Background

Differential and biased feeding by vectors on particular human demographic groups can have profound impacts on malaria epidemiology and may permit Plasmodium transmission to persist even where control measures are in place [1,2,3,4,5,6,7]. One approach to quantifying patterns of feeding on human hosts of different ages and sexes is with blood meal analysis [8]. Blood meal analysis serves to estimate the human blood index (proportion of hosts that are humans) as well as to estimate the proportion of mosquito infections originating from different demographic groups in the human population [3, 9, 10].

In Malawi, prevalence of Plasmodium falciparum infection is highest in school-age children (SAC) [11,12,13]. Furthermore, infections in this age group are more likely to contain gametocytes, the sexual stage of the parasite required for human-to-mosquito transmission [14], suggesting that SAC may be the most important reservoir of infection [12, 15]. However, to quantify the contribution of this age group to transmission requires determining the frequency with which vectors take blood meals from this age group in comparison to other age groups.

Genetic profiling of human blood meals has been applied in various transmission settings to identify the most frequently bitten population group, assess the frequency distribution of mosquito biting on individuals, and estimate the proportions of multiple blood meals on different human hosts [3, 9, 10]. It utilizes microsatellite genetic markers, or short tandem repeats (STRs) of nucleotides, which occur at thousands of Mendelian loci in the human genome with multiple alleles operating at each locus [16]. The resultant genetic variation provided by the combination of numerous loci and multiple alleles provides for various applications in research such as studies of population genetic structure [17] and in forensic sciences [18, 19].

Previous studies prior to widespread access to bed nets suggested that, based largely on differences in body surface area, a greater proportion of blood meals are taken from adults and older children than young children and infants [20, 21]. SAC are the least likely age group to report sleeping under bed nets in Malawi and elsewhere, suggesting they may be more available to malaria vector feeding [13, 22, 23]. Furthermore, Plasmodium-infected people may be more attractive to mosquito vectors than uninfected people [24, 25].

To assess the relative contribution of human age and sex demographic groups to mosquito feeding and parasite transmission, this study compared genotypes of human blood from mosquito blood meals to the blood samples taken from humans sleeping in households where the mosquitoes were collected.

Methods

Study site

Mosquito and human samples were collected as part of a household-based cohort study conducted in two districts (Machinga and Balaka) in southeastern Malawi where incidence of malaria is moderate to high throughout the year [26]. According to the Malawi National Statistical Office census report for 2018, Machinga district had a total population of 735,438 with a population density of 205/km2 on a 3582 sq. km land area, while Balaka district had a total population of 438,378 with a population density of 205/km2 on a 2142 sq.km land area. The residents of the area are mostly subsistence farmers whilst others ply trade in market centers. Malaria transmission is highest in areas of high temperature and seasonal rainfall, from October through April [27]. Long-lasting insecticidal nets were distributed via a mass campaign in September–November 2018. In the study sites 91% of households reported owning at least one net and 74% of residents reported sleeping under a net (Malawi ICEMR, unpublished data).

Household surveys

Monthly surveys of a household-based cohort were conducted to collect demographic and malariometric data in 46 households in six clusters. Clusters were created as part of a larger study (detailed methods not yet published, [28]) by randomly selecting an index household and inviting 5–7 nearest neighbours to also enroll in the study (Fig. 1). A questionnaire designed to gather information on the number of people per household, and the name, sex and date-of-birth of each household member was administered to each household head who consented to participate. Blood was collected by finger prick from consented household members, dried on Whatman filter paper, and stored under cool condition.

Fig. 1
figure 1

Map of Malawi showing the study site and sampling cluster locations

Full size image

Indoor-resting, female Anopheles mosquitoes were sampled from the sleeping spaces of the enrolled households using resting collections, pyrethrum spray catches, and CDC miniature light traps. The households were sampled one night every 2 weeks from February to April of 2020, yielding 4 to 6 nights of mosquito collection per household. The individuals who slept in the house the night prior to sample collection were considered to be at risk. The total nights at risk was calculated as the sum of the number of collection nights in which each participant was recorded as having slept in the household.

DNA extraction

Each blood-fed Anopheles mosquito was bisected into the head attached to the thorax (hereafter, “head-thorax”), and the abdomen, using sterile technique. Genomic DNA was extracted from the mosquito abdomens and from human blood on filter papers using the DNeasy® Blood & Tissue Kit (Cat. No. 69582; Qiagen, Valencia, CA, USA).

