A total of 99 clinical P. vivax parasite samples were previously collected in four locations of East and Southeast Asia, including 36 samples from Bengbu city, Anhui Province of China in 2004, 26 samples from Luodian county, Guizhou Province of China in 2005, five samples from villages near the northern border of South Korea in 2005, and 32 samples from Laiza township of northeastern Myanmar in 2006-2008 (Figure 1) 20 24 25 . Collection sites in China and Korea were malaria hypoendemic with seasonal occurrence of only P. vivax parasites, whereas the Myanmar site was mesoendemic with perennial circulation of both P. vivax and P. falciparum parasites 26 . Ethical clearance was obtained from provincial ethical review committees of Guangxi and Anhui, and Kunming Medical University Institutional Review Board, respectively. The samples used for sequencing have been identified as single-clone infections by genotyping the merozoite surface protein (msp) 3α and msp3β genes 20 25 . The highly polymorphic nature of these two genes should allow us to exclude most of the mixed-clone infections. Parasite DNA was extracted from blood filters using the QIAamp DNA Mini Kit (Qiagen, USA) with minor modifications. Primers were designed to amplify the complete P. vivax mt genome based on the sequence of the reference strain Sal-I (GenBank: AY598140). Long-range, high-fidelity PCR amplification was performed using Advantage HD DNA Polymerase Mix (Clontech, Mountain View, CA), which has efficient 3' → 5' exonuclease proof-reading activity. PCR reactions in 50 μl contained DNA template, 1 μM of each oligonucleotide primer, 10 × Advantage HD PCR Buffer, 1 mM deoxynucleosides (dNTPs), and 5 units of polymerase mix. PCR was performed at 94°C for 1 min, followed by 30 cycles of 94°C for 1 min, 60°C for 2 min, and 72°C for 3 min. A final extension was done at 72°C for 10 min. PCR products were purified with the PrepEase Gel Extraction Kit (USB, Affymetrix, USA) and sequenced using the BigDye version 1.1 on an ABI 3730XL sequencer. Overlapping sequences were obtained with the primers listed in Additional file 1. Sequences were base-called and assembled with PHRED and PHRAP 27 28 , and heterozygous bases were called with the Polyphred program 29 and confirmed by visual inspection of the corresponding trace files using the CONSED trace file viewer 30 . To remove potential nucleotide ambiguities, certain sequences were confirmed by two independent PCR reactions from the same DNA template. The sequenced fragments were assembled into complete mt sequences and deposited in GenBank (JQ240331-JQ240429).

For phylogenetic analysis, we retrieved additional 291 complete P. vivax mt genome sequences from GenBank (AY791517-AY791692, AY598035-AY598140, and AB550270-AB550280). Nucleotide diversity (π) - the average number of nucleotide differences per site between two sequences, and haplotype diversity (h) - the number and frequency of different haplotypes in the samples were estimated using DnaSP version 5.10 31 . The three protein-coding genes were identified through alignment with annotated genes from the published P. vivax mt genomes. Sequence alignments were performed using inGAP v 2.2 32 and ClustalW 33 , and were manually checked using BioEdit http://www.mbio.ncsu.edu/BioEdit/bioedit.html. To detect selection acting on these coding sequences, the number of synonymous nucleotide substitutions per synonymous site (d_S ) and number of nonsynonymous nucleotide substitutions per nonsynonymous site (d_N ) were calculated using the SNAP program http://www.hiv.lanl.gov/content/sequence/SNAP/SNAP.html. Phylogenetic trees were constructed by using the maximum likelihood method implemented in BioNumerics v5.10 (Applied-Maths, Sint Maartens-Latem, Belgium). In order to define the relationship among the isolates at a micro-evolutionary level, we performed allelic profile-based comparison using a minimum spanning tree analysis with the TCS1.21 34 as well as the BioNumerics v5.10 software. In TCS, inferences depend on the chosen probability of parsimony, for which we tried a value of 91%-95%. The number of mutational differences associated with the probability just before the 91% cutoff is the maximum number of connections between sequences. Ambiguities were treated according to the previous criteria 35 .

