemic areas surrounding Acrelândia, Acre State, between 2008 and 2011.

Single-nucleotide polymorphism (SNP) calling
Fastq files were first filtered for quality; 8 SRA samples were excluded from further analysis because of mean quality scores 30 (expected base call accuracy 99.9%). We next mapped the high quality reads onto the PlasmoDB version 10.0 of the Sal-1 reference (http://plasmodb. org/common/downloads/release-10.0/PvivaxSal1/fasta/data/PlasmoDB-10.0_PvivaxSal1_ Genome.fasta) [25], using Bowtie2 version 2.2.6 [26] with the "very sensitive" preset, allowing one mismatch per seed region; 11 SRA samples were excluded at this stage because of < 60% mapping over the reference. The resulting alignments were merged into BAM files with SAMtools [27], duplicate reads were identified and marked using the Picard version 2.0.1 MarkDuplicates tool, and files were indexed with SAMtools.
SnpEff [29] was used to identify SNPs mapping to coding sequences (further classified as synonymous or nonsynonymous), introns, and intergenic regions of the Sal-I reference genome. The resulting catalogue of 94,122 high-confidence SNPs was used to genotype each individual sample using GATK UnifiedGenotyper with default parameters except for the minimum phred-scaled confidence threshold, with a calling variant = 50 and emitting variant = 10. Heterozygote calls were converted to the majority allele if !75% of the reads in that sample were the majority allele; otherwise, the allele was undetermined. Sites with < 5× coverage in a given sample were filtered out at this stage.
The final data set of P. vivax nuclear genome sequences from the New World comprised our 9 newly sequenced samples from BRA and 75 high-quality SRA samples (2 from BRA, 23 from PER, 31 from COL, and 19 from MEX); 13 SRA samples were removed during the genotyping process because the number of SNPs identified (range: 13 to 2,889) was below the predefined minimum of 3,000 nucleotide differences compared to the Sal-I reference, consistent with poor sequencing coverage. Overall, we excluded 21 (out of 23) SRA samples from BRA and 11 (out of 34) SRA samples from PER at different stages of this analysis. Isolate codes and SRA accession numbers of samples used in this analysis are given in S1 Table. Within-population genetic diversity and recombination The average pairwise nucleotide diversity (π, average number of nucleotide differences per site between pairs of DNA sequences) was calculated within each geographic population using VCFtools [30]. Values were plotted, using R version 3.3.0, as moving averages within 1-kb sliding windows across each chromosome. We recalculated π after masking out subtelomeric regions and three hypervariable internal chromosome regions (containing sera, msp-3, and msp-7 gene families) that were more prone to sequence misalignments and poor read mapping in a previous analysis [14]. The coordinates of these regions are given in the Supplementary Table 2 of Pearson et al. [14].
Similarly, Tajima's D values [31] were calculated using VCFtools; mean Tajima's D across 1-kb windows were plotted for each population. Frequency distributions of π and Tajima's D values within 1-kb windows were plotted for each population. We defined windows with the top 50 π values within each population as highly variable genomic regions. We defined as outliers the 1-kb windows with the 50 highest and 50 lowest mean Tajima's D values within each population. We also examined the minor allele frequency (MAF) spectrum separately in each population.
We next estimated, for each population, the rate at which pairwise linkage disequilibrium (LD) decreased with increasing physical distance between SNPs due to meiotic recombination. The squared correlation coefficient r 2 between pairs of SNPs of varying distance across the same chromosome was measured using VCFtools; r 2 values were binned by distance (50-bp windows) and medians within each window were plotted against physical distance between SNPs. The level of background LD between unlinked markers within each population was estimated by calculating median r 2 between all pairs of SNPs on different chromosomes.

