Introduction

Despite widespread implementation of core malaria control measures, including distribution of long-lasting insecticidal nets (LLINs), improved case detection, and expanded treatment with effective antimalarials, malaria control has stalled in many high-transmission settings in Africa1,2. One explanation for this stagnation may be a failure of current interventions to target the key sources of malaria transmission3,4.

Transmission of Plasmodium falciparum (Pf) from humans to vector mosquitoes occurs when human infections contain gametocytes, the parasite stage that undergoes sexual reproduction in the mosquito. Infected mosquitoes then transmit sporozoites during subsequent blood-feeding on a susceptible human, thus perpetuating the transmission cycle4,5,6,7,8,9,10. Gametocytes develop 8–12 days after asexual stage parasites become detectable in the blood, and typically comprise less than 5% of the parasite biomass in an individual11. Malaria disease is mediated by asexual stage parasites and, thus, individuals who become symptomatic may be treated before gametocytes develop. In high-burden settings, individuals with partial immunity typically suppress high-density asexual-stage parasitemia, avoiding clinical malaria disease and the need for anti-Plasmodium drug treatment. However, these individuals often harbor chronic low-density infections, which can produce gametocytes10. Indeed, gametocytes are often detected in asymptomatic individuals8. Thus, people with symptomatic infection, who are often treated prior to gametocyte development, may not be the primary infection reservoirs who are most important in promoting transmission4,8,9. This can result in unrecognized human-to-mosquito transmission, with longer-lasting, untreated infections being linked to increased gametocyte density10. In high-transmission areas, targeted chemoprevention and vaccination are mostly directed toward children under the age of five (under-5) and pregnant women, the two groups that bear the greatest burden of severe disease and malaria-related mortality. However, decreasing the disease burden among these groups may not decrease population-level infection prevalence. While protecting these groups must continue to be a priority, effective population-level disease reduction in high-burden settings may require specifically targeting other groups who represent important transmission reservoirs.

To date, most studies of gametocyte epidemiology have focused on determining which people with Pf infections produce gametocytes. Parasite commitment to gametocyte production among individuals with Pf infections is associated with the people’s immune response, exposure to anti-malarial drugs, and within-host competition between different parasite strains12,13,14. However, to characterize the most important human transmission reservoir groups, it is necessary to understand the population-level distribution of gametocyte-carriage, which combines those groups most often infected with whether these infections contain gametocytes. Furthermore, transmission to mosquitoes is highly correlated with gametocyte density15,16. Prior cross-sectional surveys have compared gametocyte prevalence and density by age group with symptom status17, but few have quantified the contribution of each group to the overall population-level transmission.

Here we attempt to gain a more complete understanding of gametocyte dynamics in a moderate- to high-transmission setting by analyzing (1) the frequency of gametocyte detection over time, (2) clustering of gametocyte detection within households, and (3) the total abundance (sum of gametocyte densities) of gametocytes in the human population over time. To accomplish this, we conducted a household-based, prospective, longitudinal, open-enrollment cohort study during a 1-year period in the catchment areas of two health centers in rural Malawi. All individuals living in clusters of 5–7 households in four eco-geographic quadrants around each of two health centers were offered enrollment (Fig. 1). Using population-level longitudinal and spatial analyses, we evaluated the individual- and household-level distributions of Pf infections containing any gametocytes, high-density gametocytes, and total gametocyte abundance over time, based on routine, scheduled follow-up visits and passive case detection (PCD) when participants visited the local health center. As school-age children (5–15 years old) have previously been identified as highly prevalent gametocyte carriers, we hypothesized that they would both have the greatest incidence of transmission-relevant infections and make the largest contribution as transmission reservoirs over time.

Fig. 1: Map of Study Locations in Malawi.
Fig. 1: Map of Study Locations in Malawi.
Full size image

Map of health centers (red stars) and households (black dots) in Malawi.

In this setting, school-age children are the primary human reservoirs of transmissible P. falciparum parasites. Among a broad range of predictors and diverse analyses, we demonstrate that being of school-age is consistently the most important factor associated with gametocytemia. As a whole, our results reveal that abundant gametocyte carriage in school-age children occurs in a manner that goes beyond their higher overall rates of Pf infection. While in population-level analyses, children under five also carry a disproportionate number of gametocytes, driven by fewer infections but with higher gametocyte densities primarily detected at active case detection (ACD) visits. However, the majority of gametocytes are found in school-age children. By identifying and targeting the most important Pf transmission reservoir group(s), defined here by the prevalence and density of gametocyte carriage, malaria control interventions may be made more effective.