Analysis of population structure was performed using STRUCTURE 36 , which uses a Bayesian approach to calculate the posterior probability of the number of populations in the sample single nucleotide polymorphism (SNP) set. Replicate runs of STRUCTURE used a burn-in period of 50,000 iterations followed by a Markov chain Monte Carlo (MCMC) 100,000 iterations from which estimates were obtained. All runs were based on the admixture model without prior population information 36 . Twenty replicate unsupervised runs were performed for each value of the number of clusters K from 1 to 14. The number of population clusters was inferred according to Evanno et al. 37 and the ad hoc statistic ΔK was calculated for the presence of individuals with ancestry in two or more of the K populations. Graphs of the STRUCTURE results were made using DISTRUCT. Sequences were analyzed for maximum sequence diversity and visually inspected with highlighter tools http://www.hiv.lanl.gov.

ARLEQUIN software package version 3.11 was used to estimate genetic diversity indices and to assess population differentiation 38 . Pairwise comparisons F _ST and Φ _ST values were estimated by permutation analyses using 1,000 permutations with an assumption of no difference between populations. The P-value was considered as the proportion of permutations resulting in the higher F _ST or Φ _ST value or equal to the observed one. Analysis of molecular variance (AMOVA) was used to evaluate the extent to which sequence variation was partitioned among populations and areas.

Neutrality statistics and pairwise nucleotide differences were calculated to examine the historical demographic expansions of P. vivax. Tajima's D test 39 and Fu's Fs test 40 were used to test whether the mt genome data conform to the expectations of neutrality. Tajima's D test compares two estimators of the mutation parameter θ, Watterson's estimator θs and Tajima's estimator θπ, and significant D values can be due to factors such as population expansion, bottlenecks and selection 39 . Fu's Fs test compares the number of haplotypes observed with the number of haplotypes expected in a random sample under the assumption of an infinite-sites model without recombination, and Fs was sensitive to population demographic expansion 40 . In addition, historic demographic expansions were also investigated by examining frequency distributions of pairwise differences between sequences (mismatch distribution), which is based on three parameters: θ₀, θ₁ (θ before and after the population growth) and τ (time since expansion expressed in units of mutational time) 41 42 . The expected mismatch distribution under a constant population size was estimated by the equation F _i = θⁱ/(θ+1)ⁱ⁺¹, where F _i is the expected frequency of pairwise comparisons showing i differences, and θ is estimated by the observed mean of pairwise differences. The distribution usually has a ragged profile at demographic equilibrium due to stochastic lineage loss, but it is usually unimodal in populations following a recent population demographic expansion and population range expansion 41 43 . This reflects an underlying star-like genealogy in which all of the coalescent events occurred in a narrow time window. Both mismatch analysis and neutrality tests were performed in ARLEQUIN.

TMRCA of P. vivax was estimated using the strict-clock Bayesian MCMC method as implemented in BEAST v1.5.4 44 with non-coding region of the mitochondrial genome. We assumed a general time-reversible model of nucleotide substitution with gamma-distributed rate heterogeneity among sites and a proportion of invariant sites. In addition, we assumed an uncorrected log normal distribution molecular clock model of rate variation among branches in the tree. As previously described, TMRCA was calculated using the divergence between Plasmodium fragile and Plasmodium knowlesi, assuming that they diverged as part of the radiation of their hosts (the silenus group of macaques and the divergence of the sinica and fascicularis groups) between 3.5 and 4.7 million years ago 14 . Analyses were performed for 1,000,000 generations with the model of Yule Process and Bayesian skyline analysis. The initial 20% of each run was discarded. The effective sample size for parameter estimates and convergence was checked using Tracer http://beast.bio.ed.ac.uk/Tracer.