Between-population genetic differentiation and population structure
The Wright's fixation index F ST , a measure of population differentiation due to genetic structure [32], was calculated with VCFtools for each SNP in every pairwise comparison of populations. Values were averaged across all SNPs to estimate overall pairwise differentiation between populations. The 100 SNPs with the highest average F ST values across all populations were further characterized.
To assess population structure, we first used the PLINK software (https://www.coggenomics.org/plink2 [33]) to carry out principal component analysis (PCA); up to 10 components were analyzed. For phylogenetic analysis, a neighbor-joining tree was constructed via the maximum composite likelihood substitution model with 1,000 bootstrap pseudoreplicates using MEGA 7.0 (http://www.megasoftware.net/). To estimate the ancestry shared between individual isolates, we used the ADMIXTURE software package [34] with either all 94,122 high-quality SNPs or a curtailed set of 12,762 SNPs that are not linked. To this end, we removed each SNP that had an r 2 value > 0.1 with any other SNP within a 60-SNP sliding window advanced by 10 SNPs each time. The optimal number of clusters (K) was determined by performing multiple runs of the software under different K values (2)(3)(4)(5)(6)(7)(8)(9)(10)

Data availability
The sequence data supporting the conclusions of this article are available in the Sequence Read Archive of the National Center for Biotechnology Information, United States; accession numbers are provided in S1 Table. Results and discussion

Sample preparation and genome sequencing
We designed a single-step procedure to reduce human DNA content in P. vivax-infected blood using commercially available leukocyte-depletion filters (S1 Fig). Leucocyte depletion in 17 clinical samples, with initial parasitemias ranging between 854 and 43,177 (median: 7,566) parasites/μl, decreased the percent human DNA content from a median of 99.2% (range: 82.7-99.9%) to 23.3% (range: 0-97.6%). No human DNA could be detected by quantitative realtime polymerase chain reaction (qPCR) in three filtered samples (i.e., 0% host DNA contamination). The residual human DNA content in our leukocyte-depleted samples was similar to the median of 33.9% (range: 1.6-68.6%) found in blood samples from Indonesia after double filtration through CF-11 cellulose columns [17]. The increase in parasite:human DNA ratio after leukocyte depletion ranged between 1.7 and 3,060-fold (median: 227-fold) and appeared to be inversely proportional to the initial, pre-treatment parasite:human DNA ratio (Spearman correlation coefficient r s = -0.512, P = 0.061; S3 Fig), but did not correlate with the initial parasitemia (r s = 0.165, P = 0.412).
The next challenge consisted in selectively increasing parasite DNA yield for genome sequencing. To this end, leukocyte-depleted P. vivax samples were cultured in vitro for up to 44 h to allow uninucleate trophozoites to mature to multinucleate blood-stage schizonts, increasing parasite DNA content by !4-fold [17]. We obtained enough template DNA for library preparation and sequencing from 9 cultured samples, achieving 49.8-78.6× average genome sequencing depth; 91.8-94.7% of the reads mapped to reference Sal-I genome ( Table 1) [39]). Therefore, the sample preparation strategy described here allowed for high sequence coverage and depth, but further comparisons are limited by the use of different nextgeneration sequencing platforms across studies (Ion PGM and Ion Proton here and Illumina in other studies).