Methods

Study design

We conducted a prospective, longitudinal, open-enrollment, cohort study in two districts (Machinga and Balaka, Fig. 1) in southern Malawi from April 2019 until May 2020. In the year prior to the study, the estimated incidence of clinical malaria was 195 and 572 per 1000 people in the health-center catchment areas of Ntaja (Machinga District) and Namanolo (Balaka District), respectively. Both areas have perennial transmission, with a seasonal peak typically from January to May. The health catchment area of Namanolo is low-lying, along the Shire River, and received pyrethroid-only LLINs in the 2018 national bed net distribution campaign. The catchment of Ntaja, however, is higher in elevation with rolling hills and streams, and most recently received LLINs containing piperonyl-butoxide in addition to pyrethroid.

During September and October 2018, households in the catchment area of each health center were mapped and enumerated. Each catchment area was divided into quadrants, and four index households were sampled randomly using a random number generator from each quadrant. The “head of household” from each index household and its nearest five-six neighboring households were offered enrollment. Thus, we enrolled sixteen (four per quadrant) clusters of households per health center (32 clusters in total). If households relocated out of the catchment area, became inaccessible during the rainy season, or withdrew from the study, they were replaced. At household enrollment, individual consent was sought from/for each household member. Parents or guardians provided consent for children less than 18 years, and assent was obtained from children above the age of 13.

Participants were invited to attend monthly routine ACD visits at a convenient location every 4–6 weeks after enrollment. At ACD visits, data were collected on bed net use, symptom status, and body temperature, and a finger-prick blood sample was obtained for later detection and quantification of total Pf parasites and gametocytes. PCD for malaria was conducted at the local health center. Individuals who presented to the health center were assessed for fever and other symptoms, tested with malaria rapid diagnostic test (both SD Bioline Malaria Ag Pf, Standard Diagnostics, and Paracheck Pf by Orchid Biomedical Sciences were used based on availability at the health center), and treated as appropriate. A study nurse supported participants’ flow through the regular evaluation process provided by health center staff, and also collected the study-specific blood sample. Individuals who reported symptoms or fever at routine visits were referred to the health center.

Laboratory assays

During both ACD and PCD, finger-prick blood samples were obtained from 50 µl whole blood dried on Whatman 3 mm filter paper and stored with desiccant for Pf infection detection. For gametocyte detection and quantification, 100 µl whole blood was placed in RNA Protect cell reagent (Qiagen, CA) and stored at −80 °C. Methanol extraction was used to obtain DNA from filter papers18, and the Pf 18S ribosomal RNA gene was detected and quantified using qPCR19. Samples positive for Pf underwent gametocyte detection. To detect and quantify gametocytes, male (pfmget) and female (ccp4) gametocyte-specific RNA transcripts were detected and quantified using RT-qPCR20,21. Analyses are based on total gametocyte density. The relationship of total gametocyte density and gametocyte sex ratio stratified by age group is presented in Supplementary Fig. 1.

Statistics and reproducibility

The sample size was calculated based on the estimated prevalence of 8–16% parasitemia from previous research in the setting. Power calculations were done to determine the sample size for 80% power to detect an odds ratio of 1.45 comparing gametocytemia in children and adults.

Variables

Household data was collected within 29 days of household enrollment and included household construction materials, presence of bed nets in the house, socioeconomic indicators, household study site (Namanolo vs Ntaja), and head of household education level (no education, any primary school, and secondary school or beyond). Household construction was categorized as “finished” if at least two of the following were observed: roof, floors, or walls constructed from modern materials. The socioeconomic status (SES) index was calculated using inverse probability weights of standard socioeconomic indicators, including possession of a radio, TV, mobile phone, bike, knives, chairs, tables, bed, mattress, cabinet, livestock, and a private toilet. Elevation (<600 m or >600 m) was correlated with site and was, thus, not considered in the analysis.

Participant characteristics assessed/defined at enrollment included age group [categorized as under-5 years, school-age children (5–15 years), and over-15 years] and sex.

Visit-specific characteristics examined included presence or absence of reported or measured fever, reported use of a bed net the night prior to an ACD visit, PCD (sick visits to the health center) vs routine ACD visit, season of the visit (dry vs rainy), and presence or absence of gametocytes or any Pf parasites at previous visits.

