Numerical Distributions of Parasite Densities During Asymptomatic Malaria

Background. Asymptomatic parasitemia is common even in areas of low seasonal malaria transmission, but the true proportion of the population infected has not been estimated previously because of the limited sensitivity of available detection methods. Methods. Cross-sectional malaria surveys were conducted in areas of low seasonal transmission along the border between eastern Myanmar and northwestern Thailand and in western Cambodia. DNA was quantitated by an ultrasensitive polymerase chain reaction (uPCR) assay (limit of accurate detection, 22 parasites/mL) to characterize parasite density distributions for Plasmodium falciparum and Plasmodium vivax, and the proportions of undetected infections were imputed. Results. The prevalence of asymptomatic malaria as determined by uPCR was 27.5% (1303 of 4740 people tested). Both P. vivax and P. falciparum density distributions were unimodal and log normal, with modal values well within the quantifiable range. The estimated proportions of all parasitemic individuals identified by uPCR were >70% among individuals infected with P. falciparum and >85% among those infected with P. vivax. Overall, 83% of infections were predicted to be P. vivax infections, 13% were predicted to be P. falciparum infections, and 4% were predicted to be mixed infections. Geometric mean parasite densities were similar; 5601 P. vivax parasites/mL and 5158 P. falciparum parasites/mL. Conclusions. This uPCR method identified most infected individuals in malaria-endemic areas. Malaria parasitemia persists in humans at levels that optimize the probability of generating transmissible gametocyte densities without causing illness.

Asymptomatic malaria parasitemia is very common in areas of high stable transmission-indeed, everyone there may be infected. In contrast, asymptomatic parasitemia was considered relatively uncommon in areas of low seasonal transmission, which predominate in much of Asia and the Americas, but recent epidemiological studies using sensitive methods of parasite detection have revised this view [1][2][3][4][5][6][7][8]. In Myanmar, Thailand, Cambodia, Laos, and Vietnam, a substantial proportion of the population living in malaria-endemic areas harbor asymptomatic parasitemias [9][10][11]. These infected individuals sustain malaria over the dry season, and so they are an important source of malaria transmission and a major obstacle to elimination. However, because of the limited sensitivity of detection methods, the proportion of people in any area with chronic low-density parasitemia has been unknown. Accurate characterization of malaria epidemiology and its geographic distribution is essential in planning regional control and elimination strategies. We used a quantitative ultrasensitive polymerase chain reaction (uPCR) method of parasite quantitation in blood volumes of >200 µL [12] to characterize the numerical distributions of parasite densities in asymptomatic Plasmodium falciparum and Plasmodium vivax infections in populations living in malaria-endemic regions of western Cambodia and the Thailand-Myanmar border [11]. This method was sufficiently sensitive to identify the majority of infected persons and therefore allowed prediction of the proportion of the population with parasitemia levels below the limits of detection by the most sensitive current techniques.

METHODS
These studies took place in malaria-endemic regions along the border between eastern Myanmar and northwestern Thailand and in western Cambodia as a prelude to the assessment of elimination interventions. In these areas, malaria transmission is low and seasonal, with entomological inoculation rates of usually <3 person/year and often <1 person/year. Most clinical cases of malaria occur during the rainy season, between May and December [1,[13][14][15][16][17][18]. In the past, P. vivax and P. falciparum each caused approximately half of the clinical cases, although with recent reductions in malaria incidence, P. vivax now predominates.
Full details of these epidemiological studies have been reported recently [11]. In brief, screening surveys were conducted in 73 villages, followed by detailed cross-sectional malariometric surveys in 4 selected villages located within 10 km of the Thailand-Myanmar border that were considered representative of the area in terms of environment, ecology, population and behavior. The 3 villages in the Pailin region of western Cambodia were selected because they had the highest incidence of clinical falciparum malaria in the village malaria workers' records from 2012.

Procedures
All individuals aged ≥6 months were invited to participate in the surveys. Individual informed consent was obtained from adults, and parental consent was obtained for children aged <16 years. Demographic data and symptom information were collected from each person, and the tympanic temperature, weight, and height were measured. A venous blood specimen (volume, 3 mL) was collected from all individuals aged ≥5 years, and 500-µL blood specimens were obtained from children aged ≥6 months to 5 years.
Sample processing for quantitative uPCR analysis included separation of plasma, buffy coat, and packed red blood cells, which were then frozen and stored at −80°C and transported to the laboratory in Bangkok, Thailand, for DNA extraction and quantitative uPCR analysis [11,12]. All samples obtained for molecular analyses were handled and processed according to the standard operating procedures developed specifically for these studies [12].

