Blood meal sources of Anopheles vectors of human malaria in Malawi: Implications for malaria transmission and effectiveness of interventions

Selection of blood meal hosts by mosquitoes is a key variable in the vectorial capacity of Anopheles mosquitoes for human malaria. Blood feeding on humans is likely to be modulated by use of different types of long-lasting insecticidal nets (LLINs) and the effectiveness of LLINs is impacted by the relative intensity of insecticide resistance. The aim of this study was to test the hypothesis that LLINs containing pyrethroid and the synergist piperonyl butoxide (PBO) would lead to a reduction of human host utilization than LLINs containing only pyrethroid and that blood feeding patterns of Anopheles in Malawi compromise malaria interventions.


Introduction
Malaria is endemic in Malawi with transmission occurring throughout the country, having generally greater transmission in the south than the north [1][2][3]. The nation-wide prevalence of infection for Plasmodium falciparum among 2 to10 year old individuals, sampled from 2010 to 2017,declined from 29.4% in 2010 to 15.2% in 2017 but this change was uneven across the country's 28 jurisdictional districts [4]. Despite these successes, which are largely attributed to reduction in transmission owing to the implementation of long-lasting insecticide insecticidal nets (LLINs) distributed throughout the country, Malawi remains a high malaria burden country with meso-endemic transmission [4]. In 2017, the country-wide incidence of malaria was 247 per 1,000, with an estimated 7,077 deaths [5].
One of the drivers of persistent Plasmodium transmission is access to human blood by host-seeking female Anopheles mosquitoes. Selection of blood meal hosts by Anopheles mosquitoes is a key variable in their vectorial capacity a measure of transmission for human malaria because vectorial capacity increases with the square of the rate of human biting [6]. Further, host selection is likely to be modulated by use of different types of LLINs whose effectiveness is impacted by the extent of insecticide resistance in the Anopheles populations, and the relative availability of hosts [7,8]. These factors will enhance or limit mosquitoes' access to humans relative to non-human hosts. Emergence of pyrethroid resistance could severely compromise malaria control efforts by reducing the effectiveness of LLINs. Newer LLINs which combine pyrethroids with a synergist, piperonyl butoxide (PBO), have shown to be effective at restoring pyrethroid susceptibility [9][10][11].
Vector incrimination studies in Malawi have identi ed members of the Anopheles gambiae sensu lato (s.l.) and Anopheles funestus species assemblage as the most important malaria vectors and pyrethroid insecticide resistance have been reported (Hunt et al., 2010;Mzilahowa et al., 2008Mzilahowa et al., , 2016Spiers et al., 2002). However, only a single study has analyzed mosquito host selection and estimated the human blood index (HBI) of these vectors in the southern region of the country before LLINs were available . Accordingly, whether human host selection has changed after LLINs availability is unknown.
The objective of this study was therefore twofold. Firstly, we aimed to expand on knowledge of blood feeding patterns of malaria vectors, a key feature of malaria risk, by conducting host selection analyses and estimation of HBI and EIR. Secondly, we aimed to test the prediction that blood feeding on human hosts is reduced where LLINs containing permethrin and the synergist PBO (i.e., Olyset Net Plus, Sumitomo Corporation, Tokyo, Japan) compared to where standard, pyrethroid-only LLINs (i.e., Olyset Net, Sumitomo) have been distributed.

Study area
This study was conducted in two malaria-endemic districts of Malawi ( Fig. 1), Balaka (14° 58' 45" S; 34° 57' 20" E) and Machinga (15° 10' 6" S; 35° 18' 0" E). These districts, like the whole of Malawi, have distinct wet and dry seasons where malaria proliferates especially in the rainy wet season [2]. In 2018, residents of Machinga received LLINs with PBO (Olyset Plus net, Sumitono Corporation, Tokyo, Japan), whilst residents of Balaka received standard LLINs only (Olyset net) (Government of Malawi, National Malaria Control Programme, 2017) (Malawi Malaria Operational Plan FY, 2018). Households in rural areas located within the catchment areas of Ntaja (Machinga) and Namanolo (Balaka) health centers were enrolled in the study. Household surveys were conducted to determine the extent of LLINs ownership and utilization by householders and to quantify the number of occupants.

