Optimising systemic insecticide use to improve malaria control

Patient and public involvement

Patients and the public were not involved.

Literature review

Veterinary and human parasiticides were identified with systemic properties that affected arthropods and were not prohibited nor being phased out of use in most countries. These included avermectins, milbemycins, neonicotinoids, spinosyns and isoxazolines. Unlike pyrethroids, which target voltage-gated sodium channels, avermectins and milbemycins target glutamate-gated chloride channels, neonicotinoids and spinosyns target unique sites of the nicotinic acetylcholinesterase receptor, and isoxazolines target γ-aminobutyric acid-gated chloride channels.11–15 Further discussion of the parasiticides’ structures and modes of action may be reviewed elsewhere.12–15

To determine the relevant pharmacokinetic studies for different systemic insecticide treatments, a review of the electronic literature was conducted. PubMed was searched from inception to 24 July 2018 using the following search terms: (“systemic insecticide” OR endectocide OR ectoparasiticide OR avermectin OR abamectin OR doramectin OR eprinomectin OR ivermectin OR selamectin OR milbemycin OR “milbemycin oxime” OR moxidectin OR isoxazolines OR afoxolaner OR fluralaner OR sarolaner OR lotilaner OR neonicotinoids OR imidacloprid OR nitenpyram OR spinosyns OR spinosad OR spinetoram OR “N-tert-butyl nodulisporamide”) AND (pharmacokinetics OR “area under the curve” OR kinetics OR “half-life” OR Cmax OR Tmax) AND (blood OR plasma). The complete list of accepted and rejected studies is available on request to the corresponding author.

Study inclusion was determined in two steps. First, the titles and abstracts were screened to determine studies that were irrelevant, not primary research (eg, letter or review), or purely computational. Irrelevant studies were defined as those focusing on drug mechanism, a drug that was not systemic or ineffective against mosquitoes, or hosts that are not targeted by mosquitoes for blood meals. After the initial screen, the full papers of the remaining studies were reviewed. Inclusion required the reporting of relevant pharmacokinetic parameters in plasma, use of a standardised drug, application of a single drug, and a test-population size n>3.

From the selected studies, the following data were extracted directly into an Excel spreadsheet: host studied, drug applied, dose applied, administration route, maximum drug concentration reached in the plasma (C_max), area under the curve, absorption and elimination half-lives, mean residual time and volume of distribution. All data were summarised using basic descriptive statistics (mean and SD or SE) for each drug–host–route scenario. As different classes of drugs require different concentrations to achieve the same toxicity level, the doses for each combination of drug–host–route were not compared. Instead, the analysis focused on the pharmacokinetic metrics that impact the treatment’s efficacy and can be used to calculate the appropriate dose.

Data analysis

To determine the underlying trends between the pharmacokinetic parameters and each categorical factor (drug, host, administration route), a weighted (using the inverse of the variance) three-way analysis of variance (ANOVA) was conducted in Stata. Hosts with fewer than 10 observations were grouped into two categories based on bodyweight: other small (<75 kg) and other large (>75 kg).

Model development

To investigate strategies for applying different systemic insecticides to further limit malaria transmission, models from the literature were modified24–26 (see online supplementary text and table 1 for model development and parameter values). The proportion of bites that lead to infection, egg-laying rate and death rate of mosquitoes depend on the concentration of insecticide in LLINs (N) and systemic insecticides in livestock or humans (D_L or D_H, respectively) to which a mosquito is exposed.27–30 The model assumes the effects of the LLINs and systematic insecticides are additive and independent of each other’s concentration. Different host species and administration routes are characterised by different rates of adsorption, which were often not reported. Hence, to allow for comparison, the model represented the systemic insecticide’s pharmacokinetics as a single compartment and the initial insecticide concentration as the reported C_max.

The impact of insecticides used for livestock, human and bed net treatment diminishes as they degrade at rates   ,    and   . These rates were determined by the half-lives recorded from the review. The transmission potential from vector to human or vice versa (b_h and b_m, respectively) is a function of biting frequency (a), the proportion of bites that successfully leads to infection in humans or mosquitoes (b and c, respectively), the coverage of bed nets (C_N) and whether the LLIN’s insecticidal concentration is above the lethal concentration for killing 50% of the population in a set amount of time (  ). Mosquitoes lay eggs at a rate of   , but this can be modified with the introduction of certain systemic insecticides (  ). Similarly, the mosquito natural death rate (μ_m) is modified with the introduction of LLINs and systemic insecticides (  ). For both the egg laying and death rates, the impact of different control strategies depends on the proportion of bites on humans (  ), the coverage of LLINs and systemic insecticides in livestock or humans (C_N, C_L, C_H, respectively), and the respective concentration thresholds for reducing fecundity or survival by 50% (  , and   ).