Quantification of Malaria Parasitemia
Description, evaluation, validation, and performance characteristics of the high-volume quantitative uPCR analyses have been reported recently in detail [12]. In brief, the DNA template for Plasmodium-specific PCR detection and quantification was purified from thawed packed red blood cells. DNA extraction from a carefully measured volume was performed using a QIAamp Blood Mini Kit (Qiagen, Hilden, Germany) for sample volumes of packed red blood cells ≤200 µL, or a QIAamp Blood Midi Kit for sample volumes between 200 and 2000 µL. The purified DNA was dried completely in a centrifugal vacuum concentrator and then resuspended in PCR-grade water. The concentration factor was defined by the original blood volumes (100-2000 µL) divided by the double-distilled water volumes (10-50 µL). Two microliters of resuspended DNA was used as a template in the quantitative uPCR reaction. The number of parasite genomes in each sample was estimated by an absolute quantitative real-time PCR method, using the Quanti-Tect Multiplex PCR No Rox system (Qiagen). The 18S ribosomal RNA (rRNA)-targeting primers and hydrolysis probes are highly specific for Plasmodium species [12]. The lower limit of accurate quantitation was 22 parasites/mL of whole blood [12]. Quantification of high parasite densities (>30 000 000 parasites/mL) by means of this technique becomes progressively inaccurate.
In uncomplicated falciparum malaria, later asexual parasite stages cytoadhere and so are seldom seen in the peripheral circulation, whereas all blood stages circulate in P. vivax infections. The P. falciparum density (asexual plus sexual stages) is approximately equivalent to the genome density, as nuclear division occurs in sequestered not circulating parasites, whereas for P. vivax (which does not sequester significantly), there is 1 genome/parasite for approximately 36 hours of the 48-hour cycle, followed by genome doubling every 3 hours ( Figure 1). Thus, in a completely asynchronous P. vivax infection, there would be an approximate average of 2.63 genomes per parasite. Both asynchronous and synchronous infections were evaluated.

Species Identification
For uPCR-positive blood samples, malaria parasite species were identified using nested PCR protocols specific to P. falciparum (microsatellite marker Pk2), P. vivax (microsatellite marker 3.502), and Plasmodium malariae (18S rRNA) as described previously [12]. When there was insufficient DNA for species identification or if no amplification was obtained in this step, samples were reported as "indeterminate species."

