Arguably, the most well studied and clinically important sets of molecular biomarkers are genetic loci associated with resistance to antimalarial compounds. Resistance to aminoquinolines, such as chloroquine, and antifolates, such as pyrimethamine and sulfadoxine, is widespread, and molecular markers associated with these resistant phenotypes have been identified. Specifically, researchers have identified genetic changes in the pfcrt locus for chloroquine resistance (12), the dhfr locus for pyrimethamine and proguanil resistance (23, 24), and the dhps locus for sulfadoxine resistance (5). Other candidate molecular biomarkers for drug resistance are pfmdr1 (13, 27), associated with resistance to amodiaquine and other aminoquinolines; cytb, associated with atovaquone resistance (15, 17); and PfATPase6, implicated in artemisinin resistance (6, 16, 25). New discovery efforts, including genome-wide association studies, are beginning to identify additional candidate loci related to drug resistance (20, 26, 31); these loci will need to be validated in independent parasite populations.

It is critical that assays identify additional candidate loci related to drug resistance for known and newly discovered molecular biomarkers to be developed in order to easily and reliably examine samples from infected patients identified through positive blood smears. Because the parasite population changes readily and radically in response to selective pressure from drugs and other intervention methods, eradication policies must be equally responsive and adaptable. The data from these assays must allow real-time observation of parasite population changes to inform timely protocol changes and further global eradication efforts to reduce or even eliminate child deaths from malaria.

While a number of assays have been developed for drug resistance loci, they rely on technologies that are not suited to real-time detection and discovery of single-nucleotide polymorphisms (SNPs) associated with reduced drug sensitivity. Many of these technologies, like pyrosequencing and real-time PCR (29, 32), require extensive training and the use of expensive reagents on instruments in regional clinic settings, which means that samples collected in the field can wait months or years for analysis. Other technologies, such as restriction fragment length polymorphism (RFLP) and msp1, msp2, and glurp marker typing, are economical to implement, but interpretation of results can be complicated and dependent on the user's skill level or lead to results that cannot be directly compared between studies (21). These limitations reduce the utility of existing methodologies for global surveillance campaigns and emphasize the need for assays that are portable, cost-effective, and uncomplicated.

Our genotyping assays proved to be robust and specific. They were able to detect and differentiate alleles with low concentrations of template material even in the presence of excess contaminating genomic material. Using cultured isolates without excess human material, as would be found in filter-extracted patient samples, the limit of detection was 10⁻⁵ ng, similar to that found by other groups (Fig. 2A) (1, 14) and also similar to other PCR-based methodologies and comfortably within the yield from extraction of patient samples collected on filter paper in the field. The assays remained specific for their target regions in the presence of an excess of human genomic material that we introduced in order to mimic the ratios of human and parasite DNAs obtained from patient samples; however, the absolute limit of detection with contaminating human DNA was reduced to approximately 10 pg (Fig. 2B).

The genotyping assays are also sensitive and were made more so by our technology modifications. We were able to detect alleles comprising 2% to 5% of the total sample; by introducing the novel amplification method MAAB on a LightScanner-32 instrument using glass capillary tubes, we were able to increase the sensitivity to detect mutant alleles present at less than 1% of the mixture (Fig. 3B). There is some vertical spread between the mixture peaks; however, the inflection for each mixture does not change compared to the negative or peak curve seen with the unmixed samples. This inflection difference is consistent across repeated trials and is reliable as long as unmixed controls are present.

Remarkably, we could discriminate individual alleles even in more complex mixtures containing three or more alleles using our modified HRM approach (Fig. 3A). These results demonstrate that the assays are sensitive and specific and are able to identify alleles present at less than 1% of a mixture when multiple alleles exist in a sample. Comparable amplification technologies, such as TaqMan, fail to consistently detect mutant alleles at less than 10% of the sample mixture (8). Without the refinements to HRM using blocked probes, the typical sensitivity for detection is still an improvement at 2 to 5%.

Additional methods, such as nested-PCR amplification and RFLP analysis, rely on more subjective and complicated interpretation of gel bands. Here, we show that HRM analysis results in clearly separated and unambiguous melting peaks to determine sample genotypes.