Mosquito sampling and analysis
Blood-fed Anopheles mosquitoes were sampled in houses, selected by the randomized cluster sampling method. All households were sampled three times each period from May-June 2019, October-November 2019, and December 2019-January 2020. Beginning in February to April of 2020, only a subset of households in each site were sampled once every two weeks. The mosquito samples for each visit were accumulated. The number of humans and goats in the households was also counted, although dogs were too mobile to be counted reliably. Cattle were rare in the study area and scored as absent. Mosquito sampling tools included mouth and battery-powered aspirators, standard miniature CDC light traps (Model 512; John W. Hock Company, Gainesville, Florida, USA) and pyrethrum spray catches (PSCs) [13,14]. Mosquitoes were morphologically identi ed [15] into An. gambiae sensu lato, An. funestus sensu lato, or other Anopheles species, stored in tubes with silica gel, and kept at laboratory temperatures.
The abdomen was separated from the head-thorax of each mosquito using sterile, cross-contamination-proof technique. For each mosquito, genomic DNA was extracted from the abdomen and head-thorax separately using the Qiagen extraction kit (DNeasy® Blood & Tissue Kit; Cat. No. 69506) following the manufacturer's standard protocol.

PCR identi cation of mosquitoes
Mosquitoes of the An. gambiae (s.l.) complex were identi ed to one of two species, An. gambiae sensu stricto (s.s.) or An. arabiensis, using a published multiplex quantitative PCR (qPCR) method [16]. Mosquitoes of the An. funestus (s.l.) species assemblage were identi ed to one of three species (An. funestus (s.s.), Anopheles parensis and Anopheles vaneedeni) using another multiplex qPCR as follows.

Blood meal analysis
Individual mosquito abdominal DNA was rst tested for human blood meal using a uniplex qPCR method that involved ampli cation of a region of intron 1 of the nuclear tyrosine hydroxylase gene with primers and probes (forward: 5'-GGC CTG TTC CTC CCT TAT TT-3′, reverse: 5'-TAC ACA GGG CTT CCG AGT-3′, probe: FAM-ATG GAG TCT GTG TTC CCT GTG ACC -QSY) as described in Keven et al., (2020). Samples that did not react with the human probe in the qPCR were subjected to a standard PCR to amplify the vertebrate mitochondrial cytochrome B gene using a generic primer pair (forward: 5′-CCC CTC AGA ATG ATA TTT GTC CTC A-3′, reverse: 5′-CCA TCC AAC ATC TCA GCA TGA TGA AA-3′) [19][20][21][22][23]. The PCR reaction mixture (25 µL volume) consisted of 10 mM Tris at pH 8.3, 50 mM KCl, 1.5 mM MgCl 2 , 0.01% gelatin, 1.0 mM dNTP, 0.5 units of Taq polymerase, 50 pmol of each primer, and approximately 20 ng of DNA template. The reaction condition included one cycle of 95°C for 5 minutes (initial denaturation) followed by 35 cycles of 95°C for 1 minute (denaturization), 57°C for 1 minute (annealing) and 68°C for 1 minute (extension), followed by one cycle of 68°C for 5 minutes ( nal extension). The PCR products were visualized with 2% agarose gel electrophoresis and amplicons of positive samples were puri ed using QIAquick PCR puri cation kit (Cat. No. 2810; Qiagen) following the manufacturer's protocol. The nucleotide sequence of amplicons was determined by direct sequencing and the sequences were subjected to BLAST (Basic Local Alignment Search Tool) search for matches to the available vertebrate host cytochrome B gene sequences in GenBank (https://blast.ncbi.nlm.nih.gov/Blast.cgi). A sequence similarity of 97% or higher was used as the cut-off for an acceptable match, based on literature [20,[25][26][27].
The different hosts identi ed in the mosquito blood meals based on the results of BLAST searches aided the development of new qPCR primers and probes speci c to those host species; humans, dogs and goats were the only hosts identi ed by BLAST searches. Primers and probes (Table 1) for humans and dogs were adopted from literature [18] but those for goats were designed and validated in this study using the same procedure for humans and dogs (Keven et al., 2020). Another set of qPCR procedures was performed on all the blood meal samples. The purpose of this second qPCR was to detect presence of mixed blood meals as well as to con rm the results of the previous uniplex qPCR. The qPCR mixture (10 µL reaction volume) consisted of 1X TaqMan Universal Master Mix (Cat. No. 4304437; Thermo Fisher Scienti c), 0.5 µM of each primer, 0.25 µM of each probe, and 2 µL of DNA. PCR cycling condition (QuantStudio 7 Flex PCR System) was the same as described for An. funestus (s.l.) above. Positive and negative DNA controls were included in each experiment. Table 1 Blood meal host primers and probes used in the qPCR reactions species Forward Reverse probe human Homo sapiens