We sequenced the complete mt genomes from 99 P. vivax samples representing four populations collected in temperate and warm temperate regions of East Asia and Myanmar (Figure 1 and Table 1). Alignment of these complete mt genomes identified 26 SNPs (indels excluded), including 15 transversions and 11 transitions (Additional file 2). Sixteen of 26 SNPs are located in the three coding regions (COX3, COX1 and CYTB). Nucleotide diversity ranged from 0.00020 for China's Anhui population to 0.00096 for the combined Korean population. The d _N/d _S ratios of coding regions from all geographical sample sets were well below one, confirming that purifying selection is the predominant force in mtDNA evolution (Table 1). The 26 SNPs defined a total of 30 mt genome haplotypes. Compared with the number of haplotypes/population (8-38) identified in earlier studies, the numbers of haplotypes in our studied populations were modest, being the lowest (5) in China's Anhui population and highest (16) in the Myanmar population (Table 1). Likewise, haplotype diversity also varied, ranging from 0.584 in the Anhui population to 0.850 in the Myanmar population. The two temperate populations from China had much lower haplotype diversity and were comparable with the New World parasite population (0.58-0.68).

Analysis of P. vivax mt genomes from 99 newly collected Asian samples as well as 291 previously described isolates from 11 populations identified a total of 141 haplotypes 12 13 45 . Most geographical regions had one predominant haplotype, whereas Melanesia, China, and Southeast Asia had two predominant haplotypes. To characterize the frequencies and relationships of different haplotypes, a minimum spanning tree was constructed. This haplotype network clearly showed geographical clustering of the haplotypes (Figure 2). Inclusion of the temperate P. vivax samples did not dramatically change the overall topology of the network. Consistently, significant sharing of the haplotypes was detected between South/West Asian and African populations 12 . For the four temperate populations, most haplotypes were connected to the South/West Asian populations, which is consistent with the fact that these geographic regions are physically connected and past genetic exchanges might have occurred between these populations. Interestingly, the two temperate P. vivax populations from China's Anhui and Guizhou provinces formed a local network consisting of two predominant, related haplotypes. In contrast, the Chinese samples analyzed earlier (some from Guangxi Province in southern China, Figure 1) 12 13 formed two clusters, which might indicate discrete origins: some haplotypes were clustered in the East Asian branch, whereas others were shared and linked with haplotypes from Southeast Asia (Thailand, Vietnam and Indonesia). As reported previously 45 , most Korean haplotypes were shared with those from southern China 12 13 , but diverged from the central China populations.

To further illustrate the relationship of the mt genome haplotypes, a maximum likelihood phylogenetic tree based on the complete genome sequences was constructed (Additional file 3). All mt genomes from P. vivax were closely related to one another; no branches within the P. vivax cluster received strong bootstrap support (data not shown). In general, the topology of this tree roughly mirrored that of the minimum spanning tree. Both analyses showed that East Asian populations exhibited relatively large divergence from the remaining populations, suggesting that the temperate parasite populations in East Asia had a different demographic history.

Earlier studies using both mt genome and microsatellites have found significant differentiation of the extant P. vivax populations 7 12 . To further define genetic structure of the P. vivax populations, we used a Bayesian admixture procedure implemented in STRUCTURE to calculate the potential number of populations. Using a series of K values from 1 to 14, we detected a large incremental increase of the likelihood value as the number of genetic clusters in the model increased from 1 to 4 (data not shown). Significant genetic structure was found between parasite populations, and most parasites formed geographical clusters. At K = 2, only the East Asian cluster was separated from the rest of the populations, although this division was also supported from analysis of the distribution of ΔK using the method of Evanno et al. 37 . Posterior probability of the entire sample set at 4 populations was most consistent with a clear separation of populations according to large geographical regions of Melanesia, East Asia, Southeast Asia and the rest of the samples (Figure 3). It is noteworthy that South American and Myanmar populations appeared to be composed of a mixture of haplotypes from South/West Asia as well as Southeast Asia, suggestive of population boundaries or shared demographic histories of the parasite populations in these regions.