Local and regional patterns of genome-wide diversity
We used the Ion PGM and Ion Proton platforms to generate between 17.8 and 23.9 million sequence reads from 9 clinical samples of P. vivax from Brazil (Table 1). To explore local levels of genomic diversity, our sequence data were combined with those from two P. vivax clinical samples from Brazil that were previously obtained with Illumina platforms [11,14]. Sequence reads from these additional samples (PV4 and Brazil32) covered 86.2% and 95.0% of the reference genome, respectively (S1 Table). Since all isolates from Brazil (n = 11) were collected from sites within a radius of 120 km in the Amazon Basin (Fig 1), we define BRA as a local, nearly sympatric P. vivax population. After applying stringent quality control filters to raw sequence reads (see Methods), we uncovered 27,360 high-confidence single-nucleotide polymorphisms (SNPs) in the BRA population. The overall nucleotide diversity π was estimated at 5.6 x 10 −4 ).
We next compared BRA sequence data with those from three other New World populations of P. vivax: PER (n = 23), COL (n = 31), and MEX (n = 19) [11], from clinical isolates sampled in sites shown in Fig 1. Two populations were geographically heterogeneous: PER samples were collected in three departments (Loreto and Madre de Dios, both in the Amazon Basin, and Piura, on the northwestern Pacific coast), while COL samples came from four departments (Nariño, Valle del Cauca, and Chocó along the Pacific Coast, and Córdoba, on the Caribbean coast). MEX samples were from five different sites, but all in the southern state of Chiapas (S1 Table). We characterized 94,122 biallelic SNPs passing our high-quality filters in 84 samples; 55.0% of them are located in intergenic regions, 8.8% in introns, and 36.2% in coding regions. Most (61.2%) coding SNPs were non-synonymous (nsSNPs), as previously found in other regional P. vivax populations [14,38]. Unsurprisingly, the number of SNPs found in each population was directly proportional to sample size, being lowest in BRA (n = 27,360) and highest in COL (n = 57,262; Fig 2). Overall, 49,598 (52.3%) SNPs were unique to a population (i.e., private) and 8,529 (9.1%) were shared by all populations (Fig 2). Of the 6,891 private SNPs found in BRA, 40.0% mapped to coding sequences; 65.2% of the coding SNPs were nonsynonymous. PER and COL shared the highest number of SNPs (n = 28,667), followed by COL and MEX (n = 24,107) (Fig 2).
We defined domains with the top 50 π values in a population as highly variable genomic regions. They comprised, in addition to numerous sequences coding for hypothetical proteins, gene families such as pst-a, fam-b, fam-d, and fam-e [40] and those coding for major parasite antigens, such as the vir family (>300 genes, mostly in subtelomeric domains, on several chromosomes), the serine repeat antigen (sera) family (13 genes on chromosome 4), the merozoite surface protein (msp)-7 family (11 genes on chromosome 12), and the msp-3 family (11 genes on chromosome 10) (S2 Table) (see also [11,14,36]).
These findings are not unexpected, since natural selection favors increased diversity in antigen-coding genes to evade host immunity, but must be interpreted with caution because misalignments of paralogous sequences may have inflated nucleotide diversity estimates in gene families. We thus recalculated genome-wide π values after masking out subtelomeric domains and the internal chromosomal regions comprising the sera, msp-3 and msp-7 gene families [14], but this procedure affected our overall estimates very little; recalculated values were: BRA (π = 5.5 × 10 −4 ), PER (π = 5.2 × 10 −4 ), COL (π = 4.7 × 10 −4 ), and MEX (π = 5.2 × 10 −4 ). The single-copy msp-1 gene [41] also mapped to a highly variable genomic region in BRA. However, our nucleotide diversity estimates for msp-1 may have been affected by likely sequence misalignments in the numerous repetitive domains across this locus [42,43].