The relationship between mosquito fecundity or mortality and the concentration of LLINs or systemic insecticides is not linear, but is captured by Hill kinetics. The   ,   , and Hill coefficients (   and    for egg-laying and death rates, respectively) were calculated for lab-reared Anopheles for different systemic insecticides by fitting a Hill equation to published data28–30 (online supplementary figures 1 and 2). The    was calculated for wild Anopheles that displayed a range of resistance to LLINs, and the Hill coefficient (  ) was calculated for laboratory reared, permethrin resistant Anopheles by fitting a Hill equation to published data.27 31 The time window (  ) associated with each insecticide’s    was based on previously reported measurements.27–30

Simulations

This model was used to explore the impact of combining different systemic insecticide treatments and LLINs on permethrin-resistant mosquitoes. A new LLIN was introduced every 3 years, as recommended by the WHO.32 Systemic insecticide treatments were designed to test a range of half-lives (0.1:100 days), dose concentrations (1:10⁵ ng/mL), F₅₀s (0:480 ng/mL) and   s (7:1180 ng/mL), based on data mined from the review. The simulation begins with the initial concentration of drug in the host’s blood, approximated by C_max. The appropriate dose necessary to achieve C_max can be calculated based on pharmacokinetics associated with each host species, as previously documented.33 Although many of the systemic insecticides identified in the review remain to be characterised as mosquitocidal candidates,    were identified from the avermectin, moxidectin, isoxazoline and spinosyn classes for Anopheles gambiae or Anopheles arabiensis and used to establish a range of realistic values.    were only reported for a subset of avermectins and moxidectin. Each treatment characterised by half-life, C_max,   , and    was dosed at frequencies ranging from weekly to annually. Strategy outcome was quantified by calculating the relative reduction in malaria prevalence after 3 years of drug treatment and LLIN coverage (  ) relative to 3 years of LLINs alone (  ).

This framework was applied to evaluate strategies for areas with different levels of baseline malaria, mosquito populations with different feeding behaviours and different host coverage scenarios.

The livestock and human host coverage necessary to achieve a minimum 10% relative reduction in malaria for different dosing frequencies was estimated for each drug by holding the C_max and half-life constant (online supplementary table 2). When available, the livestock values were based on topically or orally treated cattle and the human parameters were based on topically or orally treated ‘(other) small’ (the category humans were grouped into, due to small sample size). The exception is ivermectin, where the human host parameters were based on the weighted averages from studies of humans orally treated with ivermectin (C_max=40.7 ng/mL, half-life=3.4 days). For drugs that were only characterised in dogs and small animals, those parameters were used for both human and livestock parameters.

Data retrieval

From the initial 375 articles returned by the search, 237 full-text articles were assessed for eligibility, and 139 met the eligibility criteria (figure 1A). The studies reported pharmacokinetic parameters in 8 different host categories, 10 systemic insecticides and 6 administration routes (figure 1B–D). The three most studied hosts were cattle, sheep and dogs. Systemic insecticides studied included five avermectins (abamectin, doramectin, eprinomectin, ivermectin and selamectin), three isoxazolines (afoxolaner, fluralaner and lotilaner), one milbemycin (moxidectin) and one spinosyn (spinosad). These insecticides were applied intramuscularly, intraruminally, intravenously, orally, subcutaneously or topically. Note that intraruminal and intravenous routes are experimental and not currently operationally feasible; however, they were included to help determine the full range of action possible for each drug.