Anopheles species identification

Mosquitoes morphologically identified as members of either the Anopheles gambiae complex or Anopheles funestus group [29] and subjected to polymerase chain reaction (qPCR) methods described elsewhere [26] to determine the Anopheles species.

Plasmodium falciparum detection

Presence of P. falciparum parasites in human blood samples and mosquito abdomens was tested using a qPCR method targeting the 18S rRNA gene [3].

Identification of human blood meals

DNA samples obtained from blood-fed mosquito abdomens were subjected to a qPCR method containing primers (forward 5′-GGCCTGTTCCTCCCTTATTT-3′, reverse 5′-TACACAGGGCTTCCGAGT-3′) and probe (FAM-ATGGAGTCTGTGTTCCCTGTGACC-QSY) that detects intron 1 of the human tyrosine hydroxylase gene [26, 30].

Microsatellite genotyping and profile matching

Human blood from mosquito blood meals and blood samples drawn from people were subjected to a microsatellite genotyping method [3]. The method involved PCR-amplification and genotyping of 23 well-characterized microsatellite loci, plus a sex-linked gene (amelogenin) locus to identify the sex of an individual [3, 31] (also see Supplementary Text). Amplification was performed using a validated, commercially available multiplex PCR kit (Promega, PowerPlex Fusion System; catalog no. DC2402) that contained locus-specific fluorescent-labelled primers [31]. PCR products were analysed by capillary electrophoresis (ABI 3730 Genetic Analyzer, Applied Biosystems, Foster City, CA) with LIZ 500 (Applied Biosystems, Foster City, CA) as the internal size standard. GeneMapper software version 4.1 (Applied Biosystems) was used to visualize the allelic sizes of all the loci in each sample (Additional file 1: Fig. S1). For each sample (i.e., human-fed mosquito or human blood), all observed allele sizes at each locus were listed and together served as the genetic profile of the person (Additional file 3: Table S2 and Additional file 4: Table S3). While different mosquitoes that fed on the same individual had the same genetic profile, each study participant had a unique genetic profile, except for monozygotic twins, which are extremely rare in the study communities. The genetic profile of a mosquito blood meal from a household was compared to the genetic profile of all members of that household, and to the entire human survey, to determine which person was fed on by the mosquito.

Data analysis

Genotyping data were analysed as previously described [3] using R software (Version 4.0.2; https://www.R-project.org/), and an R script that compared the genotype of each locus in a genetic profile to its corresponding locus in other genetic profiles. The similarity of the two genetic profiles was expressed as the proportion of identical alleles at all loci times 100. For example, 19 random loci with identical genotypes divided by 24 total loci multiplied by 100 returns a 79% profile similarity. For mosquitoes whose blood meals originated from the same person, the profile similarity would be 100%. However, similarities less than 100%, for blood meal samples obtained from the same person could result from the so-called “allele dropout” effect which is a failure of one or more alleles at a locus to be detected at all because of low DNA concentration or quality [32]. To account for this possible source of error, and to minimize its effect in obtaining a reliable comparison of genotypes between pairs of blood meal samples and between blood meals and human blood samples, a value less than 100% must be used as the criterion for establishing a match from same source. To determine this value empirically with confidence, a pairwise matching analysis of genotypes was performed on the genetic profiles of every individual human in the study sites who provided a blood sample. That is, for each person, his/her genetic profile was compared with the genetic profile of every other individual in the study site and the similarity value of each pairwise match was recorded. It is worth noting that unlike the blood meal samples, which are prone to allele drop out due to low DNA quantity (small blood meal volume) or quality (partially digested blood), the DNA isolated from the human blood samples had higher quantity and quality and produced quality genetic profiles that warranted their use in the pairwise analysis for the match criterion. The pairwise comparison process generated [n(n-1)]/2 similarity values for profile matches, where n is the total number of humans sampled at a site. A similarity value higher than the highest value in the pairwise match output (excluding match results for monozygotic twins, which would be 100%) was then established as the criterion below which two genetic profiles were considered to be different [3].

Blood meals from multiple humans that could not be matched to any single genotype, owing to the presence of more than two alleles at many loci, were excluded from any matching analyses. After excluding the mixed blood meal profiles, genetic profiles of single-human blood meals were compared to each other and to the number of different human individuals in the blood meal sample, to determine the number of unique genetic profiles present in that sample. The frequency of occurrence of each unique genetic profile in the blood meal sample represented the number of mosquitoes that fed on each individual human. These data were used to construct frequency distribution histograms that relate the number of different human individuals (y-axis) to the number of blood meals they provided (x-axis). The observed frequency distributions were fitted to zero-truncated Poisson and zero-truncated negative binomial frequency distribution models using the functions zerotrunc and rootogram of the R package countreg [33]. The fit of the two distributions was compared by x2 test to determine if selection of humans by mosquitoes was random (Poisson) or non-random and aggregated (negative binomial). Genotypes of human blood-fed mosquitoes (blood meals) were compared to each other, and to genotypes of human individuals.