To assess the accuracy and reliability of these assays, we applied them to genotype 44 independent, culture-adapted parasites from a number of different geographical regions and verified our results by sequencing (Genewiz, Inc., South Plainfield, NJ) (Table 2). From our assays, we found that 40% of the samples had the mutant genotype for pfcrt K76 and 18% and 40% displayed variants at dhps residues 436 and 437, respectively. For dhfr, 65% were mutants at residues 51 and 59, and 70% were mutants at residue 108. While many other SNPs associated with drug resistance were primarily wild-type genotypes, there were variant polymorphisms in this international collection for each of the following loci: pfcrt 220 (40%) and 356 (12.5%), pfmdr1 86 (32.5%) and 184 (70%), and PfATPase6 (40%). All mutations detected by genotyping assays were confirmed by sequencing (Genewiz, Inc., South Plainfield, NJ), and our assays showed 100% correlation with the sequencing results, as well as those reported on PlasmoDB version 6.3 (http://www.plasmodb.org). The genotyping assays are also sensitive and were made more so by technology improvements. These data validate the assays as able to reliably detect previously characterized mutations in the six drug resistance loci.

We next tested the correspondence of our genotyping assays with drug phenotypes among the culture-adapted parasites (2). We evaluated the K76 locus of pfcrt associated with chloroquine resistance and the N51, C59, and S108 loci of dhfr associated with pyrimethamine resistance. We observed 100% correspondence between the pfcrt K76T mutation and chloroquine resistance among culture-adapted parasites. We also observed 100% correspondence between the N51I allele (100%) and the C59R allele (100%) in dhfr for pyrimethamine resistance. For the S108N allele, four of the parasites deviated from direct correspondence with pyrimethamine resistance. Three of the parasites that had the S108N allele tested sensitive to pyrimethamine. These parasites were wild type for the N51 and C59 alleles and had pyrimethamine resistance levels near the cutoff (2,000 nM) for resistance (3). One parasite that had the S108 allele tested resistant to pyrimethamine. This parasite, however, was mutant for the N51I and C59R alleles and also had a level of pyrimethamine resistance that was just over the resistance cutoff level of 2,000 nM. Thus, we observed almost complete correspondence between the genotyped alleles and the expected drug responses for chloroquine and pyrimethamine among the culture-adapted parasites in this analysis, confirming that these molecular markers are useful for interpreting drug responses in P. falciparum parasites.

We then demonstrated the usefulness of our HRM assays on materials derived from blood collected on filter paper from patients seeking treatment for malaria at the Thies, Senegal, clinic. Table 2 shows the prevalences of mutations in the following loci: pfcrt 76 (15%), dhps 436 (26%) and 437 (48%), and dhfr 51 and 59 (85%) and 108 (93%). Other SNPs with variant alleles include pfcrt 220 (26%) and 356 (22%), pfmdr1 86 (7%), pfmdr1 184 (70%), and PfATPase6 623 and 431 (7%). Based upon comparison to data from previous years, there has been a decrease in the prevalence of mutations associated with chloroquine resistance and an increase in mutations associated with pyrimethamine resistance. The decrease in chloroquine resistance but increase in pyrimethamine resistance is consistent with the use of these drugs for malaria treatment in Senegal. While chloroquine use has largely been discontinued, use of pyrimethamine has continued (initially in a combination treatment with sulfadoxine and amodiaquine) and remains the mainstay treatment for intermittent preventative treatment for pregnant women (IPTp) in Senegal to date (30).

While traditional probe-based technology approaches fail to amplify if novel variants appear within the amplicon region, our assays allowed us to detect both known and novel haplotypes and mutations in genetic loci associated with drug resistance. Techniques such as TaqMan require prior knowledge of all sequence variants for successful genotyping in order to design labeled probes for every polymorphism. We have demonstrated that our method, on the other hand, is far more robust and can identify novel variants. First, we detected previously unreported haplotypes in Senegal for pfdhps 436/437 in three of the culture-adapted parasite lines, including the 436A/437A haplotype for two parasites (P05.02 and Th105.07), as well as the 436Y/437G haplotype for another parasite (Th28.04) (Fig. 4A). Second, we detected a new mutation in the cytB locus from a single patient sample (Fig. 4B). This novel mutation, confirmed by sequencing (Genewiz, Inc., South Plainfield, NJ), is the M270I variant, located near the amino acid associated with atovaquone resistance in the cytB gene at Y268 (17). Because this novel allele was detected in direct patient material, we were unable to test for any biological effects of the mutation on drug sensitivity. Because the probe-based analysis also allows study of the larger amplicon, we were able to discover SNPs 22 bp upstream of the probe region but still within the amplicon targeted by the ATPase6 431 assay in D10 (MRA-201) (data not shown). The additional utility of focusing on genomic regions known to be under selection is the increased likelihood that additional mutations associated with reduced drug sensitivity will be found in the regions targeted by the assay primers that may confer drug resistance. While samples with variant melting peaks must be sequenced to characterize the emerging mutation, HRM technologies offer the unique ability to quickly and economically scan samples for emerging variants.