Ever had an infection containing gametocytes

We first examined enrollment and visit-specific characteristics for associations between the following outcomes: ever detected gametocytes during follow-up, detected gametocytes at more than one visit, ever detected gametocytes at a density greater than or equal to 1/µl, and ever detected gametocytes at a density greater than or equal to 10/µl during follow-up. These cut points were chosen based on previous research indicating densities over these thresholds are associated with increased probability of transmission15. Proportions of participants with each of the outcomes were compared based on enrollment characteristics using two-sided chi-squared tests. Gametocyte density was compared by enrollment characteristics using Wilcoxon rank-sum tests.

The probability of gametocyte detection per visit by demographic and visit-specific characteristics was estimated using mixed-effect logistic regression with random effects for individual and household. Log(odds) were derived from the models and converted to probabilities using the inverse logit function. Not all individuals in all selected households were enrolled in the study, and this enrollment was not equal across groups. To account for bias in enrollment demographics, we used an initial census of selected households to create census weights. Probabilities of gametocyte detection were weighted by census-weights.

Rate of gametocyte detection

We calculated the incidence rate of (1) any Pf infection, (2) any gametocyte detection in the total population, and (3) any high-density (≥1 or ≥10/µl) gametocyte detection in the total population per individual participant number of visits by enrollment predictors among the total population. The number of observations positive for gametocytes was overdistributed (variance greater than mean) for all outcomes, so negative binomial regression with an offset for log(follow-up time) was used to estimate the association between enrollment predictors and incidence rate of gametocyte detection.

Odds of gametocyte detection over time

We examined the longitudinal odds of gametocyte detection at a given visit based on enrollment and follow-up predictors. Mixed-effect logistic regression was used to account for clustering due to repeated measures within individuals and multiple individuals per household. To examine the impact of previous Pf, gametocyte detection, or clinical malaria visits, we restricted the analysis to individuals with at least two visits where PCR for Pf was done.

Household clustering of gametocyte detection

To explore how gametocytes were distributed among households, we categorized households into three groups: High-gametocyte households had at least two household members with high-density gametocytes detected on two or more visits. Low-gametocyte households had at least one member with gametocytes detected at least once, but did not meet the criteria for high gametocyte households. No-gametocyte households had no individuals in the household with gametocytes ever detected. To compare whether the clustering of gametocytes could be predicted by the detection of any Pf infection, we compared the proportion of all observations with Pf infections detected in the three household types to the proportion of all observations with gametocytes detected in the three household types using two-sided chi-squared tests.

To better examine the drivers of household clustering, we further evaluated the level of household gametocyte detection by adding up all gametocyte detections in a household, weighting a given detection by 3 if it was a high-density (≥10 gams/µl) gametocyte detection15 to account for the non-linear increase in probability of transmission at high gametocyte densities, and dividing by the total number of person-samples a household contributed to the study (total number of visits of all people in the household). This produced a continuous index of household-level gametocyte intensity in each household. Similar methods were used to examine clustering of any Pf parasites; however, Pf detections were weighted by 5 if they were high-density (>200 parasites/µl)22.

Ordinary least squares regression was used to evaluate household SES, study site, and the proportion of school-age children in the household as predictors of household gametocyte intensity.

Parasitemia/Gametocytemia as a predictor of clinical disease

We examined whether any previous infection or detection of gametocytes was associated with subsequent clinical malaria using mixed-effect log binomial regression models, accounting for household-level clustering. These models were restricted to individuals with at least two visits. We assessed the association between any previous parasite detection, parasitemia at the previous visit, and any previous gametocyte detection or gametocyte detection at the previous visit. The outcome was whether or not a given visit was clinical malaria, as defined as having any symptoms of malaria and being RDT positive. We were unable to look at the association between high-density gametocytemia and subsequent sick visits, as too few high-density gametocyte detections were followed by sick visits.

Household frequency of sick visits was calculated as the fraction of all visits in the household that were sick visits. Ordinary least squares regression was used to evaluate household gametocyte level (as defined above), adjusting for the proportion of children under 5 in the household and accounting for spatial autocorrelation, as a predictor of household frequency of sick visits.

Gametocyte abundance in the population

We quantified the distribution of total gametocyte abundance in the population by examining first the distribution of total summed gametocytes detected across all ACD visits. For instance, the number of gametocytes detected among children under age five was calculated as the sum of all gametocyte density values for all children under five at ACD visits. To look at the difference between gametocyte carriage at ACD vs PCD visits, we calculated the average gametocyte detection for a given individual by visit type for a given season to account for the fact that people had multiple ACD visits, and the number of visits varied by individual. Summing the average number of gametocytes detected for all individuals gives us the average number of gametocytes in the population that would be detected from either sick visits or active surveillance. This allows us to account for not only the number and percent of individuals with gametocytes but also the subgroup contribution to the total gametocyte abundance in the population. To account for bias in enrollment demographics, we multiplied the age-specific numbers of total gametocytes by inverse-probability weights for the age-specific probability of enrolling in the study using census-weights as described above.