Genetic profiles of blood meals were also compared to genetic profiles of individual humans. Because a sex-specific marker was included in the genotyping process, the sex of the source of all human blood meals could be determined without the need to match the blood meal profiles to those of the study participants. However, the age of the source of human blood meals could only be determined for blood meals that matched study participants. Age was analysed categorically as ≤ 5, 6–15, 16–30 and 31–75 years old. The proportion of different sex and age groups observed in the mosquito blood meals was compared with their expected proportion (the proportion of individuals of that group in the household survey which represented the population at risk) using two-tailed binomial tests. Blood meals from multiple humans that could not be matched to any single genotype, owing to the presence of more than two alleles at many loci, were excluded. Proportions of infected individuals across age categories were compared using contingency tables and x2 test for trend. Logistic regression analyses were used to confirm associations, but data are presented in tables for clearer interpretation.

Results

A total of 5,988 Anopheles spp. mosquitoes were collected from repeated sampling of 46 households (Figs. 2 and 3). Among these, 633 mosquitoes were blood-fed with 499 having fed upon humans (human blood feeding index 78.8%). Two-hundred and fifty people slept in households in which these mosquitoes were collected. Samples were available from 190 (76%) of these people.

Fig. 2
figure 2

Flow chart summarizing the number of humans enrolled in the study, and blood samples collected, processed, and compared with blood in Anopheles mosquito abdomens

Full size image
Fig. 3
figure 3

Flow chart summarizing results of Anopheles mosquito blood meals from collections in study households

Full size image

Similarity of human genetic profiles

Quality genetic profiles were successfully generated from 165 (87%) of the 190 sampled human individuals (Fig. 2). The genetic profiles of the remaining 13% of people were of poor quality and thus excluded from subsequent analyses. A comparison was made of each genetic profile with all others, including between siblings, to determine the degree of similarity, resulting in 29,161 pairwise genotype matches. There were 20 unique similarity values ranging from 0 to 79%, each with a varying frequency (probability) of occurrence (Additional file 2: Table S1). The highest similarity value of 79% (i.e. 19 of 24 matched loci) had a probability of occurrence of 10–5 (Additional file 2: Table S1), which is extremely low and there were only two such matches. This result means that no two non-monozygotic individuals including siblings in the study sites had a pairwise similarity value greater than 79%. Thus, this value was chosen as the criterion above which two blood meal genetic profiles were considered to be from the same human source. Conversely, samples with pairwise values ≤ 79% were considered to be from different human sources. It is worth noting here that the absence of 100% match value in the data indicates that the study participants (not the entire population) did not include monozygotic twins. Although monozygotic twins could be present in the population and the blood meal profiles from the two different individuals could be confused for a single person, the effect of this error on the data is negligible given the rarity of monozygotic twins in a population.

Genotyped human blood meals

Of 499 Anopheles spp. mosquitoes with blood meals containing human blood, 468 (93.8%) yielded analyzable human genotypes (Fig. 3). Of these, 424 (90.6%) were found to contain a single genotype whilst 44 (9.4%) contained mixed blood meals from two or more humans and could not be genotyped to identify unique individuals, and hence were not used in the matching process. The 424 female Anopheles mosquitoes containing blood meals with a single genotype included Anopheles arabiensis (187, 44.1%), An. gambiae stricto sensu (66, 15.6%), An. funestus (150, 35.4%), Anopheles parensis (1, 0.2%), Anopheles vaneedeni (1, 0.2%) and unidentified Anopheles species (19, 4.5%). When the genotypes of 424 single-human blood meals were compared with the 165 human blood genotypes of the household residents, 180/424 (42.5%) blood meals matched to 46/165 (27.8%) of the study participants. The other 244/424 (57.5%) blood meals contained profiles from 119 unique humans whose genotype did not match the genotype of any of the household residents enrolled in the study and represented in the sample.