Statistical Analyses
Statistical analysis was performed with Stata 13.1 and R 3.1.1. Description of the complete distributions of P. falciparum and P. vivax densities required imputation from the measured densities and their respective distributions of the partition of Figure 1. Erythrocytes containing dividing forms of Plasmodium vivax circulate freely, in contrast to Plasmodium falciparum, in which erythrocytes containing dividing forms are sequestered. This correction factor was applied to convert densities of peripheral blood P. vivax genomes to densities of parasites in relation to parasite age. species among the lower-density parasitemias in which the species were not identified, as well as among the even-lower density parasitemias, which could not be measured. This was possible because the distributions of measured parasite densities were unimodal, with modal values well within the quantifiable range. In this analysis, parasitemia densities of >30 000 000 parasites/mL were excluded because their uPCR estimation was inaccurate, and parasitemia densities of <30/mL were excluded to reduce stochasticity close to the uPCR limit of accurate detection. Estimates of the density function for log 10 parasitemia densities of P. falciparum and P. vivax were obtained using the Epanechnikov kernel estimator (Stata command "kdensity"). These were the only parasite species identified in this study, so the probability of infection being due to P. falciparum (P PF ; ie, 1 minus the probability of infection being due to P. vivax) was calculated from the observed proportion of P. falciparum infections, assuming that the species proportions were the same in the lower-density parasitemias for which the species were indeterminate. Imputed data sets (n= 1000) were then created that comprised all of the observed P. falciparum and P. vivax cases and the indeterminate cases allocated randomly to species, based on P PF .
Adjustment to accommodate nuclear division in circulating intraerythrocytic asexual P. vivax parasites was examined. It was assumed that there is 1 genome/parasite for the first 36 hours of the 48-hour asexual cycle, followed by doubling every 3 hours, until finally there were 16 genomes per parasite in the 3 hours before schizont rupture.
Two scenarios were investigated: completely asynchronous infections (scenario A), in which there is a uniform distribution of parasite ages between 1 and 48 hours following merozoite invasion; and synchronous infections (scenario B), in which parasite ages follow a normal distribution with a mean age µ of 1, 2, 3 . . . to 48 hours following merozoite invasion, with a standard deviation of 4 hours. These scenarios require different adjustment factors to convert genomes to numbers of circulating P. vivax parasites. For scenario A, the adjustment factor was as follows: The adjustment factor for scenario B was as follows: where Φ(x, µ,4) is the cumulative normal distribution with mean µ and standard deviation 4 evaluated at value x, and D = [Φ (48, µ,4) − Φ (0, µ,4)] is the total area under the density function between 0 and 48 hours ( Figure 1). For scenario A, all P. vivax parasite genome counts were multiplied by the same adjustment factor to give estimated parasite densities, while for scenario B, each parasite count had a randomly assigned mean parasite age value µ (from the uniformly distributed values 1-48), and that count was multiplied by the adjustment factor corresponding to the value of µ ( Figure 1). The adjustments examined represent maximum deviations because they assume that all P. vivax parasites in the sample are asexual.
For each data set, the following were performed: (1) analysis with the Kolmogorov-Smirnov test to determine whether the log-transformed data conformed to a truncated normal distribution with parameters (m [mean] and s [standard deviation]), with the truncations corresponding approximately to the lower and upper limits of accurate quantitation; (2) estimation of the mean m and standard deviation s of the untruncated distributions of log 10 -transformed parasitemia values, using the maximum likelihood method; and (3) estimation of the tail of the distribution corresponding to the lower densities' truncated segment from the log-normal distribution with parameters (m; s).
The first 20 imputed data sets were also used to examine the relationship between parasitemia and study site and patient age. A linear model for log 10 parasitemia with these covariates and a random effect for participant, to account for multiple measurements. was fitted for each imputed data set. The imputation estimates of the parameter coefficients and their 95% confidence intervals [C]I were calculated [19]. Logistic regression was fitted to estimate relationships between risk of fever and parasite density.

Ethics Committee Approval
The studies were approved by the Tak  The median age of the participants was 21 years; 37% were <15 years old, and 51% were male. The age distributions were similar in the different villages [11].

Parasite Densities and Clinical Covariates
Fever (aural temperature, > 37.5°C) was detected on 73 occasions in 68 participants who did not complain of symptoms. Abbreviations: IQR, interquartile range; P PF , probability of infection being due to P. falciparum; SD, standard deviation. a Scenario A describes asynchronous infections. Scenario B describes synchronous infections (SD, 4 hours).