Signatures of selection and population expansion
The expected value for Tajima's D is zero under a neutral model that assumes random mating, no recombination, mutation-drift equilibrium, infinite sites, and constant population size. High Tajima's D values are usually due to balancing selection or recent population size reduction, while negative D values are consistent with population size expansion or purifying selection [44]. Although the P. vivax genomes from COL and PER were not part of a single population (while BRA and MEX genomes are), the Tajima Table). Only two regions with low Tajima's D values were shared by two populations; one had no annotated gene and the other had a gene coding for a hypothetical protein (S3 Table).
Similarly, the genomic regions with the 50 highest Tajima's D values within each population (cut-off values: 2.024 in BRA, 1.890 in PER, 2.078 in COL, and 2.407 in MEX) also comprised several hypothetical and housekeeping genes and a few surface antigen genes (sera, circumsporozoite protein [csp], and msp-3) that may be under balancing selection (S4 Table). However, different genomic regions giving top 50 Tajima's D values were typically found in each population and only 11 of them (none comprising antigen-coding genes) were shared by two or more populations. Whether the high Tajima's D values found in certain domains are due to random effects of the parasite's demographic history or to balancing selection on specific genes remains to be further examined in larger population samples.
MAF distributions were L-shaped in all New World P. vivax populations (S5 Fig), similar to patterns described for P. falciparum populations from sub-Saharan Africa [45,46]. The proportion of SNPs with allele frequencies 0.1 were 51.6% in BRA, 66.3% in PER, 71.5% in COL, and 55.1% in MEX. Despite the relatively small sample sizes, we interpret the clear predominance of negative Tajima's D values in PER and COL and of rare alleles in all populations, but mainly in PER and COL, as suggestive of a recent P. vivax population expansion in the Americas. Data from an extensive mitochondrial genome analysis of local parasites are also consistent with the demographic expansion hypothesis [47].
Genetic differentiation and population structure P. vivax genomic sequences from the New World were previously shown to cluster mostly according to their geographic origins, with the majority of MEX samples (that show more extensive evidence of identity by descent) clustering together in PCA plots [11]. These findings are particularly relevant to malaria-eliminating countries in the continent; if parasites can be assigned by molecular genotyping to their countries of origin, locally acquired and imported infections can theoretically be easily distinguished. Moreover, genetic analysis can also help to determine the likely origin of imported infections [10,48]. Accordingly, our PCA data revealed a clear clustering of BRA samples, which were separated from other populations by principal components C1 and C3 when the complete SNP data set was used. MEX was separated from other populations mostly by C1 and C2, while PER and COL were less clearly differentiated (Fig 4A). Plasmodium vivax genomes from the New World Moreover, a rather similar sample clustering pattern was revealed by a standard, neighborjoining phylogenetic analysis (Fig 5). All BRA samples and most (16 of 19) MEX samples formed well-supported clades, while PER and COL isolates spread across single-and multiplecountry clades.
Similar to PCA and phylogenetic analysis, the model-based clustering approach implemented in ADMIXTURE software [33] also defined BRA and MEX as distinct populations (S6A Fig, number of clusters K = 3) based on the whole genomic data set. PER and COL were nearly indistinguishable. A similar analysis with 2 clusters did not differentiate MEX from COL (S6A Fig, number of clusters K = 2). ADMIXTURE analysis using a curtailed data set (12,762 unlinked SNPs) allowed a poor differentiation between populations (S6B Fig, number of clusters K = 2 or 3). Not surprisingly, the most geographically homogeneous samples (BRA and MEX) formed well-defined clusters.
We next tested whether a recently described P. vivax SNP barcode [49] would be able to correctly assign New World isolates to their countries of origin. Of the 42 SNPs originally included in the barcode, five did not segregate in our populations. PCA with the remaining 37 SNPs was unable to cluster our parasite populations by their countries of origin (Fig 4B).
The overall pairwise differentiation between populations, estimated using F ST averaged across the entire genome, was directly proportional to the geographic distance between sites, being lowest between COL and PER and highest between BRA and MEX ( Table 2). Among the 100 SNPs with the highest average pairwise F ST estimates, only two mapped to antigencoding genes (both to sera gene paralogs on chromosome 4); all others were either noncoding SNPs or mapped to genes encoding hypothetical or housekeeping proteins (S5 Table). We next tested whether these 100 SNPs with the top F ST values could separate New World populations more efficiently than the complete genome-wide SNP set (compare Fig 4A and 4C) and the currently available 42-SNP P. vivax barcode [49] (compare Fig 4B and 4C). These results suggest that regional SNP barcodes could be further explored to track the geographic origin of P. vivax samples in the Americas.

Nucleotide substitutions and antimalarial drug resistance
Our genomic sequence data enabled the identification of SNPs mapping to several P. vivax genes known or suspected to be involved in antimalarial drug resistance (Table 3). We found nonsynonymous mutations at the dihydrofolate reductase-thymidylate synthase (dhfr) locus previously associated with pyrimethamine resistance (S58R and S117N) almost exclusively in COL (see also [35]) and PER (see also [37]), whereas antifolates are no longer recommended in these countries; two previously undescribed SNPs (H99N and H99R) were also found in COL. The dihydropteroate synthase (dhps) SNP A383G, associated with sulfadoxine resistance, was also observed in COL and PER (see also [35]), with the M205I change being found only in PER (see also [36]). Little or no polymorphism in dhfr and dhps genes was observed in BRA and MEX, likely reflecting differences in sulfadoxine-pyrimethamine use across countries in the region. Although pyrimethamine and sulfadoxine has never been recommended as a treatment for P. vivax infection, they may have been used to treat patients co-infected with P. falciparum and P. vivax, exposing the latter parasite to these drugs and likely selecting for resistant phenotypes. Moreover, the frequent use of sulfonamides to treat co-occurring, unrelated (mostly bacterial) infections may have selected sulfa-resistant strains in malaria parasite carriers from some of these countries.