Frequency distribution of human host selection

The number of blood meals taken from individual people ranged from 1 to 49. The frequency distribution of genotyped blood meals taken from individual humans showed that some people were more frequently fed upon than others. More than 60 individual humans contributed only one blood meal, each to a single mosquito, while three male humans each individually contributed blood meals to 15, 19, and 49 mosquitoes, respectively (Fig. 4).

Fig. 4
figure 4

Frequency distribution of number of blood meals on 424 individual male (n = 256) and female (n = 168) humans taken by Anopheles mosquitoes sampled in communities of southeastern Malawi, showing the number of mosquito bites per person by sex

Full size image

The frequency distribution fit the negative binomial distribution well (x2 = 9.4306, df = 9, P = 0.3985), but did not fit the Poisson distribution (x2 = 82.396, df = 6, P =  < 0.0001), indicating a nonrandom and aggregated (or clumped) pattern of human host selection.

Mosquito feeding on human hosts by sex and age

Categorization of all genotyped blood meals into male and female human blood source and comparing the data to the proportion of males and females in the population at risk revealed that males contributed a significantly higher proportion of blood meals than females (Table 1). Among successfully matched blood meals from which the age of the human blood source was known, children ≤ 5 years of age were under-represented in mosquito blood meals compared to their availability in the population at risk of being bitten. School-aged children (SAC, 6–15 years old) and young adults (16–30 years old) were represented in mosquito blood meals in equal proportion to their presence in the population. However, older adults (31–75 years old) were overrepresented in mosquito blood meals. Comparisons by sex within each age group revealed that males are the primary source of blood meals in the older age groups. Regression analysis including the interaction of sex and age group confirmed these associations (Table 1).

Table 1 Comparison of population at risk of being bitten (expected proportions) successfully matched genotypes of human individuals blood meals sources of different sex and age groups (observed proportions)
Full size table

Plasmodium falciparum infection in Anopheles blood meals

Plasmodium falciparum was detected in 151 (36%) of the 424 genotyped blood meals of Anopheles mosquitoes. The proportions of infections were similar for blood meals taken from females (n = 59, 35%) and males (n = 92, 36%) (Table 2), and results were similar if analyses were restricted to blood meals that matched study participants. Of the successfully genotyped and matched blood meals (n = 180), the proportion of Anopheles blood meals positive for P. falciparum decreased as age of the human source of the blood meal increased (p < 0.0001, test for trend). While the highest proportion of infected blood meals was found in blood meals obtained from young children (≤ 5 years old), the largest number of infected blood meals was obtained from SAC (6–15 years old). Data were too sparse to make more meaningful inferences on the distribution of infection by sex within age group (Additional file 3: Table S2). Thus, overall, SAC contributed the most infection to mosquitoes, thereby further perpetuating transmission (Table 2).

Table 2 Comparison of blood meals sources containing P. falciparum parasites by sex and age of the human genotype detected in the blood meal
Full size table

Discussion

This study shows that in the setting of current net ownership and use patterns, Anopheles mosquito feeding on humans was non-random, with more blood meals taken from older males. However, school-age and under-five children transmitted more P. falciparum infections to mosquitoes, both relatively and absolutely. These results underscore the importance of understanding both vector feeding behaviour and human infectiousness in designing targeted interventions for malaria control. Although they were more frequently bitten, adult males may not be the most important target for malaria control through decreased vector-human contact. Instead, reduced population-level P. falciparum transmission might be better achieved by decreasing mosquito feeding on the more infectious ≤ 5 year olds and SAC.

The finding that human-host blood feeding was non-random and aggregated has also been reported in other studies using genetic profiling of Anopheles, Culex, and Aedes blood meals [34,35,36,37]. Non-random feeding is likely due to differences in mosquito access to individual humans or houses as a result of differences in the distance from larval habitat, use of LLINs and repellents, late night activity, cooking location, and/or house construction that impedes mosquito entry [38, 39]. Also, humans are more likely to encounter host-seeking Anopheles where these mosquitoes are abundant, and during activities that may differ according to people's age and sex [40, 41]. Thus, human behaviour likely plays an important role in determining which humans are fed upon.

The observation that mosquitoes disproportionately acquired blood meals from older people aligns with results of studies conducted elsewhere. In Papua New Guinea, Kenya, Tanzania and Burkina Faso, Anopheles mosquitoes were generally observed to obtain more blood meals from males and older age groups compared to the youngest ages [3, 10, 42]. The youngest age groups and women of childbearing age could be receiving fewer mosquito bites than older males because they are more likely to be under the protection of bed nets.