Request permission

Data analysis

To evaluate the different treatment scenarios, weighted three-way ANOVAs were conducted for each pharmacokinetic parameter. Elimination half-life and C_max were the only parameters with significant interactions (p<0.05) with host, route and drug. The significant effectors of half-life were drug class (p=0.016), route of admission (p=0.007), and interactions between host and route (p<0.001) and drug and route (p=0.007). Regardless of route of administration, the order of drugs from shortest to longest half-lives was avermectins<milbemycins<spinosyns<isoxazolines. The median half-life for avermectins and milbemycins was <10 days for all routes of administration, whereas isoxazolines’ half-life was >10 days (figure 2A, online supplementary figure 3). Comparing median half-lives for a given drug class across hosts showed some host dependency. For instance, milbemycins had a longer median half-life in dogs (19.4 days) than in other hosts (<10 days) and avermectins had a longer median half-life in cattle than in other hosts (figure 2B, online supplementary figure 4). When comparing administration routes across different hosts, topically applied drugs typically achieved longer half-lives than orally applied ones (figure 2C, online supplementary figure 5).

Request permission

The significant factors for C_max were drug (p=0.03) and interactions between host and drug (p=0.002) and host and route (p<0.001). The order of drugs from lowest to highest C_max was different from that of half-lives: milbemycins<avermectins<isoxazolines<spinosyns (online supplementary figure 6). Cattle reported the lowest median C_max for milbemycins, whereas dogs and sheep had the lowest C_max for avermectins (figure 2D, online supplementary figure 7). There was also a dependency of C_max on host and route (figure 2E, online supplementary figure 8). Although the intravenous route resulted in the highest C_max for different hosts, due to the drug being directly delivered into the bloodstream, the order of resulting C_max for other routes varied based on host.

The spread in half-lives and C_maxs seen for a given drug–host–route combination can be attributed to host factors that may affect drug absorption and, consequently, the plasma concentration. These factors include age, gender, breed, diet, parasite infection, pregnancy, lactation, and whether topically treated hosts are restricted from self-licking.34–41 Understanding how host conditions affect basic pharmacokinetics is critical for designing optimal treatment strategies that can account for natural variations.

Basic dynamics of malaria transmission and control methods

The model captures the temporal dynamics of malaria transmission in a human population exposed to mosquitoes that bite humans and livestock indiscriminately. On the LLIN introduction, a general decline in malaria prevalence (symptomatic and asymptomatic individuals combined) is observed, followed by a steady increase as the insecticide in the net degrades and fewer mosquitoes are killed by LLIN exposure (figure 3A). The addition of treating one blood host population with systemic insecticide can further reduce malaria prevalence when applied at sufficient frequency.

Request permission

Dosing strategy design and evaluation

To quantify the efficacy of different dosing strategies, the relative reduction in malaria prevalence after 3 years of using LLINs and applying systemic insecticides to one blood host population was compared with that of using LLINs alone. For a set dosing frequency, the relative reduction increased as a function of half-life and C_max (figure 3B).

The minimum dosing frequency was calculated to achieve a target relative reduction (here, 10%) for different drugs distinguished by half-life, C_max,   , and    (figure 3C–I). The drugs with the longest half-lives and highest C_maxs needed to be dosed the least often to maintain a sufficiently high concentration to remain lethal to feeding mosquitoes. Given the same half-life and C_max, drugs with higher    needed to be dosed more frequently to compensate for the decreased efficacy of drug on mosquito fecundity or survival.

Overlaying the data gathered in the review for drugs with reported    s and    for Anopheles gambiae sensu lato predicted the frequency at which these existing drugs would need to be applied to achieve a 10% relative reduction in malaria prevalence. The avermectins, represented by ivermectin, eprinomectin and doramectin, have relatively low    s and    s, suggesting that relatively low concentrations of drug in the bloodstream would affect the fecundity and death rates of feeding mosquitoes. However, this impact is limited by these drugs’ relatively short half-lives, ranging from 0.4 to 11.1 days (table 1). Depending on the host and administration route, regimens with dosing frequencies ranging from weekly to quarter-annually would be required to achieve a 10% relative reduction in malaria prevalence. Although ivermectin and eprinomectin have a similar   , ivermectin has a stronger effect on fecundity (online supplementary figure 2). Consequently, ivermectin would require less frequent dosing than eprinomectin to achieve the same target reduction. Doramectin has a lower impact on fecundity and death rates, with a higher    and    than ivermectin and eprinomectin.

Although fluralaner and afoxolaner have higher    than ivermectin, they were predicted to achieve 10% relative reduction with yearly dosing, due to their longer half-life and higher C_max. Similarly, spinosad has a high   , a relatively long half-life of 11.3 days, a higher C_max of 1550.0 ng/mL, and could achieve a 10% relative reduction when dosed biannually. These results are conservative, as the effect of fluralaner, afoxolaner, and spinosad on fecundity was assumed to be zero until it has been characterised in mosquitoes.