Inclusion and ethics statement

This research includes local researchers throughout the research process, is locally relevant, and was designed in collaboration with local partners. All roles and responsibilities were agreed upon amongst collaborators ahead of the research, and capacity building was included in all research activities. Local and regional research is accounted for in the citations. This study was approved by the National Health Sciences Research Committee of Malawi (18/03/2011) and the Michigan State University Institutional Review Board (LEGACY17-361).

Results

From April 2019 through May 2020, 947 individuals in 238 households (median 4 people/household, range 1–11) were followed and tested for Pf infection at 6683 scheduled monthly ACD visits and 416 unscheduled PCD visits where participants presented at study clinics, accumulating 658 person-years of follow-up time (Median: 329 days, IQR: 202–342 days) with a mean of 7.5 visits per person (Supplementaryl Table 1). Due to withdrawals or household inaccessibility, 104 (11%) individuals from 18 households were sampled only once. Children under-5 attended an average of 8.0 visits/child, while males over 15 years of age (over-15) attended the fewest visits (average 5.4 visits/person). Participants had Pf infection detected by qPCR at 23% (1631/7099) of all visits: 22% (1466/6683) of routine visits and 40% (165/416) of PCD visits. Ninety-five percent (1553/1631) of Pf-positive samples were tested for gametocytes by RT-qPCR. Of those tested, 36% (560/1553) were positive for gametocytes with a mean of 0.85 gametocyte-positive visits per person-year during the study. Among all visits, 8% (560/7099) were positive for gametocytes, and only 1% (85/7099) had high gametocyte density associated with increased risk of transmission (≥10 gametocytes/µl)15 (Table 1).

Table 1 Gametocyte detection, probability of gametocyte detection, median gametocyte density, and proportion with high-density gametocytes detected in the study population (N = 947 individuals) by fixed covariates assessed at enrollment
Full size table

Gametocyte-containing infections are not evenly distributed in the study population

Gametocyte-positive observations occurred unevenly across the population, with only 23% (216/947) of participants ever having an infection with detectable gametocytes (Table 1; Fig. 2). Among the 216 participants with gametocytes detected, 118/216 (55%) were gametocyte-positive at multiple visits, and 91/216 (42%) had gametocytes detected at sequential visits with a mean of 2.6 gametocyte-positive visits per person, a maximum of 10 gametocyte-positive observations for any one individual (Fig. 2). Visits with infections containing high gametocyte densities (≥10 gametocytes/µl), which are associated with increased risk of transmission15, occurred in only 6% of the study population (54/947) (Table 1). These high gametocyte density infections were clustered among individuals who had multiple gametocyte-positive visits, with 39% of individuals with multiple gametocyte-positive visits having high-density infections vs 8% of individuals with only a single gametocyte-positive infection (aOR = 6.7, 95% CI = 3.1, 16.2) after adjusting for the number of study visits.

Fig. 2: Gametocyte-containing infections cluster among individuals and households.
Fig. 2: Gametocyte-containing infections cluster among individuals and households.
Full size image

Heat map showing all Pf infections and gametocyte-containing infections in the population at routine visits over time by quadrant and household. Each tile represents a household. Each row within a tile (y-axis) represents a single participant over time, and each column (x-axis) represents study visits over time with the infection status for a given participant at a given visit indicated by color: no Pf detected (green), Pf detected but without gametocytes (gray), low density gametocytes detected (<10/μl, red), high density gametocytes detected (≥10/μl, dark red), missing visit (white). Households are sorted by the most frequent gametocyte infections (top) to the least.

We evaluated the proportion of the study population with any gametocyte-positive infections, more than one gametocyte-positive infection, mean gametocyte density, and gametocyte density cutoffs associated with increased likelihood of transmission to mosquitoes, using fixed covariates assessed at enrollment (Table 1) and visit-specific covariates (Supplementary Table 2). Comparisons by age-group revealed that a higher proportion of school-age children (5–15 year-olds) had at least one gametocyte-positive observation during the study (37%) compared to younger children (under-5) or adults (over-15) (13 and 16%, respectively, p < 0.001). School-age children had the highest probability of having gametocytes detected at any visit compared to all other demographic and visit-specific characteristics, with a probability of 0.015, which was an order of magnitude higher than all other predictors (p < 0.001). School-age children were also more likely to have ever had a high-density gametocyte infection (23%, p < 0.001) compared to other age groups, although their median gametocyte density over the study was lower than that of younger children and comparable to adults. Participants living in households with the lowest tertile of SES and living in unfinished houses (as opposed to finished houses) were significantly more likely to have had at least one gametocyte-positive visit (p < 0.01) and to have had gametocytes detected during multiple visits (p < 0.001), respectively (Table 1). Among covariates that were measured at each follow-up visit, gametocytes were more likely to be detected when participants reported they did not sleep under a bed net the night prior to the visit (p < 0.01). Furthermore, visits with higher gametocyte densities were more likely to occur during the rainy season compared to the dry season (p < 0.05 for gametocyte densities >1 gametocyte/μl) (Supplementary Table 2).