Blood feeding on multiple humans during the same gonotrophic cycle was common and has several implications for malaria epidemiology. Arthropod vectors that take multiple human blood meals generally will increase the vectorial capacity of the vector population by increasing the chances of both acquiring and transmitting the pathogen [43]. More partial blood meals per gonotrophic cycle increase transmission above that resulting from single feedings and the feeding of Anopheles mosquitoes on two human hosts per gonotrophic cycle increases both R0 and VC by two-fold or greater [44]. The present study showed that 1 in 10 mosquito blood meals were taken from multiple individuals. This finding is consistent with other studies conducted in Papua New Guinea, Burkina Faso and Kenya, where the proportions of multiple blood meals ranged from 6 to 15%, 11–18%, 12–16%, and 0–14% [3, 10, 42, 45], respectively. Future studies should incorporate quantification of multiple feeding on malaria parasite transmission dynamics and intervention policy.

Plasmodium infection prevalence differed among age groups. The present study showed that blood meals from SAC had higher P. falciparum infection prevalence than the other age groups and that no significant differences were observed between male and females. Keven et al. [3] found that in one Papua New Guinea village, males and age group 15–30 years old had a higher probability of infection than females and other age groups. The consistency in these two studies is the observation that some human population groups have a higher likelihood of malaria parasite infection than others. As such, these population groups would be more responsible for malaria parasite transmission to mosquitoes.

An important strength of the present study was the use of 24 loci for genetic profiling, which increased the sensitivity of discrimination as compared to previous studies that analysed fewer markers [10, 34, 35, 37, 42, 46, 47]. Using 24 microsatellite loci not only increased power to discriminate between genetically related individuals, but also decreased the possibility of error due to allele dropouts [3, 48]. Thus, the assessment was more robust against matching errors. Further, the current study is among the few that have attempted to detect infection in the mosquitoes based on the sex and age of the human source of blood meal and inferring human-to-mosquito parasite transmission.

More than half of analysed blood meals did not match the genotype of an enrolled participant, which represents a limitation for analyses such as age that required a match. This might be because some indoor resting mosquitoes took their blood meal outdoors from non-enrolled neighbours, household visitors, or from non-participating household members. Harrington et al. [36] reported that most of the blood meals taken by Aedes aegypti in Thailand villages came from non-household members, a finding attributed to peddlers, visitors, and passersby in the community during the daytime hours, when Ae. aegypti most often seeks hosts. In the current study, the number of non-resident visitors in or near the study households was not assessed either during the day or the night/early morning when Anopheles mosquitoes seek hosts. However, the genotyped blood meal data suggest that such human movement was considerable. Also, the few blood meals that were determined to be from multiple humans were not analysed. Finally, the analyses included an assumption that the population distribution in other households was similar to study households. Future studies will be stronger if they maximize the genotype matches by simultaneously sampling for humans and mosquitoes in each household, obtaining blood samples from visitors, and genotypically identifying unique individuals in blood meals from multiple human sources.

This study analysed Plasmodium infections detected in the mosquito abdomen to infer infection transmission from humans to mosquitoes. While P. falciparum parasites in the abdomen do not always produce infectious sporozoites, their presence in the abdomen serves as a starting point for future parasite transmission (mosquitoes becoming infectious). Others studies have shown that a high proportion of infections in SAC contain gametocytes, the parasite stage required for transmission, [12, 14, 49] and that infections in this age group are highly transmissible in mosquito feeding assays [10, 50, 51]. The current study further suggests that the school-age group could be the primary reservoir and main driver of Plasmodium infection into mosquitoes.

Non-random vector feeding has important epidemiological implications. Several simulation modeling studies have shown that non-random human blood feeding results in a greater Basic Reproductive number (Ro), which increases the probability that transmission persists [1, 5, 45], even in the presence of malaria control programmes such as distribution of bed nets [4]. Therefore, targeted interventions focusing on those human demographic groups that disproportionately serve as Plasmodium infection reservoirs should enhance reduction of malaria parasite transmission on top of uniform interventions.

Conclusion

The Anopheles mosquito population in the study area of southern Malawi exhibited non-random human host feeding, with males and the older adults being most frequently bitten, in the setting of high levels of net ownership and moderate levels of reported use. However, the prevalence of P. falciparum infection in mosquitoes was highest when the blood meal source was from ≤ 5 years old and SAC. Because younger children were the source of fewer blood meals, SAC are the source of the majority of mosquito infections. These findings are crucial for efforts to focus malaria control and prevention at the demographic groups that are currently most important to maintaining transmission, specifically SAC.