Here is described a new high-throughput method to detect SNPs associated with drug resistance of P. vivax, simultaneously in pvdhfr, pvdhps and pvmdr1 genes. To improve the efficiency of this assay, a nested PCR approach was chosen which allowed a successful genotyping in 83.3% of the LDR-FMA P. vivax positive samples.
The main challenge faced in developing this multiplexed assay was the interpretation of fluorescence data, due to the simultaneous screening of multiple alleles at one locus and the high level of polyclonality (up to four isolates infecting a single patient). As described previously , an increase of background (or off-target) fluorescent signal was consistently observed when PCR amplification produced strong gene-specific fragments. A polar transformation of the fluorescent data measured was performed for bi-allelic datasets; the usefulness of this transformation has been shown before . In contrast, datasets with more than two alleles present required a multidimensional transformation of the data, based on the same principle as the polar transformation. This analysis was applied to three sets of multi-allelic data (n>2): pvdhfr codons 57-58-61, pvdhfr 117, and pvdhps 382-383.
As in the case of bi-allelic data sets, while the coordinate transformation/histogram segmentation approach better accounted for advancing background due to off-target hybridization, it is not a bulletproof algorithm that guarantees 100% accurate diagnosis. Indeed, given the intrinsic variability of any biological measurement process it is unlikely such a perfect algorithm exists. A significant limitation of the heuristic algorithm described here, at least from a theoretical point of view, is that independent determination of the diagnostic thresholds for each locus can lead to a situation in which a given sample is judged to be positively infected at some but not all loci, which is an inconsistent result. For example, the field data analysed here comprised 366 samples, each of which showed a positive diagnosis for P. vivax by LDR-FMA . Of these samples, 17, or 5%, were judged not to have enough net fluorescence signal to be considered positively infected based on the overall magnitude of the fluorescence vector for all seven loci interrogated. Of the 349 samples that showed a positive overall infection based on the magnitude histogram analysis, another 12.6% (44 samples) failed to have a positive single-allele diagnosis for any allele at one of the loci, leading to an incomplete haplotype determination. As a practical matter, in these cases the algorithm described under Methods was supplemented by censoring the samples with an incomplete haplotype determination.
A total of 23 alleles were screened simultaneously in the assay, with the possibility of extending the assay towards screening more alleles as 100 microspheres can be detected in the same multiplexed assay on the BioPlex liquid array reader. This high multiplexing capacity significantly reduces costs associated with diagnosing complex arrays of SNPs associated with anti-malarial drug resistance. While it would cost approximately 3 USD to genotype one sample using a real-time PCR assay screening for only four alleles (TaqMan), the costs of the PCR-LDR-FMA are similar to screen one sample for more than 20 alleles. In addition, as multiple loci of a gene are screened simultaneously, it is possible to identify new genotypes. If a positive signal is observed at one locus of the gene, but a negative signal at another, this would demonstrate the positivity of the PCR, and at the same time, the presence of a new allele not targeted by any of the LDR primers present in the reaction. In this case, it would then be necessary to sequence the PCR product in order to confirm the existence of a new genotype. Modifications to the LDR-FMA are then easily achieved by designing new sequence-specific primers.
Due to the wide use of sulphadoxine-pyrimethamine (SP) to treat falciparum malaria, P. vivax has been exposed worldwide to this anti-malarial drug. In Papua New Guinea, SP has been added to 4-aminoquinolines (CQ and amodiaquine) to treat P. falciparum and P. vivax malaria since 2000. It is therefore not surprising that pvdhfr and pvdhps mutant alleles [23, 33–36] were highly prevalent, with 65.6% of infected patients displaying a triple mutant dhfr genotype (57L-58R-61M). Because of a high rate of polyclonal infections (83.3%) and a mean MOI of 1.99 calculated from pvdhfr genotyping results, it is difficult to reconstruct haplotypes and give more than an estimation of their prevalence. As shown by the results obtained from monoclonal infections, the 117T mutation was associated with the triple mutant haplotype 57L-58R-61M in 100% of the cases. It is therefore possible to predict an overall rate of infection (among monoclonal and polyclonal infected patients) with a quadruple mutant dhfr haplotype to occur in about 66% of the cases.
In vitro results (using a yeast model to express PvDHFR) have shown a strong correlation between mutations in pvdhfr and the amount of drug needed to inhibit the growth in this system. The 50% inhibitory concentration (IC50) of pyrimethamine is increased by 100-460 fold for a dhfr double mutant 58R-117N, and more than 500-fold for a quadruple mutant enzyme 57L-58R-61M-117T in comparison to the IC50 measured for the wild-type enzyme [28, 34]. Furthermore, the quadruple mutant haplotype has been associated with in vivo treatment failure after treatment with artesunate-SP in a study conducted in Papua, Indonesia . Similarly, the quadruple mutant haplotype has been associated with SP treatment failure when compared to lower mutant parasites (triple mutant to wild type) . The pvmdr1 Y976F mutation (when found in combination with pvdhfr mutations) was associated with treatment failure in a PNG study conducted in 2004-2005 when P. vivax infected patients were treated with AQ+SP .
Very few reports have been available on the prevalence of P. vivax dhfr and dhps mutations in PNG. From 19 isolates collected in 1998 in the Wosera (East Sepik Province) , only one (5.3%) displayed a double 57L-58R
pvdhfr mutation. In the same study, of 25 isolates collected in 2000 in Liksul (Madang Province), nine displayed a pvdhfr mutant genotype (six were double mutant 57L-58R). More recently, genotyping results for 94 samples collected in three areas from PNG (Simbu, East Sepik and Madang provinces) between 2004 and 2005 showed higher prevalence of mutant genotypes and the appearance of quadruple mutant parasites (pvdhfr 57L-58R-61M-117T) in the population . This increase is confirmed here with significantly higher rates of patients infected with mutant genotypes: 57L was found in 78.4% of the participants, 58R in 83.0%, 61M in 65.9%, 117N/T in 70.8%, versus, respectively, 59.6%, 67.0%, 20.2% and 40.4% from the earlier study (p < 0.001). Data from PNG isolates on pvdhps have not been generated; only one isolate from PNG was previously sequenced, showing a mutation at codon 647 (647P) . As for pvmdr1, the 976F mutation was found at a prevalence of 39.4%  and remains at the same level here (35.1%) despite known high level of CQ-SP drug resistance in PNG (51.4% after PCR correction ). It has already been discussed that 4-aminoquinoline resistance may result from a multigenic process involving different SNPs and/or gene amplification, variation in the level of expression [24, 25]; it is therefore harder to correlate the 976F mutation frequency with an increase of the in vivo level of drug resistance.