These findings contrast with the distribution of total parasitemia in the population, where a majority (72%: 678/947) of all participants had at least one Pf infection detected during the study (Supplementary Table 3). Among individuals who ever had Pf infections, 32% had gametocyte-containing infections. School-age children were both significantly more likely to have Pf infections (p < 0.001) and, among individuals with Pf infections, were twice as likely to have gametocyte-containing infections as other age groups (p < 0.001). While the type of visit (routine ACD vs. PCD) was not a significant predictor for gametocyte detection, participants were significantly more likely to have Pf parasites detected and to have higher parasite densities at PCD visits compared to routine visits. Despite this, among visits with Pf infections detected, routine visits were significantly more likely to have gametocytes detected than PCD visits (Supplementary Table 3). These findings indicate that transmissible, gametocyte-containing infections are not a random subset of all Pf infections.

Measuring parasitemia alone underestimates the contribution of key risk factors to transmission

Longitudinal follow-up allowed us to calculate the incidence rate of Pf infections, gametocyte-containing infections, and infections with high-density gametocytes over time. We estimated the importance of predictors for the incidence rate of each outcome, accounting for the individual number of visits, using negative binomial models (Fig. 3; Supplementary Tables 4–6). Overall, the average incidence rate was 0.85 gametocyte detections per visit and the predictors of the incidence rate of infections containing any gametocytes or high-density gametocytes were similar to predictors of the rate of any Pf infection; however, the magnitude of association, the incidence rate ratio, was significantly larger for predicting gametocytes than Pf infection across all predictors (Fig. 3). School-age children had significantly higher incidence of infections containing gametocytes than other age groups, with a rate ratio (RtR) of 2.50 (95% CI 1.6–4.0) compared to under-5s, and this relationship was nearly double the effect size for the association between age and any Pf infection. While school-age children had more frequent high-density gametocyte infections than children under 5, this relationship was not statistically significant. Reported net use at enrollment and higher SES were both significantly associated with lower rates of any gametocytes or high-density gametocytes. Incidence rates, rate ratios, and confidence intervals for all three outcomes are reported in Supplementary Tables 4–6.

Fig. 3: Incidence rate ratios (box at center of error bar) and 95% confidence intervals among the total study population (N = 947) for P. falciparum detection (gray), gametocyte detection (orange), and high-density gametocyte detection (blue) over time from negative binomial models accounting for individual number of visits (Supplementary Tables 3–5).
Fig. 3: Incidence rate ratios (box at center of error bar) and 95% confidence intervals among the total study population (N = 947) for P. falciparum detection (gray), gametocyte detection (orange), and high-density gametocyte detection (blue) over time from negative binomial models accounting for individual number of visits (Supplementary Tables 3–5).
Full size image

Education levels refer to the head of household’s education level. SES socioeconomic status, REF reference group.

Visit-specific longitudinal analyses showed increased odds of gametocyte-containing infections at PCD visits and visits with fever (Table 2). However, visits with fever and PCD visits were rare (12% and 6%, respectively), limiting their contribution to the population-level gametocyte reservoir.

Table 2 Association among the total population (N = 947) from 7099 observations between having any gametocytes detected and visit-specific predictor variables from mixed-effect logistic regression, accounting for repeated measures and clustering by household
Full size table