Linkage disequilibrium
Here we found a sharp LD decline within 100 bp of distance between pairs of SNPs in BRA and MEX, with very low LD in COL and negligible LD in PER (Fig 6). Interestingly, the LD decay in BRA and MEX populations, originating from the sites with the lowest malaria transmission [53], was even faster than that described for P. vivax populations from Southeast Asia, where P. vivax endemicity is substantially higher [14], and that described for P. falciparum populations from areas of much higher endemicity in Sub-Saharan Africa [54,55,56] and Southeast Asia [56]. These results are consistent with a very high outcrossing rate in P. vivax populations and imply that SNP array-based genome-wide association studies (GWAS) would require a very high marker density (inter-marker distance < 100 bp) to help find genetic loci that determine phenotypes of interest, such as drug resistance and virulence, in BRA and MEX populations. SNP-based association studies are unlikely to be successful in PER and COL populations, given the negligible LD. In contrast, P. falciparum populations from the Pacific Coast of Colombia showed a much more gradual LD decline, reaching background LD levels only at an inter-marker distance of 240 kb [57].

Conclusions
We found high genome-wide diversity and relatively little geographic structure in P. vivax populations from areas of relatively low malaria transmission in the Americas. These findings contrast with the low genetic diversity and clear subdivision into local, highly divergent populations that is typical of New World populations of P. falciparum [57][58][59][60][61]. Interestingly, local P. vivax populations are nearly as polymorphic as their P. falciparum counterparts from Africa [45,46,54] and their P. vivax counterparts from Southeast Asia [11,14]. Moreover, the low levels of between-population differentiation in P. vivax from the Americas are reminiscent of those in P. falciparum populations from hyperendemic sub-Saharan Africa [45,46,59,62].
Demographic history might explain, at least in part, the observed differences between P. vivax and P. falciparum populations from the New World. The original diversity of the ancestral P. falciparum population may have been partially lost during (a) its migration to the Americas in post-Columbian times [63] followed by the adaptation to entirely different local vectors [64], (b) recent selective sweeps driven by antimalarial drugs, such as chloroquine [65] and pyrimethamine [66], and (c) local extinctions and clonal expansions following eradication attempts in the Americas between the mid-1950s and late 1960s [53,60]. These events would have resulted in small and fragmented, mostly inbred P. falciparum populations being scattered throughout the region. The significantly greater genome-wide diversity currently found in P. vivax populations worldwide, compared with P. falciparum [11,67], suggests that (a) this species retained more ancestral diversity than P. falciparum when transferred from apes to humans and following subsequent spread out of Africa and/or (b) P. vivax is simply older than P. falciparum and therefore has accumulated more mutations over time. Moreover, these species might also differ in the ways they colonized the New World. The high diversity of presentday P. vivax lineages across this continent is consistent with an admixture of parasite lineages originating from geographically diverse regions [5,7,47], including now extinct European lineages [12]. The success of colonization events and the recombination between these introduced strains may have been favored by the ability of P. vivax to stay dormant in the liver as hypnozoites, prolonging the duration of parasite carriage in human hosts and increasing the probability of superinfections leading to the co-occurrence of genetically diverse clones that may recombine once transmitted to mosquito vectors. Moreover, genome-wide P. vivax diversity in the region appears to have been little affected by recent selective sweeps driven by pyrimethamine [35,37] in Brazil and Colombia. Interestingly, P. vivax resistance to chloroquine remains relatively infrequent in the New World [68] and is unlikely to have induced a major selective sweep in local parasites.
Improved methods for clinical sample preparation and next-generation sequencing now enable further genome-wide analyses of additional P. vivax isolates from the New World, to test the demographic scenarios suggested by our data.