Despite some of the moxidectin studies reporting relatively long half-lives and high C_maxs, its high    and    mean that it would have to be dosed more frequently (>weekly) or at higher doses to provide an effective complement to LLINs.

Application to different scenarios

This framework can predict the extent to which systemic insecticides could aid in the reduction of malaria prevalence in scenarios with different endemicities (online supplementary figure 9a–c). For instance, in a low-level mesoendemic environment (malaria prevalence=25%),42 LLINs alone played a significant role in reducing transmission; however, additional treatment of livestock and humans further reduced prevalence and could theoretically break transmission. In a high-level mesoendemic environment (malaria prevalence=50%), LLINs alone were not as effective and the additional treatment of livestock and humans could significantly reduce malaria transmission. When malaria is hyperendemic (prevalence=65%), the addition of systemic insecticide treatment reduced malaria prevalence relative to LLINs alone; however, to bring malaria transmission under control, longer, sustained treatment and/or the use of drugs with longer half-lives and higher C_max, such as the isoxazolines, would be necessary.

Similarly, this framework can help evaluate control strategies for mosquitoes with different blood meal preferences (online supplementary figure 9d–e).43 Malaria transmission in regions with zoophilic mosquitoes was largely controlled by LLINs because the mosquitoes do not target humans as frequently. Treating livestock with systemic insecticides would be more effective than treating humans, requiring less frequent dosing for a drug with a given half-life and C_max. Malaria in regions with anthropophilic mosquitoes was reduced the most with the treatment of both livestock and humans, with most of the reduction due to the treatment of humans. Dosing frequencies were increased, given the need to maintain high enough lethal systemic insecticide concentrations to affect a greater number of mosquitoes targeting humans for blood meals.

Estimating coverage

While 100% coverage is ideal, it is difficult and often unrealistic to achieve in practice. The minimum dosing frequency of each drug was calculated to achieve a relative reduction >10% for a range of humans and livestock coverage (figure 4). Intuitively, as the coverage of both hosts increased, the dosing frequency necessary to reach the target reduction generally decreased. Ivermectin treatment of human and cattle appeared to contribute equally to malaria reduction (figure 4A). If at least 50% of both humans and cattle were treated with ivermectin every 2 months, then at least a 15% relative reduction in malaria could be achieved over 3 years. If 75% of both humans and livestock were treated, then the dosing frequency could be reduced to once every 4 months and still achieve a 10%–12% relative reduction.

Request permission

Eprinomectin appeared less effective at reducing malaria than ivermectin, with the lowest frequency dosing option being every 2 months (figure 4B). This is largely due to ivermectin’s greater impact on fecundity (online supplementary figure 2). Assuming that the pharmacokinetic properties of eprinomectin in ‘small’ hosts are a proxy of those in humans, the malaria prevalence was more sensitive to increasing the eprinomectin coverage of humans than of livestock. This is because the eprinomectin reached a higher C_max in humans than cattle (20.1 vs 11.0 ng/mL, respectively) and had a slightly longer half-life in humans than in livestock (4.8 vs 3.5 days) (online supplementary table 2). Thus, human blood meals would remain toxic to a mosquito for a longer period of time and could remain effective at lower dosing frequencies.

Doramectin was the least effective of the avermectins, requiring weekly dosing for the majority of the coverage scenarios to result in a 10% relative reduction in malaria prevalence (figure 4C). This is reflective of its    being larger than the C_max.

Assuming that the pharmacokinetic properties of both isoxazolines in dogs and ‘small’ hosts are a proxy of those in humans and cattle, fluralaner and afoxolaner were predicted to achieve the 10% reduction with yearly dosing and very low coverage (20% for fluralaner and 30% for afoxolaner) (figure 4D–E). With 70%–80% coverage of both populations, fluralaner was predicted to reduce malaria prevalence by 50%. However, afoxolaner applied at the same coverage would only reduce malaria prevalence by 25% because of its lower C_max and higher   , relative to fluralaner. The isoxazolines’ C_max is one to two orders of magnitude greater than the avermectins’, which results in the drug remaining at toxic levels in the blood for longer periods of time.