Gametocyte-containing infections cluster at the individual and household levels

To quantify the clustering of gametocyte-containing infections among certain individuals over time (Fig. 2), we examined the association of previous Pf infection or gametocyte detection on subsequent gametocyte carriage, accounting for repeated observations among individuals and clustering by household. Analyses were restricted to the 843 individuals with at least two visits with laboratory results, resulting in 6152 observations where we could assess previous parasite or gametocyte detection (Table 2). Overall, neither ever had any prior Pf infection, nor having Pf infection at the previous visit was associated with having a gametocyte-containing infection at the current visit. Ever having previously had symptomatic malaria was associated with a significant decrease in later gametocyte detection (OR: 0.57, 95% CI: 0.35, 0.93), likely due to parasite clearance after clinical treatment. However, having any prior infection containing gametocytes was strongly associated with having gametocytes at the current visit (OR: 3.82, 95% CI: 1.75, 8.33), confirming clustering of gametocyte-containing infections among a subset of individuals. Among the 216 individuals with gametocytes detected, 128 (59%) had gametocytes detected on the same date as their first detected Pf. infection, with a mean of 56 days and a maximum of 346 days between first Pf. detection and subsequent gametocyte detection, with no difference by age, symptoms, or any other characteristics.

We assessed whether increased incidence of gametocyte-positive visits within certain individuals was explained by clustering at the household level and found that most gametocyte-containing infections clustered within a few households, with high-gametocyte households clustering by study site (Fig. 2). After restricting analysis to households with at least 2 follow-up visits (nhousehold = 220, nindividual = 890), we found that only 16% (34/220) of households had gametocytes detected at multiple time points among multiple members. These 34 “high-gametocyte” households contained 22% (195/890) of the study population and 63% (349/557) of gametocyte-positive visits (Table 3). This was significantly different (p < 0.001) from the corresponding distribution of Pf infections in the population, with 25% of all Pf infections occurring in households where there were no gametocytes detected at any time during follow-up.

Table 3 Distribution of households, individuals, P. falciparum infections, and gametocyte-containing infections by household types among all households with at least two completed study visits
Full size table

To examine the drivers of household-level clustering of gametocyte containing infections, we created a continuous variable for “household level of gametocyte detection” which accounted for the number of people in a household, the number of visits each person contributed, and the density of gametocyte-containing infections detected (Fig. 4). In adjusted linear regression models accounting for household Pf infection levels, neither household SES nor site were associated with household level of gametocyte detection (Supplementary Table 7). However, the increasing proportion of school-age children in the household remained highly predictive of increasing household level of gametocyte detection, even after adjustment for Pf infection levels (p < 0.01). This suggests that school-age children drive household-level gametocyte clustering, independent of Pf infections in the household.

Fig. 4: Spatial clustering of proportion of the household comprised of children aged 5–15 (left), household gametocyte levels (center), and total household Pf infection levels (right) in Namanolo (top) and Ntaja (bottom).
Fig. 4: Spatial clustering of proportion of the household comprised of children aged 5–15 (left), household gametocyte levels (center), and total household Pf infection levels (right) in Namanolo (top) and Ntaja (bottom).
Full size image

Red star indicates the location of the health center. Dot size represents household size. Gametocyte and Pf infection levels are a combination of density and frequency of detection (data points have been jittered for visualization to improve visibility and maintain anonymity).

No association between gametocyte detection and subsequent sick visits

We examined the association between detection of gametocytes or Pf and whether or not subsequent visits were clinical malaria. Neither any Pf infection, Pf infection at the previous visit, or previous gametocyte detection were associated with increased risk of subsequent sick visits (Table 4).

Table 4 The association between the detection of any P. falciparum or detection of gametocytes on risk of the subsequent visit being clinical malaria vs a routine visit (ACD), restricting to participants with at least two visits, using mixed-effect log binomial models accounting for household clustering
Full size table

We next examined whether or not increasing household-level gametocytemia was associated with increased risk of household sick visits. As expected, households with higher proportions of children under 5 in the household had significantly more sick visits. However, there was no association between either household levels of gametocytemia or household levels of Pf infection, with or without accounting for spatial autocorrelation.

The majority of gametocytes are detected in school-age children

Finally, we estimated the distribution of gametocytes in the population over the total study period, accounting for gametocyte density, frequency of gametocyte-containing infection, frequency of follow-up visits, and probability of enrolling in the study. Overall, 82% of people counted in our census of the study area were enrolled in the study. However, the probability of being enrolled in the study differed significantly by age, which our results showed was highly associated with gametocyte carriage. Individuals over 15 years of age were the least likely age group to enroll in the study (24% not enrolled). After accounting for age-specific probability of enrollment, plots of total gametocyte abundance over the study period during ACD visits by population characteristics were produced (Fig. 5).