DISCUSSION
In malaria-endemic areas of western Cambodia and the Thailand-Myanmar border, >85% of asymptomatic P. vivax infections and nearly 75% of asymptomatic P. falciparum infections could be detected by uPCR. The distributions of parasite densities were unimodal and log normal, as they are for symptomatic malaria and indeed for many other bloodstream infections. There was no evidence for discrete subpopulations of lower densities. If it is assumed that the malaria species proportions in the lower parasite density range (in which species could not be identified) were consistent with the proportions in the range where species identification was possible, then the geometric mean densities for all detected asymptomatic P. falciparum and P. vivax infections were similar, at approximately 5 parasites/µL (5000 parasites/mL; 2.5 × 10 7 parasites/adult). The uPCR DNA-based measurement can overestimate parasite densities slightly, particularly in P. vivax malaria, as parasite genomes in merozoites in developing schizonts and those free in plasma (<0.01% of concomitant intraerythrocytic numbers) and phagocytosed by leukocytes may be quantitated [20]. DNA assessment also does not distinguish asexual from sexual parasite stages, unlike highly sensitive messenger RNA methods [21,22]. As gametocytes are cleared more slowly than asexual stages from the peripheral blood, particularly in P. falciparum infections, a significant proportion of the parasites detected by uPCR in asymptomatic individuals are likely to be gametocytes. In this study, the potential errors related to schizontemia and synchronicity of infection in P. vivax infection were examined and found to contribute little to the estimates of geometric mean parasite density or undetected proportions, and measurement errors from free DNA or merozoites were minimized by red cell separation. If the distribution of parasite densities in asymptomatic individuals living in low-transmission settings is visualized as an iceberg, only the top of which is revealed by conventional microscopy or rapid diagnostic tests (RDTs), then high-volume uPCR reveals most of the iceberg and thereby  allows prediction of the proportion of individuals who have parasite densities below the level of detection.
These data also suggest that the majority of infected individuals should be detected by widely used methods involving PCR analysis of capillary blood specimens on filter paper, which typically contain only 5-10 µL of blood [23][24][25]. However, because the limit of detection for these methods is in the parasite density range that contains a large proportion of asymptomatic parasitemias (1-10 parasites/µL), small interassay or intraassay changes or variations in assay performance could have large effects on their malaria prevalence estimates.
Asymptomatic malaria infections in low-transmission settings may be harbored in an individual for long periods in a quasi-steady state [26][27][28][29][30][31][32][33]. If the longitudinal profile of submicroscopic parasite densities mirrors that in the range countable by microscopy, then regular waves of higher densities occur with the sequential emergence of new antigenic variants [27][28][29][30]34]. In P. vivax malaria, persistence is enhanced by relapse, which occurs in the majority of infections in Southeast Asia [35][36][37]. Some of these waves of asexual parasitemia generate transmissible densities of gametocytes [27-30, 33, 34, 38, 39]. Because the prevalence of asymptomatic malaria at any time is orders of magnitude higher than that of symptomatic malaria, the transmission reservoir from asymptomatic malaria is likely to be large. Assuming that a vigorous host response to malaria reduces the probability of onward transmission, particularly in the modern era of antimalarial drug treatment, then, from a parasite perspective, the optimum density to persist at is one that does not generate illness but has the greatest chance over time of generating transmissible gametocyte densities.
Are these parasite density distributions from areas of low, unstable malaria transmission in Southeast Asia relevant to highertransmission settings, where a much larger proportion of asymptomatic individuals have parasitemias detectable by microscopy? Immunity to blood stage parasites is greater in higher-transmission settings, and splenic function is augmented and, thus, clearance of parasitized erythrocytes is likely to be more rapid, so the net result of higher inoculation rates will be a truncated lower end of the parasite density range, with proportionally fewer densities below the limit of uPCR detection. This is because, with a median sporozoite inoculum of approximately 10 and liberation of approximately 35 000 merozoites from each hepatic schizont, the median blood parasite density following hepatic schizogony of each inoculated brood in an adult (blood volume, approximately 5000 mL) approximates 70 parasites/mL (which is above the current level of uPCR detection). The greater proportion of very low-density P. falciparum infections, compared with P. vivax infections, in this study may reflect the currently lower transmission of this species in the study areas and, thus, the greater probability that an individual was being sampled while their infection was being eliminated, whereas P. vivax infections were acquired both from inoculation and more-frequent relapse [35]. In contrast, in high-transmission settings, the upper range of densities of asymptomatic infections is known to be extended as premunition elevates the pyrogenic density [27]. In such high-transmission areas, a higher proportion of the population is parasitemic on the basis of microscopy or RDT findings, and asymptomatic parasite densities often extend up to 10 000 parasites/µL. Age is a greater determinant of parasite density distributions than in lower-transmission settings. A recent cross-sectional survey conducted in an area of previously high transmission in Papua New Guinea two years after distribution of insecticide-treated bed nets used sensitive nucleic acid based methods of parasite detection and estimated geometric mean parasite densities of 42 genomes/µL for P. falciparum and 8 genomes/µL for P. vivax in identified infections [21]. However, densities were substantially higher in young children (for P. falciparum, approximately 650 genomes/µL; for P. vivax, approximately 140 genomes/µL). Overall, it is likely that more of the "iceberg" of the parasite density distribution in asymptomatic people will be detectable in the microscopy or RDT detectable ranges in settings of high transmission, compared with settings of low transmission.
These data suggest that uPCR assessment of parasite densities in population cross-sectional surveys provides accurate characterization of the true prevalence of malaria in the community in all transmission settings. If the density distributions reported here are general properties of malaria, then the regional and total global burden of malaria parasites can be calculated, and these data used to estimate accurate selection pressures provided by drugs and other interventions.