To further determine whether population differentiation exists, we estimated F _ST (only based on haplotype frequency) and Φ _ST (based on both haplotype frequency and genetic distance) for each population using the mtDNA sequences. The majority of pairwise F _ST values were greater than 0.25 (ranging from 0.25 to 0.85), suggesting that strong population differentiation exists among the populations. Notable exceptions included the following pairwise comparisons: South/West Asia and Africa, South/West Asia and South America, South/West Asia and Southeast Asia, as well as Korea and southern China (Table 2). In Southeast Asia, parasite populations from the Greater Mekong subregion, Indonesia and Melanesia were also distinct with F _ST values of 0.32 - 0.62 and Φ_ST values of 1.58 - 4.84 (P = 0.01). For the temperate parasite populations from Asia, which include northeast Myanmar, Korean, and three Chinese P. vivax populations, between-population genetic differences were strong (F _ST = 0.32-0.75, Φ _ST = 0.97-5.13) and statistically significant after sequential Bonferroni correction (P < 0.0001). The only exception was between the Korean and southern Chinese population (exact origin of the samples unknown), which showed little genetic differentiation (F _ST = 0.05, P = 0.08; Φ _ST = 0.20, P = 0.08) (Table 2). We further grouped all studied populations into eight geographical groups consisting of Africa, America, South/West Asia, Southeast Asia, Myanmar, East Asia, Indonesia, and Melanesia, and performed a hierarchical AMOVA analysis. The three covariance components (within population, among population/within group, and among groups) could explain 22.7%, 32.3% and 45.0% of the variance, respectively. Apparently, considerable variation was preserved at the population level.

The star-like shape of the haplotype network is consistent with significant population expansion (Figure 2). To test this hypothesis, we performed mismatch distribution and skyline analyses. Mismatch distribution was unimodal for the entire P. vivax sample set (Figure 4A), a pattern characteristic of populations that have undergone large-scale expansion. To obtain more precise estimates for the temperate parasite populations, we divided our recent samples into two large groups represented by the Myanmar and East Asian populations (Chinese and Korean samples combined), and performed neutrality tests and mismatch distribution analysis. Although mismatch distributions for these two populations did not perfectly fit the expected pattern under the sudden expansion model (Figure 4B and 4C), they did not differ significantly (P > 0.05) from this model and therefore were suitable for the analysis of demographic patterns. The Fs test and Tajima's D test agreed well with the mismatch analyses. The Fs values of the entire P. vivax populations as well as the East Asian and Myanmar populations were negative and significant (P = 0.025 for East Asian populations and P = 0.005 for Myanmar population). Tajima's D test compares two mutation estimators θs and θπ, which are affected by the size of the present-day and original population, respectively 39 . Consequently, a history of population growth would inflate S relative to π and generate a negative value of D 46 47 . Consistent with a past population expansion, Tajima's D test was negative (-2.05) and statistically significant (P = 0.001) for the entire P. vivax populations analyzed. For the Myanmar and East Asian populations, the D statistic was negative and only significant for the Myanmar population (P = 0.021). The τ value, which reflects the location of the mismatch distribution crest, provides a rough estimate of the time when rapid population expansion started. The observed value of the age expansion parameter (τ) was 4.18 units (95% CI: 2.56-10.41) of mutational time in P. vivax. In comparison, the τ value of the Myanmar population was smaller (4.10, 95% CI: 1.36- 6.59). Yet, the East Asian populations had a τ value of 5.13 (95% CI: 2.01- 13.3), which was even greater than that for all vivax populations, suggesting that the East Asian parasite populations may be at demographic equilibrium in recent evolutionary history. The results of the skyline analysis further supported P. vivax population expansion showing a rapid increase in effective population size starting ~ 200,000 years before present (ybp), followed by a more recent increase in population numbers starting at ~50,000 ybp (Figure 4D).

We estimated the divergence time of Myanmar P. vivax population to be 385,000 (± 3,800) years ago, whereas the East Asian P. vivax population diverged at 383,000 (± 7,300) years ago. Our estimate also indicated that the TMRCA for the whole vivax populations was around 346,000 to 452,400 years ago (95% highest posterior density).