Fig. 5: Distribution of total gametocyte abundance and total parasite abundance in the population.
Fig. 5: Distribution of total gametocyte abundance and total parasite abundance in the population.
Full size image

Distribution of total gametocyte abundance (left) and total parasite abundance (right) in the population by characteristic, restricted to routine visit (ACD) observations with reported net use (missing = 15), weighted for age-distribution in the census. N = 6292 visits. Footnote: *Among children under 5 during the rainy season, reported net use was sufficiently common that the number of gametocytes among non-net users was negligible (0.2% of total) and cannot be seen.

Of the total detected gametocyte abundance (sum of gametocyte densities) determined from routine ACD visits, and after adjusting for age-related enrollment bias, 53% of gametocytes were detected among school-age children who made up only 33% of the census population. A proportionately higher fraction of gametocytes was detected among under-5 children, with 40% of gametocytes detected among under-5 children who represented only 18% of the census population. The remaining 7% of gametocytes were detected in adults (49% of the population) (Fig. 5a).

Among school-age children and adults, most gametocytes were detected during the rainy season; however, among younger children, more gametocytes were detected in the dry season. This suggests that clearing all infections from school-age children during the rainy season would lead to a 67% decrease in the total gametocyte abundance in the population at that time, compared to similarly treating all under-5 children during the rainy season, which would decrease total gametocyte abundance by only 25%.

To examine the difference in gametocyte abundance between ACD and PCD visits, we calculated the average number of gametocytes per individual by season and type of visit to account for the repeated ACD visits and the differential number of ACD visits between participants (Fig. 6). PCD visits comprised only 6% of gametocyte-positive visits. Furthermore, gametocyte densities were low at these health center visits; thus, gametocytes detected during PCD visits were only 11% of all gametocytes in the population.

Fig. 6: Average gametocytes and total parasites detected per visit.
Fig. 6: Average gametocytes and total parasites detected per visit.
Full size image

Seasonal average gametocytes (A) and total parasite abundance (B) detected by season and visit type from 947 individuals, calculated from individual visit-type averages and weighted by age-specific enrollment probability. ACD active case detection, PCD passive case detection.

We compared the distribution of gametocytes to the distribution of total Pf abundance using total Pf parasite density. At routine visits 84% of all Pf parasites were detected among school-age children, with only 7% detected among adults, and 10% among under 5 children (Fig. 5). While the number of both gametocytes and total parasites was highest in school-age children, when comparing the treemaps for gametocytes and all Pf infection, the distribution of total parasite distribution by age group were significantly different than the distribution of gametocytes in the population (p < 0.0001). Among school-age children, most parasites were detected during the dry season, in contrast to the distribution of gametocytes.

When examining the difference between average ACD visits and PCD visits (Fig. 6B), most Pf parasites in the population were still detected in school-age children. In contrast to gametocyte abundance, more than half of all parasites among school-age children, and more than half of all parasites found in other age groups, were detected at PCD visits (Fig. 6B). While clearing parasites among school-age children at PCD visits alone would clear 45% of the total parasite abundance, it would decrease only 4% of the total gametocyte abundance in the population.

Discussion

At a population level, school-age children in our study sites were the primary human reservoirs of transmissible P. falciparum parasites. Among a broad range of predictors and diverse analyses, we demonstrated that being of school-age was consistently the most important factor associated with gametocytemia. As a whole, our results reveal that abundant gametocyte carriage in school-age children occurs in a manner beyond their higher overall rates of Pf infection. Both school-age children and children under five carry a disproportionate number of gametocytes, which were primarily detected at ACD visits. Thus, including school-age children in interventions focused on clearing asymptomatic infections would remove a substantial portion of the gametocyte reservoir, suggesting such targeted interventions could disproportionately reduce transmission in the community.

Defining the contributions of specific human reservoir groups of Pf transmission is complex and, as we demonstrated here, not simply a reflection of Pf infection risk. While Pf infection is, of course, necessary for gametocyte development, the distribution of gametocytes in the population is not random among people with Pf infections. While 72% of our study population had Pf infections detected at some point during follow-up, only one-third of these infected people ever had gametocytes detected. Likewise, previous Pf infection was not associated with later gametocyte detection, while previous gametocytemia was predictive of subsequent gametocyte detection. Although age is a strong predictor of both Pf infection and gametocyte detection, the association between age and gametocyte detection was stronger than the association between age and any Pf infection. School-age children with Pf infection were also more likely to have gametocytes than other age groups. These findings agree with prior studies evaluating gametocyte carriage among individuals with Pf infections, which have shown that people’s age is a strong predictor of gametocyte detection17,23,24. However, by using longitudinal data of a carefully characterized cohort, we showed that school-age children contribute disproportionately more to gametocyte carriage than would be expected by Pf infection rates alone, because gametocyte carriage occurs more frequently, and at transmission-relevant densities, than seen in other age groups.