In addition, the assays differentiated between copies of a genetic locus. The pfmdr1 locus is known to be present in multiple copies in some drug-resistant parasites, particularly those resistant to mefloquine (13, 28). As part of this analysis, we observed different genotypes for the pfmdr1 locus in the Dd2 parasite and confirmed the presence of the newly reported pfmdr1 N86F mutation in Dd2 (4, 9). This demonstrates the ability of these assays to detect mutations among loci subject to copy number variation, which has been implicated in some drug-resistant parasites (11, 18, 22). Copy number variation has also been identified as a source of reduced sensitivity to antimalarial drugs. Instruments with real-time capacity (Idaho Technology LightScanner-32 and Qiagen Rotorgene-Q) offer additional utility of these assays for direct detection of these variations.

Several other groups have presented HRM assays applied to P. falciparum (1, 7, 14); however, the techniques presented here show clear advances over the previous works. We have implemented a probe-based technique that improves the sensitivity of these assays over the whole-amplicon methodology presented in the work of Andriantsoanirina et al. and similar to the technology presented by both Gan and Loh and Cruz et al.

Compared to the previous publications, we have developed and validated an extended set of assays for SNPs related to reduced drug sensitivity and have made notable advances in the ability to detect mutant alleles present in small amounts within a sample, an improvement on both studies, which noted a 10% limit of detection of mutant alleles while we have detected mutant alleles at less than 1%. While MAAB is most effective in instruments that use glass capillaries for their improved heat transfer properties, the method could be especially useful for surveillance of emerging resistance where the primary causative SNP is known. The ability to detect these mutations at very low levels can inform control efforts. MAAB could be applied following larger-scale plate-based screens where the results may indicate the presence of a minor population of mutants. MAAB could confirm these observations.

Further, we applied these assays to determine the SNP genotypes of 27 patient samples with both single and mixed infections and were able to detect the emergence of haplotypes new to Senegal. Like Gan and Loh, we identified new mutations within the probe region but also found evidence of new mutations outside the probe but within the larger amplicon.

With this technology, results are robust, consistent, and repeatable; however, several caveats exist. Care must be taken in preparing the reaction mixtures. Salt concentrations have a strong effect on the melting temperature of the complexes, so it is advised that all samples be in the same buffer for the most consistent results. Not all samples have to be at the same concentration, but the reactions are most reliable with low concentrations of template. They become less so with an excess of template (more than 10 ng/μl), and the melting peaks may shift even if all samples are in the same buffer if any of them are at a very high template concentration relative to the other samples. While most groups report femtogram levels of detection within culture-adapted strains without human genomic material, the sensitivity of the assays decreases when this human contaminant is present in the sample. Still, we observed universally robust amplifications and unambiguous genotypes from 0.005 to 0.01 ng of P. falciparum template from patient-derived samples that had not been depleted of white blood cells.

HRM is a promising technology for field-based studies; however, the success of the results also depends on the quality of the analysis software that is included with the HRM analysis instruments. Currently, no standalone application exists for melting analysis. We have had the best success with the suites offered by Idaho Technology, Inc., as a combination of sensitive analysis and ease of peak calling without extensive and complicated user training. User training will be further reduced as researchers publish additional validated assays that require no further optimization. Our initial panel of 23 assays is a strong start toward this goal.

Because our HRM assays are sensitive, work on mixed infections, and are easily deployed to study field samples, they have the capacity to track both current and emerging infections. Our genotyping assays can be used to monitor the frequencies of known drug resistance alleles or parasite types within a population. Furthermore, they can elucidate changes in these frequencies as interventions or other factors that impact transmission dynamics in a given population. For example, as molecular markers of emerging drug resistance are discovered, these markers can be monitored across a population to determine the likelihood that current drug therapies would remain effective or to inform policies to contain the spread of emerging drug resistance and detect the importation of drug-resistant parasites. Because these assays do not require fluorescence-based probes, the reagent cost per sample is much lower than for alternative technologies, and storage and preparation conditions are much less stringent. In our hands, HRM costs approximately 2/3 of what TaqMan-based SNP genotyping costs and have less than 1/10 the cost for sequencing all of these regions. In addition, the assays are reliable for samples collected on filter paper, a low-cost method for sample preservation, transportation, and storage. The fast readout of the melting analysis is clear and unambiguous, allowing easy interpretation of results and less specialized researcher training.

With increased efforts to establish global networks of research centers based on-site across the globe, such as the International Centers of Excellence for Malaria Research (ICEMR) and the collaborative efforts of the Gates Foundation, the initial capital expenditures for equipment capable of high-resolution melting analysis can be distributed across sites. In addition, many of these instruments are multipurpose and are able to offer real-time analysis in addition to HRM.