Data analysis
Mosquitoes that had fed on one species of host, as determined by blood meal analysis as described above, were classi ed as single host blood meals. If the blood meal analysis revealed two or more species of vertebrate hosts, then the blood meals were classi ed as mixed blood meals. The HBI of an Anopheles population (i.e., mosquitoes of a particular species from a particular village) was calculated as the proportion of blood-fed mosquitoes that fed on human hosts. Human-fed mosquitoes included both single human blood meals as well as human-nonhuman mixed blood meals. [18, [29][30][31]. Mosquitoes whose blood meal hosts were not identi ed were excluded from the analysis.
Differences in host utilization by district with different LLINs were analyzed by contingency table analysis. To test for variation in the propensity of the three main malaria vector species to feed on humans, non-humans and humannonhuman mix blood meal types, a 3 x 3 contingency table analysis was carried out, and the percentage deviations of observed from expected frequencies were calculated. Host selection of a vector population was quanti ed using the theta statistic (θ = π 1 /π 2 ), which tested whether the ratio (θ) of the proportion of a host species in mosquito blood meals (π 1 ) and proportion of the host in the village (π 2 ) is different from unity [32]. A host species was considered over-selected by the vector population if theta was signi cantly greater than 1.0 or under-selected if theta was signi cantly less than 1.0. A host species was considered to be fed on by the mosquitoes in proportion to its relative availability in the village if theta was not signi cantly different from 1.0. The theta statistical analysis was performed using the ci.prat.ak function of asbio package in RStudio Version 1.1.456.
The sporozoite rate (SR) was estimated as the proportion of mosquito heads-thoraces that tested positive for Plasmodium sporozoites. The entomological inoculation rate (EIR) can be estimated indirectly from samples obtained by indoor resting mosquitoes, and also directly from samples obtained by the human landing catch (HLC) method [33].