Assuming that spinosad’s pharmacokinetic properties in dogs can serve as a proxy for those in livestock and humans, it was also predicted to require less frequent dosing at lower coverage than the avermectins (figure 4F). Although spinosad had one of the highest C_max values of the drugs simulated here, it also had one of the highest   s. Thus, for a given coverage level, spinosad needed to be dosed more frequently than isoxazolines to reach the target reduction level.

Moxidectin was not capable of reducing malaria prevalence, regardless of dosing frequency or host coverage, because its    is three orders of magnitude greater than its C_max.

Emergence of resistance

As the frequency of insecticide administration is increased to break the malaria transmission cycle, it is imperative to consider how these new treatment regimens will contribute to the emergence of resistance in mosquitoes and other parasites. Resistance is considered the greatest threat to malaria vector control.59 The two main mechanisms of insecticide resistance observed so far in mosquitoes can be categorised as metabolic or target site mutations.60 Studying how other arthropods have formed resistance to systemic insecticides will shed light on the potential pathways to resistance formation in mosquitoes. For instance, macrocyclic lactone resistance has been shown to arise in cattle ticks, fruit flies and body lice due to increased expression of adenosine triphosphate (ATP)-binding cassette transporter, P-glycoprotein and P450 genes.61–63 With strategic use of systemic insecticides that target different mechanisms, monotherapy could be avoided and the development of resistance in mosquitoes could be delayed.

Selecting for resistant, non-mosquito parasites is also a concern. One study reported ivermectin-resistant Rhipicephalus microplus found on 50% of cattle being treated regularly with ivermectin for gastrointestinal nematodes.64 Similarly, repeated treatment of ivermectin for onchocerciasis in humans has selected for resistant Onchocerca volvulus in Ghana.65 As these systemic insecticides are commonly used to control other parasite infections in humans and livestock, it is important to consider this in drug selection and actively screen for unintended consequences. Refugia, or leaving a small portion of the population untreated, offers a possible solution for reducing the selective pressure on parasites and arthropods to form resistance.64 66

Presence in food products

Considering how different systemic drugs are eliminated from a host’s system is also important in evaluating a new treatment programme. For instance, lactating hosts treated with a single dose of eprinomectin or ivermectin produced milk with detectable drug levels for weeks, thus exposing the nursing young.67 68 Due to the unknown effects of the drug on a newborn, nursing human mothers in the first week after delivery have been excluded from MDAs of ivermectin.69

Additionally, systemic insecticides from treated hosts becomes incorporated into dairy and meat products.70 Regulations have been established for levels of acceptable residues of a few drugs, such as eprinomectin; yet, most other drugs have not been licensed for use in dairy animals and do not have an acceptable limit for drug concentration found in milk.71 It was surmised that the extent of drug excretion and residence time in milk depends on a drug’s lipophilicity and route of administration.72 73 However, more studies are needed to characterise the different drugs’ excretion in milk and, subsequently, establish safety limits for suckling young or consumable items. In the meantime, refugia offers a possible solution to ensure that the meat and milk produced by a portion of the livestock are safe to consume.

Limitations

Since the model parameters were limited to the pharmacokinetic/pharmacodynamic values reported in the literature, there are a number of limitations to this approach. To achieve a large enough sample size, hosts were categorised by species. As a result, the impact of subspecies details, such as breed and health status, on pharmacokinetic properties were no longer distinguished. For instance, ivermectin was observed to be less available in zebu Gobra, a Western African cattle breed, than other cattle breeds74; however, the pharmacokinetic behaviour of doramectin and moxidectin were similar.75 Additionally, parasitised animals were reported to have faster absorption times and shorter circulation times for ivermectin than their healthy counterparts.38 Another constraint was being limited to pharmacokinetic data from one species as a proxy for how the drug might act in cattle or humans. Although interspecies variation in drug pharmacokinetics is significant, this was the most relevant data available for simulations until further studies are conducted. Furthermore, the effect of the isoxazolines and spinosyns on mosquito fecundity remains to be quantified. The proposed model conservatively assumed that those drugs had no effect on egg-laying rates. Finally, although lotilaner was identified as an isoxazoline with a particularly long half-life and high C_max, its mosquitocidal effects remain to be measured and thus could not be included in the simulations. These limitations reveal the need to further investigate these drugs’ safety and pharmacokinetics in relevant hosts and their effects on mosquitoes. Nevertheless, this framework can help identify scenarios in which systemic insecticides should be considered for malaria control and develop an effective MDA strategy for a range of systemic insecticides.