Our finding that gametocyte-containing infections occur more often within certain individuals and households is intriguing. This gametocyte clustering was consistent across an array of longitudinal analyses that compared prior gametocyte carriage and prior Pf infection as predictors of subsequent gametocyte carriage, and age was the strongest predictive factor. We anticipated that the age group would explain most of the variation, as age is also a surrogate for developing immunity, with most school-age children having achieved immunity to clinical malaria disease, permitting chronic infections with higher parasite densities. Development of immunity to early-stage gametocytes25 could result in some individuals, especially older children and adults, being less likely to have gametocyte-containing infections. Environmental factors that elevate mosquito vector density will inherently influence the distribution of Pf infections, but could additionally contribute to gametocyte carriage, as they reflect local heterogeneity in exposure and longer-term development of immunity. We did find that the number of school-age children in a household was associated with increased household gametocyte carriage, but not associated with Pf infection. Other studies have also described how a few highly infectious individuals can drive transmission10.

This study was able to quantify the relative contribution of clinical disease and asymptomatic infections to gametocyte abundance in the population because we collected information from a large number of community members of all ages and both sexes across high- and low-transmission seasons, using both routine ACD spaced regularly over time and PCD of study participants presenting to the health center. Specifically, we observed that most gametocytes were detected during routine ACD visits. This was most stark among children under 5 who carried a disproportionately large fraction of gametocytes, detected almost primarily at ACD visits. These results are consistent with prior studies of human Pf infection showing that gametocytes are frequently detected in asymptomatic infections26,27,28 and indicate that individuals who present for sick visits are generally less likely to harbor gametocytes. Our results suggest that interventions aimed at improving case management or adding transmission-blocking drugs to treatment regimens are unlikely to substantially reduce transmission, as most gametocytes in the population are detected outside of health center visits.

This population-based study with intensive follow-up is one of the largest and most comprehensive examinations of population-level gametocyte distribution in a moderate-to-high transmission setting. However, there were a few limitations. First, the study was designed to capture, to the best of our ability, the total population, allowing household members to enroll over time if they missed the initial enrollment visit. Despite this, a significant fraction of adult males in the census population did not enroll in the study, and among those who did enroll, adult males contributed fewer observations than any other group. We attempted to correct for this by weighting the study observations by the census population distribution. However, the baseline census was done during the household registration of the national net distribution campaign, which may have led to the inclusion of non-permanent household members. Furthermore, many men in the study area work outside of the district or country. Thus, we may have overestimated these individuals’ contributions to transmission if they were truly not present in the study area during the study period. Second, easier access to care at the local health center due to the presence of a study nurse to support participants may have increased the likelihood that participants would seek care. This may have biased our results toward having more PCD visits than in a non-study population. Third, our study took place in two districts in rural Southern Malawi, where malaria transmission is perennial with a seasonal peak. While transmission patterns are geographically specific, the epidemiologic patterns we observe in terms of the distribution of infection are similar to those in other areas in Malawi and across the region. Thus, our results are likely generalizable to these areas. Lastly, we only tested for gametocytes in visits that participants had previously tested positive for Pf from dried blood spots. Given the lower sensitivity of DNA detection from dried blood spots compared to RNA detection from whole blood, it is possible we may have missed some low-density Pf infections, which were thus never tested for gametocytes; however, this would lead to missing only very low-density gametocyte-containing infections and is unlikely to substantially bias our results.

Previous research indicates that targeting malaria interventions to school-age children can lead to indirect decreases in transmission among other groups29,30,31. Large cluster-randomized trials have found that, even with imperfect coverage, targeting parasite-clearing treatment to all school-age children leads to decreased community-level malaria prevalence29,30. The community-level impacts of clearing infections in school-age children have also been predicted by agent-based simulation models across transmission settings31. Previous research has suggested that school-aged children are additionally significantly more likely to be bitten by mosquitoes than younger children32, and future analyses are planned to estimate the specific contribution to transmission, accounting for biting rates as well as actual human-to-mosquito transmission rate. Here, we provide evidence from a large, well-characterized population suggesting that clearing infections in asymptomatic school-age children could reduce the total population of transmission-relevant malaria parasites in the community by more than half, and that it will be critical to include school-age children in such interventions. In combination with previous trials, we believe this is strong evidence to support recommendations to target school-age children in moderate-to-high transmission settings with additional interventions to clear and prevent infections. Indeed, progress toward further malaria control and elimination may not be achieved until infections in this reservoir group are reduced.