Host selection and HBI
Of the blood-fed Anopheles mosquitoes (n = 633), the blood meal host of 541 (85.5%) were successfully identi ed either by qPCR or direct sequencing ( Fig. 2 and Table 2). The remaining 92 (15%) were either non-reactors in PCR reactions (n = 42), or amplicons generated by standard PCR failed to match any feasible host (n = 50). Of the 541 mosquitoes whose blood meal host was successfully identi ed, 436 (81.0%) were solely human blood meals, 28 (5.2%) were solely goat blood meals, 11 (2.0%) were solely dog blood meals, and mixed blood meals were: 1 (0.2%) dog-goat, 5 (0.9%) dog-human, and 60 (11%) goat-human ( Fig. 2 and Table 2). Humans were the most frequently identi ed blood-meal host for all three of the most abundant Anopheles species (An. arabiensis, An. funestus s.s. and An. gambiae s.s.), at approximately 70% of blood meal samples for each species (Table 2). Mixed blood meals comprising human and goat were present for all three mosquito species and ranged from 7.1 to 11.7% among species. Blood meals identi ed solely from goats ranged from 1.4 to 5.8% and were also found in all three species. Dog-only, and mixed dog-human or dog-goat blood meals, were present but uncommon. Differences in human or non-human (goat and dog) and mixed (human-nonhuman) host feeding by these three species, and the percentage deviations of observed from expected frequencies are shown in Table 3.
Although the chi-square test was not signi cant (X 2 = 6.4, df = 4, p = 0.17), the percentage deviation values were suggestive that An. arabiensis tended to feed on nonhuman hosts more so than did An. gambiae (s.s.) and An. funestus (s.s.), whereas An. funestus s.s. tended to underutilize nonhuman hosts compared to An. gambiae (s.s.) and An. arabiensis. Additionally, An. funestus (s.s.) tended to have fewer human-nonhuman mixed blood meals compared to the other species (Table 3). The results for analysis of host selection by the theta statistic are shown in Fig. 3 In Ntaja, An. arabiensis and An. gambiae (s.s.) overselected goats and under-selected humans, whilst An. funestus (s.s.) selected these two host species in proportion to their relative abundance in the village. In Namanolo, by contrast, all three vector species selected both hosts in proportion to their relative abundance, although there was a nonsigni cant tendency for overselection of goats compared to humans.  (Lindblade et al. 2015). Despite these variations in mosquito species abundances between sites, An. funestus (s.s.) and An. gambiae (s.s.) are generally considered epidemiologically more important than An. arabiensis due to their welldocumented anthropophilic and endophilic behaviors [38][39][40][41]. This study found high rates of feeding on human blood by all three species, regardless of these variable phenotypes.
These ndings are more evident in the fact that human blood comprised most blood meals in unmixed conditions, regardless of species or study site, and that blood meals from other potential sources (goats, dogs) were secondary, with goat blood meal being more frequent than dog. This observation is not surprising because it was commonly observed that people kept goats indoors in special rooms at night, probably for protection against theft, while dogs were left outside as guard dogs. This may explain the higher number of goat blood meals compared to dog blood meals. Killeen et al., (2001), in modeling mosquito populations of Tanzania and Kenya, demonstrated that there is a relationship between host availability and the amount of time that vectors spend seeking blood meals; by inference, hosts that require less time to locate will be fed upon more frequently. Orsborne et al., (2020), studying mosquito populations in Ghana, reached a similar conclusion, emphasizing that local host availability even for known anthropophilic malaria vectors, is a powerful driver for host selection. In Malawi, there have been no previous studies that consistently quanti ed relative availability of potential blood meal hosts. The high prevalence of human host blood-feeding by Anopheles species observed here is consistent with 2002 ndings from southern Malawi, in which blood meals were nearly entirely from humans and secondarily from bovines [13]. In northern and southern Zambia, similar high human host selection (> 90%) and comparatively lower goat selection (< 5%) by An. gambiae and An. funestus (s.s.) were observed [42,43]. In contrast, the dominant blood meal of malaria vectors around Lake Victoria in western Kenya was humans for An. gambiae (s.s.) and An. funestus (s.s.), but for An. arabiensis was predominantly bovine [44] or equally bovine and human [45]. Through application of the "ratio of ratios" method of host selection, this study approached the problem of variation in host selection semi-quantitatively in order to assess host selection tendencies of these often behaviorally stereotyped species.
The high frequency of human blood meals detected in this study can be attributed to several factors, in particular bed net use practices. The higher HBI and EIR in Namanolo compared to Ntaja (96.4% vs. 88.9% and 0.11 vs. 0.06, respectively) could be due to widespread use of PBO-containing Olyset Net Plus in Ntaja, which have been shown to be more effective than convention LLINs against pyrethroid-resistant Anopheles populations [11,14]. Lindsay et al., (2021) have suggested that the underlying mechanism of PBO-containing LLINs may simply be that they are more toxic, rather than overcoming insecticide resistance. Regardless, other randomized eld trials in Tanzania and Uganda have shown signi cantly lower human infection prevalence where LLINs with PBO were distributed [9,10].
Although the entomological mediators of these reductions were not investigated, they are likely due to reduced transmission intensity. In the present study, the lower HBI of An. arabiensis and An. gambiae (s.s.) in Ntaja than Namanolo could also be explained by the use of LLINs with PBO in Ntaja, which apparently increases their susceptibility to the insecticide as explained above [14,47].
In the only other study analyzing blood meals of Anopheles vectors in Malawi, conducted in Chikwawa district (southern Malawi) during 2002 prior to any mass distribution of insecticide-treated nets, most blood meals were from humans, with relatively few coming from bovine or mixed human-bovine feeding [13]. Molecular-based approaches to blood meal analysis to detect vertebrate host feeding have advanced since the review of this topic by [18,25,44,49]). At the forefront of this advance has been development of qPCR methods using host-speci c probes by either TaqMan or SYBR green detection [18,50]. However, host species-speci c probes in multiplex qPCR targeting Anopheles blood-meal hosts were developed only recently [18]. The use of speciesspeci c probes, designed within a qPCR format here, favored the detection of single and multiple (i.e., mixed) blood meals in this study. By screening all blood meals for human blood, and then analyzing by PCR, amplicon sequencing, and BLAST search matching those blood meals not reacting to the human probe, it was possible to reveal the narrow breadth of dominant hosts being utilized by the Anopheles community, and then using qPCR to reveal the extent of multiple feeding. This combinatorial approach indicated that ~ 12% of the blood meals were mixed feeding of human and goat (10.8%), human and dog (1%) or goat and dog (0.2%).
The host-selection analysis showed that, in both sites, two of the Anopheles species fed more often on goats than humans in proportion to availability of these hosts. An. arabiensis and An. gambiae (s.s.) over-selected goats and under-selected humans, while An. funestus (s.s.) selected the two hosts about equally (i.e., randomly) in proportion to their availability. These results are not surprising. Although An. arabiensis is reportedly more zoophilic, while An. funestus (s.s.) is more anthropophilic [51]. Plasticity and or opportunistic tendencies have been observed in various Anopheles species [7,[38][39][40][41]. The explanation to the relatively higher goat feeding in indoor mosquito samples is consistent with goats being kept indoors at night, and malaria control interventions in the area (LLINs) (Malawi Malaria Operational Plan FY, 2018). These two activities provide easy accessibility to goat blood meal and makes it more di cult to access human blood meal due increased mosquito-goat contact and reduced mosquito-human contact [38]. The switch in host utilization is indicated by the reduced access to human blood meal and increase in the utilization of the non-human hosts. The comparable increase in mixed blood meal may also suggest disruption in feeding either due to LLINs' activity or otherwise, pushing the mosquito to get a full blood meal from other hosts.
Either way this is one of the few studies in Malawi to report blood meal analysis including detection of mixed blood meals. This study will prompt more research in blood meal studies in Malawi to document the range of blood meal hosts, especially those involving goat blood meal which is rarely reported in literature.
The ndings of this investigation suggest important implications for Plasmodium transmission and malaria control. Multiple host feeding by some Anopheles females might allow for increased survival and reproduction [49,52]. The presence of multiple malaria vector species that successfully obtain human blood meals could lead to an increased Plasmodium transmission by increasing the basic reproductive number (R o ) [42,53]. More widespread use of LLINs, particularly with PBO, could help reduce transmission, but this intervention alone is unlikely to reduce malaria incidence in this meso-endemic setting to acceptable levels where elimination can be contemplated. Residual Plasmodium transmission and weakened intervention efforts [54] are likely to persist into the future.

Conclusion
This study has shown that, in southern Malawi, human blood comprises the bulk of the blood meals of the three species of Anopheles vectors, yet dog and goat blood meals are also present and commonly mixed with that of humans. Host selection analysis revealed that goats were over-selected compared to humans in proportion to host availability. This could be a direct result of LLIN usage. The presence of mixed-blood meals showed the adaptability of these vectors to switch hosts to obtain a full blood meal, possibly reducing effectiveness of malaria control interventions. The frequent use of humans as a blood meal source elevates the human blood index and consequently the entomological inoculation rate, sustaining malaria incidence. Some evidence suggests that pyrethroid-based LLINs containing PBO reduced mosquito-human contact, as the HBI was signi cantly lower at the site using these nets. Nonetheless, estimates of EIR remain high, indicating that transmission is well sustained despite the use of PBO-containing LLINs. Other factors, such as net durability and use may be reducing effectiveness of interventions.

Figure 1
Map of Africa showing the study sites in Namanolo and Ntaja in Malawi.

Figure 2
Blood meal identi cation to vertebrate host, including mixed meals, for Anopheles mosquitoes two sites combined.

Figure 3
Anopheles species blood meal host selection in relation to number of available hosts. Top panels; goat selection in relation to human, bottom panel; human selection in relation to goats. The black shaded circles are theta values with 95% CI bars. Red dotted line at 1.0 represents random selection in relation to availability of both hosts. The 95% CI bars represent deviation from random selection pointing to over-selection (theta signi cantly > 1.0) or under-selection (theta signi cantly < 1.0).